Mecway PC benchmarks

antte

Hello everyone!

I decided to start a thread for benchmarking different pc's using mecway. I think this could be helpful for someone looking to buy a new computer for FEA.

I used mecway 12 versions ccx solver(not pardiso for easier benchmarking on different systems).

File used for these benchmarks was this excellent bolt assembly posted by prop_design, thanks!
http://mecway.com/forum/discussion/737/bolt-assembly-example#latest

CPU: Xeon 1230v3
Memory: DDR3 4x8GB Hyperx Fury 1600MHZ CL10
Run time: 6 Min 41 Sec

CPU: i7 8700k(Stock clock)
Memory: DDR4 2X8GB Hyperx Fury 2400MHZ CL16
Run time: 5 Min 18 Sec

I hope that someone finds this helpful, and I'm looking forward for other users benchmarks and suggestions for other possible test cases. Personally I'm wondering how well AMD Ryzen systems compare to i7-series.

mmartin

I've calculated the bolt assembly:

CUP i7 6700 CPU 3.40GHZ. 8 cores
MEMORY: 64GB DDR4
Run Time: 2minutes +45 seconds. (MW+calculix 2.15 PARDISO)

antte

Maybe it's better to compare pardiso results, because that's probably the most used solver.

Solver: Ccx Pardiso 2.15
CPU: Xeon 1230v3
Memory: DDR3 4x8GB Hyperx Fury 1600MHZ CL10
Run time: 1 min 43 seconds

It's almost unbelievable how much faster it is!

Sergio

Solver: CCX Pardiso 2.16 using 18 cores
CPU: Intel(R) Xeon(R) CPU E5—2697 v4 © 2.306Hz 2.30 GHz
Memory: 32.0GB Unknown @ 1197MHz

Run time: 1 min 46 seconds


I was doing some CAD stuff whit Solidedge and Teamcenter while solving.

JohnM

Intel i7-4790 3.6GHz 24GB
1min 44sec
ccx 2.16 PARDISO

tk1537

Solver: CCX Pardiso v2.16
CPU: Intel Xeon W-2123, 3.6GHz, 4 cores, 8 logical processors/threads
Memory: 16G, 2666 MHz

Run Time: 1 min 43 seconds

Sergio

Have tried with an old Lenovo T420 laptop:

SOLVER: CCX Pardiso 2.16 using 4 cores
CPU: Intel Core i5 2520M @ 2.50GHz
RAM: 8.00GB Doble-Canal DDR3 @ 664MHz
HD: Hitachi HTS723225A7A364

Run Time: 9 min 45 seconds (!!!)

Looks like the 1:43 mark is very consistent no matter if is run over 8 or more cores.

prop_design

it took 9 min 37 sec on one core

ccx pardiso solver v2.16
intel mkl 2019 update five
amd a9-9400 laptop apu runs at about 2.9-3.2 GHz for this case (boosts so not a constant number)
RAM 8GB DDR4 2400 MT/s
HDD 5,400 rpm

in the past i tested using two cores. it's hard to do a comparison because of the boost frequency. but generally it runs at slower rates when using two cores. it also makes the computer not usable for other tasks. so it's better for me to just use one core.

i'm also interested in the new amd processors. this is an old one. i'm sure the new ones are much faster.

antte

Mecway inbuilt ccx(not pardiso)

CPU: i5 9600k(Stock clock)
Memory: DDR4 hyperx fury ddr4 3200 mhz cl18 2x8gt
Run time: 5 Min 0 Sec

I was surprised that this i5 pc was faster than the i7 8700k. Maybe its the memory speed causing this? i7 8700k is hyperthreaded 6 core with 12 threads and i5 9600k is 6 core without hyperthreading.

barmin

Same Mecway inbuilt ccx:

core i3-8100, stock clock (4 cores, no HT)
24GB RAM Ballistix Sport LT 3200Mhz
Run time: 6min 51 sec.

As far HT goes- Ansys even tells to shut the HT off in BIOS, due to performance issues of how threads "compete" with each other for the data

prop_design

@antte

although it takes a little work, it's worth it to install the ccx pardiso version and the intel mkl. you get a huge speed up. it also allows you to solve larger files. at least that's my experience with it. there is another thread on the topic here;

https://mecway.com/forum/discussion/750/propeller-hub

you can configure it to use as many threads as you like.

antte

I do have pardiso on my pc, and it is really good!
The reason why I did some of these test without pardiso was that I didn't want to spend time to get it working on my friends computer just for testing purposes.

I dont know if pardiso can use more cores than non pardiso, but at least this kind of testing gives some idea of the speed differences between different cpu's. Like I said earlier it's probably best to compare pardiso results if possible because thats what most people use anyway.

disla

CPU: i7 4700MQ (2.4GHz -3.40 GHz) (Refurbished 2015)
Nvidia Gforce745M
Cores-4 Threads-8
DDR3 2X8GB PC3 12800 Dual Chanel (800Mhz)

SSD Update:SanDisk Ultra II 480GB Update
Laptop Total Investment 760,00€.

CCX Pardiso 2.16 using 4 cores
Intel mkl 2019 update five
Run time: 5 Min 17 Sec

disla

Very good numbers, JohnM.

i7-4790 looks really optimized performance versus price.

