Luminous Landscape Forum
Equipment & Techniques => Computers & Peripherals => Topic started by: Christopher on November 03, 2009, 09:48:08 pm
-
Ok soon it will be time for me to build a new workstation at home. My current one is three years old and while it was upgraded throughout the past years, it is dated and has problems handling larger files.
First the budget, which is really flexible. I don't want to name a exact amount, because I don't have one. I am willing to spend 3000 or 5 times the amount, AS LONG as it makes SENSE.
By that I mean, i will not Spend let's say 5k on a workstation which is 15% faster when a 2500 one, but 2 years later a new 2500 one is 50% faster, because everything moved on.
So what I need it for:
- Lightroom
- Capture One
- Photoshop
- PtGui (less important, because I will use my current one for rendering panos)
Now I have a few questions let's start.
CPU:
Xeon VS Core i7
Here are my thoughts:
The choice is between a dual Xeon System running two 2,8Ghz Quad cores which can manage 16 threads.
OR
One single Core i7 OCd at around 3,8Ghz which can manage 8 threads.
So I know PTGui would benefit from 8 real cores, but will the rest ? Will it REALLY make a difference, or will the real speed determined by Memory and HDs ?
RAM
- I think 24GB should be my choice. More is only possible with a Xeon System and would get realllllly expensive. Any thoughts ?
HD Setup
This is probably the most difficult part.
Here I have quite some questions. Let's say you have Set Up 1, which is 1TB in size and has a read and write speed of 500MB/s. Let's say Set Up 2 has 2 x 500GB drives with a Speed of 300MB/s
Now what is actually faster ? Splitting everything up on two disks or one faster one ?
By that I mean putting System, Programs, working files and Scratch disk on Set Up 1
OR
Putting System, programs working files on one of the two drives of Set Up 2 and the Scratch disk on the second one
I know the second option would be the faster one if all drives had the same speed, but that is just not the case anymore.
I am thinking about using a few SSDs as System disks, I'm just not sure if it is smarter to split them up or do one large fast RAID 0 array. (I'm not worried about dater loss, there is never just one copy of an important file on that array.
As Image storage I was thinking about using RAID 5 with 4-6 HDs. So that in the End I can get around 4-8TB of storage. (I can't say how much I need until I am back home and can calculate what is really needed)
Any other thoughts or suggestions ?
-
You don't mention Mac or PC?
In PC land, I'd recommend an OC'd i7 CPU (shoot for 4Ghz), SSD (or pair of RAID 0 SSD's) for the OS, 2-4 7200RPM HD's in RAID 0 for scratch, plus several big TB HD's for storage. 24gb of RAM is overkill on a 64bit OS, since it's all directly accessible. 12-16GB should be fine. The storage hard drives don't need to be fast, but having SSD for the OS will improve responsiveness and having high data throughput (RAID 0) for scratch disk will improve speed on big file handling. This system could easily be done for sub $2k.
That would be a smoking fast machine. Dual CPU's doesn't really get you anything for most of Lightroom/CS4/etc compared to a fast quad core.
-
You don't mention Mac or PC?
In PC land, I'd recommend an OC'd i7 CPU (shoot for 4Ghz), SSD (or pair of RAID 0 SSD's) for the OS, 2-4 7200RPM HD's in RAID 0 for scratch, plus several big TB HD's for storage. 24gb of RAM is overkill on a 64bit OS, since it's all directly accessible. 12-16GB should be fine. The storage hard drives don't need to be fast, but having SSD for the OS will improve responsiveness and having high data throughput (RAID 0) for scratch disk will improve speed on big file handling. This system could easily be done for sub $2k.
That would be a smoking fast machine. Dual CPU's doesn't really get you anything for most of Lightroom/CS4/etc compared to a fast quad core.
First of all thanks. I'm talking about a PC, not seeing any sense in getting a mac. (no benefits)
I want to add something about file size, I'm when i sad large I meant often something around 4-10GB, sometimes even above 15GB, that was the main reason why I wanted to aim for a full 24GB of RAM.
One of the main question is still, does it to make sense to split "System, programs, current image files and scratch disk"? I mean if you have two SSDs in RAID 0 for the OS, and it gives you let's say 300Mb/s and 4 normal HDs in RAID 0 which would give you something like 300-400Mb/s, is that really FASTER than to put everything on one large RAID 0 out of 4 SSDs, giving you a speed of 600Mb/s ?
(All speed numbers are not correct, just picked them to illustrate my point)
-
I'm talking about a PC, not seeing any sense in getting a mac. (no benefits)
You need to watch more 'Get a Mac Adds (http://www.apple.com/getamac/ads/)'
On topic - As the last thing the internet needs is another mac v. pc thread - I would go with 32 gig of RAM. I am running the same applications as you are Christopher on a Mac Pro 8 Core Xeon and I noticed a performance increase that was substantial from 24 to 32 gig when stitching multiple large DSLR 20+ MPX files. Your P65+ files with love you for the extra RAM.
-
You need to watch more 'Get a Mac Adds (http://www.apple.com/getamac/ads/)'
On topic - As the last thing the internet needs is another mac v. pc thread - I would go with 36 gig of RAM. I am running the same applications as you are Christopher on a Mac Pro 8 Core Xeon and I noticed a performance increase that was substantial from 24 to 36 gig when stitching multiple large DSLR 20+ MPX files. Your P65+ files with love you for the extra RAM.
When stitching yes, you would even notice more with 64GB or 128GB of RAM. PTGui for example requires disk space for rendering anything from 20GB to 100GB. So with more ram it is a lot faster. However once the pano is done I think the difference won't be that big anymore. (working with large files in lightrrom or PS) Here I think a fast storage solution will speed up the workflow much more.
I mean if I have a 10GB pano for example, than loading that into PS is not really paced by the RAM but on the reading speed of your drives. Here 16, 24, 32 or 64 GB won't make a difference, but a a simple RAID 5 with only 200MB/s against a SSD Raid 0 with 600-800MB/s would make a huge difference.
The thing that really pisses me off the most that there are no real benches and reviews for LR, PS, C1 and so on. I mean in the End the question is always, what do you get for your money. Will a 8 Core system with 36GB of RAM, really be THAT much faster than a Quad core sytsem with 16Gb of RAM ? I don't have the answer to that, but I know the price difference is around certainly over 4000$, if not more.
EDIT: I mean I would love to test a few systems out there, and check how they perform for us photographers, but well I don't have these options. I mean the only thing one can sometimes find is PS CS4 and a test with a 70MB tiff file, there they stop the time for 6 filters to run. The one thing it tells us that CS4 is still very poor in handling multi cores. The biggest difference in time is clock speed and not cores. a lot of Dual Core Systems outperform a quad or 8 core system. However, that sadly tells us nothing about speed for opening and saving files and other stuff.
-
Chrsitopher,
You know as much about this as I do. I'll describe what I did - with improvements for my "slips" - FWIW. (I do have a degree in photography, and one in computer science.)
My target is usually about 85% on a cost/effectiveness basis. Beyond that it is just cheaper to wait 6 months and buy a better base machine for $600 instead of $2,500 today, as you mention. Plus, I over obsess and overengineer when designing, but on a day-today basis, that extra 5% to 10% is totally irrelevant.
