Luminous Landscape Forum
Equipment & Techniques => Computers & Peripherals => Topic started by: alatreille on August 10, 2019, 09:46:23 pm
-
Hey all,
Just wondering if anyone is considering the new Ryxen processors for a new build?
Seems like an amazing all round set and the 3900x or 3950 are on my short list at the moment.
I'm wondering if C1 might benefit from thw multi core/threads of these cpus.
Let mw know your thoughts.
Cheers.
A
-
Hey all,
Just wondering if anyone is considering the new Ryxen processors for a new build?
Seems like an amazing all round set and the 3900x or 3950 are on my short list at the moment.
I'm wondering if C1 might benefit from thw multi core/threads of these cpus.
Let mw know your thoughts.
Cheers.
A
I would but I convinced myself that I want thunderbolt for my next computer, be it built by myself or bought as is, so I think AMD is a no show here.
-
I would but I convinced myself that I want thunderbolt for my next computer, be it built by myself or bought as is, so I think AMD is a no show here.
What are you looking to get from TB3? The advantage of a desktop is standard PCIe slots, and with USB-C (3.2) external speeds are still nice.
I'm looking at doing a video workstation build - stitching video (Insta360 Pro2) sucks with my laptop.
-
I would but I convinced myself that I want thunderbolt for my next computer, be it built by myself or bought as is, so I think AMD is a no show here.
Some of the X570 mbs have thunderbolt don't they?
-
What are you looking to get from TB3? The advantage of a desktop is standard PCIe slots, and with USB-C (3.2) external speeds are still nice.
I'm looking at doing a video workstation build - stitching video (Insta360 Pro2) sucks with my laptop.
Yes, but I want my storage separated from my computer, partially for easier upgrades down the road and partially for size of storage. With enough drives I should get at least as much speed as from my SATA drives (even as they are SSD only) but I don't think USB will cut it. I know I can add drives in the desktop and I did, but I feel it would be easier to separate the upgrades. Another reason is that I want my computer to be silent; if I want a lot of cheaper storage I would need to go back to regular HDD, thinking of 6 to 8. Putting them in a box that I can move farther away from me would decrease the noise.
I would also need to upgrade my NAS in the next year or so, therefore I contemplated using a NAS with dual 10GB, or with the newer 40GB or 100GB but it looks a little too complicated for my level, particularly if I need to troubleshoot it. I could try a NAS with both TB3 and 10GB for best of both worlds but I would like to see some real world testing.
Some of the X570 mbs have thunderbolt don't they?
I didn't check but I know on the native Intel side there is a paucity of motherboards that can reliably provide TB3. So far the Gigabyte Z390 Designare is the best candidate that offers built-in TB3, without additional cards. On the built-in side is a Dell XPS Special edition with the add-in card, or the latest iMAC 27 if you want Mac.
Searching now it seems that only ASRock has a few that will need an add-in card.
-
Yes, but I want my storage separated from my computer, partially for easier upgrades down the road and partially for size of storage. With enough drives I should get at least as much speed as from my SATA drives (even as they are SSD only) but I don't think USB will cut it. I know I can add drives in the desktop and I did, but I feel it would be easier to separate the upgrades. Another reason is that I want my computer to be silent; if I want a lot of cheaper storage I would need to go back to regular HDD, thinking of 6 to 8. Putting them in a box that I can move farther away from me would decrease the noise.
I would also need to upgrade my NAS in the next year or so, therefore I contemplated using a NAS with dual 10GB, or with the newer 40GB or 100GB but it looks a little too complicated for my level, particularly if I need to troubleshoot it. I could try a NAS with both TB3 and 10GB for best of both worlds but I would like to see some real world testing.
I can appreciate the idea of having storage separate from your computing. I would argue it's better to go with NASes than an external disk array, as they're about the same cost if not cheaper. QNAP actually does the TB3/10gb NAS, where the TB3 is a network adapter that's connected to a 10gbps switch inside the NAS, but I wouldn't go that route. The only storage I see needing TB3 is the ThunderBlade or Express 4m2 by OWC - 4x NVMe slots. My most wanted device right now is a TB3 -> 8-12 M.2 SATA or 2.5" SATA enclosure, but at that size they're going to want to have a RAID function built in :( .
What amount of space are you looking at needing? Dual 10gb is fine, but what are you plugging it into that'll have the bandwidth to match it? 25/40gb would be a fiber cross connect between the NAS and your workstation. Solo 10gbps should get you around 1,2000MBps with Jumbo Frames if the NAS can handle it (and they're loaded with SSDs).
-
If I don't go the NAS route I was looking at the OWC Thunderbay RAID 6; it doesn't have built in RAID but with a new processor (in my computer) I hope I won't feel it that much. It has room for 6 HDD and 1 NVME SSD. I still get a feel they are optimized for Mac though and I'm not convinced that I will switch.
I currently have around 3.5TB but with the Z7 I wan't to have enough storage for some years, so I would like something in the 16-18TB range at least. My current NAS has 8TB and I'll soon have to delete some of the windows backups.
I knew Synology was supposed to come up with a TB3 NAS but it's missing in action (the DS1817T).
-
I would love a ThunderBay 6 designed for 2.5" drives! Too bad the Synology DS620slim doesn't have dual 2.5/5gbps ports, and a touch more horsepower - it'd make a great 10tb (SHR/F1 RAID5 over 2tb SSD's). Plug in an external 14tb drive over USB3 & backup to it.
-
Back on topic.
Does anyone else have any insite, thoughts on these new processors?
Thanks.
Andrew
-
These days I’d definitely go for an 3rd gen Ryzen. I think the AMD Ryzen 9 3900X, with 12-cores and 24-threads and boost clocks up to 4.6GHz is a high-performance achiever while at about $ 499,- still very cost effective. Also many PCIe lanes, you won’t have limitations here on the graphics- and m.2 storage pipelines.
Combine this with a motherboard containing a X570 chipset supporting the latest PCIe 4.0 standard, giving you up to 51% faster SSD performance, along with 69% faster graphics performance (could be more something for the nerds or benchmark measurebators, as end user you may not notice that much from it).
Also something to consider: various security leaks with Intel processors and performance degradation after a ‘fix’ in software.
On a personal note: I now have a PC containing an Intel i7 processor but my next system will definitely be based on the above mentioned AMD configuration, in combination with an Nvidia RTX 20xx graphics card together with the new 10 bit studio driver (no expensive Quadro required anymore).
Regards,
Jaap.
-
Intel is not supposed to have an answer to Ryzen 3000 processors until sometime in the second half of 2020 for desktops and workstations. Now would be the time to give AMD some support so that there will still be competition in the x86 processors space in the future.
I plan to build a new PC later this year with either a Ryzen 3000 or a yet to be announced Threadripper processor of the same generation. This will depend on the price difference for a 16 core processor. I don't need more cores but could use the extra PCIe lanes offered by the latter processor in a few years.
The only thing bothering me with X570 chipsets is their need to be actively cooled with a 4 cm fan. That could be noisy and will likely fail at some point down the line, with no standardised replacement part. There is only one existing motherboard which has a passive heatsink but with a ridiculous price tag and tons of features that are not needed by most people.
Cheers,
Fabien
-
I am really interested in the 3rd Gen Ryzen and need to build a new system soon...... but I am scared silly .....
as it looks like heavy duty cooling might be needed and I do not want fans running all the time.
I am confused as to what would be the optimum graphics card to use ... my interest is Photoshop and Lightroom processing... not video editing.
-
A good read pertaining to photography: https://www.pugetsystems.com/labs/articles/Photoshop-CPU-Roundup-AMD-Ryzen-3-AMD-Threadripper-2-Intel-9th-Gen-Intel-X-series-1529/#AretheRyzen3rdgenerationCPUsgoodforPhotoshop?
-
Thanks for the link to Puget Systems. I missed that post.
AMD looks like the way to go until mid 2020 unless Intel change their pricing model.
I have watched several builds using the 3rd gen Ryzen and also the Intel i9. These builds all include significant cooling options, with a few indicating performance throttling and needed to upgrade their cooling hardware. I do not mind putting these into the build, but I hate the thought of fans running beside my desk for the next 5 years or so.
-
I'll probably go Intel when I do it. I built my current system with a focus on low noise and I would like to keep it that way.
