PDA

View Full Version : New article compares performance gain from software upgrades, Hyper-Threading

DigiHound
08-01-2012, 05:23 PM
All,

I recently wrote the following article (http://www.extremetech.com/computing/133121-maximized-performance-comparing-the-effects-of-hyper-threading-software-updates) comparing the performance gain users might see from upgrading to newer software packages against the performance advantage/disadvantage of using Hyper-Threading on a 16-core / 32 thread system.

I realize this might sound like link spam, and it is my first post, but that's not why I'm here. I wrote this story after a great deal of research into the types of measurements that might be useful to CG artists and system builders; I'm hoping to find out more from this forum.

If the data sets I measured don't provide the sorts of comparisons people are looking for, where do they need to be improved? What rendering plugins do you generally care about? Between the initial review and the further deep dive that I've linked here, I'm hoping that I captured a useful suite of data -- but I'd like concrete feedback on what might be done more effectively.
sentry66
08-01-2012, 06:25 PM
thanks for the article.

I don't have a 16 core xeon system handy to run any tests myself, but from my maya 2009 tests on a 4.2ghz 4 core i7, having hyperthreading on gains 18% rendering performance on my files. Then on my 4.7ghz 6-core i7 machine, hyperthreading gains 25% rendering performance when enabled with maya 2009.

Maybe it's my particular scene files, or maybe there's some seriously diminishing returns for hyperthreading with so many threads with the dual xeon machines - especially when they have such a low clock speed
DigiHound
08-01-2012, 06:33 PM
Sentry,

That's precisely the sort of question I wanted to answer. Clock speed, though, shouldn't impact relative scaling The cores I tested were E5-2690s -- 2.9GHz base, 3.9GHz turbo, 20MB L3. It's possible that the SPEC Maya test itself doesn't capture the benefit of HT very well, but I had very limited resources to evaluate Maya performance. There's the SPEC benchmarks from 2009, and then not much else publicly available.
tswalk
08-03-2012, 03:29 AM
i'ld be interested in knowing why these performance differences in HT between Max versions are the way they are... has anyone from Autodesk provided reasons?
DigiHound
08-03-2012, 03:44 AM
No, and my requests to them went unanswered. I was actually hoping that someone here might have more insight.

Also, it looks like the forum ate my previous response. HT scaling shouldn't be impacted by core count *unless* the program doesn't allocate resources very effectively. That was one reason I tested both HT and software updates -- I was curious to see if older software versions would see speedups thanks to more efficient use of resources.

The 3ds Max disparities are significant but erratic. I have spoken to the authors of the SPEC 3ds Max 11 test -- they're at something of a loss to explain it as well.

I'd love to look into it further, but have a limited portfolio of scenes on hand.
Panupat
08-03-2012, 04:23 AM
Interesting article. I'm very interested to see Vray comparison too.
DigiHound
08-03-2012, 04:46 AM
I don't have scene files or the plugins I'd need to test vray, or I'd likely have done so.