So, I built a core network of 4 machines with 2 gigabit lans. One for "local" data - photoshop macgine, server, nas - and one for "external" data - wirelress lan, laptops, etc.
I would suggest 2 machines as a start:
1) A Photoshop/Lightroom **only** box that may not even have a daily ionternet connection! Probably an i7 920, and
2) An application/data/storage server for **everything** else - any core applications, data storage, offload processing, printing, backups, etc. Probably a Dell Poweredge quad Xeon 300 or 500 series, no highg-end video required. With dual gigabit network cards and a hardware Raid controller (a 3rd box would be an actual NAS, in addition to this server.) Base cost $500 + RAID card + RAM + disks.
In the photo machine, I personally would put 3-4 320GB WD 7200 drives in Raid 0 or similar for OS. Apps, and Scratch disk (2 or more separate logical drives of course. As many as you can squeeze. Cloned for instant re-install) I am ambivalent on SSD - a bit too pricy yet. (I got the WD Caviar Blue for $30 each. They area great size for OS, aps.)
Then a couple of 1 to 2TB HDD's - one for RAW files (local working copy) and one for WORKING files (lightroom database, etc.) Backed up to the Ap/Data server often over gigabit.
For RAM, I'd do as much as I could with 2GB chips. The 4GB chips are just too effing expensive! Might as well buy an extra box instead for $600!!!
My 3rd box is the NAS with 6TB and Windows Home Server. It is off 90% of the time.
My 4th box is my "everythinhg else" machine -a dual core laptop with 4GB for watching TV, music listening, e-mail, Outlook, etc. I keep the server as a pure server (Running Hyper-V: was ESXi from VMWare.)
Everything else - bluray players, tv's, 3 other laptops, etc run off of the "external" lan to avoid traffic. There is actually a bridge to a 3rd router, which is the internet gateway for all.
Just thoughts. Have fun! (Sorry for typos, hand pain ....)
Cheers,
Michael
-
Wow, those are large files! With that size of file I would agree that 24GB of RAM is totally reasonable.
Yes, I think there is a significant benefit for splitting the Scratch disk off from the rest of the other functions. The scratch disk would definitely become involved in working on 5-15GB file sizes, even with a lot of RAM.
From what I've seen, using an SSD for scratch doesn't really gain any benefits over a conventional hard drive. The sustained write speeds on SSD's aren't any better than a good 7200RPM drive, so there's no need to spend the extra money for use in a scratch disk. Take 4 1TB or 640GB drives such as the WD RE3 and put them in a RAID 0, partition off the first stripe for scratch. So many of the operations on huge file are totally dependent on the data throughput rate for a scratch disk. If you have any other operations trying to access the scratch disk when it's being used, that would slow down the scratch throughput. That's why people usually create a dedicated scratch volume.
The data transfer speed of the image storage volume would matter when loading up a file, but I don't think that the storage volume gets accessed much once the file is completely loaded. I think you might be able to get away with using the remainding partition space of the RAID 0 array as short term image storage for working files (with some sort of backup arrangement for longer term storage). The fast RAID 0 array would speed up opening up a large working file, and also would serve as fast scratch.
SSD is ideal for the OS/Program drive, because of its fast read and seek times. Faster overall operation, faster loading of programs, plus no interference of the OS with scratch disk usage.
I agree with your comments about multi-core usage. LR 64 bit does use a quad core fairly well, but clock speed is more important overall. I don't think that an 8 core dual CPU machine would help much for LR/CS4/C1 usage. Get a fast OC'd i7, and focus the rest of the build on RAM and hard drives with fast data transfer rates.
-
Well, this is the big problem, as far as I can see: (all Newegg prices)
* CORSAIR XMS3 6GB (3 x 2GB) 240-Pin DDR3 (PC3 12800) $170 ($30 GB MIDPOINT PRICE)
* Kingston HyperX 12GB (3 x 4GB) 240-Pin DDR3 (PC3 12800) $1,200 ($100 GB )
This is a bit better:
* G.SKILL Ripjaws Series 16GB (4 x 4GB) 240-Pin DDR3 (PC3 8500) $800 ($50 GB)
If you really need more than 12 GB, then:
* SUPERMICRO MBD-X8DAE-O Dual Intel Xeon 5500 sequence Server Motherboard - Retail $430
Number of DDR3 Slots: 12 × 240pin
DDR3: Standard DDR3 1333
Maximum Memory Supported: Supports up to 96 GB 1333 / 1066 / 800MHz DDR3 ECC Registered memory
Supports up to 24 GB 1333 / 1066 / 800MHz DDR3 ECC Unbuffered memory
-
Well, this is the big problem, as far as I can see: (all Newegg prices)
* CORSAIR XMS3 6GB (3 x 2GB) 240-Pin DDR3 (PC3 12800) $170 ($30 GB MIDPOINT PRICE)
* Kingston HyperX 12GB (3 x 4GB) 240-Pin DDR3 (PC3 12800) $1,200 ($100 GB )
This is a bit better:
* G.SKILL Ripjaws Series 16GB (4 x 4GB) 240-Pin DDR3 (PC3 8500) $800 ($50 GB)
If you really need more than 12 GB, then:
* SUPERMICRO MBD-X8DAE-O Dual Intel Xeon 5500 sequence Server Motherboard - Retail $430
Number of DDR3 Slots: 12 × 240pin
DDR3: Standard DDR3 1333
Maximum Memory Supported: Supports up to 96 GB 1333 / 1066 / 800MHz DDR3 ECC Registered memory
Supports up to 24 GB 1333 / 1066 / 800MHz DDR3 ECC Unbuffered memory
That's one reason I'm still thinking on it. 4GB modules are really expensive compared to the rest. On solution would be to get 16GB now and wait for the last 8 until prices drop in half a year or so.
-
wait for the last 8 until prices drop in half a year or so.
That was my call. It is just that incremental price of 3x the cost for teh same amount of RAM that I choke on ..... I can take a sip of coffee for the next 6 months to save $1K - plus the Xeon 5400 - plus the MB, etc. ....
But - my files are rarely above 1GB. So I went i7 920.
Maybe the SSD Raid 0? Phil pointed me to a $90 SSD recently I think.
"The network is the computer" was where I turned for second tier (stroage, backup, etc.) functions. Too bad we can't parcel all this out in chunks yet for core functions!
Best,
Michael
-
Ok soon it will be time for me to build a new workstation at home.
...
As Image storage I was thinking about using RAID 5 with 4-6 HDs. So that in the End I can get around 4-8TB of storage. (I can't say how much I need until I am back home and can calculate what is really needed)
Any other thoughts or suggestions ?
RAID 5 tolerates single drive failures. Rebuilding the failed drive when the array is built using terabyte drives requires a LONG time, and during the rebuild a second failure looses the entire array. Rebuilding a RAID 5 array after the failure of a 1TB drive is likely to require at least 12 hours. (There's experience behind that statement!) Other RAID levels, e.g., RAID 6 and RAID 10, tolerate two simultaneous drive failures, meaning the array isn't vulnerable while rebuilding from a single drive failure.