-
Speaking of thermal dissipation, it should be noted that recent Intel processors rated with a Thermal Design Power (TDP) of 95 W will draw up to 200 W when all cores run at turbo speeds. This is unlike Ryzen processors, which draw only slightly more than their advertised TDP of up to 105 W at turbo speeds. Newer Ryzen processors are more power efficient than their Intel counterparts.
If a quiet setup is the goal, this has to be accounted for. I am not sure that an Intel based system will be the quieter option (unless speaking about laptop processors) because of the need to dissipate more heat.
The active fan on X570 chipsets for Ryzen builds (mentionned in previous posts) now seems to run quieter with recent BIOS updates. Those apparently correct issues where the cooling fan was always running at full speed, independently of the system load.
Cheers,
Fabien
-
Newer Ryzen processors are more power efficient than their Intel counterparts.
they are indeed , https://www.tomshardware.com/news/amd-ryzen-3000-boost-clock-controversy-intel-attack,40231.html & https://www.tomshardware.com/news/amd-ryzen-3000-not-hitting-advertised-boost-speeds-survey,40291.html
-
they are indeed , https://www.tomshardware.com/news/amd-ryzen-3000-boost-clock-controversy-intel-attack,40231.html & https://www.tomshardware.com/news/amd-ryzen-3000-not-hitting-advertised-boost-speeds-survey,40291.html
Boost clocks not being achieved has nothing to do with power efficiency. The potential issue described in the linked articles has already been corrected in an upcoming BIOS update, as stated in this article (https://arstechnica.com/gadgets/2019/09/amd-improves-real-world-boost-and-idle-on-ryzen-3000/). More relevant to the theme of power efficiency is the lower idle power consumption also introduced in this update. To quote the article:
AMD's new reference firmware adds an activity filter that lets the CPU's own boost algorithm ignore "intermittent OS and background application noise" that might otherwise ramp the CPU up into its boost clock. Certain bursty but overall lightweight tasks, such as video playback, application launch, and monitoring utilities, make regular requests for a higher boost rate, even though their overall activity level is low. The activity filter smooths things out for these lightweight tasks, keeping the CPU out of boost mode without harming overall or perceptible performance.
As for the hypothesised premature failure of batches of Ryzen processors mentioned in the linked articles above, we will know for sure in a few years. Intel was badly affected by this problem three times in the past decade (Intel 6 series chipsets, Avoton/Rangeley and Apollo Lake Atom processors) and should be in no position to spread FUD about its competitor on this topic.
Cheers,
Fabien
-
Boost clocks not being achieved has nothing to do with power efficiency.
I run i7-9700K @ all 8 cores locked all the time @ 4900 MHz... and yes when I stress all 8 cores at the said freq it is not 95W, but almost double (and more with AVX2)... but intel goes (I mean running stress test) with aircooling by Noctua NH-D15S for hours with slight noise from the fans ... try to overclock all the cores on Ryzens ;D - so what is the point that it power efficient if you can't really clock it ? you are not running a datacenter ...
-
I run i7-9700K @ all 8 cores locked all the time @ 4900 MHz... and yes when I stress all 8 cores at the said freq it is not 95W, but almost double (and more with AVX2)... but intel goes (I mean running stress test) with aircooling by Noctua NH-D15S for hours with slight noise from the fans ... try to overclock all the cores on Ryzens ;D - so what is the point that it power efficient if you can't really clock it ? you are not running a datacenter ...
We were talking about designing quiet systems, so having less heat produced (due to the higher efficiency) will help in achieving this goal.
If your workload can use all cores for hours, you might as well use a 12 cores Ryzen at base frequency with 105 W, which will still outperform a current Intel 8 core system that outputs 200 W. ;)
Cheers,
Fabien
-
I will go for the Ryzen 3900X or 3950X with the X570 Motherboard.
All the test results on the internet shows the Ryzen has much more processing power when it comes to productivity.
I see no reason to go with the Intel unless you do heavy gaming and aimming for Frame/Sec for First Person Shooting game.
-
We were talking about designing quiet systems, so having less heat produced (due to the higher efficiency) will help in achieving this goal.
my system is quiet... and as I am not running a 1 liter case in the middle of Sahara in a 10 sq ft room w/o airconditioning ... high end air cooler with can deal with with up to 250w w/o much noise in a regular home ...
If your workload can use all cores for hours, you might as well use a 12 cores Ryzen at base frequency with 105 W, which will still outperform a current Intel 8 core system that outputs 200 W. ;)
nope - my workload a single raw at at time work using either ACR or C1 and then minor finishing PS and Ryzen efficiency (or more cores) brings nothing because all heavy lifting is done by a single thread on CPU + GPU ;D ... plus tests you possibly saw, don't use overclocking and once you do ;D ... as I noted - I have no noise (I do notice noise from GPU in real workloads, not from CPU cooling) or heat issues with my computer case, my cooling, and ambient temperature in the room where I work
(https://www.pugetsystems.com/pic_disp.php?id=56165&width=800)
-
I seems there are finally AMD motherboards supporting TB3, this one does it with a add-in card but otherwise looks good (outside of price): https://www.newegg.com/p/N82E16813145174?Description=TRX40&cm_re=TRX40-_-13-145-174-_-Product
Hopefully there will me more mainstream motherboards with similar specs showing up over the next half year.
-
I would definitely choose a PCI 4.0 board, and a CPU that can handle a lot of PCI lanes needed for M2 memory and the GPU.
AMD ryzen seems a fast choice (AMD Ryzen 9 3900X) cheap and very fast for LR. It can handle only 24 PCI lanes, so you have to calculate if that is enough.
-
I'm definitely consider the CPU upgrade to the 3900 or 3950. I found out my older Asrock MB can handle the chips with a BIOS upgrade. I've been using a Ryzen 1800x since it was first introduced (2 years??). While running on a MB with the x370 chipset might hobble things a little, by comparison my system should scream. I find no matter what I do, Lightroom is a slug, so I'm looking at both the CPU and maybe replace my GTX1080, though it's still a good GPU.
While Thunderbolt would be nice, it's a non-issue really. My Dell XPS15 2-in-1 has it, but unless I get a TB-centric external drive, it's no faster than the USB 3 Gen 2 ports.
-
With respect to “Lightroom is a slug” you’ll probably get no improvements after a CPU upgrade (at more or less equal clock speeds). And your GTX1080 is already more than fast enough.
The sluggishness must be caused by something else (apart from the crappy coding of Lightroom itself – poor multi processor support, poor GPU processing). Still applying mechanical HDDs with platters? Storage over the LAN? In that case you may expect significant improvements after transferring to local SSDs.
Regards,
Jaap.
-
Agreed Jaap. I have a PC (Windows 10) with a 7200 speed HD and now have a MacBook Air. The Air was got for portability and fast boot and, dare I say it, I like the screen. I know this is not in the league of these builds, but I now have a couple of Sandisk 2TB portable ssds. They come formatted for both PC and Mac, and with the adapters. With a hard shell that holds the SSD and adapters each weighs 115g (4oz) and is small. 47GB of images went from an ssd to the mac's ssd in less than 2 minutes, magic for me! I bought them from Amazon when there were offers.
Of course it all depends on need. I am not dealing with big video files nor with 16bit TIFFs from a GFX100. My RAWs are about 50M.
Best wishes,
Jonathan
-
The sluggishness must be caused by something else (apart from the crappy coding of Lightroom itself – poor multi processor support, poor GPU processing). Still applying mechanical HDDs with platters? Storage over the LAN? In that case you may expect significant improvements after transferring to local SSDs.
I use two SSDs -- one is my boot/C: drive, the other is my scratch disk for Photoshop and Lightroom. I do have a mechanical drive (a WD Black) on which my programs are stored, but once LR is loaded (I have 32GB RAM), it should be running off RAM plus the scratch disk. Of course, who knows how they really have these things programmed. A faster CPU does help with performance, even LR and Photoshop. Some of my issues may be due to driving two monitors -- an HD and a 4K -- with "only" 8GB on the GTX1080. (I have some other issues that have cropped up with my 4K, but that's a post for another time.)
-
...I find no matter what I do, Lightroom is a slug, so I'm looking at both the CPU and maybe replace my GTX1080, though it's still a good GPU.
On my computer i use a GPU half as fast as the GTX1080 and it works fast with LR.
