Luminous Landscape Forum
Equipment & Techniques => Computers & Peripherals => Topic started by: Steve Weldon on August 28, 2011, 01:24:06 am
-
If you guys were building an ultimate workstation for image processing and video rendering 50/50.. what components would you choose?
CPU/chipset/RAM. I'm also curious as to your thoughts on the 2600k/z68 vs. the i7-980/x58 for the intended purpose, and how the price differential once you include the 2600k's max of 32gb of RAM vs the i7-980's 24gb of ram (about $200 for the same quality/speed).. For this build we want max processing/speed. Do you think the i7-980's six cores for $600 and $200 for 24gb of ram will outperform the 2600k's four cores ($349) and 32gb of ram (about $1300)? What motherboards would you use? Are dual CPU systems still viable when compared? I'm not interested in the 2600k's build in GPU, hybrid controller, or any of that sort of thing. Only features/performance that would render video/images better/faster.
GPU's.. The displays will be 3 PA301w's which are 10 bit displays. Adobe isn't listing CS5 as 10 bit compatible and from what I can read between the lines CS6 won't be either. Considering that, will the greatly superior processing power of say a dual GPU ATI HD 6990 ($729) serve better than a mid-grade workstation card like the 7800 for about the same price for image processing and video rendering? I know I'd choose a workstation card for CAD/CAM aps.. an ISV card. But is there any "visual" or "performance" benefit for images with a workstation card? I'm especially interested in the 10 bit pros and cons..
Storage: We're going for a Revodrive 3 x2 480gb for system/programs/caches, and a Vertex 3 240gb for a work drive, and a Vertex 3 60gb coupled with a WD R4 Black 1tb and a hybrid controller for LR indexes. WD R4 Black 2tb's will be internal storage, Drobe via esata external. Not interested in internal raids. Would you do it differently if money wasn't an issue?
We're also looking at an Lian-li V1020R case.. it's a work of art.. with a Enermax 1350 GOLD which I know is overkill, but this PS has really high quality componentry including very high quality Japanese capacitors so the $70 increase over the 900watt model doesn't bother me. Would you choose a different case or supply and why?>
I'm building this for a good client and I want to explore all options. Input is greatly appreciated.
-
I'm looking at building out a system with very much the same objectives as the above and am looking forward to suggestions as well. Since I also plan to do audio recording as well as photo and video work, those offering suggestions please add to the list of desirable features in the case and power supply as close to absolutely silent performance as possible - rock-bottom minimum possible fan and drive noise noise, etc. My current Dell desktop sounds like a 747 spooling up for takeoff >:(.
-
I'd say that's pretty close to an ultimate system. I like the bevy of SSD's.
For the 2600k system, I assume you meant $300 for memory, not $1300.
In a number of months, the 2600k will be superseded by a faster chip. Plug it in, and performance goes up. I don't know, but I might guess that the i7-980 is getting close to end of line and may not offer much growth potential.
-
I'd say that's pretty close to an ultimate system. I like the bevy of SSD's.
For the 2600k system, I assume you meant $300 for memory, not $1300.
In a number of months, the 2600k will be superseded by a faster chip. Plug it in, and performance goes up. I don't know, but I might guess that the i7-980 is getting close to end of line and may not offer much growth potential.
No mistake, at Crucial.com 32gb (4 8gb sticks) will run you about $1300 (http://www.crucial.com/store/listparts.aspx?model=GA-Z68X-UD7-B3&Cat=RAM) and that's for only 1333.. There are cheaper 8gb sticks, $800 for 32gb being the cheapest I can find with a real name.
And yes "Ivybridge" will be out towards mid-late 2012.. it will be interesting to see what that brings. And there has been talk of another 1366 series with 8 cores coming out before that. But so far it's all vaporware.
-
I'm looking at building out a system with very much the same objectives as the above and am looking forward to suggestions as well. Since I also plan to do audio recording as well as photo and video work, those offering suggestions please add to the list of desirable features in the case and power supply as close to absolutely silent performance as possible - rock-bottom minimum possible fan and drive noise noise, etc. My current Dell desktop sounds like a 747 spooling up for takeoff >:(.
A quality Lian-li case will be very quiet, especially if you use an aftermarket cooler like this one (http://www.bangkokimages.com/Articles/Equipment/entryid/1015/Arctic-Cooling-Accelero-XTREME-Review.aspx)on your GPU. 140mm fans at mid-speeds are almost silent in the Lian-li cases. The loudest noise is the power supply.
For audio recording NO noise is the goal, and that means one of the few fanless very high quality power supplies. The problems with these is they're in the 400-500watt range which isn't enough power for a well equipped image workstation.. so you might want to consider two machines, one for each purpose, depending how serious you are about your audio recording.
-
or cut a hole in the wall and put the machines in a different room from the recording.
