Is AMD potentially risky? (Threadripper)

[DaJuice]

On 9/13/2017 at 6:31 AM, VFXArabia said:

@DaJuice Thanx man, cant wait

@VFXArabia, okay back with some numbers.

 

Threadripper 1950x @ 3.8GHz, memory @ 3,200MHz

Pyro @ 0.02: 14m 55s

Pyro @ 0.05: 54s

Pyro @ 0.1: 24s

 

Grain @ 0.01: 6m 58s

Grain @ 0.02: 49s

 

FLIP @ 0.05: 45m 10s

FLIP @ 0.15: 1m 39s

 

Interesting to see the CPU utilization at different resolutions. For example the FLIP scene had distinct spikes and dips at regular intervals at 0.05 particle separation, whereas at 0.15 it was a pretty even ~50% CPU utilization.

 

On 9/12/2017 at 1:34 AM, Diorn said:

Sure i can do the Tabasco Mantra benchmark for statistic's sake but i can already bet TR will beat the 7900x  in this department  

Heh, I expect so too but I'm curious if Cinebench is really that accurate an indicator of CPU rendering performance. For some reason I don't think the gap will be as big with Mantra.

[Diorn]

Wow gap really got wider this time. It's great to see some clear results about these two CPUs. Looks like i would need to push the i9 more than i thought. Added the TR results to my CPU benchmark as well.

I'll try some OC tuning this weekend but already i'm considering returning the CPU and Mobo for a swap  , there is more to take into account besides these 3 kinds of sims but i was aiming for their gap to be smaller than 33% with both OCed within their respective comfort zone

 

@DaJuice Just to be clear, you used the perf monitor like this: hit record, go back to scene view, type in frame 400, then read out the time from the Performance Monitor displayed at "Total Statistics" ?

Edited September 15, 2017 by Diorn

[DaJuice]

That's exactly right Diorn.

[Guest tar]

Anyone running TR or 7900x on Linux or is this all for Windows? For Ryzen it's almost impossible to get the cpu temps in the 4.10 kernel whilst the 4.11 allows it the Nvidia drivers are borked there.

[Diorn]

2 hours ago, marty said:

Anyone running TR or 7900x on Linux or is this all for Windows? For Ryzen it's almost impossible to get the cpu temps in the 4.10 kernel whilst the 4.11 allows it the Nvidia drivers are borked there.

Win10 here

 

[VFXArabia]

@DaJuice Really interesting numbers, I was a bit disappointed with TR but now those numbers are fair enough. 7980XE will outrun those numbers I guess but that extra 1000 is not worth in my opinion.

Thanx man for your time and effort

https://www.vfxarabia.co/single-post/2017/09/10/Houdini-Benchmark-Cores-vs-Clockspeed-updated

Edited September 18, 2017 by VFXArabia
Added a link

[Guest tar]

5 hours ago, VFXArabia said:

7980XE will outrun those numbers I guess but that extra 1000 is not worth in my opinion.

Lets wait to test it before saying it will  Also Threadripper is the start of the new line of AMD HEDT, so what comes next will be interesting, i.e. more ram support. Also EPYC is out there for serious builds.

[VFXArabia]

@marty I agree let's wait, but as for Epyc as long as it has a low core speed it wont help in our everyday tasks as the results have shown with my dual xeon test. On the other hand it will sky rocket when it comes to render, Intel will eat its dust.

[kemijo]

Very interesting results in that spreadsheet. It was clear that TR would win in rendering but I was not expecting to see TR come out ahead in the longer Pyro/Grain tests yet tying with FLIP. And not all that much extra performance brought by overclocking, considering the power/heat/hassle required, esp on the 7900x. 

Really appreciate the time and effort from everyone running these tests and making that spreadsheet! So hard to find data like this anywhere else. Still waffling over which CPU to go with (and monitor - separate discussion lol).

[Diorn]

Hi guys, i'm back with an update. I managed to dig up an older project and turn it into a benchmark scene. I've added more details on a note inside the project, in short it uses a few for each loops, attribute transfers, a sop solver inside a DOP network, an RBD solver over some packed geometry. The last run of SIM benchmarks sure do the extra cores justice, but Houdini is more than those 3 types of sims, i'm still curious to find out how it behaves in different circumstances.

I've tuned my 7900x some more this weekend: the basic OC profile i ran from the BIOS was overvolting the CPU unnecessarily, hence larger temps on lower speed. I managed to get 8 of the cores running@ 4.7 and the other two@ 4.6( these two where slightly underperformers and got hotter than the others at the same speed), now so far in my particular case this is what i would call a stable and workable rig with the AIO and a good chasis with fans. the CPU is really an overclocking champ but the temperatures are the main issue. (Disclaimer, overclock only if you know what you are doing, don't take my word for it, please understand this still remains risky for any CPU, intel or AMD)

So far i've run benchmarks of intense workload for just under the one hour mark. Now it is worth mentioning i did get to see it reach 4.8 ghz on all cores, it was stable enough for a few benchmarks but the system froze on mantra rendering tasks, this is getting into delidding teritory.

i've added the new scores in the spreadsheet ( https://docs.google.com/spreadsheets/d/1jQFxPdOBcWbYtJJkVafr4VYojhkV5MqInZmEsT_zmZA)

So all this speed got it closer to the OC threadripper for these kinds of sims with PYRO seein the TR a clear champion.(lots of multithread love going on in PYRO it seems, kudos!)

