Help to decide which CPU purchase for scientific calculations

[IkSak]

Hi.

I am a researcher in Nonlinear Dynamic Behavior of Reinforced Concrete Buildings and my lab just received funding to upgrade our existing computers which are more than 5 years old. We work with a software called OpenSees is an Open Source software that can be really taxing for any system. It can even use GPUs for computing purposes. We are looking for information on the CPU performance in OpenSees in order to make the best choice at the time of buying the new PCs, unfortunately we have not found any in the OpenSees forums.

For this reason, I decided to ask for help on the tom's hardware forum. I created an small building model that is more than enough for benchmarking purposes. I uploaded the necessary files to this link:[like removed]
Would you be so kind to help us? it won't take more than two minutes. All you have to do is download:

https://app.box.com/s/ngp27x0maniz3ju3i6ev2lda0swmnmko

1) download the file, extract it in a folder.
2) Right click and open with administrator privileges: ActiveTcl8.5.18.0.298892-win32-x86_64-threaded. Make sure that the install directory is C:\Program Files\Tcl and not C:\Tcl as appears as default.
3) run OpenSees.exe (included in the file)
4) in the window that will open type: source runpushoverz.tcl. It will return a time in seconds in the bottom of the window.
5) post your result indicating which CPU you have.

I kindly appreciate your help.


It will use one core of your CPU, so you can run it in the background.

Thanks in advance to all, your help is appreciated.

 

[BadActor]

I tried to run your benchmark and came up with an error that I'm missing a .dll file. In any case, if your software performs better with stronger single cores, you can check the PassMark benchmarks to compare the relative performance between processors. Does the software use multiple threads and do you require high precision?

http://www.cpubenchmark.net/singleThread.html

 

[rgd1101]

http://opensees.berkeley.edu/community/viewforum.php?f=2&sid=1f2bfccac67db3f16764e44987be8cba
did you try asking the opensees forum?

 

[IkSak]

Hi.

I also uploaded the file there a couple of hours ago, I'm trying to get as much information as I can and in this forum people have better rigs than there. Research budgets are not as good these days.

Thanks for the tip about Passmark BadActor.

I will look at the dll missing file.

 

[IkSak]

Thanks all of you for your help. I manage to perform some benchmarking by myself and some help from some friends and I have found there is a positive correlation between the PassMark and the OpenSees, though apparently, OpenSees shows a better scaling with frequency and possibly architecture.

Here are some of the results I've found:

i5 3570K at stock = 145s
i5 4440 at stock = 141s
i7 4790 at stock = 114s
i5 3570K at 4.2GHz = 132s
i5 2400S at stock = 175s
i7 3720QM (Macbook Pro Retina) = 239s

Based on the advantage between the i5 4440 I think that maybe some of the improvements between Ivy and Haswell may have some influence and I wonder if Broadwell or Skylake processors may bring further improvements.

If someone has one of these processors, would you be so kind to download this file:

https://app.box.com/s/ngp27x0maniz3ju3i6ev2lda0swmnmko

1) download the file, extract it in a folder.
2) Right click and open with administrator privileges: ActiveTcl8.5.18.0.298892-win32-x86_64-threaded. Make sure that the install directory is C:\Program Files\Tcl and not C:\Tcl as appears as default.
3) run OpenSees.exe (included in the file)
4) in the window that will open type: source runpushoverz.tcl. It will return a time in seconds in the bottom of the window.
5) post your result indicating which CPU you have.

I kindly appreciate your help.

 

[PreevBR]

Thanks all of you for your help. I manage to perform some benchmarking by myself and some help from some friends and I have found there is a positive correlation between the PassMark and the OpenSees, though apparently, OpenSees shows a better scaling with frequency and possibly architecture.

Here are some of the results I've found:

i5 3570K at stock = 145s
i5 4440 at stock = 141s
i7 4790 at stock = 114s
i5 3570K at 4.2GHz = 132s
i5 2400S at stock = 175s
i7 3720QM (Macbook Pro Retina) = 239s

Based on the advantage between the i5 4440 I think that maybe some of the improvements between Ivy and Haswell may have some influence and I wonder if Broadwell or Skylake processors may bring further improvements.