-
At this point, Socket 2011 is nearly here. It will replace Socket 1366. Although Steve is well aware of the differences between the X58 Chipset (socket 1366) and Z68 (socket 1155) - I thought it would be usefull to compare:
The chief difference is the method of memory access - the Z68 chipset continues to use the tried and true south bridge, resulting in serial memory access over a bandwidth constrained bus. Granted, with the Z68 that bandwidth has dramatically increased resulting in better performance figures than the x58.
The X58 and the soon to be released X79 (socket 2011) offer on die memory access controllers- thats exactly why the pin count on these CPU's is so high. On die memory access control offers some significant advantages:
1) Lower latency -transfer of data from RAM to the CPU's L3 Cache occurs in fewer "clocks" - to put it another way, the CPU simply spends less time waiting for it's data.
2) Interleaving - The X58 uses 3 channel interleaving, basically data is "striped" across 3 memeory modules and sequentially accessed (the X79 uses 4 channel interleaving). This also dramatically reduces the performance difference between various memory clock speed - in addition, the performance hit you take with ECC memory is also reduced.
So given the question - if I were to build that ultimate workstation today (for photography) - I would tend to choose the Z68, as it actually outperforms the X58, with the advantage of additional memory capacity. All that being said, I'm *not* personally replacing my X58 (reference Intel mainboard) workstation with a Z68....
In a very unscientific comparision, I tried LR/CS5 processing a 14 shot pano on both my X58 and a Dual Xeon CPU S5500 Workstation built for a client, I really saw no significant difference in either speed or core utilization (again strictly for photgraphy)
My recommendation would actually be the X79 populated to 32GB (I would use ECC, but that won't be required - the X79 chipset will support both). If you look at prototypes from this year's CES, you see very short memory traces with the 4 memory sockets immediately adjacent to the CPU.
http://motherboardnews.com/2011/05/30/x79-motherboards-galore-at-computex-2011-pre-show/
For Video processing, it's a different story: Here's a nice source of information, providing Premiere CS5 benchmark's and results for various machine configurations:
http://ppbm5.com/DB-PPBM5-2.php
Note that machines with Dual Xeon, Socket 1366 *and* Z68 are among the top 5........
Finally on cases - you may want to check out Mountain Mods:
http://www.mountainmods.com/monticle-18-cyo-p-501.html
They actually make the Lian-Li seem kinda thin..... ;)
-
At this point, Socket 2011 is nearly here. It will replace Socket 1366. Although Steve is well aware of the differences between the X58 Chipset (socket 1366) and Z68 (socket 1155) - I thought it would be usefull to compare:
So given the question - if I were to build that ultimate workstation today (for photography) - I would tend to choose the Z68, as it actually outperforms the X58, with the advantage of additional memory capacity. All that being said, I'm *not* personally replacing my X58 (reference Intel mainboard) workstation with a Z68....
Note that machines with Dual Xeon, Socket 1366 *and* Z68 are among the top 5........
Finally on cases - you may want to check out Mountain Mods:
http://www.mountainmods.com/monticle-18-cyo-p-501.html
They actually make the Lian-Li seem kinda thin..... ;)
1. In all good conscience, I can't recommend not only a brand new motherboard model, but a brand new chipset as being "reliable" to a client. It might be something I'd play around with for myself just because it's the latest and greatest and therefore cool.. but not to a client.
2. The 2600k/z68 does outperform the 980/x58 "per core" to the tune of what.. 7-9%? But you've got two more cores with the 980. Considering this to you still think the z68 outperforms? And that 32gb memory comes with a steep pricetag, though money is a secondary consideration on this build.
3. Nice list on Premier! Sure, the z68 is in the top 5, but it's pushing 5.2ghz to get there.. that's smoking. Wonder what it's lifespan would be? And it's certainly water cooled. We'll be using stock speeds and air cooling. The 980's seem to dominate that list along with it's 970 brother (same six cores, but a tad slower stock speed). coming in 2nd. Kinda surprised the z68 isn't better represented, but it hasn't been out that long either.
4. Thanks for the tip on the Mountain Mod's cases.. they look work spending a bit to read up on. Have you personal experience with them? I've not seen much (other than a Mac Pro) in the same class consistently as Lian-li's.. but I'm open to new stuff. I've always considered cases as a vital component because of air flow and ease of service, while most still look at them as a box to hold the parts together.
Any thoughts on 10 bit workflows?
-
When it comes to workstation class machines, I have no hesitation adopting Intel's "reference" implementation right away. Just look at Apple lately.... I'll be building an X79 as soon as it's available and will be happy to report on it.
In regards to Z67/68 offerings, Intel really messed up on the first two releases of the Z67 chipset (SATA controller). They've changed process' as a result - I've had *no* issues with a first run Asus Z68 board. In fact, I have yet to see a single BIOS update posted -*very* unusual, especially for Asus.....
