High performance NX CAM workstation. Parallel Generate and multi threading. 1

[Dani_Andres]

Hi,
We are gathering information for our new Workstation.
We do not need rendering performance, neither we use to have big assemblies, but we use to make high precision complicated machining programs.
Therefore I think our most important factor would be single thread benchmark as in NX specs there is no CAM reference on SMP or multi threading performance.
But that's only my guess as I could not find any detailed information for Workstations mostly dedicated to CAM process.

Finally we would like to know if the "Parallel generate" option in CAM does use as many cores available as possible or it just uses one.
It would be useful to use multi threading in this case as you can calculate several machining processes in each core at the same time.
It's something to have a thought about as AMD and Intel handle clock speed and multi threading very differently.

By the way, in the last NX update we have problems with this option, sometimes parallel generate works perfectly and sometimes it just refuses to calculate anything
without giving any error message. Is there any workaround?

Thanks in advance,
Dani.

 

[Tingsryd]

Hello,
There is a lot of SMP and parallel processing in NX CAM. However there is still some single threaded processing so you will have to balance your decision between core speed and core count.

Z-level and contour operations use both single and multi threading.
Gouge check is multi threaded.
Verify Tool Path uses a lot of multi threading.
Parallel Generate uses as many cores as are available, with caveats. For example if have 10 operations that do not use IPW they will generate using all cores. Operations using IPW will generate one at a time because subsequent operations rely on the IPW from the previous operation.


Make sure to disable hyper threading in the computer bios.
Make sure to enable smp in the Windows variables with UGII_SMP_ENABLE=1 (there are some modeling functions which can use smp as well)
Make sure to enable Create Multiple Processes in customer defaults-Manufacturing-Operation-Tool Path tab.
Make sure to set Maximum Concurrent Processes in customer defaults-Manufacturing-Operation-Tool Path tab to set the max number of cores to use. (I use all cores or all but one)

My company has always used Xeon processors but the new AMD Ryzen 9 cpu's with 12 or 16 cores look promising so I just might build a computer to test the Ryzens.




NX 12.0.2
Testing series 1899

 

[Dani_Andres]

Hi Tingsryd,
That's a very good answer, thanks!
Actually all you say it has a lot of sense and I don't know why there is so few information regarding that points.
I've tested your parameters and I've seen how NX uses more cores than I expected.

My guess was that at least it will use SMP when using Parallel Generate, but now that we are sure that is much more than that I think
we will go for the new AMD Thread Ripper series, probably the 3960X. With 24 cores and 3,8Ghz base clock I think will be a tough challenge for any Xeon workstation.
It would be nice if anybody, using NX CAM, has an experience with that system and wants to share it.

Where did you gather all these information by the way?
I tried to search the web and I found nothing about it.

Thanks again,
Dani.

 

[Tingsryd]

Hello,
Much of the information is in the What's New documentation for every version and I agree this is an area which is lacking in the docs. Even the Find Default search function in Customer Preferences does not reveal these important settings. As mentioned earlier there are some design benefits with smp such as boolean operations and perhaps silhouetting. I reiterate disabling hyperthreading in the bios; supposedly the code isn't optimized for it and the performance is slightly lower with it turned on.

Generally Parallel Generate works very well but occasionally an operation will be shown as generating but the cpu is idle and it never finishes. This is annoying but thankfully it is not a common issue. Also, sometimes the benefits of Parallel Generate seems to have diminishing returns on very large, 3D workpieces as the prt becomes huge. Thankfully it generally works great for most users and I highly recommend using it. Where it really shines is regenerating a large program with large and/or many operations; you can start parallel generating then be free to work on other things.

In my previous post I mentioned the setting for Maximum Concurrent Processes. If you need to do other things while generating many operations you might want to set this option to one less than the total number of cores you have so your desktop isn't sluggish. Of course if we are doing something which is fully multi threaded we won't have a choice in the matter.

I think Xeons used to have better FP performance than desktop processors but now Core i9's and the new AMD's have done very well and I think some of the new AMD's support ECC memory as well but I am not sure about multi-socket support. This situation reminds me of my early days with Unigraphics. We used HP RISC workstations with RISC processors and at the time were much more powerful than their PC CISC counterparts. As time went on PC's caught up and surpassed RISC processors. Thankfully, so. The latter were so expensive! I recall paying $25K for used RISC computers!

