[Carlos Wilson]

Hello,

I’m new at CFD. Looking into investing on a workstation for my simulations and image processing any suggestion? Planned is to use Abaqus and maybe also Ansys and matlab. To analyze the mechanics of building structures for example solid composite beams with regard to friction impact between different material layers.

I wish to use GPU computing if possible but my budget is limited. I have found information about that the Quadro K5000 is supported by the software’s but it costs too much for my budget. Is there may be a cheaper graphics card for GPU computing?


Precision tower 5810
CPU: Intel Xenon E5-1650 v3 @ 3.5GHz (6-core) or 8-core
Graphics: NVIDIA K620 2GB, but looking for some alternative which support GPU computing
Memory 4 x32G (128GB)
SSD 1TB
and a cheap 5 TB disk.

Thanks in advance.
Best regards,

[flotus1]

Whatever you do, do not buy a Quadro K5000 for GPU computing. While it may be officially supported, it is not a compute card at all with a DP peak performance of only 90 GFLOPS. Plus it is outdated by now and has a rather low amount of memory
One of the last Nvidia Graphics cards with full DP compute capabilities was the Titan Z. You can buy them used for a fairly reasonable price. However, I have no idea if it is (officially) supported by your software or if it can be used to accelerate the simulations you want to run.
My advice would be to maximize CPU performance first. If you want to use GPU acceleration you need to investigate further if it can be used for your specific type of simulations. Just ask the software vendor.

[Carlos Wilson]

Hello,

thanks for your reply Alex. I kept looking for parts, probably can the choice be optimized more without increasing the cost?

Basically will I use the computer to make various mechanical simulations in Ansys - static structural in my studies.

Thanks in advance for comments.
Best regards,




Precision 7810 MT
Processor Intel® Xeon® E5-1650 v4 (6 cores, 3.6 GHz, 4.0 GHz Turbo, 2400 MHz, 15 MB, 140 W)
Or is it suitable with two 4 cores so totally 8 cores?

Graphics NVIDIA Quadro® K6000 at 12 GB (2 DP, 2 DL DVI-I) (2 Adapter DP to SL-DVI)

Memory 128 GB (4 × 32GB) DDR4 RDIMM memory at 2400 MHz ECC

Hard disk 2.5-inch SATA SSD 1TB, Class 20

Secondary hard drive 3.5-inch SATA hard drive 4 TB and 5400 r / min

[zx-81]

Originally Posted by Carlos Wilson

I wish to use GPU computing if possible

The problem with GPU computing is the very limited RAM compared to CPU computing. In my experience, the amount of avialabe RAM is at least equally important than processing speed of the compute core(s).

While GPUs have the potential for ten times as much raw processing speed than CPUs they also have ten times less RAM (very roughly 10 GB for large GPU vs. 100 GB CPU) In this regard, a CPU based approach is a more balanced approach to high performance computing.

Of course, the one situation where GPUs are really superior is when used in a multi-GPU cluster with domain decomposition where each GPU only holds part of the domain within its limited (but very fast) RAM

So, unless one does not planning for multi-GPU scenarios I would recommend a CPU-only based approach for a work station.

It can also be assumed that in the near future (think AVX512) the gap in processing power between CPU and GPU will close, whereas the advantage of more RAM and simpler proramming model will remain with the CPU