In regard to cores, thats where i'm frankly unsure - in my comparison i did not see the additional cores make any appreciable difference, at least with LR/CS5. In addition, I remember a post by Eric Chan mentioning that LR really doesn't benefit beyond 6 cores....
ppbm5 is a fascinating site! Obviously, I don't overclock either, but it's quite amazing how well "enthusiast" class machines compare with workstation class. Of course, with Premiere the video card is also a huge factor.....
I recently received the Mountain Mods "monticle" case, I'll be building the X79 in it - they actually ship flat. Panel thickness is much closer to a Mac Pro
As far as a 10bit workflow, obviously displayport is required - nVidia Quadro class cards appear to be the way to go. Apparently the GTX580 is capable of 10bit, but the driver doesn't allow it. I saw another thread on Abobe, where a user mention that the GTX470 was supported:
http://forums.adobe.com/message/3062686
Honestly I'm waiting for a bit of clarity from the display makers.......
http://www.nvidia.com/docs/IO/40049/TB-04701-001_v02_new.pdf
Finally, Eizo documents 10bit display support for CS5:
http://www.eizo.com/global/support/compatibility/monitors/20.pdf
-
In regard to cores, thats where i'm frankly unsure - in my comparison i did not see the additional cores make any appreciable difference, at least with LR/CS5. In addition, I remember a post by Eric Chan mentioning that LR really doesn't benefit beyond 6 cores....
ppbm5 is a fascinating site! Obviously, I don't overclock either, but it's quite amazing how well "enthusiast" class machines compare with workstation class. Of course, with Premiere the video card is also a huge factor.....
I recently received the Mountain Mods "monticle" case, I'll be building the X79 in it - they actually ship flat. Panel thickness is much closer to a Mac Pro
As far as a 10bit workflow, obviously displayport is required - nVidia Quadro class cards appear to be the way to go. Apparently the GTX580 is capable of 10bit, but the driver doesn't allow it. I saw another thread on Abobe, where a user mention that the GTX470 was supported:
1. Adobe will an unlimited number of cores. At least this is the conventional wisdom I've read in a few tech reports from Adobe. Something I've always felt they should provide is a more complete area for equipment giving the system builder more information to work with. I feel it's poor support to just ignore the subject (or give some generic answer which is just as good) and let the guys in the forums figure it out. This goes for 10 bit support too. Nothing official from Adobe that I can find either other than which GPU's and systems are ISV certified for specific applications.. which is really nice, but they're not addressing the 10 bit question directly. If you buy an ISV certified GPU which supports 10 bit, does that mean CS5 Photoshop now supports 10 bit? And I'm wondering if we even want 10 bit, what I really want is the closest thing to the capabilities of my own printer or my print shops printer.
In any case, I do think 6 cores with 24gb of RAM, all else being equal, should out perform 4 cores with 32gb of RAM which run 7-9% faster. And that benchmark page you link shows the 6 cores dominating the top spots.
2. I'll over clock when I feel I'm getting enough out of it for my trouble AND I know I can do it reliably. With the 920/930/950/960's I can do this comfortably to 3.8ghz.. but I'm happy with them at 3.6..
3. Lian li has a more serious series of cases with thick panels and a brick build quality. But if I can get the same quality or better built in the USA at a comparable price then I will.
4. Did you read that thread you linked? It was like the blind leading the blind, in the dark. No one was sure of anything.
I suppose the answers we need to know concerning 10 bit:
a. Do we benefit from it with our current printers?
b. How to achieve it throughout 100% of our workflow.
-
1) The issues raised by Eric and others are very relevant. Simply throwing more cores at a given task doesn't automatically imply "better performance", at some point processor cache coherency, and resultant memory faults are going to begin to swamp any benefit offered by multiple cores; this is why I take his statement regarding 6 cores for Lightroom very seriously. In my experience, comparing CS5 rendering a 14 shot pano on my X58 and a new Dual Xeon (12 total cores), revealed a suprisingly small performance difference. Running Windows Resource Monitor during the render on both machines, showed relatively little CPU utilization.
For video, I would expect a different result, as ppbm5's statistics bear out. But again, Premiere relies heavily on nVidia's CUDA GPU methods.
In any case, it would be tough for me to recommend the X58, in the face of the X79 coming out. I did pose the 8-core 1366 question to my Intel Rep - he simply responded with another question; "Now why would we cannibalize 2011 sales?" 8 cores on 32nm process in the face of the newer Ivy Bridge 22nm process with trigate transistors?
http://www.pcmag.com/article2/0,2817,2384909,00.asp
Probably not... In my opinion the forthcoming Z79/Socket 2011 hardware is going to be every bit the "monster" that the X58 was when *it* came out 3 years ago.
2) I'll never overclock; I guess our conservatism shows in different ways....
3) Mountain Mods are made in my backyard (well close) in Hoods River, OR. I'm quite knocked out by the quality, and modularity. A bit industrial at first glance, I'm planning on replacing exterior hardware with recessed, smoothing appearance out - none the less, the one consistent comment I get is "wow".
4) LOL - that exactly why I posted it. The comment about the nVidia support engineer (if true) is telling. Lots of confusion....