Also note that RAID is for improving availability, not as a substitute for backups!
Reliable storage is clearly subject to "Fast, cheap, good--pick two." RAID controllers, desirable for RAID levels using striped parity, cost more than a simple JBOD controller, and redundancy requires additional storage, ranging from less than 10% to 100%.
Mac Performance Guide (http://macperformanceguide.com/index.html)offers some suggestions for RAID (http://macperformanceguide.com/Storage-RAID.html) use, as well as some interesting benchmarks showing Photoshop performance (http://macperformanceguide.com/OptimizingPhotoshop-TestResults.html) with various memory sizes. While focused on Apple systems, I believe the benchmark results can be relevant to Windows boxes.
Adaptec offers a white paper (http://www.adaptec.com/NR/rdonlyres/874D145E-F64F-4804-9E27-037BC5A9DCE0/0/3994_RAID_WhichOne_v112.pdf)describing the common RAID levels.
Bob
-
I agree with your comments about multi-core usage. LR 64 bit does use a quad core fairly well, but clock speed is more important overall. I don't think that an 8 core dual CPU machine would help much for LR/CS4/C1 usage.
C1 Pro makes a very good usage of multi-core at least on OSX, my 8 core Mac Pro is near 100% (all 8 cores maxed out) most of the time during C1 conversions, same thing with Autopano pro except for the smartblend part (but that will be fixed soon).
Since I typically do both at the same time, I would personally gain a lot from having 16 cores instead of 8.
I also did some 32GB vs 16GB RAM comparisons for my typical heavy paralleled tasks (C1 Pro conversion, Autopano pro/PTgui pano computation and PS work) and found up to 40% speed increase for some of the apps (no gain for PS).
PS is years behind but many other apps have move ahead.
Cheers,
Bernard
-
* Kingston HyperX 12GB (3 x 4GB) 240-Pin DDR3 (PC3 12800) $1,200 ($100 GB )
You can get 32GB of 1066 with special Mac Pro specs for 1250 US$...
http://eshop.macsales.com/item/Other%20Wor...g/85MP3S4M32GK/ (http://eshop.macsales.com/item/Other%20World%20Computing/85MP3S4M32GK/)
Cheers,
Bernard
-
Thanks Bernard! Interesting!
I did find one nasty comment at macrumors on that memory. I'll post the link without comment - no real input/knowledge on my part. About 3/4 of the way down:
http://forums.macrumors.com/archive/index.php/t-697621.html (http://forums.macrumors.com/archive/index.php/t-697621.html)
Now I want to buy a box just to shove that RAM in it. Is that ass-backward or what?
Cheers,
Michael
-
Just did some very rough calculations on my part, to get a better feeling of the actually money it would cost. Here are my results, for Motherboard, CPU and RAM nothing else. Prices in EUR
Core i7 3,0 Ghz
X58 board
12Gb RAM
-----------
900EUR
Core i7 3,0 Ghz
X58 board
16Gb RAM
-----------
1300EUR
Core i7 3,0 Ghz
X58 board
24Gb RAM
-----------
1600EUR
1 x Intel Xeon 2.7Ghz
ASUS board
24Gb RAM
------------
1750EUR
1 x Intel Xeon 2.7Ghz
ASUS board
32Gb RAM
------------
2000EUR
2 x Intel Xeon 2.7Ghz
ASUS board
24Gb RAM
------------
2500EUR
2 x Intel Xeon 2.7Ghz
ASUS board
36Gb RAM
------------
2700EUR
The rest would be more or less the same for both systems. HDs, SSDs graphic card and so on.
-
Thanks Christopher, nice summary!
Pretty linear. Put it up on the wall and throw a dart?
Too bad you can't order it all up, test, then send back the rejects. Those files of yours are really pushing the RAM hard.
Maybe Lightroom 3 or PS 5 or Capture One 5 will ease the requirements a bit? At least they may make more use of the multiple cores? Have not looked yet in detail. (OK, just re-tread Bernard's excellent comments on this. Thx! m.)
-
Thanks Christopher, nice summary!
Pretty linear. Put it up on the wall and throw a dart?
Too bad you can't order it all up, test, then send back the rejects. Those files of yours are really pushing the RAM hard.
Maybe Lightroom 3 or PS 5 or Capture One 5 will ease the requirements a bit? At least they may make more use of the multiple cores? Have not looked yet in detail. (OK, just re-tread Bernard's excellent comments on this. Thx! m.)
Well Yes and it really starts one thinking. I still haven't decided. The main aspect why I am so unsure is if the Xeon makes sense. I have a lor of experience with OCing desktop CPUs and I know what I can get out of the i7. I built two machines for some friends, one a gaming guy the other doing a lot of video work. Both have a i7 running at 3,9Ghz, without any problems. (Not hotter than with normal or anything else)
This makes one really think, especially NOW that I know that CS4 still uses only ONE CPU for saving and opening files. ( Which by the way ADOBE WTF ? ) I'm not even sure if they are gonna change it in CS5. Next point being I have a separate computer for just rendering panoramics with PTGUI.
I'm still trying to fine people, who have access to many different computers, main problem right now being most don't have a dual Xeon with 32 Gb sitting around ;-)
Still i think I narrowed down one topic. Right now I am thinking about the following drive setup:
System, programs and workfiles on:
3 x SSDs in RAID 0 (probably 256Gb each)
4 x HDs with XXXGb as Scratch. I would use 40 or 80Gb, problem being I can't find new fast HDs in the sub 500GB category. So perhaps it will become a 2TB scratch ^^
6 x 2TB disks running in RAID 5 or 10, still trying do decide. (I know that Raid 5 is faster and raid 10 more secure, any actual info on how big the speed difference is ?)
1 x 500Gb disk to store all the crap, like installers and stuff.
I still would prefer SSDs as scratch, however the prices are still high, especially the there is ONLY the Intel E which has high write speed with low GB count. Most SSDs offer only 130 write speed till they hit 128 or 256Gb and than jump up to 170.
-
4 x HDs with XXXGb as Scratch. I would use 40 or 80Gb, problem being I can't find new fast HDs in the sub 500GB category. So perhaps it will become a 2TB scratch ^^
Raptors/Velociraptors are still your best bet IMHO.
I still would prefer SSDs as scratch, however the prices are still high, especially the there is ONLY the Intel E which has high write speed with low GB count. Most SSDs offer only 130 write speed till they hit 128 or 256Gb and than jump up to 170.
You might want to check out OCZ Vertex Turbo series, pretty close to Intel E performance I hear. Not cheap, but cheaper.
Besides, 3x60GB in Raid 0 will result in one 180 GB disk that is large enough for most applications. Even my largest panos didn't generate a scratch file larger than 100GB if I recall correctly. That would you less than 1000 US$:
http://www.newegg.com/Product/Product.aspx...N82E16820227469 (http://www.newegg.com/Product/Product.aspx?Item=N82E16820227469)
You could also get 3x Intel E 32GB for about the same price: http://www.newegg.com/Product/Product.aspx...N82E16820167013 (http://www.newegg.com/Product/Product.aspx?Item=N82E16820167013)
That would still be enough for many cases.