NX 12.0.2
Testing series 1899

 

[looslib]

I think the reason Xeons are the 'preferred' processor is not just clock speed or FP operations, but internal cache. A few years ago the internal cache on a Xeon was beyond reach of the I-series processors.
I always have IT turn off hyper-threading as it basically splits each core in half.

"Wildfires are dangerous, hard to control, and economically catastrophic."

Ben Loosli

 

[UserCFAO]

I would be a little more reserved on the Nx cam performance , they have declined significantly a lot over the years, especially if you have a lot of operations , if you have the opportunity to use Nx12 (I don’t know for the higher versions...) the access to some menus are very slow (especially access to non-cutting move menu) if I have a tip to give you not to waste your money in RAM Nx does not use much, prefer to put more money on a good processor
Great post Tingsryd , thank you

 

[Dani_Andres]

Hi again,
Today we have deactivated the Hyper Threading option from the BIOS as Tingrsryd suggested and it seems it has improved the speed somehow.
Like UserCFAO wrote, we also have the same speed problem accessing menus from tool path options.
I haven't been able to guess where is the bottle neck on this issue, it does not seem to be HDD related and it's way to slow to come from CPU.
But after HT is no longer active it has been improved and it's now not bothering us like before.
By the way, we have seen that Adaptive machining it's also using multi threading and that's also a big CPU consuming type of toolpath.

Going back to CPU's, If you compare the performance of High End CPUs at

https://www.cpubenchmark.net/high_end_cpus.html

you get quite confused of why the Workstations are still using Xeon processors.
The closest models to AMD are double or triple the price of them. The only one that could be close to AMD 3960X is Intel Core i9-10980XE @ 3.00GHz and it's still not at the same level
at general performance. When you look at caché size, as loolib mentioned, AMD is also quite ahead of Intel too.
At the same price range you have Intel Xeon W-2295 @ 3.00GHz and AMD Threadripper 3960X, the first having 24.75 MB of cache and the second having 128MB.
You can take a look at here:

https://www.cpu-monkey.com/en/compare_cpu-intel_xeon_w_2295-1169-vs-amd_ryzen_threadripper_3960x-947

https://www.cpubenchmark.net/compare/Intel-Xeon-W-2295-vs-AMD-Ryzen-Threadripper-3960X/3701vs3617

I'm not in any shape of form an AMD fan, in fact I think it would be my first or second machine with a CPU from this manufacturer in all my live.
But it's so strange to me that no big workstation manufacturers like DELL or HP are going in this direction that I'm trying to find a good reason for that.

Thanks for all of your comments and suggestions.
When I build our new workstation I'll try to post our experience here.

Have a great weekend,
Dani.

 

[Tingsryd]

Dani,
I think Dell has had a deep partnership with Intel for a long so perhaps there are contractual reasons for them to stay with Xeons for their Precision workstations. The last time I checked I was able to specify an AMD processor for a desktop computer although it was not with the latest Ryzen processors which are really the only cpu's to serious consider for a workstation IMO. I was not able to specify any AMD cpu's for a Dell Precision configuration.

I have seen some data at cpubenchmark.net and I cannot see floating point performance, only integer, single and multi thread performance. I know cad/cam/cae relies on a lot of FP performance so I would be very interested in a real world test of NX with a Ryzen processor. If I come across any NX-Ryzen benchmarks I will be sure to post it in this forum thread.

Best regards.

NX 12.0.2
Testing series 1899

 

[Dani_Andres]

Hi again,
I've been taking a look at Floating Point performance.
Here there are two good reviews about it:

https://www.anandtech.com/show/1504...60x-and-3970x-review-24-and-32-cores-on-7nm/6

https://www.pugetsystems.com/labs/h...0x-Compute-Performance-Linpack-and-NAMD-1631/

My guess if that in case NX is Intel optimized (using AVX512 library in example) the best processor in terms of price/performance right now could be Intel Core i9-10980XE.
Otherwise the new AMD Threadripper 3 series will outclass Intel by far. But that is something that only Siemens NX or an user with both systems platform can answer.

I'm going to open a post about AVX library to see if somebody knows what use NX has of it.
We will probably make the bet of building a 3960x machine and I will post our results as well as a part machining to use as a Benchmark for other users interested.

Best regards,
Dani.