From my understanding, PS-CS5 doesn't really have to "support" 10bit color, any more than it "supports" 16bit printing. Internally, it's 15 (+1) bit of course - the interface to outside devices and their attendant colorspaces would be dependant on the the individual device capabilties;
Eizo has definately documented 10bit display color for CS5 for Windows 7 to it's panels via ATI adapters in the link I provided above.
Which begs the question; do we need it? For me, certainly not - i rarely, if ever, see display banding, and fully understand that it will have no effect on my printed output. I've read that 10bit color *is* important for rendering and evaluating broadcast color - but to my knowledge that is normally done on dedicated hardware/applications (ie: Davinci Resolve).
-
1) The issues raised by Eric and others are very relevant. Simply throwing more cores at a given task doesn't automatically imply "better performance", at some point processor cache coherency, and resultant memory faults are going to begin to swamp any benefit offered by multiple cores; this is why I take his statement regarding 6 cores for Lightroom very seriously. I
For video, I would expect a different result, as ppbm5's statistics bear out. But again, Premiere relies heavily on nVidia's CUDA GPU methods.
In any case, it would be tough for me to recommend the X58, in the face of the X79 coming out. I did pose the 8-core 1366 question to my Intel Rep - he simply responded with another question; "Now why would we cannibalize 2011 sales?" 8 cores on 32nm process in the face of the newer Ivy Bridge 22nm process with trigate transistors?
Probably not... In my opinion the forthcoming Z79/Socket 2011 hardware is going to be every bit the "monster" that the X58 was when *it* came out 3 years ago.
2) I'll never overclock; I guess our conservatism shows in different ways....
3) Mountain Mods are made in my backyard (well close) in Hoods River, OR. I'm quite knocked out by the quality, and modularity. A bit industrial at first glance, I'm planning on replacing exterior hardware with recessed, smoothing appearance out - none the less, the one consistent comment I get is "wow".
4) LOL - that exactly why I posted it. The comment about the nVidia support engineer (if true) is telling. Lots of confusion....
Which begs the question; do we need it? For me, certainly not - i rarely, if ever, see display banding, and fully understand that it will have no effect on my printed output. I've read that 10bit color *is* important for rendering and evaluating broadcast color - but to my knowledge that is normally done on dedicated hardware/applications (ie: Davinci Resolve).
1. In this instance we are talking six cores which everyone seems to agree on ARE utilized, and we are talking an equal amount of video. I see the six cores of the 980 superior for these purposes to the four cores of the 2600k. "How much?" Only testing with specific applications will tell. And with the cost of 32gb in the 2600k system the x58 system is significantly less expensive.
2. For me it would be tough.. well I just wouldn't.. recommend a system that isn't consumer tested and well on it's way to it's first revision. Much less a system that isn't even out yet. This system is being built next week, not sometime next year. So the z79 while interesting isn't a possibility.
3. It will be interesting to see how the z79 evolves. Though, there is still plenty of time for a major upset. These days you can't count on anything until it's actually sitting on your dealers shelves.
4. Some CPU's overclock without penalty, all the major manufacturers have put out their versions of overclocked 920/930/950/960's and backed them with their standard or extended warranties. We've had this conversation before and while I agree it's an area I normally wouldn't touch, it also doesn't make sense to not take advantage of free power when it's indeed free.
5. Hoods River? I owned a small ranch in Umpqua for some years and just recently relocated what stateside assets I had from the Medford area to the midwest. I'd love to support a local Oregon company so I'll be checking them out carefully.. I might even arrange to visit their factory on my next visit home.
6. Exactly.. I'm waiting for someone to tell us why we need or even want 10 bit. NEC, HP, Eizo, Dell.. they're all jumping on the 10 bit bandwagon and the GPU manufacturers are as well. AFAIK Adobe has no ISV 10 bit certifications.. so I wouldn't build one in a clients system anyway unless they had a specific need and understood the risks and penalties.
On the other hand maybe there is a reason we need it.. and someone will put it in the next post.. :) It's definitely significant technology, but what will we use it for?
Interesting subject.. I know most don't really care, they just want a box and they want it to work. But I like to understand a lot deeper and when I build a system I like to know I'm spending my clients money wisely. Okay, and to me it's fun. I just find this stuff interesting.. Great conversation!
-
I haven't read everything. But I think I can give some information.
I have two computers right here.
1.
Intel i7 980 at 4Ghz (Running on air. At around 60-70 on full load) - 6 cores
X58 board
24Gb RAM
For drives see below
2.
Dual Xeon running at 3,6Ghz - 12 cores
48GB of RAM
1 x 128GB System SSD
1 x 512GB Lightroom SSD
2 x LSI Mega RAID cards
(1 x 4 - 500GB HDDs in RAID 0 as Scratch. Way better than any SSDs. Still cheaper and not much slower. And at least i know the performance will stay the same)
(1x 10 - 2TB HDDs in RAID 10. Amazingly fast. Huge performance and I can keep all my current files in the computer.)