If someone has one of these processors, would you be so kind to download this file:

https://app.box.com/s/lhd7jo64915xgurrq9hmxwp3hax7jigu

1) download the file, extract it in a folder.
2) Right click and open with administrator privileges: ActiveTcl8.5.18.0.298892-win32-x86_64-threaded. Make sure that the install directory is C:\Program Files\Tcl and not C:\Tcl as appears as default.
3) run OpenSees.exe (included in the file)
4) in the window that will open type: source runpushoverz.tcl. It will return a time in seconds in the bottom of the window.
5) post your result indicating which CPU you have.

I kindly appreciate your help.

FX-6300 @3.9GHz, 194s
FX-6300 @4.1GHz, 187s

 

[BadActor]

I'm curious to see the results of the benchmark on my system, but I'm getting an error after running OpenSees.exe, "PushInfo.tcl" - no such file or directory exists. Any suggestions?

 

[Inkiad]

I am curious how an xeon and skylake handle this.

 

[IkSak]

I'm curious to see the results of the benchmark on my system, but I'm getting an error after running OpenSees.exe, "PushInfo.tcl" - no such file or directory exists. Any suggestions?

Did you download the latest file?

I will perform the steps I wrote when I get to my office.

By the way, I'm also curious about how it goes in a Xeon

 

[IkSak]

I'm curious to see the results of the benchmark on my system, but I'm getting an error after running OpenSees.exe, "PushInfo.tcl" - no such file or directory exists. Any suggestions?

My mistake, I apologize.

I fixed the link in my previous posts adding the Pushinfo.tcl file to the .zip download. Please tell me if it works this time.

The new link is this: https://app.box.com/s/ngp27x0maniz3ju3i6ev2lda0swmnmko

 

[IkSak]

I've been looking for information about Skylake and in scientific benchmarks such as LinPack and Euler3D it shows important benefit of i6700k over i4790K and I wonder, is there some in the forum with a Skylake that could do the test I placed in the first post?

Thanks in advance.

 

[PreevBR]

As question is still not solved, unmarked best answer (Which I think was also placed wrongly on me?)

 

[BadActor]

I've tried to run it on two different machines and still get the file missing error even though I can plainly see the file in the folder I extracted to (pushoverinfoz.tcl).

 

[IkSak]

I'll try to see if there is something wrong with the file. Did you install the ActiveTcl as administrator? I'll post the individual Pushinfo.tcl file tomorrow to see if there's something wrong with the one I uploaded.

Thanks for your help

 

[IkSak]

I've tried to run it on two different machines and still get the file missing error even though I can plainly see the file in the folder I extracted to (pushoverinfoz.tcl).

Hi.

Lots of thanks for your interest. You can download the file pushinfo.tcl from this link: https://app.box.com/s/z6op5salzin6owsriwyb8n1vq60az070

If it doesn't work, I think that before trying that, the easiest way for you to run the file is this:

1. Open the runpushoverz.tcl file using any text editor, even the notepad is ok. Just right click and Open with...
2. Go to the line where it says: source PushInfo.tcl and delete the line.

3. Paste the following text:
set IDctrlNode 1000006
set FreeNodeID 1000006
set LBuilding 18.000000

4. Try again.

Let me know how it goes for you.

 

[IkSak]

I got an i7 3770 (non K) processor to run the benchmark and it scores 140s. Based on this result and the 145s of the i5 3570k, do you think there is some benefit of the Hyperthreading or maybe from the larger L3?

 

[BadActor]

OK, third time was the charm.

i5 6600K @ 4.4 ghz returned 95 seconds.

Not sure if RAM frequency is a factor, but I'm running DDR4 @ 2400 mhz.