I have a very fast CPU 10core intel i9 that runs on 4.5hz... and very fast m.2 memory - so it must be that ( 32gig ram- but that will not be the problem)
LR uses the CPU tot the max.
-
That's why I'm considering a "brain swap" in putting in the new 3900x.
-
I understand you considering a brain swap. I do hope however not to put your expectations too high here. First look at the CPU performance during Lightroom, I don't think it's hitting 100% for a considerable amount of time. If performance is low then the CPU is waiting for other processes or peripherals. Here a faster CPU would not solve anything. As with your GTX1080, this one should fly, also with 2 monitors attached and 8Gb video memory.
I suspect Lightroom doing something on your mechanical drive. I sometimes see weird things going on with Photoshop (I don’t use Lightroom), building up a scratch file while there is still more than sufficient unused RAM available, communicating to the HDD in 4K datablocks, etc. Please be aware that HDD performance significantly drops, becoming terribly slow, when dealing with 4K data blocks.
As long as you haven’t found the root cause of Lightroom’s sluggishness I’d be a bit careful spending money and efforts on exchanging parts that supposed to be already fast in the first place. As for now my suspicion goes into the direction of the mechanical drive.
Regards,
Jaap
-
I use two SSDs -- one is my boot/C: drive, the other is my scratch disk for Photoshop and Lightroom. I do have a mechanical drive (a WD Black) on which my programs are stored, but once LR is loaded (I have 32GB RAM), it should be running off RAM plus the scratch disk. Of course, who knows how they really have these things programmed. A faster CPU does help with performance, even LR and Photoshop. Some of my issues may be due to driving two monitors -- an HD and a 4K -- with "only" 8GB on the GTX1080. (I have some other issues that have cropped up with my 4K, but that's a post for another time.)
Where is the Lightroom Catalog? It should be on one of the SSD disks, since that's where most the disk I/O happens (between the catalog, preview & smart preview files). The referenced files can be in spinning rust without issue, but he catalog really has to be on something faster. NVMe > SATA SSD > SATA SSHD > SATA HDD
The way you think of everything being loaded into RAM is actually not correct. Have you uninstalled it from the current location & installed it to the OS SSD?
Have you run task manager and kept an eye on the performance tab while using Lightroom? If you're not working the CPU hard as shown there, doing a brain upgrade isn't going to give you a huge jump.
-Joe
-
Having just assembled a new PC with a Ryzen 3900x processor, I can say that Lightroom flies compared to my previous build (7 years old Xeon E3-1245 V2).
The interesting thing is that I haven't changed the GPU (NVIDIA GTX 780) and the HDD hosting the RAW files. The improved behaviour is probably related with the increased single thread CPU performance and with the move of the catalogue from a SATA SSD to a NVMe SSD.
So add one more data point to what Joe said above: Lightroom seems to be I/O bound for the catalogue and the CPU will probably help in the develop module. Since my GPU is limited to DirectX 11, it only works in basic mode and doesn't assist much with the processing.
To come back to Nemophoto's point, a "brain swap" between a Ryzen 1800x and 3900x would at most bring a 30% improvement to single thread CPU performance. Not bad but that alone might not solve the performance problems.
Cheers,
Fabien
-
Despite my initial intent to have TB3 with my next update, the significantly better results the Ryzen has is forcing my hand.
I'm probably 80% decided on what to get, see here: https://pcpartpicker.com/list/87BZzN
I have the rest of the components, including a decent case and a video card (RTX 2060). If the video card doesn't get along I'll change it to a Radeon 2700 XT.
For storage I'll likely keep my collection of SATA SSDs in a storage pool (currently @ 5.4TB effective and occupied @ 67%) instead of going all in with NVME SSDs because I would need them to be larger than 2TB and currently there is only one option, Sabrent Rocket 4TB (I would get 2), with some interesting performance choices.
Where I'm not decided is the CPU; initially I was set on the 3900x and it is the rational choice when you think of money/performance compared to the 3950x but the latter would make the system more future proof. I doesn't make sense to get the 3900x now and upgrade later to the 3950x so I will stick with the decision. If there will be any CPU upgrade it will be for the next gen Ryzen, if the motherboard will still support it of course.
Any thoughts on this?
Thanks
-
In my opinion a 3900x (12 core @ 3.8GHz) is a better choice than a 3950x (16 core @ 3.5GHz). For now, and also in the long term. And a 3900x is even significantly less expensive.
Investments for the future are always tricky. In a few years’ time there will be much faster CPU’s, sweeping the current generation off its feet.
Keep in mind though that CPUs are in many cases not the limiting factor as they are waiting for peripherals to finish their tasks. Think of a relatively slow CPU performing its task with a 50% load versus a fast one at a 25% CPU load (so doing nothing for 75% of the time).
Regards,
Jaap.
-
3800x $340 / 3900x $470 / 3950x $750
I've got the 3800x and it is keeping up with my tasks. If anything I'd look at the matchup in RAM capacity v speed.
Remember the Ryzen4 will drop in 10 months. The idea of doing a huge upgrade and running it for years is kind of old school. It's better to do incremental upgrades on a rotating cycle. Figure you'll replace at least half of the system in 2-3 years.
-
Based on this: https://www.pugetsystems.com/labs/articles/What-is-the-Best-CPU-for-Photography-2019-1620/ for photo the 3950x doesn’t add more than 5% performance for 50% increase in price. Interestingly for video, particularly Adobe, the Intel 9900K holds its own. I don’t do much video (mostly because it’s too time consuming) but it’s good to know you don’t lose everything going Intel route.
I’ve been building my systems for a while but truth being told, while I occasionally do small incremental updates, if you wait several generations between updates you end up changing most of the system. That’s why buying a prebuilt system (desktop or laptop) and just replacing it every 4-6 years makes sense for many; having your storage solution separate helps also.
-
I like the looks of those CPUs but I usually build my systems on the cheap. I just built a new AMD box for myself that started with a ASUS Ryzen 2700 since they got down to $150. So I built a nice "gaming" system around that with a ASUS ROG STRIX B450-F GAMING mobo, 32 GB DDR4 3200 MHz RAM, a 2TB M.2 PCI SSD, and a MSI Radeon RX 580 video card. It's a nice upgrade from my previous i7 system that was getting pretty old.
My goal with this build was not only to have a new system but also to get on the same platform of that new Ryzen 9 chip for when that gets cheaper.
It will be a nice upgrade in a couple/few years to double the ram and pop a new CPU in.
-
I like the looks of those CPUs but I usually build my systems on the cheap. I just built a new AMD box for myself that started with a ASUS Ryzen 2700 since they got down to $150. So I built a nice "gaming" system around that with a ASUS ROG STRIX B450-F GAMING mobo, 32 GB DDR4 3200 MHz RAM, a 2TB M.2 PCI SSD, and a MSI Radeon RX 580 video card. It's a nice upgrade from my previous i7 system that was getting pretty old.
My goal with this build was not only to have a new system but also to get on the same platform of that new Ryzen 9 chip for when that gets cheaper.
It will be a nice upgrade in a couple/few years to double the ram and pop a new CPU in.
I was contemplating a new MacBook Pro 16 so this is actually quite a lot cheaper.
I ran the Puget benchmark for Photoshop and I am roughly at 557 (out of their 1000 for a 9900K test station), see attached. I couldn't get the Lightroom benchmark to run but somebody with a slightly better processor (4790k vs 4790) got an overall score of 553. So on average I can expect a much faster computer.
As a side note the Ryzen test results that I've seen there aren't that much faster, a least not on everything they test.
-
I was not aware of this benchmark- so i also ran it.
Must say somethings may differ: History state, percentage of RAM used etc...
Also it only runs on an English Photoshop ( you can alter it) and with psd saving set to compressed. ( otherwise the file will exceed 2GB and will not be saved- script stops)
Anyway my system does 853. I like the benchmark ; it works for windows and Apple because it is a java written script and seems very comprehensive. (it takes about 25 minutes to run 3 times)
PS i do not have an iMac-pro but a hackintosh with a 10 core i9 7900x running at 4.5 GHZ
-
Ok, I did not know this benchmark either so I ran it too. 691.2 here. I could probably do a little more tuning but it seems acceptable for my daily needs for what it cost.
(https://live.staticflickr.com/65535/49453450596_2eb9cc8e3c_z.jpg) (https://flic.kr/p/2im2RVW)
-
Did any here built or saw a system with the latest Ryzen or with the Intel 9900K that has 128GB RAM?