All is in one PC tower. I hate external storage solutions, especially when I can put everything into one easly. I have room for around 8 more drives.
All in all you could say that they are quite different when it comes to price. The X58 costs arund 1/4 or less compared to the dual board.
Some testing obersavtion.
- Capture One the dual Xeon System is faster, but only around 15-30% debending on many different factors.
- Why would I spend more on a dual Xeon System ? Pretty easy. Panoramics. When using PtGui I rendered three different panoramic in batch mode. Each having around 8-16 IQ180 files.
Here the Xeon System kills the X58 system hands down. The X58 needs 10 minutes and 55seconds for them to finish. The Dual Xeon System does the same job in 4 minutes and 44 seconds.
However, there is one much mor important point why I still would spend so much money on a very expsnive system.
I can do many things at once. Render panoramics, process raw files and still Photoshop feels great. Fast and responsive. I can't say that with the X58, which feels more like it wants to quite any second.
If you have any other questions just let me know.
I would not recomend Mountain Mods, I don't find the quality to good. I would suggest ( http://www.overclock.net/case-labs/ ) I think they offer some amazing options and cases. Yes they are expsnive but very well made. )
-
Have you tested the OCZ RevoDrive 3 for yourself ? I haven't but I heared it is great but as system drive you will not be able to see any real difference between it and a lot cheaper SSD. (Sure you see it in benches, but that doesn't matter really)
-
Have you tested the OCZ RevoDrive 3 for yourself ? I haven't but I heared it is great but as system drive you will not be able to see any real difference between it and a lot cheaper SSD. (Sure you see it in benches, but that doesn't matter really)
I currently have a OCZ Revodrive original.. and yes, you can feel a huge difference over the C300's or Vertex 3's on a SATA III port. I've read testers who say what you said, but I just can't agree. Everything is noticeably faster. I can only hope the Revo 3x2 at three times the speed will be even better.. we'll see. They're down to $559 now for a 240gb version and I'll be ordering one for my own workstation next week. I'll also be ordering one of their new 1tb Hybrid Revo's to just for my LR indexex/previews..
-
I know you probably have lot to do, but I would really be intersted in some comparsions tests or user experiences. I am always looking to make my system even faster, but only for sepcifc tasks. For example I dont give a **** about boot time. I don't care if it takes 15 or 25 seconds.
I mean on here everything feels instant . Progam loading times are amazing my LR catalouge is quite fast. SO I would be interested how you feel or see the differences.
Thx
-
Take a look at http://ppbm5.com/
Although application specific, the the results bear out some trends:
Cores - Amdahl's Law (http://en.wikipedia.org/wiki/Amdahl's_law) is clearly beginning to take effect - more cores aren't a guarantee of better individual application performance
Memory - Obviously, more is better, but access times are crucial - it's amazing how much time CPU is simply waiting..... The perfect computer would either be at 100% CPU utilization or 0.
Disk Access - Crucial, but not the deciding factor in many situations.
Like you, i could care less about boot times, etc - whats more important is determine what is the system/application waiting for? Adobe is taking a fascinating approach by embracing the CUDA cores that nVidia offer in their display cards. It's an elegant approach that nicely circumvents Amdahl's Law - I wonder how soon others will embrace this?
-
Adobe is taking a fascinating approach by embracing the CUDA cores that nVidia offer in their display cards. It's an elegant approach that nicely circumvents Amdahl's Law - I wonder how soon others will embrace this?
When they started doing this, it reminded me of the old days of adding a maths co-processor :-)
-
I know you probably have lot to do, but I would really be intersted in some comparsions tests or user experiences. I am always looking to make my system even faster, but only for sepcifc tasks. For example I dont give a **** about boot time. I don't care if it takes 15 or 25 seconds.
I mean on here everything feels instant . Progam loading times are amazing my LR catalouge is quite fast. SO I would be interested how you feel or see the differences.
Thx
Actually, this is part of what I have to do.. :) So I'll share my results. Boot time interests me to a degree, but what I'm really hoping for is to see decreased overall wait times for everything I do, including the programs my caches serve.
I have a Crucial C300 in a faster computer than I have an original Revo in.. the C300 is not a slow SSD, but the slower computer with the Revo feels faster.. significantly faster. Hopefully the Revo3x2 will be even better, and hopefully their Revohybrid will really serve my LR indexes/previews much better..
-
Adobe is taking a fascinating approach by embracing the CUDA cores that nVidia offer in their display cards. It's an elegant approach that nicely circumvents Amdahl's Law - I wonder how soon others will embrace this?
I think it's worth noticing that Adobe is supporting the CUDA's to such as extent. Kinda curious too, we're at 2 PA301w's vs. 3 which to me is still overkill.. but now we can look at the Nvidia dual GPU card. And since half of his use is Adobe Premier will all thos extra CUDA cores translate to decreased task times?