 

[IkSak]

Thanks everybody in the forum. Up to this point we have some very interesting possible conclusions? we can draw, so let's recap the data we have so far:

i5 3570K at stock = 145s
i5 4440 at stock = 141s
i7 4790 at stock = 114s
i5 3570K at 4.2GHz = 132s
i5 2400S at stock = 175s
i7 3720QM (Macbook Pro Retina) = 239s
i5 6600K at 4.4GHz = 95s
i7 3770 at stock = 140s
FX-6300 at 3.9GHz = 194s
FX-6300 at 4.1GHz = 187s

Based on these results, here is my take about CPU performance OpenSees:

- Hyperthreating has a minimal influence (evidenced by i5 3570 nd i7 3770)
- CPU frequency has a good influence (evidenced by overclocking the i5 3570k and the FX 6300)
- CPU architecture has the highest influence of all factors (evidenced by the jump from i7 3770k to i7 4790k and by the differece between the i5 3570k and the i5 6600k).
- AMD CPU fall behind Intel in performance, though they do it well if we considered that their price is competitive.

What are your thoughts?

 

[Inkiad]

I am seeing ownage from Skylake. Hyper Threading is also not that big of a deal. Go with an i5 6600K then.

 

[rgd1101]

HT is only good if the thread is not using all the resource for that core, look like it might be using most of it.

 

[IkSak]

Thanks everyone for your help.

Apparently the best price to performance ratio in this benchmark is offered by the i5 6600k, and the latest PC builds suggest that this CPU is a solid choice, however, as seen by the benchmarks, clock speeds are very important for OpenSees, so, I wonder:

Is it possible to buy a Z170 motherboard and configure the i5 6600K cores to have a turbo for all cores of 4.4GHz (as the OC of BadActor), but to change reduce the frequency if necessary. I apologize if this is a simple question, but I have to consider that we'll be buying more than 20 computers and energy bills these days are not cheap .

Another quick question. The computers to be bought are not meant for gaming, so I wonder, I saw in the PC build article that the Gigabyte Z170 Gaming 5 is a popular choice, but I wonder if the ASUS Z170-P or other board can do the job, since saving US$40 on the MB of +20PCs is important.

Thanks in advance.

 

[cbag]

Why did you not conduct a GPU test? Your first post said it can use GPU. They might be much faster.

 

[IkSak]

Currently, most of our models are small and they do not benefit much from GPU. Our priority is upgrading the PCs we currently have on the lab, which are outdated for today standards (Core 2 Duo E8400) and are really slow, even running the small benchmark I put in this forum.

The GPU accelerates big models, but only when you work really big ones. If someone wants to see how much your GPU can crunch numbers: http://opensees.berkeley.edu/wiki/index.php/Mega-Tall_Building_Benchmark_Models

 

[BadActor]

Thanks everyone for your help.

Apparently the best price to performance ratio in this benchmark is offered by the i5 6600k, and the latest PC builds suggest that this CPU is a solid choice, however, as seen by the benchmarks, clock speeds are very important for OpenSees, so, I wonder:

Is it possible to buy a Z170 motherboard and configure the i5 6600K cores to have a turbo for all cores of 4.4GHz (as the OC of BadActor), but to change reduce the frequency if necessary. I apologize if this is a simple question, but I have to consider that we'll be buying more than 20 computers and energy bills these days are not cheap .

Another quick question. The computers to be bought are not meant for gaming, so I wonder, I saw in the PC build article that the Gigabyte Z170 Gaming 5 is a popular choice, but I wonder if the ASUS Z170-P or other board can do the job, since saving US$40 on the MB of +20PCs is important.

Thanks in advance.

There's not been a whole lot of "real world" difference in performance from one board to another in the reviews I've seen, and pretty much any Z170 motherboard will get a Skylake CPU to 4.2 to 4.4 GHZ with proper cooling. No need for the high end overclocking boards. The Asus Z170 P would probably be a good board for your application. Gigabyte also has the GA Z170X HD3P which is a budget board and gets good reviews, you might consider it.

You can overclock the CPU and have the frequency and voltage drop when at idle, be sure to set the power plan to "Balanced."