In theory it's supported but I have yet to see or hear about a system like this and when I looked through the supported RAM list on motherboards there is a singular 2666MHz Crucial @ 32GB that might work in a 128GB setting but it's not clear. Everything else that I saw was up to 64GB RAM and even there many had issues with 32GB sticks, particularly faster than 2666MHz. The closest that I've seen that makes think it should be possible it's the configurator from Puget for a LR machine where they have the option for 128GB but even there they change to a Samsung memory which I have yet to see for sale.
-
Made me curious how my system stacked up against Puget 9900k. I have an 6 core/12 thread 8086k which I have a daily overclock of 5.1 GHz all cores, 2070 Super and 64 GB memory for which I allocated 40 GB to photoshop. Pretty darn close but I wonder why it is worse than their stock 8700k, especially on the filter score. is it different versions of their benchmark, photoshop?
At the moment, ff I were going for a new build from mostly scratch, I think I would go with AMD.
(https://photos.smugmug.com/photos/i-LvRpJBd/0/c37c1941/L/i-LvRpJBd-L.jpg)
EDIT: FWIW, a lot of the difference is in Spectre patches (disabled below), some from increasing memory for photoshop. So there is something like a 5% hit for Spectre/meltdown for Intel that may or may not be accounted for if you compare results.
(https://photos.smugmug.com/photos/i-GSzBCvH/0/ed198c6c/L/i-GSzBCvH-L.jpg)
-
My intention is to build my Ryzen computer with 128gb.
Aiming for the middle of Feb as a finished build.
Did any here built or saw a system with the latest Ryzen or with the Intel 9900K that has 128GB RAM?
In theory it's supported but I have yet to see or hear about a system like this and when I looked through the supported RAM list on motherboards there is a singular 2666MHz Crucial @ 32GB that might work in a 128GB setting but it's not clear. Everything else that I saw was up to 64GB RAM and even there many had issues with 32GB sticks, particularly faster than 2666MHz. The closest that I've seen that makes think it should be possible it's the configurator from Puget for a LR machine where they have the option for 128GB but even there they change to a Samsung memory which I have yet to see for sale.
-
I‘m currently building a Ryzen 3960 system with 256Gb.
I choose the G.Skill Ripjaws V DDR4-3200 32Gb each.
Did any here built or saw a system with the latest Ryzen or with the Intel 9900K that has 128GB RAM?
I would need to check for x570 boards.
In theory it's supported but I have yet to see or hear about a system like this and when I looked through the supported RAM list on motherboards there is a singular 2666MHz Crucial @ 32GB that might work in a 128GB setting but it's not clear. Everything else that I saw was up to 64GB RAM and even there many had issues with 32GB sticks, particularly faster than 2666MHz. The closest that I've seen that makes think it should be possible it's the configurator from Puget for a LR machine where they have the option for 128GB but even there they change to a Samsung memory which I have yet to see for sale.
-
I‘m currently building a Ryzen 3960 system with 256Gb.
I choose the G.Skill Ripjaws V DDR4-3200 32Gb each.
That's on TR40, right?
-
My intention is to build my Ryzen computer with 128gb.
Aiming for the middle of Feb as a finished build.
Great, we are almost in sync. I will reuse my PSU, case and most of the storage.
So I've been looking everywhere for a 128GB build and have yet to find it. I know Puget offers it with a 2666MHz Samsung RAM that's not really on the market and I hear that going above 2666MHz for 128GB total might cause stability headaches.
In the pcpartpicker lists of completed builds with my tentative motherboard I have yet to find somebody with a 128GB. I did find a couple of build with 64GB @ 3200MHz and 3600MHz.
So far I'm at this stage, trying to figure the RAM problem: https://pcpartpicker.com/list/PJGhmg
PS. I should have been more clear, Ryzen 9 3rd generation, 3900x in particular. Willing to look again at 9900K if it will offer 128GB RAM and Ryzen won't. Ryzen Threadripper are just too expensive and the Intel i9x don't offer similar performance to the 9900K unless you pay triple.
-
That's on TR40, right?
Yes that is correct. My current intel system is running 128Gb and I’m running out to often. Or let’s say often enough so that I want more on my next build.
I’m currently building with MSI boards because I have very good recent experiences with them. However, Asus and Gigabyte are fine as well.
For example the MSI PRESTIGE X570 CREATION Supports ( would choose it for the 10g LAN support) same goes for the MSI MEG X570 ACE
Corsair CMV32GX4M1A2666C18 2666MHz 2666MHz 2666 MHz Spectek PP038-093TP 1.2V DUAL 32GB √ | √ | √ (The third ok is for all 4 memory banks)
Or
G.SKILL F4-3200C16Q-128GVK 2666MHz 3200MHz 3200 MHz SK hynix MFR 1.35v DUAL 32GB √ | √ | √
So it should be 100% fine to go with either Memory on that board.
I can check the Gigabyte as well just don’t have time right now.
A few more points to your list. I would go with Noctua NF-A12x25 as in my testing and opinion these are by far the best fans out there currently. I would combine it with one noctua NF-A14 ULN.
I would choose a SSD without any heat sink on, as you will put it under the mainboards cooling system. I’m not sure if you can remove it easily on the SSD.
If you have any question just ask. I will try to help where I can.
-
Yes that is correct. My current intel system is running 128Gb and I’m running out to often. Or let’s say often enough so that I want more on my next build.
I’m currently building with MSI boards because I have very good recent experiences with them. However, Asus and Gigabyte are fine as well.
For example the MSI PRESTIGE X570 CREATION Supports ( would choose it for the 10g LAN support) same goes for the MSI MEG X570 ACE
Corsair CMV32GX4M1A2666C18 2666MHz 2666MHz 2666 MHz Spectek PP038-093TP 1.2V DUAL 32GB √ | √ | √ (The third ok is for all 4 memory banks)
Or
G.SKILL F4-3200C16Q-128GVK 2666MHz 3200MHz 3200 MHz SK hynix MFR 1.35v DUAL 32GB √ | √ | √
So it should be 100% fine to go with either Memory on that board.
I can check the Gigabyte as well just don’t have time right now.
A few more points to your list. I would go with Noctua NF-A12x25 as in my testing and opinion these are by far the best fans out there currently. I would combine it with one noctua NF-A14 ULN.
I would choose a SSD without any heat sink on, as you will put it under the mainboards cooling system. I’m not sure if you can remove it easily on the SSD.
If you have any question just ask. I will try to help where I can.
Thank you.
It's a little weird because I looked at those MSI supported RAM and I missed it, wonder if I was looking at something different than Matisse or maybe because when I sort by stick size 16GB is shown to be bigger than 32GB. So they support 128GB up to 3200MHz (certified). For Gigabyte and AsRock I can only find the Crucial 2666MHz, for Asus up to 64GB. In the same time the MSI list is the only one that looks that it was updated after the board was released (12/5/2019) while the others are on the launch date (8/2/2019) and they didn't bother updating them after.
Until now I've used only Asus boards but on my last there are small nagging things (mostly software related) that I'm willing to look somewhere else; Gigabyte because that's what Puget uses but I can look at others.
Regarding the fan; I already have case fans right now, those in the build are in addition to (or maybe replace the older ones), because I hear X570 is hot. I only have Noctua that I added my my existing Fractal Design that came with the case. Noise is my main concern, currently my computer is practically silent as in I can't hear it if when I listen. When I overclocked the CPU I had an initial faint whoosh upon starting, for few seconds, but that's about it.
As for the SSD with heat sink, it depends on the board. On the Gigabyte the M2 shields are separate from the cooling so I can just have SSD with the built in heat sink and not cover it; on the MSI that you referenced I can see that they are part of the cooling system.
-
Still don't understand why people need more than say 32 GB RAM...
I check my ram use frequently and hardly ever i need more. ( and i work with large GB panorama's)
If i do need more, I just close a program that is not in use and have enough again.
It is mostly unused cache that piles up, also because some programs just gather and gather without releasing it. ( photoshop for one)
I am on a Mac 13.6
-
Still don't understand why people need more than say 32 GB RAM...
I check my ram use frequently and hardly ever i need more. ( and i work with large GB panorama's)
If i do need more, I just close a program that is not in use and have enough again.
