Personal supercomputer recommendation

yeti75

Distinguished
Apr 2, 2010
6
0
18,510
I need to build up a new "supercomputer" for electromagnetic simulation. My present computer, a dual core, dual processor Opteron, takes as much as 5 days to solve a problem. I need a better solution. The primary software that I use takes advantage of between 16 and 255 cores. So, I figure the best solution is to maximize the number of cores (but I might be wrong.) Currently, I am looking at the AMD G34 processors, because they offer 12 cores per processor. I am running Windows XP X64 Professional, and don't want to switch to Windows 7, unless someone can give me a REALLY good argument to switch. (I have not heard one yet.)
I realize that with xp x64 I cannot access more than 2 processors, so that limits things. But, a motherboard with 2 12 core chips would probably do the job. The problem is that most of the motherboards out there are either vaporware or do not support high end video cards, like the ATI HD5970, or do not have cases available.

Any suggestions would be appreciated. I need to buy something in the next few days.
 
There are no mobos available (or in existence) that meet your criteria. The best you can do is something like the Supermicro H8SGL unless you go strictly high-end server grade equipment. Bottom line is that the AMD G34 is made for the high-end market and any mobo made for it will not (possibly never) support a high end gpu like the 5970. Seems to me that you need to re-assess your expectations and/or compromise on the type of equipment you are looking at.
 

yeti75

Distinguished
Apr 2, 2010
6
0
18,510
I would think that high end workstations would be the other target for chipsets like the G34. After all, things like electromagnetic simulation and computational fluid dynamics take all the computing horsepower they can get, as well as all the graphics power available. In fact, some of the software is now running on arrays of many GPU's, because they are faster, for computation, than even the fastest computer cores, like the G34 and Intel 7500 series.
 
If the applications you are using could be run on a gpu, then multiple GTX480's or 5870/5970's would offer more computational power than multicore procs. And, you could get an enthusiast level mobo with the X58 chipset that supports up to 24GB DDR3 memory and an i7-930 at a far more reasonable price than server type equipment.

Perhaps another option could be a Supermicro H8Dli+ which supports 2 six core Opterons but is limited to DDR2 800 memory and only PCIe 2.0 x8 for graphics but would still give you 12 cores.
 

yeti75

Distinguished
Apr 2, 2010
6
0
18,510
Well, I've made my mind up and ordered. I ordered a Tyan S8230 motherboard. This will support 2 12 core AMD G34 2.1 GHz processors. I also ordered 32 GB of ram, in 8x4 GB format. This allows me full advantage of quad channel ram, and allows me to add another 32 GB of ram in the future, if I need to do that. (Highly unlikely.) In this way, I get the maximum ram speed, using quad channel 1333 MHz ram. I also ordered an ATI HD5970 graphics cards. This is one of the fastest cards available today, and will support 3 monitors. I wish the motherboard would allow me to put two hd5970's in crossfire mode, but this is not going to happen. Realistically, one card is probably fine for my needs, which are mainly lots of computing power, rather than extreme graphics power. At the moment, I only have 2 24 inch monitors, but I will probably add a 30" Dell monitor for the main monitor, and use the 2 24" monitors as auxiliary monitors.
For hard drives, I ordered a 160 GB solid state Intel drive for the main system drive. Then, I ordered a 300 GB Velociraptor drive for the data drive. I ordered a 2 TB Cavalry external sata drive for the backup and archive drive. For a case, I ordered a Thermaltake Armor+ VH6000SWA case and a W)1333RU 1200 watt supply.
In addition, I ordered a 2.5 to 3.5 inch drive holder, plus a few sata drive cables.

Did I forget anything? Did I make the best choice for a personal supercomputer? My normal use is running dozens of programs, both high end engineering simulation packages, as well as generic things like firefox, wordperfect, word, excel, powerpoint, etc. The biggest problem was EM simulation programs that took 5 to 6 days on a dual core, dual processor computer. These programs support 15 to as many as 255 cores.
 
I don't see the 5970 ....a killer GFX card for gaming but of no real use in your stated situation. Your stated usage screams GPGPU and CUDA computing

http://www.acceleware.com/index.cfm/our-products/em-solvers/

"The Acceleware Electromagnetic (EM) Solvers connect the industry leading EM software applications to the massive parallel processing capabilities available with NVIDIA graphics processing units (GPUs). This amalgamation delivers unprecedented computational capabilities in a form factor that features a very low Total Cost of Ownership and a high ROI. The Acceleware EM solutions bring flexible capabilities directly to engineers and product designers and delivers accelerations from 10X to 50X."

http://www.remcom.com/video-center/2010/1/14/remcoms-xstream-gpu-acceleration-for-xfdtd.html

"This video shows a side-by-side comparison of cell phone simulations with and without Remcom's XStream GPU acceleration. The simulations are run on an HP XP9400 Workstation 2x Dual-Core Opteron 2216. The XStream simulation has the added power of 2x NVIDIA CUDA-enabled Tesla C1060 external boards."

http://www.remcom.com/press-releases/remcom-announces-massive-speed-increase-with-updated-gpu-acc.html

Remcom announces that its XStream® GPU acceleration now provides the capacity to solve electromagnetic simulation problems up to 300 times faster than a modern 64-bit CPU. In addition, Remcom announces this updated version of XStream will now be bundled at no additional cost with XFdtd Release 7.0 (XF7), the latest release of Remcom’s 3D electromagnetic simulation tool.

XStream tremendously improves EM simulation performance by leveraging the powerful NVIDIA graphics processing units (GPUs) available on modern video cards to make ultra-fast FDTD numerical computations. Remcom’s new generation of XStream enables XF7 calculations to finish in minutes as compared to hours using a CPU only. This solution has the potential to save customers months of simulation time in delivering products to market.


http://www.nvidia.com/object/personal_supercomputing.html
http://www.nvidia.com/object/cee.html
 

yeti75

Distinguished
Apr 2, 2010
6
0
18,510
If I could afford the CUDA and/or the GPGPU solution, that would be the way to go, but I don't have that kind of budget. Also, I don't do enough time-consuming EM simulation to justify a dedicated solution. I also do a lot of circuit board layout, and generally have my computer tied up with at least 8 to 10 different open programs, some of which are compute intensive engineering simulation or circuit board layout programs, and some are just generic e-mail or web browsers, although I usually have at least 7 or 8 internet explorer and firefox sessions open at any given time, bringing the total number of processes into the hundreds, and open programs into the dozens. So, to keep everything moving along, the more cores the better. Also, circuit board layout is very, very demanding on video cards, as is the visualization of the EM simulations. Also, 3d cad is demanding on the video board, especially since I run 2 monitors, and plan to add a 3rd. So, overall, I expect to be happy with my solution, although I agree that if all I did was EM simulation, and if I had the budget, the CUDA GPGPU solution would be better.