Remember that system sizing should not be done with a heavy emphasis on peak usage, that would result in specs way overkill for 90% of your usage.
Cheers,
Bernard
-
Edit: I know I only would need a scratch around 100Gb, however there are no smaller HDs. Even Raptors/Velociraptors don't make so much sense. Their read and write speed is not faster than current 7200 HDs.
Yes that is a option, and I would do it, IF all sized SSDs had the same performance. However it looks like this:
(all average values)
OCZ Vertex Turbo 120GB = 230 Read and 220 Write = 550US
OCZ Vertex Turbo 60GB = 200 Read and 130 Write = 315 US
OCZ Vertex Turbo 30GB = 200 Read and 120 Write = 180US
Seagate Barracuda 7200.12 320Gb = 100Read and 100 write = 45US
For a Scratch disk the most important part is Write speeds, now the perfect choice would be a few OCZ 120GBs, but well, I don't have nearly 2k just for a scratch disk ;-)
So I think for me there are two options.
Getting 4 x OCZ Vertex Turbo 30GB, which would cost me 720 US
or
Getting 4 x normal HDs which would cost me 180US
I mean if the smaller SSDs had the same write speed as the larger one, I wouldn't even argue about it ;-)
For my System I will use something like three the OCZ, just not sure if 120Gb each or 256. Still deciding between some companies, but best is to wait till January and see what is happening.
-
I wanted to add some notes to this thread.
Regarding RAM, you can’t have too much with a 64-bit machine. In an ideal world your computer would have enough ram so that it almost never needs to access the hard drive.
Investigate using a RAM disk as a paging file/scratch disk. This is not to be confused with a SSD drive.
Anti-Virus/Anti-Spyware software: Configure these programs so that they exclude the directory structure where PS/LR are installed, where your data is stored, and especially the drive where the paging file/scratch disk(s) are maintained. A/V software checks most every file you load and save. It adds a lot of processing time.
Page file/scratch disk. For best performance put these on their own drive or as mentioned above, a RAM drive. A SSD drive is also a good choice.
-
I wanted to add some notes to this thread.
Regarding RAM, you can’t have too much with a 64-bit machine. In an ideal world your computer would have enough ram so that it almost never needs to access the hard drive.
Investigate using a RAM disk as a paging file/scratch disk. This is not to be confused with a SSD drive.
Anti-Virus/Anti-Spyware software: Configure these programs so that they exclude the directory structure where PS/LR are installed, where your data is stored, and especially the drive where the paging file/scratch disk(s) are maintained. A/V software checks most every file you load and save. It adds a lot of processing time.
Page file/scratch disk. For best performance put these on their own drive or as mentioned above, a RAM drive. A SSD drive is also a good choice.
I'm not sure, but wasn't there a test a while back, whiched showed that PS can't use "a created / not physical" RAM disk very well ?
-
Chris: on x58 chipset boards you do not want to populate all 4 (or 8) slots - this incurs a performance hit, populate either 3 or 6, leave the 4th (8th) slot empty.
You also may want to consider going with a platform that supports ECC memory, like the Xeon 5500 or 3500. Memory errors are an increasing problem:
http://blogs.zdnet.com/storage/?p=653&tag=col1;post-653 (http://blogs.zdnet.com/storage/?p=653&tag=col1;post-653)
I'm personally evaluating an Intel Workstation board with a Xeon 3500 Quad Core CPU - the only downside is it's limited to 16GB (make that 12) memory.
http://www.intel.com/products/workstation/...bp-overview.htm (http://www.intel.com/products/workstation/motherboards/wx58bp/wx58bp-overview.htm)
hth - John
-
In regard to a RAM disk on a 64bit platform; maybe someone familiar with PS memory access can correct me but, PS will address *all* of the memory presented to it. Unless there is some artificial boundary built in - it makes no sense to use any memory to create a RAM disk for scratch purposes. Why add the artificial overhead of a filesystem to memory access?
-
> I'm not sure, but wasn't there a test a while back, whiched showed that PS can't use "a created / not physical" RAM disk very well ?
I don’t know of this test.
> …it makes no sense to use any memory to create a RAM disk for scratch purposes. Why add the artificial overhead of a filesystem to memory access?
For CS3 it is worthy for CS4 (x64) it is not. The overhead of a ram disk is trivial. Here is a specification on how CS3 uses memory:
When you run Photoshop CS3 on a computer with a 64-bit processor (such as a, Intel Xeon processor with EM64T, AMD Athlon 64, or Opteron processor) running a 64-bit version of the operating system (Windows XP Professional x64 Edition or Windows Vista 64-bit) and with 4 GB or more of RAM, Photoshop will use 3 GB for it's image data. You can see the actual amount of RAM Photoshop can use in the Let Photoshop Use number when you set the Let Photoshop Use slider in the Performance preference to 100%. The RAM above the 100% used by Photoshop, which is from approximately 3 GB to 3.7 GB, can be used directly by Photoshop plug-ins (some plug-ins need large chunks of contiguous RAM), filters, or actions. If you have more than 4 GB (to 6 GB), then the RAM above 4 GB is used by the operating system as a cache for the Photoshop scratch disk data. Data that previously was written directly to the hard disk by Photoshop is now cached in this high RAM before being written to the hard disk by the operating system. If you are working with files large enough to take advantage of these extra 2 GB of RAM, the RAM cache can speed performance of Photoshop. Additionally, in Windows Vista 64-bit, processing very large images is much faster if your computer has large amounts of RAM (6-8 GB).
http://kb2.adobe.com/cps/401/kb401088.html (http://kb2.adobe.com/cps/401/kb401088.html)
Accordingly CS4 x64 will us all ram available to the computer
http://kb2.adobe.com/cps/404/kb404439.html (http://kb2.adobe.com/cps/404/kb404439.html)
-
I think you may be crossing into that 10% bleeding edge catagory with your drive choices. It might be faster to do that many SSD's, but the real world price/performance for it just isn't worth it IMHO, even on an unlimited budget.
I'd keep the system drive smaller/leaner. It's probably better to run a simple OS/programs setup than trying to build a big SSD RAID 0 array with lots of programs. You might have a faster system with a 3 drive RAID 0 setup, but if you load it up with so many programs then you'll probably run slower than if you ran smaller/leaner. There's not a huge need for fast write speeds or sustained reads, mainly this drive would be doing a lot of small random access seeks for the OS.
I'd recommend a pair of Intel X25M 80GB drives in RAID 0 as your system drive. You shouldn't need more than 160GB for an OS drive. Going to 3 drives for the OS just introduces another layer of possible failure and uses up ports on your motherboard.
For scratch disk, I would stick with conventional HD's. The ones that are the leaders right now are the WD Caviar Black or RE3 drives. Look at the 640GB and 1TB sizes, they have the highest platter densities and will give you the best results. 4 640GB drives in a RAID 0 would give a very nice scratch disk and would cost just $300.
