What is the best expansion for a Workstation computer to get more out of Particle systems? (fumefx,realflow)

Joe Presswood

Honorable
May 11, 2013
7
0
10,510
What is the best expansion for a Workstation computer to get more out of Particle systems? (fumefx,realflow)
To get more out of particle systems like Fumefx, Thinking particles, Realflow, what would be the best direction?
1.Add more Ram? If so, how Much... 64 to 128 gb?
2.Add another Graphics Card?
3.Add another E5-2660 Processor to make it a 16 core 32 hyperthread?
I have about $2000.00 to play with. Where should it go?

My Computer config:
Z820 Workstation
Windows 7 Professional 64bit OS US
HP Z820 1125W 90 Efficient Chassis
Intel Xeon E5-2660 2.2 20M 1600 8C 1 CPU
NVIDIA Quadro K2000 2GB 1st GFX
24GB DDR3-1600 4x4GB+4x2GB 1CPU RAM
1TB 7200 RPM SATA 2nd HDD
500GB 7200 RPM SATA 3rd HDD
16X DVD-ROM SATA 1st ODD
256GB SATA 1st SSD
 
Solution
Joe Presswood,

The answer to the best optimization of your HP Z820 for fluid / gas simulation is more complex than formerly as more applications are using GPU acceleration. This kind of simulation is CPU based, and formerly I'd suggest simply to add the 2nd CPU, but today there is the amazing option to add a GPU coprocessing unit. Formerly, you needed to be using NAMD (molecular biology, protein folding etc) or whether simulation applications, writing custom C++ code to use it, but today the list of GPU-accelerated software is becoming mainstream and Autodesk and Adobe plus many others can utilize NVIDIA CUDA-based co-processing units>

http://www.nvidia.com/object/gpu-accelerated-applications.html

> If your applications are...
Joe Presswood,

The answer to the best optimization of your HP Z820 for fluid / gas simulation is more complex than formerly as more applications are using GPU acceleration. This kind of simulation is CPU based, and formerly I'd suggest simply to add the 2nd CPU, but today there is the amazing option to add a GPU coprocessing unit. Formerly, you needed to be using NAMD (molecular biology, protein folding etc) or whether simulation applications, writing custom C++ code to use it, but today the list of GPU-accelerated software is becoming mainstream and Autodesk and Adobe plus many others can utilize NVIDIA CUDA-based co-processing units>

http://www.nvidia.com/object/gpu-accelerated-applications.html

> If your applications are on this list, I'd suggest looking into an NVIDIA Tesla GPU unit. The great feature of these is that no one knows what to do with them, so the previous models can be amazingly inexpensive >

http://www.ebay.com/itm/IBM-NVIDIA-TESLA-M1060-VIDEO-CARD-4GB-PCI-E-X16-VIDEO-CARD-43V5909-/190842360798?pt=PCC_Video_TV_Cards&hash=item2c6f1753de

> an eBay listing for someone with 10 Tesla M1060 4GB coprocessors- new $2,200, but today "Buy It Now" for $85! For comparison, the current 6GB Tesla K20 costs, $3,300 and the K20X, (18,688 of which were used in the "Titan" supercomputer) is $7,500. This is not to say you can use these to advantage- or at all, but obviously if so, buy as many as you have slots for. There is also the Tesla C1060, C2050, C2070, C2075, and K20 to consider. The K20 is fantastically capable- for the right software, new is $3,400 and used is still $1,200- not impossible in your budget.

Given the processor emphasis of your use, it makes sense too to add the second CPU used for about $800-900. That is an extremely good CPU too > no. 10 on the Passmark CPU benchmark list- out of about 1,700 CPU's. ( About $1,300 new.)

The Quadro K2000 is a very good graphics card for your use, if in combination with something like the M1060. If so, add as many as you have slots, add the second E5-2660, upgrade RAM to 32GB. Remember to buy a matching second fan and/or heatsink for the second CPU. I would have no qualms about buying a Xeon used, and an E5-2660 today costs about $850-900.

Another strategy would be to buy a used K20 ($1,200)or C2050 /2075, and a second, used E5-2660 ($900 with second heatsink), +8 GB RAM to make 32GB ($75).

If you're not able to use GPU co-processing, for your $2,000, you might buy a used K5000 (about $1,300) which will act as a GPU accelerator, used E5-2660 ($900), and if your file size can be large, perhaps upgrade to 32GB RAM. -a bit over your budget,.. You don't mention ECC, but with particle / gas simulation, the error correcting can be your friend- certainly not faster, but more accurate.

One component not mentioned is the monitor, and in my view, a very useful upgrade wold be to consider- it you don't have it, adding a 27'" 2560 X 1440 monitor for the image window and running a second monitor for menus and viewports.

Cheers,

BambiBoom

[Dell Precision T5400 > 2X Xeon x5460 quad core @ 3.16GHz, 16GB ECC, Quadro FX 4800 (1.5GB), WD RE4/ Segt 500GB] [Windows 7 Ultimate > AutoCad, Revit, Solidworks, Sketchup, Adobe CS, Corel Technical Designer, WP Office, MS Office] [ Monitor> HP 2711x / 27" 1920 X 1080)
 
Solution

Joe Presswood

Honorable
May 11, 2013
7
0
10,510
Thank you so much for the answer, gives me a solid blueprint for upgrading.
I have the double monitor set up and love it, I can't work any other way now.
I appreciate your detailed response.
Thank you,
Joe
 
Joe,

I've not used these GPU processors, and don't know the subtleties, but I believe the M1060 is the server version of the C1060 > I think it uses the Tesla (definitely not Fermi) C1060 architecture with 240 CUDA cores, 512-bit, about 100GB/S memory bandwidth, and 4GB RAM. This kind of processor will only make a substantial difference with GPU- accelerated applications that include drivers that can manage the CUDA cores as parallel processing cores- which are running at about 1.3GHz. If the module works, it will massively extend the double precision calculation and in a wide bandwidth, which in my limited understanding is exactly what particles, fluids, and gas simulations have for breakfast, lunch, and dinner. Of course, the newer models have many more cores at higher speeds- a a C2075 has 1,500, a K20 has whatever it is , 2,500 and so on. Except for the C2075, these do not have any video outputs- the monitors have to be connected to a conventional graphics card like your K2000.

I wish I knew enough to be more specific about the world of GPU co-processing as all I read about it suggests it's the future of high performance workstations, moving them into the realm of Personal Supercomputers. Intel is getting in to this too, with it's Phi something or others.

Cheers,

BambiBoom
 

TRENDING THREADS