Now back to my initial point, here is the link to my SOPs benchmark proposal. Of course i am open to opinions about it if it can be improved somehow or if there are better types of tasks that touch general use. 

https://drive.google.com/open?id=0B-EHM5diHgrERHAwanZCdXZON28

@DaJuice eager to see some scores from your TR for this also 

 

Edited September 17, 2017 by Diorn

[VFXArabia]

@Diorn @4.8 wow your machine must be screaming, was it so hard to make it that stable. And yeh Pyro seems to love more cores.

Thanx for updating the numbers

[Diorn]

12 hours ago, VFXArabia said:

@Diorn @4.8 wow your machine must be screaming, was it so hard to make it that stable. And yeh Pyro seems to love more cores.

Thanx for updating the numbers

Yes, at 4.8 it was actually asking for more current than i was giving it for it to become stable under load. But more current would mean larger temps at 4.8ghz i was getting auto throttling from the CPU hardware fail-safe mechanisms. 8 cores at 4.7 and 2 at 4.6 runs indeed hot, but no throttling and no instability so far for renders or sims or any other of the benchmarks, and some of them run hotter(90+) some of them run cold (70+).

PS: if you don't mind, and have the time, could you please try out my sops/Bullet benchmark on your machines?

@Beatnutz still around for some more tests?

 

Edited September 18, 2017 by Diorn

[Noobini]

surely PathFinding should love 'parallel' processing (I'm doing poorman's PathFinder, not necessarily shortest, in Max MCG...sloooooow as hell but maybe someone could whip up a decent FindShortestPath test  for H16, that'd be nice...)

[VFXArabia]

@Diorn Sure I will run it on both systems.

[VFXArabia]

@Diorn okay came back with some numbers:

6800k @3.8 14:56

2x E5-2683v3 @2.5 22:09

Well it seems your test doesnt like multicores at all, barely one or two cores were used.

[Diorn]

15 minutes ago, VFXArabia said:

@Diorn okay came back with some numbers:

6800k @3.8 14:56

2x E5-2683v3 @2.5 22:09

Well it seems your test doesnt like multicores at all, barely one or two cores were used.

yes this was supposed to be a more single threadded collection of use cases (i'm sure the methodology can be improved), i'll update the running spreadsheet. Thank you for running the test!

in related news, mr Timucin Ozger, lead FX TD for MPC, has kindly provided a through benchmarking scene file with the bonus of having a Deadline implementation:

Quote

With Deadline, I can tell a task to be divided into 10 subtasks for 100 frames. Then I can tell Deadline to run each of these tasks simultaneously, It would lunch 10 Houdinis in the background and process the information. As long as you can fit it into your ram you get almost full threading.
This makes a huge difference in scenarios like, cleaning up caches, meshing geometry, caching collision geometry or VDBs before pyro or flip simulations.

Link to the SideFX forums post:

https://www.sidefx.com/forum/topic/51730/

Edited September 18, 2017 by Diorn

[VFXArabia]

@Diorn you are most welcome, yeh yesterday we were talking about it in the Houdini Artists FB page he mentioned sth about releasing it today.
It's a great idea to use deadline to distribute single threaded tasks.

[Diorn]

9 minutes ago, VFXArabia said:

@Diorn you are most welcome, yeh yesterday we were talking about it in the Houdini Artists FB page he mentioned sth about releasing it today.
It's a great idea to use deadline to distribute single threaded tasks.

Deadline is great and i never used it like this - but i am running the test right now, i am not entirely sure this is such a universal benchmark. my I9 has finished chewing on all the sim-based tasks, with the only ones left being 3 mantra ones (wich supposedly will take a few days to render?) so again it looks to me like it is is a mostly CPU rendering test. Still no grain, liquid (with proper meshing setup), complex sops opperations with no simming involved, or FEM.  So i am preparing a FEM benchmark next...

Edited September 18, 2017 by Diorn

[VFXArabia]

@Diorn I agree sim and sops are what Houdini is all about, FEM is a night mare for me it takes alot of time to finish.

[tricecold]

hey guys, i was going to post a link here too, but I see that its already shared. To be honest Grain solver is a disaster on any CPU and definitely would not lunch it on CPU if a GPU is available. From experience I can say that Flip will scale as well as Pyro. 1950X has an average speed per core , so I don`t expect it to shine on smaller tasks also, however there is an FEM simulation in there with a some sop solver involved, although I believe pcOpen would perform a lot better than an attribute transfer. I can probably speed up this setup another %30 with lots of compile  sops where possible, but yet again it will be limited to sop operators.

However the purpose of the Benchmark is to simulate a production environment. The goal is how efficient a 1950X would be if you were to deliver to the client all the way to the renders as a freelancer, This CPU is a waste in a studio environment as an artist machine.

Diorn, I am however curious to see those numbers on all the tasks that are completed

I can also try a setup of some complex SOP operators one with and without compile sop to see the diffrence

 

Edited September 18, 2017 by tricecold