[POLL] What is your ZBrush multicore performance?

pasha_sevez

Hey guys! I was wondering if all of you who use ZBrush could post below some of your stats:

1) What CPU (mem, mobo?) do you use?
2) What is your benchmark score in ZBrush (single thread, multi thread)?

I'm really curious about the criteria of selecting a proper CPU for ZBrush. I've thoroughly googled for any real performance stats and surely scrutinized the topic Benchmarking In Zbrush at Zbrushcentral. I've realized that the CPU choice is not that simple and straightforward as it seems.

Recently we've got some multi-core consumer CPUs like 32-core @ 3-4.2 GHz Threadripper 2 2990WX, 8-core @ 3.6-5 Intel Core i9 9900K, 18 cores @ 3-4.4 GHz Intel Core i9 9980XE. And there's completely no information about how they perform in ZBrush. 

The problem is that CPU core frequency decreases as core count grows. And when it comes to ZBrush, this starts to impact performance in a pretty ridiculous way. On the one hand, ZBrush seems to easily utilize all available threads and cores, on the other hand, it looks like ZBrush doesn't benefit from it a lot! Adding extra cores doesn't increase the actual performance proportionally. 

I'll put some stats from the aforementioned forum thread:

• Dual Xeon E5-2690 - 32 threads @ 2.9-3.8 GHz ~$385
  single - 3.52 | multi - 0.443
  
  
• Core i7-980X - 12 threads @ 3.33-3.6 GHz ~$800
  single - 3.1 | multi - 0.68
  
  
• Core i9 7900X - 20 threads @ 3.3-4.3 GHz  ~$990
  single - 1.935 | multi - 0.31
  
  
• Threadripper 1950x - 32 threads @ 3.4-4 GHz ~$566
  single - 2.461 | multi - 0.365
  
  Threadripper 1950x - 32 threads @ 4.3 GHz ~$566
  single - 2.1 | multi - 0.34
  
  
• Core i7 8700K - 12 threads @ 3.7-4.7 GHz ~$360
  single - 1.94 | multi - 0.54
  
  
• Core i7 7700k - 8 threads @ 4.5 GHz ~$350
  single - 1.92 | multi - 0.588
  
  Core i7 7700k - 8 threads @ 5 GHz ~$350
  single - 1.785 | multi - 0.575
  
  
• Ryzen 7 2700x - 16 threads @ 3.7-4.3 GHz ~$330
  single - 2.4 | multi - 0.5
  
  
• Core i7 - 7820x - 16 threads @ 3.6-4.3 GHz ~$600
  single - 2 | multi - 0.45
  
  Core i9 - 7820x - 16 threads @ 4.9 GHz ~$600
  single - 1.9 | multi - 0.39
  
  
• Ryzen 7 1800x - 16 threads @ 3.6-4 GHz ~$270
  single - 2.742 | multi - 0.618
  

As you can see, the multi-core performance almost linearly depends on the core frequency, but core count doesn't improve the performance that much. So it's very interesting to understand what is most important for ZBrush top performance - core count or frequency - with respect to the price point?

thomasp

Core i7 4790k - 4 cores/8 threads @ 4 Ghz
single - 2.12 | multi - 0.69

(4 years old system, running Win 7)

thomasp

Xeon E5 1650 v2 - 6 cores/12 threads @ 3.5 Ghz
single - 2.83 | multi - 0.7

(Mac Pro 6,1 running mac OS)

Aabel

1. Cores don't increase performance linearly

2. Windows thread scheduler has issues, especially above 10 cores.

pasha_sevez

Aabel said:

1. Cores don't increase performance linearly

2. Windows thread scheduler has issues, especially above 10 cores.

Yes, you are completely right. That’s why I decided to start this thread. ZBrush is too important tool in the pipeline of many artist while you can’t find any meaningful comparison, stats or something which could help you choose the CPU properly. I’ll try to gather as much information as possible and eventually compile a kinda table with all of data.

thomasp

Updated my numbers since the first time around I did not allow ZBrush to initialize the session before running the test. Canvas size does seem to make a difference.

Anyway that looks like pretty decent multi core utilization to me. Would have expected way worse!  

pasha_sevez

@thomasp Thanks for your input Right now the pure multi core test leaders are 10 core Core i9 7900x with 0.31 and overclocked Threadripper 1950x with 0.34. I’m curious is there any CPU which gives, say, 0.25 or below. We’ll see!

thomasp

Well, looking at this benchmark ZBrush seems like a real reason to go multi-core but don't disregard all the other tools you'll end up using which may not thread so well and rather benefit from single core performance.