It is mostly unused cache that piles up, also because some programs just gather and gather without releasing it. ( photoshop for one)
I am on a Mac 13.6
For one thing I'm lazy and keep many things open, including a Chrome with 10-30 tabs, and if Photoshop is opened then LR is too (I almost always make a trip only from LR). Lately my PS has been giving me errors in regards to memory, I had to do some registry editing to run that Puget test. I didn't test enough after to see if it really fixed the problem. My idea is that I want to maximize my build to keep me the longest and it's easier to add the max RAM now than to try to find identical sticks later.
I'm not earning a living out of photography (on the contrary) so I want to have this flexibility.
The question mark with X570 and 128GB is that while theoretically is supported I don't see builds with that much RAM. I do see many problems with 4 sticks/faster memory through, at least with XMP enabled, that I'm a little apprehensive about it.
Christopher quoted above some G.SKILL RAM at 3200MHz (not seen supported on my Gigabyte info) but shown as supported on MSI motherboard and shown as compatible with the Gigabyte board on G.SKILL compatibility list; problem is it suddenly went out of stock everywhere in US, quite weird, makes me think it was pulled off the market that actually sold.
On a side note the Aorus Master remains on the list of motherboards and I added the MSI Prestige X570 Creation (10GbE LAN, tons of USB 3.2 ports) and ASRock X570 Creator (10GbE LAN, TB3 that might work).
-
From what I've seen, there never has been much supply of the high-density 128GB kits. It is new and there is not a lot of demand.
Generally, if someone needs that much RAM. they are looking at motherboards with more than 4 dimm slots and more than 2 channels.
-
Still don't understand why people need more than say 32 GB RAM...
I check my ram use frequently and hardly ever i need more. ( and i work with large GB panorama's)
If i do need more, I just close a program that is not in use and have enough again.
It is mostly unused cache that piles up, also because some programs just gather and gather without releasing it. ( photoshop for one)
I am on a Mac 13.6
You probably don’t work on 200-1000Mp images.... sure you. An do that with 32 or even 16Gb of ram. However, it’s painfully slow if Photoshop hast to use a scratch disk all the time. Even if it’s a modern m2 SSD.
-
You probably don’t work on 200-1000Mp images.... sure you. An do that with 32 or even 16Gb of ram. However, it’s painfully slow if Photoshop hast to use a scratch disk all the time. Even if it’s a modern m2 SSD.
I do - not slow at all- certainly in relation to how slow it is to make these giant photographs.
https://www.beeld.nu/beeld/
-
As I said the definition of slow and fast is difficult to judge for people not experiencing both. I can say that the difference between 64GB and 128GB is there on larger images and it’s very significant.
All in all Photoshop shows very nicely if you are at full speed or not, as when it’s dropping efficiency it’s not as it’s fastest and this is happening to me quite often. I do work with lots and lots of layers at some points of my processing.
If you don’t run out of memory then it’s absolutely true you don’t need more. It’s also true that 64GB is probably enough for 99% of us all.
-
I was contemplating stepping up to the Creator/Creation series from ASRock or MSI because they have 10GbE LAN and TB3 (for ASRock). The rational for having those is to move my fast access storage from inside my computer to a NAS/DAS. The reason would be easier computer upgrades and easier access for other computers.
However, I think I'm having an epiphany in regards to how much I truly need this.
Currently my fast storage is on SATA SSD in a storage pool in Windows (basically JBOD). I can reasonably expect somewhere around 500 MB/s.
Moving to 10 GbE LAN I'm looking at 1000 MB/s or so (if the NAS can provide that reading speed). A higher level Synology can have up to 2000 MB/s although I have my doubts how they got to that (probably dual aggregate 10 GbE LAN card and SSDs). A higher level QNAP can provide 1000 MB/s or so, through a 10 GbE or TB3.
So in theory with one of those options I can double my speed. If I look at the cost though I will pay more than $1500 either way (likely more for QNAP). A TB3 DAS isn't that easy to come by, not one that will easily exceed my current 500 MB/s.
Now, if I look again at the internal fast storage option, I can add 2 more Gen4 (even Gen3 will be good) at 4TB each for a total of 8TB at a speed reaching 5000 MB/s and currently at $1500. I don't like that there is only one option though, the Sabrent Rocket. I could go with a add-in PCI-E NVME card and have 4 2TB drives but it's more complicated. My current video card doesn't saturate even a PCI gen3 x8 and I don't need much SATA ports either (sharing bandwidth with the add-in card) so it's possible.
If I don't need a fast NAS I can just keep up with what I have right now for a little longer, so no extra cost here.
So, in conclusion, I think the priority has become for me how many M2 gen4 slots the motherboards has and how easy is to use all (and add more).
-
An OWC ThunderBay6 would give you 6x SATA plus a M.2 NVMe slot off a single TB3 port. Their OWC Express 4M2 is 4x NVMe drives off a single TB3 port.
There are a couple of PCIe 3.0 16x cards to 4x NVMe adapters out there, but I haven't seen a PCIe 4 one yet.
The ThunderBay6 seems to be the best option out there. I haven't dealt with Windows Storage Pools yet, but if you can use a NVMe as a caching device in front of a bunch of SATA SSD's, with it tuned for performance it'll be fast & mobile. I actually think checking on the SSD units themselves as to their performance & tech might actually get you better results. Some SSD's just aren't up for the higher workloads.
-
The ThunderBay6 RAID was the main option that looks reasonable, the OWC Express not so much (see my previous post about getting the NVME drives internal) with a max 8TB capacity.
The problem is that it can go up to 1150 MB/s or so, but is similar to a 10GbE would do so not much advantage here. And that's the maximum speed which you rarely achieve in real world. IF I were on a Mac, then this would still be a decent option. However, the SoftRAID that they base their speed calculations on (https://www.softraid.com/pages/features/raid_levels.html#raid_speeds_chart ) is not available for Windows, not yet at least. So in Windows you have to use storage pools, the best would be the "parity" option, similar to a RAID5.
Then it's the big elephant in the room, TB3 on windows on an AMD processor. Basically one option, ASRock X570 Creator (I hear Taichi with a add-on PCIE card works too but then the cost difference is much smaller and not worth it). I don't know how reliable the TB3 connection is on this motherboard. And going back to actual speed, getting a similar speed on a 10 GbE makes me less interested in TB3 in a desktop if I can add enough NVME SSD internal capacity. I would have an opposite opinion if I had only a laptop as a primary computer though.
-
Upon further reading it looks like the write performance of Storage Pool with parity in Win10 is pretty bad so for fast SSDs you are probably better with simple storage pools (similar to JBOD), you lose some read speed but gain a lot more write speed. Not too mention that if a drive fails with the first option it will take much longer to recover in comparison to restoring from a backup.
-
It was once said that Mac's were just Photoshop machines, as in they were used for that purpose alone. Much like Mac folks had Wintendo's, pc's just to play games. I like NAS'es because they separate my complex & large storage from my day to day technology uses.
That the storage pool setup on Win10 has limitations like that sucks big time - especially when you have invested in SSD's.
-
Upon further reading it looks like the write performance of Storage Pool with parity in Win10 is pretty bad so for fast SSDs you are probably better with simple storage pools (similar to JBOD), you lose some read speed but gain a lot more write speed. Not too mention that if a drive fails with the first option it will take much longer to recover in comparison to restoring from a backup.
Any software based raid with parity will perform badly on writes because of the parity calculation, including Intel RST. For raid with parity, you need a dedicated RAID card in order to get decent performance.
-
Indeed, software based raid performs badly and should be omitted. However, many motherboards have on-board hardware raid controllers, configurable from the BIOS, not drawing performance from the CPU, supporting raid-5 with parity.
Regards,
Jaap.
-
Indeed, software based raid performs badly and should be omitted. However, many motherboards have on-board hardware raid controllers, configurable from the BIOS, not drawing performance from the CPU, supporting raid-5 with parity.
Regards,
Jaap.
Those aren't hardware based - they are software based. They merely let you recognize the software RAID in BIOS, so you can boot from it (unlike Microsoft software raid, which you can't boot from). They are still software based even though you configure it in BIOS. They have been dubbed "fake raid". Parity is still calculated in software by drivers.
Software RAID 0 performs very well, but doesn't provide any redundancy.
-
Any software based raid with parity will perform badly on writes because of the parity calculation, including Intel RST. For raid with parity, you need a dedicated RAID card in order to get decent performance.