I think you could skip the extra 500GB drive for miscellaneous stuff, you'd have 2 TB of spare storage on the RAID 0 arrary after partitioning off for scratch.
The 6 drive RAID 5 or RAID 10 array sounds good for storage. It should also be fast enough not to be a significant bottleneck for loading up a file on open. I don't know whether it would be faster to keep working files on the OS, on the extra space on the RAID 0 scratch array, or on the storage RAID array. I'd imagine there is a huge IO demand to load up a 10gb file into system memory, but I don't know which one would have the fastest throughput. It would be close between the SSD RAID 0 array and the 4 drive scratch array. Either one of them might have conflicting uses, the OS accessing the drive at the same time the file was loading, or the scratch drive trying to read the file and write into scratch if the physical RAM ran out.
Another thing to think about is the speed of your RAID controller. 12 drives in three RAID arrays might be more than a normal controller could handle, this could end up being your bottleneck.
-
Thx for all the input. Yes I need a extra RAID controller. Most boards just have 6 Intel Sata ports and some have 4 more extra. So I would need at least a RAID Sata card with 4 ports.
Chris: on x58 chipset boards you do not want to populate all 4 (or 8) slots - this incurs a performance hit, populate either 3 or 6, leave the 4th (8th) slot empty.
Do you mind expending on that ? it is the first time I have heard of it and I read a lot on computer sites. The X58 has a triple channel RAM interface, so one needs to use at least 3 slots. (for example 3 x 1GB) However o achieve 24Gb of ram one has to fill all 6 slots with 4GB mem. So far I haven't heard any negative comments. The one reason some people have problems filling them is, because they are using standard dual channel Mem, which isn't really the best choice for a triple channel board. (Yes dual channel RAM can work, but does not have to)
Here are my thoughts about why using three larger SSDs instead of smaller ones.
First it is speed. To get the best out of writing and reading speed one needs to use at least 128GB per SSD. (As I posted above, below that you loose speed especially write) I really don't worry about a drive failure. I have all my presets and stuff stored away and could set up a new system from scratch in around 1-2hours.
For the scratch disk, I probably will go with HDs around 1000 for SSDs is just to much for scratch. I can always change that in the future if the price drops a bit. (Still thinking on it)
In regards of CPUs, well the one main thing that puts me off the Xeon rout is, that the current generation isn't really new and still is really expensive if one wants to get the same speed.
I'm pretty sure, that one can't buy a dual Xeon System with 2,26ghz and expect it to hold up against a 3.6ghz i7 system. It just won't. Even if you have twice the cores. Especially in Lightroom and PS :@
So to get a real advantage of a Xeon workstation, these CPUs have to be around 2,7 ghz and here it get's really expensive.
-
I think you may be crossing into that 10% bleeding edge catagory with your drive choices. It might be faster to do that many SSD's, but the real world price/performance for it just isn't worth it IMHO, even on an unlimited budget.
....
For scratch disk, I would stick with conventional HD's. The ones that are the leaders right now are the WD Caviar Black or RE3 drives. Look at the 640GB and 1TB sizes, they have the highest platter densities and will give you the best results. 4 640GB drives in a RAID 0 would give a very nice scratch disk and would cost just $300.
...
Another thing to think about is the speed of your RAID controller. 12 drives in three RAID arrays might be more than a normal controller could handle, this could end up being your bottleneck.
The samsung F3 drive have a plattersize of 500Gb, which is for big writes/reads about 30% faster than 334Gb platters.
I would take a good look at good caching raid controllers. I suspect big gains there.
-
Sure! Populating the 4th memory slot will disable triple memory (interleaved) access. It is only there for backward support - you are correct in populating by 3's....
Here's a link including a block diagram of the x58 chipset:
http://www.intel.com/Assets/PDF/prodbrief/...oduct-brief.pdf (http://www.intel.com/Assets/PDF/prodbrief/x58-product-brief.pdf)
-
Sure! Populating the 4th memory slot will disable triple memory (interleaved) access. It is only there for backward support - you are correct in populating by 3's....
Here's a link including a block diagram of the x58 chipset:
http://www.intel.com/Assets/PDF/prodbrief/...oduct-brief.pdf (http://www.intel.com/Assets/PDF/prodbrief/x58-product-brief.pdf)
All clear. My mistake, I was reading your original statement wrong. At that time I had in mind that 8 slots are all and you said, one could not fill all slots. However that wasn't what you said ;-)
-
Hi Christopher
For a really fast SSD : OCZ Z-Drive m84 PCI-Express SSD
SSD info page (http://www.ocztechnology.com/products/solid_state_drives/ocz_z_drive_m84_pci_express_ssd)
Sustained write : 600 MB/s
But price is about $1000
-
Hi Christopher
For a really fast SSD : OCZ Z-Drive m84 PCI-Express SSD
SSD info page (http://www.ocztechnology.com/products/solid_state_drives/ocz_z_drive_m84_pci_express_ssd)
Sustained write : 600 MB/s
But price is about $1000
Thanks for the info. I was thinking about using something like that, however the only real benefit I can see is that one saves up some SATA II ports. When it comes to speed I think one is on the cheaper side combining 3 x 128GB SSDs and gets away cheaper. I'm still considering this option for my scratch disk. If it would sell for around 600-700 I would go for it, but 1000 is a lot ^^
-
Thanks for the info. I was thinking about using something like that, however the only real benefit I can see is that one saves up some SATA II ports. When it comes to speed I think one is on the cheaper side combining 3 x 128GB SSDs and gets away cheaper. I'm still considering this option for my scratch disk. If it would sell for around 600-700 I would go for it, but 1000 is a lot ^^
Well it's way above my budget. I did compare it with the other OCZ specs and with those you need 6 in RAID-0, just looked at the AnandTech tests and then it compares to 3 in RAID-0.
If the motherboard sata II controller can drive those full speed it will be a lot cheaper to use 3x ssd.
-
Well it's way above my budget. I did compare it with the other OCZ specs and with those you need 6 in RAID-0, just looked at the AnandTech tests and then it compares to 3 in RAID-0.
If the motherboard sata II controller can drive those full speed it will be a lot cheaper to use 3x ssd.
Yes however, there is one point in favor of the PCI card. A good X58 board has 6 SATA II (Intel) and 4 additional SATA II (third party) ones. So I have ten in total. However for my set up I need around 13 ports, which means I have to get a extra RAID SATA controller. These start around 200 and top end one can cost around 400, which than put together with the 3 x SSDs isn't any cheaper than the other solution.
-
Yes however, there is one point in favor of the PCI card. A good X58 board has 6 SATA II (Intel) and 4 additional SATA II (third party) ones. So I have ten in total. However for my set up I need around 13 ports, which means I have to get a extra RAID SATA controller. These start around 200 and top end one can cost around 400, which than put together with the 3 x SSDs isn't any cheaper than the other solution.
Well even if there are enough connectors it's doubtful that the controller will be able to push them all to it's limits, not much desktop users use 3x ssd in RAID-0.
On the other hand some dedicated RAID SATA controllers have 512 MB memory. It could make write speeds of the storage device much less important, maybe not for you but for most users. A 400-500MB write buffer with 2x SSD is maybe be comparable to a bufferless with 3x SSD.