Also, not sure how applicable this test is to scenarios where you have a complex model up-close, filling a large canvas, perhaps using several matcaps - and you are interacting with the application. My guess is single core will play a larger role there, as may memory bandwidth.

Anyway, one difference I notice between my systems - Windows does not multi task well - if some job is running in the background I can feel the slowdown - enough CPU utilization will even make the mouse pointer stutter. On the other hand on the Mac I can have a render going on all cores in Cycles and still be able to sculpt in the foreground with hardly a hiccup.

pasha_sevez

I'm trying to assess ZBrush multi-core performance in it's best realistic scenario - the benchmark (procedures like Decimation master are optimized much better, but they don't directly affect your actual sculpting). The worst case - non-optimized pieces of software, utilizing mainly only single core (like Maya does in modeling). That's why I'm checking a single core performance as well - and it seem to be directly related to the CPU frequency, the higher, the better. The point is to probably find the golden mean - the mixture of CPU characteristics which give ZBrush the best performance to bucks ratio. So, since it's more or less clear on single core performance, multi-core is still inconvenient. Especially, due various other factors like mem, OS, etc.

brainchildpl

Hey, any progress on this? 

pasha_sevez

brainchildpl said:

Hey, any progress on this? 

Well, people are not much willing to share their stats on their performance, unfortunately (and not only on polycount). I'm not giving up on this stuff anyway but this research will definitely take more time than I expected.

brainchildpl

I am buying 9900k next week. I can test some stuff for you, just tell me what you need. I have Ryzen 1700 atm, so it would be nice to test it now and compare later to i9 9900k.

pasha_sevez

brainchildpl said:

I am buying 9900k next week. I can test some stuff for you, just tell me what you need. I have Ryzen 1700 atm, so it would be nice to test it now and compare later to i9 9900k.

Hey! Any contribution is welcome ^_^ 
In general, just run the built-in ZBrush benchmark and put your single and multi core score in this thread.

brainchildpl

Wasn't it removed with the newest version?

pasha_sevez

brainchildpl said:

Wasn't it removed with the newest version?

Doh! I just have noticed that benchmark is no longer available in 2019.1 (( That's really upsetting news (at least this explains why so few responses), this makes further research simply impossible or requires making some side-party testing plugin ('ll check the possible ways of it).

gnoop

That test run with some rolling star or something as I recall.  Would be much more interesting to see baking and projecting speed differences. 

brainchildpl

Rendering is basically faster with any high core CPUs (like 3900x or Threadripper) but brushes and UI (Active work) stuff works better with higher frequency CPUs like i9 9900k.

You can basically open your task manager and see what happens with CPU usage when you do certain things in zbrush or any other app.

I usually use 30-90% of my CPU, 50-90% is pretty rare that's why I want i9 9900k because it has 5.0ghz single core.

pasha_sevez

Yep, good point. I’m curious if zbrush can utilize more than one core while regular sculpting with brush. I’ve noticed that regular strokes on quite dense mesh cause it to somewhat lag. Utilizing multiple cores on decimation or rendering is obvious, more interesting the performance of operations we do 90% of all time.

OccultMonk

So generally Intel is still better than AMD clock for clock? Still thinking about a 12 or 16 core AMD or Intel processor. And ZBrush performance is the deciding factor for me. 

Aabel

OccultMonk said:

So generally Intel is still better than AMD clock for clock? Still thinking about a 12 or 16 core AMD or Intel processor. And ZBrush performance is the deciding factor for me. 

No. Ryzen 3000 now surpasses Intel in IPC. Clock for clock, Ryzen is faster. Intels new 9900ks is actually lower IPC than the 9900k because Intel put some of the HT exploit mitigation's into the silicon. 

pasha_sevez

It seems like ryzen 3950x should be the sweetspot in terms of raw performance to buck ratio. This cpu is not the fastest gaming cpu, but in cpu-centric tasks it really rocks. I'm really looking forward to see any feedback from upcoming owners if this cpu. Unfortunately since built-in zbrush benchmark has been deprecated, the only way to evaluate it before buying is to just ask guys who can actually compare .