Building another system, you might know this answer. A day trader contacted me to build a several systems that all support 8+ monitors. The V9800's are outside his budget at $3300.. so thinking of the ATI Eyefinity 6 with a bevy of Dell panels that support DP and S-IPS because with 8 monitors your viewing angles will be crowded. Do you know if two of the Eyefinity 6 cards can co-exist and provide a total of 12 video outs?
-
Windows supports up to 64 displays, so no problem - my only concern would be heat, I'd take a carefull look at the PCIe slot spacing....
-
Windows supports up to 64 displays, so no problem - my only concern would be heat, I'd take a carefull look at the PCIe slot spacing....
Right.. but was wondering if you've any experience with the Eyefinity 6 cards? I've only used one, never two. At ATI I get the "they should work", so I guess I'll need to just order them in and see. Really they should, but I like to be sure.
Heat is certainly an issue. His last system with (4) 9800's cooked itself. And it was sloppy at best when it worked. Not to mention noisy.
I'm thinking (2) Eyefinity 6 cards will take less than half the power (less heat), and there's enough spacing on the slots to support Arctic Coolers which will keep things cool and quiet. Should be more than enough power for his needs to keep things zippy. With support for up to 12 displayport monitors he'll have room for expansion.
I can't think of any other clean way to support 8 monitors from the same box.
-
Only 1 here as well - driving 4 displays as one image....
Question: why would a "daytrader" need any form of high performance display card to display (as opposed to render) data, let alone "discovering" the mouse cursor amongst multiple individual desktop displays.
Why not a KVM solution (obviously we'd only be utilizing the Keyboard/Mouse) involving multiple reasonably configured CPU's? A very rational way of determining display focus. Figure 4 medium performance spec machines with dual display adapters, sharing a common IP gateway - need more displays? Add a machine.....
Another advantage - redundancy. Any free form internet "research" results in significant risk. This can be partitioned to individual machines, leaving others dedicated to reporting unnaffected...
Patch/Update management for this cluster? Setup one box with WSUS and instrument it to email you results daily...
Good for you, Good for your customer....
-
Only 1 here as well - driving 4 displays as one image....
Question: why would a "daytrader" need any form of high performance display card to display (as opposed to render) data, let alone "discovering" the mouse cursor amongst multiple individual desktop displays.
Why not a KVM solution (obviously we'd only be utilizing the Keyboard/Mouse) involving multiple reasonably configured CPU's? A very rational way of determining display focus. Figure 4 medium performance spec machines with dual display adapters, sharing a common IP gateway - need more displays? Add a machine.....
Another advantage - redundancy. Any free form internet "research" results in significant risk. This can be partitioned to individual machines, leaving others dedicated to reporting unnaffected...
Patch/Update management for this cluster? Setup one box with WSUS and instrument it to email you results daily...
Good for you, Good for your customer....
Good questions and some good ideas. This customer I think wanted to game and enjoyed the thought of a high performance box as much as he wanted the day trading capabilities. We ended up with Sapphire 5670 Flex cards which can run 4 monitors using an active adapter which Sapphire sells relatively cheaply. I'd still have rather went with the Eyefinity 6 as it's not that powerful a card and I like Displayport cables/ports/monitors for multi displays..
Anyway, I wrote a short review on the system you can see here, (http://www.bangkokimages.com/Articles/Equipment/entryid/1094/Day-Trading-PC-Build.aspx) and have since shipped half a dozen more and twice that many readers have given feedback that they've built their own using this build as a model. Not a great success, but not bad considering I normally do imaging machines. It seems more than a few day traders like a more powerful machine than they really need.
-
I've just been through this process myself for a new workstation, built on current consumer components. It dosn't fit the author's exact spec as it is built on the new x79 motherboard, with the new 2011 processor, but that could be replaced easily with the generation prior MB.
With regards performance,
Once the basics processor and memory is taken care of, the single biggest impact on system performance is DISK. This is even more important for video editing that photo's, but it can be just as important when browsing through that large image library.
My System has been speced with the expectation of editing eithe 80mp files off MF backs, or working though 1000's of files from a wildlife shoot. I was after the best performance I could get from consumer grade components.
Note, this is yet to be built, so I don't have performance specs. come next month.
PC:
ASUS PX79 WS LGA 2011 Motherboard. (Workstation board)
G.SKIL 64gb DDR3 1600 memory in 8x 8gb modules.
Intel core i7-3960X 6 Core Processor
2 x NVidia Quadro 4000 graphics cards
Thermaltake Toughpower 1050W 80+Gold PSU
Thermaltake Chaser MK-1 Case (Though looking at other cases still)
LG 12x DVD/BlueRay Burner
1 x Samsung 512gb SATA III SSD. (Double checking these compared to others for speed, may go OCZ yet) - OS Drive.
2 x 250gb SSD (existing) 1x PageFile, 1 x Scratch disk
Price: Approx US$5k from NewEgg for the new parts. (everything except the 2x 250gb SSD's)
Note: I am deliberatly avoiding INTERNAL raid storage for the image library.
