Dell T5500 2.66ghz faster than T7500 3.46ghz

[finalight]

Anyone run into this when comparing the same sim on a slower vs 'supposedly' faster box? Here are the details....

Sim file is identical. 3ds max 2015. FumeFX 4.1.1. Machine A is spec'd slower but sims faster (exponentially as more ram is used). Regular rendering is as expected comparing machine specs (T7500 about 25-30% faster). All bios settings are the same, write to comparable drives, and have Avast AV installed but disabled for sim. The boxes..

Machine A:
Dell T5500
2 x 6core @ 2.66ghz Xeon 5650
72GB ram
Windows 7 x64, installed years ago but updates not run since 2017-ish and sporadically chosen
DirectX 9 (if this even matters)

Machine B:
Dell T7500
2 x 6core @3.47ghz Xeon 5690
Windows 7 x64, fresh install, all updates applied
192GB ram
DirectX 11

One of the primary goals with the T7500 was to make use of the ram to do extreme detail at 4k (not looking for optimization tips or how much ram you are able to get away with - thx though . We can max out the T5500's 72GB without a problem on detailed mushroom sims. Enter, the T7500 - but it chokes badly when it's nearing 110GB of ram usage during WT. Haven't pushed it during default simming yet.

I'm wondering if the slightly differing Win7's have anything to do with it.

Thanks in advance!

[finalight]

Also, in the bios, will setting ram/cpu to NUMA (from SMP) speed up fumefx simming? Thanks.

[sitnisati]

Hello,

Could you please check if you have exactly the same memory chips in both computers?
I also suggest to have Viewport or GPU diasplay disabled during the sim.
There is a nice utility called CPU ID - Hardware Monitor where you can check each core clocks at any time.

Best regards,
Kresimir Tkalcec

[finalight]

Hi Kresimir,

Thanks for that info. I was trying out CPU-ID but I didn't run it during simming - I'll try that out. I've been trying to determine exact data from the memory chips. I could have sworn I had a utility that would tell me all the specific breakdown of serial numbers, etc. for memory back when I picked up the T5500 - might have been command-line. I'll keep digging for it. Great tip on the GPU/viewport off - will try that. I'm usually remote desktopp'ed into my workhorse machines for simming and I get a lot of white screen flashing during heavy sims as well - it may help that issue also to have viewport closed.

Do you have a take on the NUMA vs SMP setting in the bios? I read that perhaps a program would have to be written to utilize NUMA - otherwise SMP is the suggested default.

Thanks!

Best,
Brian

[sitnisati]

Hello,

he NUMA architecture was designed to surpass the scalability limits of the SMP architecture. With SMP, which stands for Symmetric Multi-Processing, all memory access are posted to the same shared memory bus.
This works fine for a relatively small number of CPUs, but the problem with the shared bus appears when you have dozens, even hundreds, of CPUs competing for access to the shared memory bus. NUMA alleviates these bottlenecks by limiting the number of CPUs on any one memory bus, and connecting the various nodes by means of a high speed interconnect.
Thus, I doubt that switching from NUMA to SMP would improve anything.

Interesting to note that one our of beta testers had a nice machine, but they've got a really low MVOPS from the QCG solver.
Once they have increased system memory amount, the QCG solver speed was almost 50% faster with no apparent reason (sims were not hitting tyhe RAM limit).

It would be interesting to find out why those two machines sims so diferently.

Regards,
Kresimir Tkalcec

[finalight]

Very interesting - thanks for that explanation. Both of these nodes were shipped with SMP active. I'll try switching them over to NUMA and compare sim speed.

Yes, I'm not sure why I'm not seeing a consistent 30% increase in speed which was what I expected. Well, I guess I'm keeping it - I'll call it another T5500 equivalent since that's how it's often behaving.

Over the years, all my hardware upgrades have been predictable regarding speed increase - I always knew someday that formula would fail due to some form of machine voodoo.

Thx again for your time!