-
Well even if there ate enough connectors it's doubtful that the controller will be able to push them all to it's limits, not much desktop users use 3x ssd in RAID-0.
On the other hand some dedicated RAID SATA controllers have 512 MB memory. It could make write speeds of the storage device much less important, maybe not for you but for most users. A 400-500MB write buffer with 2x SSD is maybe be comparable to a bufferless with 3x SSD.
Both internal RAID controllers are connected through the same lines as any PCI port. In total they can handle more than 3GB/s
-
Both internal RAID controllers are connected through the same lines as any PCI port. In total they can handle more than 3GB/s
Christopher
I have my doubts about the controller itself, but I can be complete wrong.
It's maybe worthwhile to wait for SATA 3 MB's and the first SATA 3 SSD's. I've seen some info to the first MB's. Asus has something special to connect to one off the PCe x16 lanes.
I expect that the SSD's will quickly have much better read speeds.
-
I have never read so much BS in my entire life. Build the fastest system you can afford and in 6 months you will be able to have a system 50% faster for 50% less, IOW, you are spending more time researching trying to gain less than 10% for what? 2010 Q1 will see newer faster and higher capacity SSDs, that still are unproven technology. Set a budget, stick with it and upgrade a little every 6 months. Why spend $10,000.00 today when $5,000.00 tomorrow will be twice as fast? Be smart! Hint: You are NOT using any programs that need "State of the Art" computing power.
-
Hint: You are NOT using any programs that need "State of the Art" computing power.
Perhaps not.. but he is working with P65+ files; which can easily exceed 2 gig as layered files in PS. P65+ files 'NEED' a fair amount of computational power - end of story.
I use an 8 core Mac Pro with 32 Gig of Ram and RAID10 FAST HD's - and I am processing and working with 1DS MK3 files. I can easily make my puta chug with these files. Christopher needs a high performance machine for the files he is dealing with - period.
-
Well that is the main point why I am arguing about different solutions. I'm not planing on buying the most expensive stuff out there. I will get the things which will actually boost my performance. As example, going with a i7 System I would get the i7 920/30 which is quite a low end model, but can easily be clocked to the same speed as the high end models. And you argument is just old. with electronic you can always wait and get something faster, however I need something which comes close to the 100% which are possible now. Not in 6 months. (besides that your argument is flawed because you just can't get something 6 months later for half the money which is twice that fast. (only if you would buy ONLY Top End products, which would be stupid.)
Some might be able to work with larger files on a medicore system and get coffee between each mouse click, but I certainly can't.
-
Perhaps not.. but he is working with P65+ files; which can easily exceed 2 gig as layered files in PS. P65+ files 'NEED' a fair amount of computational power - end of story.
I use an 8 core Mac Pro with 32 Gig of Ram and RAID10 FAST HD's - and I am processing and working with 1DS MK3 files. I can easily make my puta chug with these files. Christopher needs a high performance machine for the files he is dealing with - period.
Lots of Ram & a couple fast HDs does NOT equal state of the art. Your 1st problem is you are using a MAC, so you are overpaying (a factor of 2) for a machine and receiving no benefits. If this were a discussion about 3D animation or even 3D CAD programs, then we could talk, but static images using Photoshop might take some ram, but that's about it. It's a total waste of cash to overpay today for negligible returns. What's high performance? Period? Have you ever tested various machines with various configurations to see what works best? I doubt it, if you had, then you wouldn't be using a MAC. If this is a discussion about best bang for the buck, or even highest performance computing than a MAC doesn't even enter the discussion.
-
Lots of Ram & a couple fast HDs does NOT equal state of the art. Your 1st problem is you are using a MAC, so you are overpaying (a factor of 2) for a machine and receiving no benefits. If this were a discussion about 3D animation or even 3D CAD programs, then we could talk, but static images using Photoshop might take some ram, but that's about it. It's a total waste of cash to overpay today for negligible returns. What's high performance? Period? Have you ever tested various machines with various configurations to see what works best? I doubt it, if you had, then you wouldn't be using a MAC. If this is a discussion about best bang for the buck, or even highest performance computing than a MAC doesn't even enter the discussion.
Well I agree on you on the mac part when it comes to costs. I also really doubt that 8 cores bring any real advantage over 4. (Main point between Xeon and i7) However I think there is no better way to speed up working with larger files than fast disks / Raid setup and lots of ram. I agree as well that PS is not state of the art, it is a joke when it comes to usage of modern technologies, than however it is a lot cheaper than 3D animation software and similar stuff. The main problem I really see is that there is no competition which can actually put pressure on Adobe. I mean there really is nothing out there besides PS.
If you look over the Mac Performance Guide one can see some clear benefits to some fast components, however it would be better to actually try out some other system combinations. I would love to see what the speed difference between a 8 core system with 32GB ram compared to a 4 core system with 24Gb ram is. Is it 10 or 15 % ? Which In could care less about, or is it 30 or even 40% which for example would be a lot for me. I haven't found the answer and I fear I won't because there aren't to many test out there.
-
Well that is the main point why I am arguing about different solutions. I'm not planing on buying the most expensive stuff out there. I will get the things which will actually boost my performance. As example, going with a i7 System I would get the i7 920/30 which is quite a low end model, but can easily be clocked to the same speed as the high end models. And you argument is just old. with electronic you can always wait and get something faster, however I need something which comes close to the 100% which are possible now. Not in 6 months. (besides that your argument is flawed because you just can't get something 6 months later for half the money which is twice that fast. (only if you would buy ONLY Top End products, which would be stupid.)
Some might be able to work with larger files on a medicore system and get coffee between each mouse click, but I certainly can't.
I guess you kind of answered your own question, huh? I'll agree that if one is using his/her machine for business (as I do with 3D CAD) and the faster you can process the file the more money you can make then it's just a question of payback. If a machine which costs $10,000.00 more saves you 30 minutes per day at $100.00 per hour then your payback is 200 days or less because you can section 179 the thing so there's also a tax savings as well as a time savings. If it's 10 minutes savings per day, then the payback is 600 days and if it's 2 minutes per day......................................................... I'm sure you get the point. Computers are like cars, the final 2% of performance is the most expensive, is it worth it? If racing cars is your business and winning is the goal, it's worth it. How many images do you process a day? How much time can you save? Is it better to have 2 or 3 medium priced machines running rather than one super high performance machine? I've been building high performance computers for 12 years and I currently run 6 machines and any two of those machines is faster than my fastest machine. Some people tend to overspend on what they don't understand.
-
I guess you kind of answered your own question, huh? I'll agree that if one is using his/her machine for business (as I do with 3D CAD) and the faster you can process the file the more money you can make then it's just a question of payback. If a machine which costs $10,000.00 more saves you 30 minutes per day at $100.00 per hour then your payback is 200 days or less because you can section 179 the thing so there's also a tax savings as well as a time savings. If it's 10 minutes savings per day, then the payback is 600 days and if it's 2 minutes per day......................................................... I'm sure you get the point. Computers are like cars, the final 2% of performance is the most expensive, is it worth it? If racing cars is your business and winning is the goal, it's worth it. How many images do you process a day? How much time can you save? Is it better to have 2 or 3 medium priced machines running rather than one super high performance machine? I've been building high performance computers for 12 years and I currently run 6 machines and any two of those machines is faster than my fastest machine. Some people tend to overspend on what they don't understand.