Storage
1 x Burly Systems 8 bay SATA Enclosure with 8x1 TB Hitachi Drives. Configured in Raid 0
1 x 2 Channel SAS Card
This produces 800/MB/S sustained transfer rate for data and costs less than a bunch of SSD's. (Approx $2.5) and is well proven.
Main Image Library and Lightroom Library / Previews live here.
Online Backup
DROBO-PRO 8 Bay populated with 6TB of storage. (This is existing device, and will be the backup array)
I am still working out the best configuration of the SSD"s inside the PC, and that will require some testing between OS, Page, Scratch and other working. I may throw in another SSD for working drive yet, and reduce OS drive size. The bulk storage however, lives ont he external SAS array, and provides very high speed streaming for video editing.
Burly systems devices were in fact mentioned recently here in the Camera-Print-Screen tute when going around Jeff Schewe's lab. The guys there have been extremley helpfull in specing this for me. I havn't tested it myseflf as yet, but will be end of Jan.
Having just gone through the process of carefully selecting components for a similar system, perhaps it is of some use to someone.
regards
Mark
-
Mark: Similar to what I'm currently working with; some issues I've had:
Asus MB as shipped has some minor issues - be sure to update to latest BIOS, also check latest QVL (http://usa.asus.com/Motherboards/Intel_Socket_2011/P9X79_WS/#download) (qualified vendor list) for your selected memory.... I have this board, but received Intel's reference board and built using it...
Why the Quadro 4000's? You can do much better performance and price wise - hardware is increasingly getting more and more application specific.
This obsession over the the windows "pagefile" on a separate disk, don't bother, with that amount of memory, you can turn it off. Take that pair of SSD's and set them up as RAID0 0 for your PS scratch - speaking of that, make sure your devices support TRIM in that mode.
You'll find the overall platform to be very similar in "feel" to a solid X58 - but with longer legs. If your applications are not particularly memory bound, then a good Z67/68 platform will be comparable
Honestly, If you insist on building the ultimate today, this would be the platform, but the cost/performance aspect is difficult to justify. A Z68 platform with a 2600 CPU is *really close* (again, keep application memory access in mind) - I'm waiting on Ivy Bridge with it's trigate transistors (and presumably Thunderbolt).
-
Why the Quadro 4000's? You can do much better performance and price wise - hardware is increasingly getting more and more application specific.
This obsession over the the windows "pagefile" on a separate disk, don't bother, with that amount of memory, you can turn it off.
Take that pair of SSD's and set them up as RAID0 0 for your PS scratch - speaking of that, make sure your devices support TRIM in that mode.
Honestly, If you insist on building the ultimate today, this would be the platform, but the cost/performance aspect is difficult to justify. A Z68 platform with a 2600 CPU is *really close* (again, keep application memory access in mind) - I'm waiting on Ivy Bridge with it's trigate transistors (and presumably Thunderbolt).
a. Having built a number of these now I tend to agree with most of your points, where I don't agree I'm just inclined to do things a bit differently. I agree about the 4000's. Unless the application is directly supported by that board there will be no significant gains in performance over a gaming series. If running multiple GPU cards I really like Lian-li's V1020 case with their optional BS-06x GPU cooler. I've measured an easy 20-25c difference running that cooler with zero increase in noise levels, actually there is a decrease because the GPU fans are running at lesser speeds.
b. Or leave it on.. it won't matter with that much RAM and the SSD's. There will be so little measurable difference either way it's not even worth thinking about.
c. Without a doubt I'd go with a Revo 3 x2 for the system, especially with all that PCIe bandwidth. Even the Revo Hybrid would be a better choice in my mind than the two 256gb's. If I did this I'd also go with an extended ATX board for the extra PCIe slots and extra room for cooling.
d. Agreed. The only reason to go with the 3960x six core is if you're rendering large video files routinely. And the same with the memory, 64gb is way overkill for anything but heavy video rendering.
I'd also much rather have one of the new Seasonic Platinum's PS's over the Thermaltake by any measurement of comparison. I've had both side by side naked and the differences in componentry and build quality are significant.
A lot of this comes down to being more about individual comfort levels/experience with certain gear more than about the small differences in performance. Though, the Revo's provide a performance level significantly better than you're likely to achieve with individual SATA III SSD's. Think about the cooling more than anything, on such a pricey box you want to get your mileage out of it.. and that means thoughtful and complete cooling.
-
A good workstation for a photographer should have a 10-bit capable video card. From direct experience I can tell you that only ATI graphic cards can run Photoshop CS5 using this feature. Of course you will need proper OS, cable, software. http://forums.adobe.com/thread/506853?start=0&tstart=0
10 bit at the moment is a little bit tricky but there are some advantages. If you are using a recent monitor (such as Nec PA271W and PA241W) then you will have noticeable benefits using the 10bit mode in Photoshop.