Indeed, software based raid performs badly and should be omitted. However, many motherboards have on-board hardware raid controllers, configurable from the BIOS, not drawing performance from the CPU, supporting raid-5 with parity.
False, the world runs on software RAID. NetApp, Dell/EMC/EqualLogic, HP, IBM, Oracle/Sun - every enterprise storage vendor uses software RAID extensively, and exclusively once you get into larger systems (for which they charge millions of dollars). Hardware RAID chips are still around, but they're not used heavily anymore. Things like thin provisioning, data deduplication & data tiering require software, not some Adaptec or LSI chip. Hardware RAID is only used when software RAID can't - like boot volumes. Yes, a lot of the 'RAID' in BIOS chips is fake software RAID. Look at things like ZFS and mdam - software RAID that requires direct access to the disks, not an abstracted hardware controller. Every cloud provider runs on software RAID - AWS, Azure, Google, etc... YouTube learned early on that disk I/O is much better when it knows there's 10 disks to talk to, rather than the 1 disk that a RAID controller would present.
Microsoft had to architect the Storage Pools to work in a way that's safe enough for consumers who don't do things like have UPS'es & are prone to accidently unplugging something.
I would be interested to see if adding 1/2 NVMe drives to the pool would improve the write performance, but there are limits built in to handle consumer level issues.
-
False, the world runs on software RAID. NetApp, Dell/EMC/EqualLogic, HP, IBM, Oracle/Sun - every enterprise storage vendor uses software RAID extensively, and exclusively once you get into larger systems (for which they charge millions of dollars). Hardware RAID chips are still around, but they're not used heavily anymore. Things like thin provisioning, data deduplication & data tiering require software, not some Adaptec or LSI chip. Hardware RAID is only used when software RAID can't - lie boot volumeks. Yes, a lot of the 'RAID' in BIOS chips is fake software RAID. Look at things like ZFS and mdam - software RAID that requires direct access to the disks, not an abstracted hardware controller. Every cloud provider runs on software RAID - AWS, Azure, Google, etc... YouTube learned early on that disk I/O is much better when it knows there's 10 disks to talk to, rather than the 1 disk that a RAID controller would present.
Microsoft had to architect the Storage Pools to work in a way that's safe enough for consumers who don't do things like have UPS'es & are prone to accidently unplugging something.
I would be interested to see if adding 1/2 NVMe drives to the pool would improve the write performance, but there are limits built in to handle consumer level issues.
I am referring to software raid that runs on host computers, not dedicated external raid. And software raid can run on boot volumes - like Intel RST, which has a BIOS component (driver) in addition to OS driver.
Data Deduplication, thin provisioning and tiering aren't what we are talking about here. They are not low latency high performance systems.
ZFS is about the only mature local software storage system with redundancy that does perform well. It's write speeds are still slow compared to a dedicated external arrays.
-
ZFS is about the only mature local software storage system with redundancy that does perform well. It's write speeds are still slow compared to a dedicated external arrays.
Wrong. ZFS caches write transactions to RAM or to a SLOG (Separate Intent Log) for high end systems. So as long as these aren't full, your underlying RAID storage performance will not impact write performance. This is the reason why write performance is typically higher on ZFS systems than read performance (if the data is not in read cache). When a zfs system reports write completed, data may not have yet been copied to its final destination on disks. This aggressive caching masks the latency of the actual write operation.
Aggressive caching is also what hardware RAID cards perform behind the scene to avoid async write penalties to spinning rust arrays. This is the reason why such hardware might increase write performance, not offloaded RAID parity calculation. The latter is no longer an issue with a modern multicore CPU, as pointed by Joe above. A typical HW RAID card processor (such as a PPC 440 @ 800 MHz on Avago/LSI cards) is not that powerful compared to a modern CPU, which often has idle cores which could be put to use for software RAID.
Aggressive write caching can be done on a Windows client as well: just tick that box to turn off write-cache buffer flushing on the disk. Don't do this unless the PC is backed by an UPS, otherwise some transactions might get lost in case power loss occurs.
Cheers,
Fabien
-
Wrong. ZFS caches write transactions to RAM or to a SLOG (Separate Intent Log) for high end systems. So as long as these aren't full, your underlying RAID storage performance will not impact write performance. This is the reason why write performance is typically higher on ZFS systems than read performance (if the data is not in read cache). When a zfs system reports write completed, data may not have yet been copied to its final destination on disks. This aggressive caching masks the latency of the actual write operation.
Aggressive caching is also what hardware RAID cards perform behind the scene to avoid async write penalties to spinning rust arrays. This is the reason why such hardware might increase write performance, not offloaded RAID parity calculation. The latter is no longer an issue with a modern multicore CPU, as pointed by Joe above. A typical HW RAID card processor (such as a PPC 440 @ 800 MHz on Avago/LSI cards) is not that powerful compared to a modern CPU, which often has idle cores which could be put to use for software RAID.
Aggressive write caching can be done on a Windows client as well: just tick that box to turn off write-cache buffer flushing on the disk. Don't do this unless the PC is backed by an UPS, otherwise some transactions might get lost in case power loss occurs.
Cheers,
Fabien
Well my experience with ZFS has been with a large global file system consisting of hundreds of Linux file servers each running local files systems. For XFS + RAID6 hardware raid vs. ZFS RAID Z2, and the raid servers won in both performance and price/performance. Writing to RAM or to the ZIL/SLOG catches up in this use case (always full). And, as you say, for this performance the data has not yet been moved to its final location from RAM or from the ZIL/SLOG. This only benefits you if the SLOG faster storage and if I recollect, will be a detriment if not significantly so (having to make a second copy in case of larger files). But that may have been a bug - it's been a while.
I have no idea how well ZFS performs on Windows or Mac OS. I expect not as well as Linux.
Yes, I agree both hardware raid arrays and dedicated raid arrays benefit mostly from caching. That also helps in performing the parity calculation. Good raid-array controllers also use FPGA for faster processing.
But this is far astray of the topic which was local file systems and software raid (thanks to my loosely generalizing). I don't think anyone had in mind ZFS etc, but Windows pools (or MAC OS equivalent) and BIOS based software RAID, neither of which perform well with parity. And that performance has nothing to do with caching/buffering. I will admit I haven't looked at this in a while, but I expect it is still the case.
Cheers
-
With reference to my previous email my motherboard has a IHCR8/IHCR9 Raid Controller, configurable in the BIOS. This very much looks like a hardware controller to me, not software.
I don’t like software RAID, bus this is my opinion, not a fact ;)
Regards,
Jaap.
-
I just want to swing this back on topic.
Ryzen 3 - RAM options.
I am looking at 128 gb. (Probably the MSI creation mb but may go a Gigabyte board)
Having never done a build before, how do I determine which Ram kits I should use?
Can I buy 2 x 64 gb kits?
I saw the other posts above but they just had answers, not a process for determining.
Thanks everyone.
-
With reference to my previous email my motherboard has a IHCR8/IHCR9 Raid Controller, configurable in the BIOS. This very much looks like a hardware controller to me, not software.
I don’t like software RAID, bus this is my opinion, not a fact ;)
Regards,
Jaap.
It is Intel software raid with a BIOS support so you can boot from the RAID. Striping and parity are performed by software running on the host (device driver).
-
I just want to swing this back on topic.
Ryzen 3 - RAM options.
I am looking at 128 gb. (Probably the MSI creation mb but may go a Gigabyte board)
Having never done a build before, how do I determine which Ram kits I should use?
Can I buy 2 x 64 gb kits?
I saw the other posts above but they just had answers, not a process for determining.
Thanks everyone.
You can buy two kits but it is generally better to get 1 kit as the DIMM may be better matched.
Motherboard vendors generally have QVL lists (qualified vendor lists) which list RAM kits they have verified to be compatible with the motherboard. If a kit is not on the list, that doesn't mean its not compatible, but just that the motherboard vendor hasn't verified it (and probably won't support it if it doesn't work). You can find the QVL memory list on the motherboard's support page as a start.
You can work the opposite way with some memory vendors which have their own QVL lists for motherboards that they support their RAM on. g,skill does this for instance. The memory vendors I would recommend are g.skill, Corsair, Kingston, and Crucial.
-
I just want to swing this back on topic.
Ryzen 3 - RAM options.
