[Echidna]

This is the workstation i am going to build. The system is quite old but i was in need of a good workstation while waiting for the new technology to mature and perhaps go down in price (AMD Epyc, Xeon Scalable).

SuperMicro X8DTi-F motherboard
2 x Intel Xeon X5670 (LGA1366)
6 x 16GB DDR3-1366 RAM ("catching" the 6 memory channels)
AMD Firepro W7100 GPU
EVGA Supernova GQ 650w (80+GOLD)

The total cost of the above system was a bit more than 600Euros, so i think it was a good vfm.

I will use it mainly for FEA, maximum 3million elements models.
Do you think it was a good choice?

[Echidna]

I have just fit 6x16GB DDR3-1333Mhz ECC DIMMs.
For some reason the system cannot recognise the full amount of RAM, it either reads 80GBs or 64GBs. I will see how i can come over this.

I tried to compare my system to the "slowest" configuration that took part on the official Ansys benchmark.
Here is the benchmark: https://tinyurl.com/y8k32afl
My case was similar but i was running only with 64BGs of RAM. The case only demanded 33GBs so no paging at all.
Linear Static, 12M DOFs, PCG solver.

Here is the slowest configuration from the ANSYS benchmark.

Lenovo ThinkStation P510 Workstation, E5-1680v4 CPU, 128GB (4x32Gb) 2400MHz GB RAM, Win10 Px64, NONE, DMP config, 8 Total CPU Cores, 0 GPUs

This is one modern 8-core CPU versus an "ancient" (8 y.o.) 12-core dual Xeon configuration.

The result for the modern machine was around 200runs/24h which translates to an elapsed time of 432sec.
My result was 710sec with less than ideal memory configuration.
And this is a 600$ workstation. What do you think?

[evcelica]

I'm surprised, that is very good performance! Especially for something so old, those Nehelem and Westmere CPUs were pretty slow I thought. I'm also using old equipment, 6 year old i7-3930K, which isn't as old as yours.
For comparison I ran the benchmark Distributed on 4 machines:
4 machines; 8 cores: 372 sec
4 machines; 12 core: 328 sec
Single machine 4 cores: 661 sec
See if you can get that memory working properly and see how much of a difference it makes.

[Echidna]

It performed pretty good on Calculix FEA benchmark. Still no solution for the memory issue though...
170.64 which is a bit worse that the i7-7800X!
http://www.tomshardware.com/reviews/...u,5167-12.html

[evcelica]

I found the benchmarks online, (after a lot of searching) you can download them from the ANSYS customer portal if you create an account using your customer number.

https://support.ansys.com/AnsysCusto...cal+Benchmarks

Here is the page with Fluent+CFX+Mechanical Benchmarks:
https://support.ansys.com/AnsysCusto...ort/Benchmarks

I guess I should add, the easiest way I found to run them is to copy the input and DB file for one benchmark into a working directory, then run it with the "Mechanical ADPL Product Launcher" in batch mode. Just pick "ANSYS Batch" for the simulation environment at the top. Then choose the working directory where you copied the files, leave the job name as is, and pick your input file as input. I name the output file "solve". Underneath that, uncheck the "include input listing in output" box, as this does nothing but make the output file too large. Finally, pick the CPU options under the HPC Setup Tab, and hit run. When finished, check the solve.out in the working directory and it will tell you the CPU time and the "Elapsed Time" at the bottom, which is what we are interested in.

[Echidna]

Hello i have bought a used server to build a cluster with my workstation. Right now i am searching for an Infiniband card. I have found the QLogic QLE7340 single port 40 Gbps card. Do you thi k it'a good option? Is it a problem that it oonly has one port per card? And finally what type of cable i will need?

Sent from my SM-A310F using CFD Online Forum mobile app