Running 10bit can help you achieve better results giving better visualization of your work.
A good (and cheap) card is the V4800 FirePro. It works fine, not a very powerful GPU but it does the job. If you need more GPU power (I can easily work with P30+ and P65+ files with the V4800. For my personal photographic needs I am fine, but of course you may prefer a different solution for other reasons.
-
A good workstation for a photographer should have a 10-bit capable video card. From direct experience I can tell you that only ATI graphic cards can run Photoshop CS5 using this feature. Of course you will need proper OS, cable, software. http://forums.adobe.com/thread/506853?start=0&tstart=0
10 bit at the moment is a little bit tricky but there are some advantages. If you are using a recent monitor (such as Nec PA271W and PA241W) then you will have noticeable benefits using the 10bit mode in Photoshop.
Running 10bit can help you achieve better results giving better visualization of your work.
A good (and cheap) card is the V4800 FirePro. It works fine, not a very powerful GPU but it does the job. If you need more GPU power (I can easily work with P30+ and P65+ files with the V4800. For my personal photographic needs I am fine, but of course you may prefer a different solution for other reasons.
a. Are we really at the "should" stage with 10 bit? I dare say the overwhelming majority of pros haven't yet completed a full 10 bit work flow and of course they're putting out professional quality work.
b. This is the key. Can you really see the difference (anyone?) and if so, does it translate to your final media destination, which would be prints, web output, and whatever else you're creating. I don't see it for web output, nor for the lower end lab quality prints, but what about the best inkjet prints and any other higher end type prints?
I suppose what I'm getting at is are we getting the 'value' from 10 bit. After you spend for a pricey workstation card, perhaps you have to upgrade your monitors, maybe your version of CS, prints, etc.. are you seeing a different in your final output which makes that money well spent? I do spent for speed, speed from faster SSD's , fast RAM, machine, etc.. equate to less hours on the workstation which to me is worth money. At least when I'm working.. If I take money from something else to make the investment in 10 bit which for me would require at least a decently fast workstation GPU (I wouldn't want less speed/power than I have now) and two new PA271W's.. would I see a difference in output which I could market for an increased ROI? Sure, if I was replacing my monitors today I'd buy 10 bit, it doesn't make sense not to. But that workstation card is still a pricey chunk of cash.
Sorry for the disjointed paragraph, I need more information to better focus my questions on this..
-
Such a fun thread......
Steve, I finally bit a few days ago and received a Revo 3x2 240GB (http://www.newegg.com/Product/Product.aspx?Item=N82E16820227744) early today ($700), holy s*** .....
Folks: If you are considering messing with anything more than 2 SSD drives, STOP! Just put your O/S, and point your PS scratch to this thing. Save those SATA channels for *storage*.
My only concern with this is De-Duplication (http://storagemojo.com/2011/06/27/de-dup-too-much-of-good-thing/)*, but for O/S this is a good risk.
*De-Duplication is an optimization technique that Sandforce uses to limit the number of writes to MLC blocks. On the face of it, it a great idea, but the Filesystem itself depends on redundant blocks of information, if there is a read error in a critical block, the filesystem simply fails over to a a redundant block. De-Duplication circumvents this; the result is that a single failed block can result in the entire loss of the filesystem.
-
Such a fun thread......
Steve, I finally bit a few days ago and received a Revo 3x2 240GB (http://www.newegg.com/Product/Product.aspx?Item=N82E16820227744) early today ($700), holy s*** .....
If you are considering messing with anything more than 2 SSD drives, STOP! Just put your O/S, and point your PS scratch to this thing. Save those SATA channels for *storage*.
My only concern with this is De-Duplication (http://storagemojo.com/2011/06/27/de-dup-too-much-of-good-thing/), but for O/S this is a good risk
"Awesome" doesn't even begin to describe it eh? Or get a 512gb and put everything on it but your storage, your OS, scratch, work drive, indexes.. it wouldn't be a stretch would it.. :)
Did you get the max IOPS version, the latest in decadent storage needs?
I started 6-7 months back with 1 Revo on the highest end rigs, and since the hybrid came out 2.. which pushed me towards the extended ATX boards. And by using the two Revo's now you can use those 3.5" slots in the case for (8) 4tb SATA III drives with a Promise Tech card for a super fast 24gb RAID 50.. Money seems to be no object with so many.. "if you build it they will come.."
Meanwhile I'm looking forward to Ivy Bridge and upgrading my personal workstation. I'm still using my 'old' Crucial C300 256gb SSD.. with an original Revo in my workshop station. With only 12gb of 2133 RAM, a dual GPU 5970, and an ancient 950 I'm way behind the times. Ivy Bridge will be my next frontier and I'll have got 3 years from my current workstation. A few more months.. :)
Personally though.. I feel like the 3960x is pricey for those who aren't waiting for Ivy Bridge. Ah well.
Share an AS-SSD screen capture.. don't be shy.. ;o)