I am looking at 128 gb. (Probably the MSI creation mb but may go a Gigabyte board)
Having never done a build before, how do I determine which Ram kits I should use?
Can I buy 2 x 64 gb kits?
I saw the other posts above but they just had answers, not a process for determining.
Thanks everyone.
The process is there even if it might not be apparent.
I did a lot of reading these days, motherboards and memory qvl included. The qvl isn't that helpful, most are not up to date.
The conclusions that I have so far:
- on the better motherboards chances are you will be able to get 128 GB at 2666 MHz. Corsair LPX/RGB Pro or GSkill Trident Z Neo or the Ripjaws V are the best bets. The timings are slower on the more compatible ones.
- if you want to get the faster RAM things get trickier. On ASRock I saw a response from them that they will work at 2933 if you want 128 GB and some reports about 3000.
I've only seen one build with 128GB that used a slower 3200 MHz 18-20-20-38 I think, likely a Corsair LPX (was called Gigabyte LPX which doesn't exist).
PS. There are no 64 GB sticks, only 32 GB
-
I forgot to add, if you go with "only" 64 GB RAM I've seen numerous builds using 3600 MHz RAM on mainstream to high end X570 boards.
As for the 2x 64 GB, in which the 64 is 32x2, the option is better if you are happy with 64 GB. If 128 GB won't then you can send back just one of the kits. The problem, as said above and which makes me a little reluctant to go this route, is that the RAM producers use various sources (I think there are 3 or 4) even for the same type of memory, so 2 kits might be different enough to cause problems.
-
I forgot to add, if you go with "only" 64 GB RAM I've seen numerous builds using 3600 MHz RAM on mainstream to high end X570 boards.
As for the 2x 64 GB, in which the 64 is 32x2, the option is better if you are happy with 64 GB. If 128 GB won't then you can send back just one of the kits. The problem, as said above and which makes me a little reluctant to go this route, is that the RAM producers use various sources (I think there are 3 or 4) even for the same type of memory, so 2 kits might be different enough to cause problems.
^This is good advice. 3600MHz is a sweet spot in performance for DDR4. 3200 is good too. I am running 64 GB 3600 CL16 g.skill (on Intel). I would say for the safest bet, first look at g.skill and their QVL list of motherboards, which is kept up-to-date. If you then have problems, you can then go to g.skill support.
-
Thanks Armand.
I wasn't meaning 64gb sticks...2x 64gb KITS so 4 32gb modules.
You've answered my question though so I appreciate that!
-
^This is good advice. 3600MHz is a sweet spot in performance for DDR4. 3200 is good too. I am running 64 GB 3600 CL16 g.skill (on Intel). I would say for the safest bet, first look at g.skill and their QVL list of motherboards, which is kept up-to-date. If you then have problems, you can then go to g.skill support.
More than 3600 MHz would likely lower performance on Ryzen 3000 CPUs. This is due to the fact that the internal bus between memory I/O chip and processor chiplets caps at 3600 MHz and has to operate at half the memory speed when the latter is clocked above 3600 MHz. Anything above 3200 MHz is considered overclocking by AMD and is not guaranteed to work.
To put things into perspective: all memory sticks operating at speeds above the JEDEC standard (currently 2666 MHz for DDR4 UDIMM, with 2933 coming next) are in fact factory overclocked and are mostly operated in systems using two memory sticks (max. 64 GB memory, more commonly gamers with 32 GB memory). Trying to operate them at their advertised overclocked speed with four populated slots to reach 128 GB memory is tough since they were not meant to be used in this configuration. That explains why almost no such OC memory sticks are advertised as compatible with 4 slots use in the QVL of motherboad manufacturers.
I would be curious to see if non OC memory at speeds higher than 2666 MHz will be made available in the near future or if manufacturers will go straight to DDR5. The former would be ideal for people wanting 128 GB in Ryzen 3000 systems, since the performance of this architecture scales well with higher memory speeds.
Cheers,
Fabien
-
There has been only one company I am aware of that has announced native DDR4-3200 - last summer I think. They are not a major player IIRC and it has yet to materialize, so I would expect the major memory companies to move on to DDR5. In any case I expect the minor differences in RAM frequency don't have much effect on Photoshop performance. It is mostly processor speed and instructions per cycle and to a lesser extent, the number of cores.
-
... In any case I expect the minor differences in RAM frequency don't have much effect on Photoshop performance. It is mostly processor speed and instructions per cycle and to a lesser extent, the number of cores.
I agree about small effect of the speed of ram, but see a trend that software uare getting to use more cores , even Photoshop, not to forget plugins like topaz denoise, helicon focus and panorama programs like ptgui that use all cores...
Lightroom9 uses all cores - it also uses the hyperthreading ; on my 10 core going to 1800% of use.
-
I agree about small effect of the speed of ram, but see a trend that software uare getting to use more cores , even Photoshop, not to forget plugins like topaz denoise, helicon focus and panorama programs like ptgui that use all cores...
Lightroom9 uses all cores - it also uses the hyperthreading ; on my 10 core going to 1800% of use.
If you look at the PugetBench results for different core count Intel or AMD processors, the individual scores aren't really much different from each (e.g. AMD 8 vs 12 vs 16 core https://www.pugetsystems.com/pic_disp.php?id=58263). Maybe it is just the operations this specific benchmark does are mostly single threaded.
-
More than 3600 MHz would likely lower performance on Ryzen 3000 CPUs. This is due to the fact that the internal bus between memory I/O chip and processor chiplets caps at 3600 MHz and has to operate at half the memory speed when the latter is clocked above 3600 MHz. Anything above 3200 MHz is considered overclocking by AMD and is not guaranteed to work.
To put things into perspective: all memory sticks operating at speeds above the JEDEC standard (currently 2666 MHz for DDR4 UDIMM, with 2933 coming next) are in fact factory overclocked and are mostly operated in systems using two memory sticks (max. 64 GB memory, more commonly gamers with 32 GB memory). Trying to operate them at their advertised overclocked speed with four populated slots to reach 128 GB memory is tough since they were not meant to be used in this configuration. That explains why almost no such OC memory sticks are advertised as compatible with 4 slots use in the QVL of motherboad manufacturers.
I would be curious to see if non OC memory at speeds higher than 2666 MHz will be made available in the near future or if manufacturers will go straight to DDR5. The former would be ideal for people wanting 128 GB in Ryzen 3000 systems, since the performance of this architecture scales well with higher memory speeds.
Cheers,
Fabien
Puget only uses 2666 for their Lightroom system.
My target is 128GB RAM @ 3200 MHz and the GSkill QVL for the 128GB kit says it's compatible with most higher end boards (both Trident Neo Z and Ripjaws V). Then I still be happy enough with 2933.
If the above won't work then I'll settle for 64GB @ 3600 MHz.
-
I can’t comment on 128GB, as my second ryzen System just uses 64GB, but it uses 4x16GB running at 3200 on a x570 Board.
In a few weeks I can tell how my 256GB are running. They are also 8x32Gb specs at 3200.
-
I finally decided to bite the bullet and order the stuff: https://pcpartpicker.com/list/MyDqwh
Because I decided to give my current computer to the kids for work/games I added more stuff than I needed.
So the changes that reflect my actual configuration will be:
- I'll keep my current video card, RTX 2060 (instead of the 1660 Super from the list)
- I'll keep my current dual monitor setup (instead of the gaming Dell monitor)
I should have all the parts by early next week and I will know in a day or two if everything goes well. My only concern would be the 128 GB RAM but I'm cautiously optimistic.
-
Looks great though wondering why the 120mm case fans instead of 140mm - the 140mm will move more air with less noise.
-
Looks great though wondering why the 120mm case fans instead of 140mm - the 140mm will move more air with less noise.
The case already has some fans and I'm not sure how much room I have, that's why I ordered o combo of 120 and 140. This is the easiest thing to change though.
-
The case already has some fans and I'm not sure how much room I have, that's why I ordered o combo of 120 and 140. This is the easiest thing to change though.
Ah, thought you were replacing the Fractal fans that came with the case (the Noctua would be a worthy replacement).
Looks great and a pretty decent price for a system with 128 GB.
I am running Gigabyte Master z390, g.skill (64GB), Noctua NH-D15, Noctua fans, Fractal Design R4, RTX 2070 Super. Similar except Intel and 1 Gen behind you. This build should perform well.