¿Are you running on Windows platform + Mecway or ccx pardiso on Linux?
Seems that the mem amount is more important than the mem frequency.
¿Don't you think?

¿Why is some people going to Xeon. Is it because the reliavility?

Sergio

I have a DELL workstation at work with Xeon and Nvidia Quadro, I don't see the benefits today (in fact since years) compared with standard good hardware, but, it's a enterprise decition. In the past there was huge difference between performance of workstation hardware and standard (or better, gamer hardware), even CAD vendors certificate his products on workstation configurations (and workstation level drivers), so using standards pc could leave you in a situation where the vendor can't give you a solution due to the hardware. But today is not more the case.

JohnM

I'd like to say that was deliberate - it's just the machine we bought 4 years ago that "made sense".
In the past, I have benchmarked mult-processor speedup, with spooles on CCX2.8. I found that 8 processors was the point of diminishing returns, so even on our servers we do not run more than 8-10 processors.

cmmcnamara

SOLVER: CCX Pardiso 2.16 using 10 cores
CPU: Intel Core i9 9900K @ 3.60GHz
RAM: 64.00GB GSkill RipJaws DDR4-3200
HD: Samsung 970 Pro 512 GB SSD

Run Time: 01:19

prop_design

@JohnM

that fits my vague recollection of ansys plots, when multicore was first coming out. they did comparisons of speedup versus cores. it's nice to have something updated for what we are using.

cmmcnamara

Did another with my laptop:

SOLVER: CCX Pardiso 2.16 using 4 cores
CPU: Intel Core i7 6700HQ @ 2.60GHz
RAM: 16.00GB Kingston DDR4-2400
HD: Samsung 850 EVO 500 GB SSD

Run Time: 01:56

prop_design

interesting results @cmmcnamara. that's really fast for a laptop. interesting not much difference from 4 to 10 cores.

cmmcnamara

@prop_design I agree. I was expecting much lower than that.

I did my home workstation as well now, so that's all the PC's I have:

SOLVER: CCX Pardiso 2.16 using 12 cores
CPU: Intel Core i7 7800X @ 3.50GHz
RAM: 32.00GB Corsair DDR4-3200
HD: Samsung 970 EVO 500 GB SSD

Run Time: 01:34

3rav

May I ask anybody to generate .inp file (full) of this test case.

disla

I was very surprised about cmmcnamara results on his i7.

i7 6700HQ @ 2.60GHz
RAM: 16.00GB Kingston DDR4-2400
Run Time: 01:56

My spects are very close apart from ram speed.

i7 4700MQ @ 2.4GHz
RAM: 16.00GB DDR3-1600
Run time: 5 Min 17 Sec

I have made a second run after a clean restart, antivirus shut down and found time has drop to 2 Min 06 Sec.

That confirms my suspect on the huge impact that the amount of available mem has, probably much more than its frecuency.

JohnM

Intel Xeon CPU E5-2698v4 2.2GHz 64GB
2min 13sec
ccx 2.16 PARDISO
using 10processors out of 40 ~6%CPU reported

JohnM

@3rav
CCX input file of case

cmmcnamara

@disla I think this may also be related to the RAM spec too. DDR3 has nearly half the bandwidth that DDR3 does if I recall correctly and may explain some of the difference as well.

mmartin

I've calculated the benchmark bolt assembly using MW13beta1 + ccx2.16 PARDISO solver:

CUP i7 6700 CPU 3.40GHZ. 4cores:8 threads.
MEMORY: 64GB DDR4
HARD DISK: SSD
Run Time: 2 minutes 30 seconds. (MW+calculix 2.16 PARDISO)
(this is the average time after 3 calculations)

antte

Bolt assembly calculated using Mecway 12 inbuilt ccx.

CPU: Ryzen 5 3600
RAM: 3200 MHz cl16
Run time: 5 min 8 sec

Just 8 seconds slower than the more expensive i5 9600k, amd cpu's seem to work well with ccx.

German

Benchmark runtime 3:42 with Acer Aspire V7
MW13+CCX 2.15 Pardiso solver
Intel i7@2.60GHz 8 threads
8 GB RAM

antte

Bolt assembly test calculated with Mecway 13 + ccx pardiso:

CPU: AMD Ryzen 3700X 8 core 16 threads
RAM: DDR4 Ripjaws V 3600 MHz cl16
Run times:
16 threads 1 min 3 sec
8 threads 1 min 3 sec
6 threads 1 min 5 sec

Above 6 cores there doesn't seem to be a big difference, so I think that ryzen 3600 6 core could be really good value for these kind of calculations.

I used these instructions to get this AMD cpu running well with MKL:
https://pugetsystems.com/labs/hpc/How-To-Use-MKL-with-AMD-Ryzen-and-Threadripper-CPU-s-Effectively-for-Python-Numpy-And-Other-Applications-1637/

Basically you just need to use MKL_DEBUG_CPU_TYPE=5 environment variable to get it running well.

3rav

CPU: Intel Core i7-4800MQ @2.7GHz 4 core 8 threads
RAM: DDR3 32GB @3500MHz
Run times:
1. Intel MKL: 1 min 53 s (best of 3 test)
2. PARDISO Project 6.0: 2min (best of 3 test)