Yes I understand you, the for me the frustrating part is that one can find so little information about real performance. I mean yes I can find LOTS of infos on 3D stuff, Video stuff, raw power, games, audio only very little on image software.
I mean even though I use the computer to make money, I would never buy it as a complete package. I mean if I even start to put together a DELL or whatever, I reach a price around 2twiche or 3 times what I would need to pay otherwise. Besides that I love to build computers. I think if I could make digital cameras i would build them myself, too.
I think there is one perfect example. (I know there are many but it is something I love) Intel has their great Extreme edition of every CPU generation. Which in most application is not even 10% faster, but costs twice the money compared to the next CPU in line.
-
Chrsitopher,
You know as much about this as I do. I'll describe what I did - with improvements for my "slips" - FWIW. (I do have a degree in photography, and one in computer science.)
My target is usually about 85% on a cost/effectiveness basis. Beyond that it is just cheaper to wait 6 months and buy a better base machine for $600 instead of $2,500 today, as you mention. Plus, I over obsess and overengineer when designing, but on a day-today basis, that extra 5% to 10% is totally irrelevant.
So, I built a core network of 4 machines with 2 gigabit lans. One for "local" data - photoshop macgine, server, nas - and one for "external" data - wirelress lan, laptops, etc.
I would suggest 2 machines as a start:
1) A Photoshop/Lightroom **only** box that may not even have a daily ionternet connection! Probably an i7 920, and
2) An application/data/storage server for **everything** else - any core applications, data storage, offload processing, printing, backups, etc. Probably a Dell Poweredge quad Xeon 300 or 500 series, no highg-end video required. With dual gigabit network cards and a hardware Raid controller (a 3rd box would be an actual NAS, in addition to this server.) Base cost $500 + RAID card + RAM + disks.
In the photo machine, I personally would put 3-4 320GB WD 7200 drives in Raid 0 or similar for OS. Apps, and Scratch disk (2 or more separate logical drives of course. As many as you can squeeze. Cloned for instant re-install) I am ambivalent on SSD - a bit too pricy yet. (I got the WD Caviar Blue for $30 each. They area great size for OS, aps.)
Then a couple of 1 to 2TB HDD's - one for RAW files (local working copy) and one for WORKING files (lightroom database, etc.) Backed up to the Ap/Data server often over gigabit.
For RAM, I'd do as much as I could with 2GB chips. The 4GB chips are just too effing expensive! Might as well buy an extra box instead for $600!!!
My 3rd box is the NAS with 6TB and Windows Home Server. It is off 90% of the time.
My 4th box is my "everythinhg else" machine -a dual core laptop with 4GB for watching TV, music listening, e-mail, Outlook, etc. I keep the server as a pure server (Running Hyper-V: was ESXi from VMWare.)
Everything else - bluray players, tv's, 3 other laptops, etc run off of the "external" lan to avoid traffic. There is actually a bridge to a 3rd router, which is the internet gateway for all.
Just thoughts. Have fun! (Sorry for typos, hand pain ....)
Cheers,
Michael
This is simply more logical and is close to the way I do things. I have a cheapo machine that does nothing but internet and email, then I have another cheapo that has all my financial information(SCSI HDs). I then have a dual xeon configuration (5520) that has all the trimmings, this is my main 3D box it includes all SCSI HDs, (I might add SSDs next year) & 24GB RAM then I have two 3.06 I7 running on an Intel board (X58) with 12GB again all SCSI HDs. I mostly stick with Intel or Asus MBs & SCSI HDs for their incredible speed and reliability. I also own a couple Imacs, an HP laptop and I just purchased a MBP 15" which I took to the Riviera Maya with NO dual boot Windows 7 crutch, I wanted to learn the MAC OS and see what the differences were. I will tell you this, if you boot up everyday a MAC will save you about 1-2 minutes in boot time It certainly boots and shuts down quicker than any windows machine that I've ever built. I just ordered an X25-E for the MBP and will be taking it with me to Bali in 4 weeks. SSDs have less issues with higher altitude, they consume less power and produce less heat, that's why I was an idiot and spent $700.00 for 64GB. I also have a NAS, but that's irrelevant.
If you want to send me some files that I can test on my various machines I could let you know what kind of time savings you'd be looking at? I don't use that one program you mentioned to do panos, but I do have PS (both Win & MAC versions), LR and Nikon Capture NX2.
-
I am thinking about an 8 core Mac with 32 Gb memory, with stitching and stacking for macro in mind.
-
Yes I understand you, the for me the frustrating part is that one can find so little information about real performance. I mean yes I can find LOTS of infos on 3D stuff, Video stuff, raw power, games, audio only very little on image software.
I mean even though I use the computer to make money, I would never buy it as a complete package. I mean if I even start to put together a DELL or whatever, I reach a price around 2twiche or 3 times what I would need to pay otherwise. Besides that I love to build computers. I think if I could make digital cameras i would build them myself, too.
I think there is one perfect example. (I know there are many but it is something I love) Intel has their great Extreme edition of every CPU generation. Which in most application is not even 10% faster, but costs twice the money compared to the next CPU in line.
Hi Christopher
I found a small program that can test disk write speed for specific size. If there's some write caching, it will take it in account. Continous tests don't do that.
Small HDD test software (http://www.raymond.cc/blog/archives/2008/02/28/measure-actual-hard-disk-perfomance-under-windows/)
-
I just finished building a Core 2 Quad system and have been very happy with its improved performance. My simplistic take on the matter:
* load up on as much RAM as you can afford. This will mitigate the effect of any other slow IOs (HDs, USB, ethernet, etc...).
* Include SSD in your solution. They are substantially faster in real world apps.
* Don't pay the premium for bleeding edge.
* I haven't seen it mentioned. Get a decent video card with lots of VRAM.
Another note, I use PTGui extensively, and one thing I was disappointed to see was that even in its 64-bit version, it never uses more than 2GB of RAM. I am working with 15MP images from a 50D that are saved in TIFF format, but I have never observed PTGui use more than that 2GB of RAM. The greatest single improvement I have seen with PTGui was the use of a small (30GB) OCZ Vertex SSD as both the source-file location and the temp file location. This has increased speeds by (estimated) three-fold.
-
I also think that a fast I/O subsystem is more worth than 8 or more core cpus.
At the moment most software is still not there. Take a look at Mr. Chambers tests (or any other site doing benchmarks on Photoshop). PS and LR have still a lot place for improving performance on many core machines. They don't scale well. And I estimate that it'll still take some iterations of both programmes before the can efficiently use every core in 8 or more core machines. This isn't only PS's and LR's problem. The vast majority of software is not optimized for high degrees of cpu parallelization. And the major desktops oses (MacOS and Windows) are only now beginning to improve in this respect. Vista/7 and MacOS 10.6 are a big step forward in that regard but both have still plenty of room left. So, hardware at the moment is quite some way ahead and software is only beginning to catch up (3D seems to be on the fore in this regard).