-
Ah, thought you were replacing the Fractal fans that came with the case (the Noctua would be a worthy replacement).
Looks great and a pretty decent price for a system with 128 GB.
I am running Gigabyte Master z390, g.skill (64GB), Noctua NH-D15, Noctua fans, Fractal Design R4, RTX 2070 Super. Similar except Intel and 1 Gen behind you. This build should perform well.
Will see. I have the same case as you, its own Fractal Design fans were good enough for me; when not trying a mild overclock the entire thing is silent (I use only SSDs for usual work).
-
I finally got time to install the basics. I already posted the last configuration: https://pcpartpicker.com/b/NFWD4D
As I built my last computer more than 6 years ago I got a little rusty. Took me longer than I recall to setup the case and get everything in order.
As an example of rustiness while I was trying to figure out how the Ryzen went into the socket (not that obvious, at least less than the Intels I used in the past) I managed to drop the processor which resulted in multiple bent pins. When I tried to fix the first with some pliers I overbent it in the opposite side and I thought I broke it. Lesson learnt, I used only a blade and after few tense minutes and ~10 straightened pins I managed to get it back in shape and it's working.
Otherwise it all worked fine from the beginning, no issues. I'm happy to report the the Gigabyte X570 Aorus Master supports the 128 GB of GSkill Trident Z Neo; by default it sees it at 2667 MHz but enabling the XMP profile is one click away and worked without issues and now it's 3200MHz, I don't see the need to overclock it beyond this
This time I got the PugetBench for LR to run, I attached the results and a couple of cellphone shots of the interior. I'll probably open it again at some point, I'm not convinced I got the thermal paste properly and after I properly read the manual and look through the motherboard cables I see that there is a noise monitor also.
What you don't see in those pictures are 5 SSDs and 1 HDDs, all on the other side of the case. One thing that drove me nuts for 30 min or so was a sound as if the Windows kept detecting a drive, I thought one has a loose cable. It was just the external card reader.
-
Congratulations !
Seems like a great speedy machine for a very decent price... well done.
am sure it will be as fast as a new macpro costing $$$
-
Congratulations !
Seems like a great speedy machine for a very decent price... well done.
am sure it will be as fast as a new macpro costing $$$
Thank you.
I run again the benchmark for Photoshop and it's less impressive but still quite significantly better than before.
I do have some questions regarding its validity because on the new computer the graphic card score is much higher despite being the same card.
-
Thank you.
I run again the benchmark for Photoshop and it's less impressive but still quite significantly better than before.
I do have some questions regarding its validity because on the new computer the graphic card score is much higher despite being the same card.
First, thanks for posting these. I was thinking about getting a 3900x, but my current scores are about the same so that won't be a justification.
I found the final scores are very sensitive to the individual ones. A small, insignificant IMO, individual score results in a larger overall score. I do not think you will notice tens to 50's difference in the overall score, but will in the hundreds.
What really makes a significant difference is processor speed and instructions per clock cycle, less so number of cores. So when you jump generations of processor you will see significant differences.
As for GPU scores, you are running a different architecture (AMD) which will have a different PCI implementation, different graphics drivers, different photoshop graphics interface implementation to some level, all of which can make a difference in the GPU score.
I look to the individual test for differences. You know what they are doing and the result (seconds to run), instead of dome concocted overall score. I think these numbers are valid.
I should add too that there are a number of patches in the OS for Intel vulnerabilities that affect performance. These are not present for the AMD processor and may account for the GPU score difference.
-
Looks great.
I'm about ready to pull the pin on this - I need a project while at home...
Think I'm going to go the Define 7 and put the MSI Creation in it. It'll be a squeeze and limit harddrive installs, but lot's of room for M.2 drives and 2.5SSDs
-
Looks great.
I'm about ready to pull the pin on this - I need a project while at home...
Think I'm going to go the Define 7 and put the MSI Creation in it. It'll be a squeeze and limit harddrive installs, but lot's of room for M.2 drives and 2.5SSDs
It will definitely be tight in there. If you have the room there is also the Define 7 XL.
-
It will definitely be tight in there. If you have the room there is also the Define 7 XL.
That E-ATX board might not fit in the Define 7 case.
EDIT: NM, looks like it will but will be tight.
-
Yeah...found these two builds with the 6...tight yes, but not something that isn't possible.
https://pcpartpicker.com/builds/by_part/KBmFf7#e=1875
Do you think I'm crazy?
That E-ATX board might not fit in the Define 7 case.
EDIT: NM, looks like it will but will be tight.
-
Yeah...found these two builds with the 6...tight yes, but not something that isn't possible.
https://pcpartpicker.com/builds/by_part/KBmFf7#e=1875
Do you think I'm crazy?
ME? I would do what Armand suggested and get an XL define case. As far as the MSI board, with the quad channel memory it should scream, but I have no feel on the quality of the MSI board.
-
Yeah...found these two builds with the 6...tight yes, but not something that isn't possible.
https://pcpartpicker.com/builds/by_part/KBmFf7#e=1875
Do you think I'm crazy?
I just opened mine to replace the thermal past and rearrange some fan cables before you posted this and yes, it can be done. I forgot the E-ATX is wider and not taller. That means however that all your cables will have to come from above or below, the side ports will be covered.
Even with mine, accessing some of the small fan connections on the board has been a headache, I would advise having some extra connectors and plug them before you get the motherboard in the case.
With the proper fan connections I did decrease a little the noise but unfortunately there is no comparison between this and my older Intel build, Intel was practically silent while this is easily audible, particularly when you do heavier lifting.
-
Regarding PC cases, what do you guys think of the Dune Pro case, the look-alike from the fruit company?
Regards,
Jaap.
-
Regarding PC cases, what do you guys think of the Dune Pro case, the look-alike from the fruit company?
Regards,
Jaap.
Didn't hear about it until now (not that means much). Looks ok but unless you want to build a Hackintosh, why this one?
A case I contemplated before staying with Fractal Design was this one from Corsair: https://www.corsair.com/us/en/Categories/Products/Cases/Carbide-Series%E2%84%A2-Air-540-High-Airflow-ATX-Cube-Case/p/CC-9011030-WW
I liked how you have 2 distinct compartments, well separated: power source and storage in one, the high heat producing stuff on the other side. You can optimize the airflow per compartment.
-
I finally got time to install the basics. I already posted the last configuration: https://pcpartpicker.com/b/NFWD4D
So, after two months, are you still happy with this build? Real world, Photoshop and Lightroom?
-
So, after two months, are you still happy with this build? Real world, Photoshop and Lightroom?
Overall it meets expectations.
It's definitely faster than my old one but that was not in question. The difference is bigger on import and previews, and Photoshop work (I don't go often in Photoshop though).
Would I notice the difference compared to a i9? Probably not enough.
It is a little noisier than I would like, although it's not that bad per se, but my old one was practically silent.
-
Overall it meets expectations.
Thanks, Armand.
-
Although i said that 32Gb ram is enough for most work i ran into some limitations with large files and 80 layer stacks so went up to 64GBRAM.
Ram is very cheap at the moment ( i paid 140€ for 2x 16GB) so it is not something to worry about.
i ran again the Photoshop Puget 0.8 test again ( the 09 did not work) and came to better results.
I noticed more cache helps, larger tiles helps and historie set to 1 helps if your system is limited in RAM.
the same system
originaly i had a score of 853 with 32GB ram
optimum i scored 923 with the same 32GB Ram system but tuned as above and Mojave instead of High Sierra.
with 64 gig Ram and 100 historie ( or 1 history) i got to a score of 973. ( enough ram for all the historie saved)
the 'weak' part of the hardware is now the GPU, but it is good enough for what i do.
BTW what the test does not include and adds significantly to the speed of Photoshop when using many large files is the storage solution and to disable compression in the preferences for tiff,psd and PSB.
-
That's not a bad GPU score. It scores slightly better than my Nvidia RTX 2070 super. AMD has better OpenCL performance, more than I expected.
I was playing around to see if I could increase performance. I switch my Nvidia and photoshop to use 8 bit instead of 10 bit display. I expected it might increase the puget score a little bit. It actually decreased it a significant amount. Hmmm.
Anyhow that is a solid system. I score 1018 on my 8086k with hyperthreading disabled, but that is overclocked to 5.2 GHz! A big sinificant difference I see between them is the Photomerge.
Will be an AMD for me next round I think.