So, a high clocked 4 core system is in most cases as fast (or at least not that much behind) as an equally clocked 8 core system. Yes, of course, depending on how many programmes are running in parallel, responsiveness will likely be better, but the individual programmes will most likely not execute (much) faster.
So, fast SSDs (maybe in RAID 0 or similar configurations), a high clocked (i7) cpu, as much RAM as one needs to fit the largest files in (plus some), is the current way to go for a photographer's workstation in my opinion.
A server for centralized storage of documents and other data is also a good idea when one syncs between the workstation and, say, one's laptop. So even if (or rather when) the workstation is under full load one can still get all the other work done.
Ah, and the advice of getting a card with much vram is especially true with MacOS based machine but also with Windows based ones since both systems try to keep the graphics on the card. For MacOS see: AnandTech (http://www.anandtech.com/video/showdoc.aspx?i=3602&cp=2) and Ars Technica (http://arstechnica.com/apple/reviews/2005/04/macosx-10-4.ars/13) with the later going quite deep into the matter. In the AnandTech article you can scroll down to the end where he gives an example of memory usage in PS CS4. I haven't found anything similar for Windows 7 but this: Ars Technica (http://arstechnica.com/microsoft/reviews/2009/10/windows-7-the-review.ars/8). Still, more vram helps
-
So, a high clocked 4 core system is in most cases as fast (or at least not that much behind) as an equally clocked 8 core system. Yes, of course, depending on how many programmes are running in parallel, responsiveness will likely be better, but the individual programmes will most likely not execute (much) faster.
So, if one uses Word to write a post for this forum, while phocus is working its way through a large number of 60 Mpx files and Photoshop is busy, then you would get benefit from an 8 core machine?
I expect that two networked machines would be faster.
-
So, if one uses Word to write a post for this forum, while phocus is working its way through a large number of 60 Mpx files and Photoshop is busy, then you would get benefit from an 8 core machine?
I expect that two networked machines would be faster.
I have been known to have my laptop sitting right next to my desktop--using both simultaneously. It seems ridiculous to many people, when the PC is really churning on an HDR mosaic panorama, it can be necessary. Though, with that said, my latest desktop build has yet to be overwhelmed with anything I have thrown at it.
8GB RAM
PCI-ex 1GB VRAM
Intel Cord 2 quad Q9550 (2.8GHz)
30GB SSD for temp files and scratch disk
Internal RAID 0 for storage
-
Have you gotten your new workstation bought/built yet?
-
Have you gotten your new workstation bought/built yet?
Nope still traveling for a few more weeks, but I will certainly write some feedback once I have finished it.
-
So, if one uses Word to write a post for this forum, while phocus is working its way through a large number of 60 Mpx files and Photoshop is busy, then you would get benefit from an 8 core machine?
I expect that two networked machines would be faster.
Basically, yes. See here (and other articles of Mr. Chamber's (the Mac Performance section is very good and thorough and largely applies to Wintel machines as well): Many Core Perfomance w PS CS4/MacOS (http://macperformanceguide.com/Reviews-MacProNehalem-MoreIsLess.html)
From a software development perspective we're at a point now where most of us need to learn new skills to take advantage of the processing capabilities of modern cpus, mainly that we now have more than one cpu core. And we're at the beginning of this process which will take years to fully settle in the minds of the development community.
I guess that the next generations of programmes such as Photoshop will be much better at parallel computing than they are now. But of course especially with such big software like Photoshop you can't make major design changes (and I bet there is need for quite some major overhaul) in one or two iterations. You'd basically need to make something in the way Apple did with MacOS 10.6. Not many new user visible feature but a major overhaul of the underlying architecture (such as introducing (and using) thing like Grand Central as a technology to help developers write concurrent code faster (and easier)). But I guess this won't happen.
And indeed: Is the workstation already built and if so, what's the final configuration, if I may ask?
-
Sure! Populating the 4th memory slot will disable triple memory (interleaved) access. It is only there for backward support - you are correct in populating by 3's....
The manual for an Intel motherboard states that when all four slots are filled, it may result in "less than optimal performance." It does not state that triple channel is disabled. What would happen, instead, is that memory usage that extends into the range of that of the fourth module -- Channel A, DIMM 1 -- performance in that range will only have a single channel while everything else that fits in the first three -- Channel A, DIMM 0; Channel B; and Channel C -- will still have the performance of triple channel goodness.
You are right, though, that for best performance populate in matching sets of three.
Additionally, I haven't found any X58 motherboards that have eight memory slots. Just three, four and six. (You've stated in another post four and eight slots for an X58 chipset. I'm still figuring out how this forum works as it is very different from phpBB.) And, I have found no evidence to back up your statement regarding reduced performance from having all of the slots filled with a Core i7, with exception to the "less than optimal performance" which I've explained. Though, I haven't really found anything to refute that claim either.
More to the point, however, is that the impact to performance is not noticeable to the end user. We're talking about a difference of seconds or at most a few frames per second in games. A greater impact on performance would be running out of memory and going to the page file because of a ridiculous suggestion to not utilize all of the memory slots. When working with such large files, as is being suggested, it is more important to have the most memory possible available for use. After all, it isn't as if the OP is running Sphinx where memory latency is the determining factor of performance.
A four year old article, but still relevant to this discussion:
Exploring the performance impact of memory latency (http://techreport.com/articles.x/8966/1) at The Tech Report.
-
More to the point, however, is that the impact to performance is not noticeable to the end user. We're talking about a difference of seconds or at most a few frames per second in games. A greater impact on performance would be running out of memory and going to the page file because of a ridiculous suggestion to not utilize all of the memory slots. When working with such large files, as is being suggested, it is more important to have the most memory possible available for use. After all, it isn't as if the OP is running Sphinx where memory latency is the determining factor of performance.
A four year old article, but still relevant to this discussion:
Exploring the performance impact of memory latency (http://techreport.com/articles.x/8966/1) at The Tech Report.
Great report, but irrelevant to the new architecture the X58 Chipset offers. The 4th Slot offered on some mainboards is simply there for backward compatibility. Populating it will turn off on-die QPI memory access and revert back to Front Side bus (northbridge) operation. QPI (triple channel interleaved memory access) offers up to 25.6 GB/sec bandwidth, effectively doubling that of 1600Mhz FSB. The architecture furthermore reduces latency. Populating that 4th slot turns all this off. If you are looking for an X58 board with that extra slot look no further than Intel's WX58BP:
http://www.intel.com/products/workstation/...bp-overview.htm (http://www.intel.com/products/workstation/motherboards/wx58bp/wx58bp-overview.htm)
-
Remember in all this discussion of memory performance that the worst DRAM memory performance will always be faster, by many magnitudes, than the fastest hard drives or flash. I think people overthink RAM speed as a major contributor to overall system performance. For our work, we are always dealing with scratch disks and pagefiles. Those are our worst enemies.
MORE RAM will almost always be better for image processing.