AMD™ Ryzen© Blender® Benchmark Scores™©® Thread

since there was plenty of intel benchmarks i decided for shits and giggles to try this on my phenom II x4 940 @ 3Ghz(was suppose to be at 3.9Ghz no clue why it was reset to stock settings since it was last restarted) just to see how far processors have come along in the last 8 years.. render took 3:08.58

the sad part is that it's only about 20 seconds slower than the FX 8350 time some one else posted..

Phenom II was at least decent at the time compared to lower end Intel C2Q. Vishera/Bulldozer was just a massive fail. I'd say the Phenom II X6 is probably a better CPU overall. Even a high clocked Phenom II X4 isn't bad for less demanding applications and games today.
However it goes to show what a massive leap AMD made with Ryzen.
 
upload_2017-3-8_10-23-56.png


I don't think i'm going to upgrade for a while yet, as my score isn't too bad for a 2009 machine:
57.86 - Xeon X5650 @ 4Ghz, 18GB DDR3 - GTX 980 TI

Looking at some of the Ryzen scores though, it's not bad :D

hmmmmmm
 
interesting with the scaling here, my 4x E7 8880 v2 did only slightly better with 4 times more cores, scaling seems to hit a wall here :)

Best regards Otto

Can't believe I missed your post! I wanted something to compare to. That is very interesting. Perhaps Blender wasn't using your additional CPU's? (In 2005 you had to set that option manually haha!)
 
i think the problem is that once you reach a certain threshold of threads it's completing them faster than it can saturate all the threads available to it. you would probably need a render that takes closer to a minute to to completely saturate all the threads.
 
i think the problem is that once you reach a certain threshold of threads it's completing them faster than it can saturate all the threads available to it. you would probably need a render that takes closer to a minute to to completely saturate all the threads.

Actually... depending on how the renderer functions I'm betting what's happening is proc 0 is being tasked with handing out threads to the other processors. Them being as fast as they are may be finishing their thread before they are given another one so you are missing out on potential throughput.... MAYBE.

That's actually something that happens on Oracle servers when on non Oracle hardware/hosting OS's. The proc 0 gets so busy handing out worker threads the system plateau's and can't go any faster even with incredible physical I/O available. you wind up with non proc 0 CPU's just sitting there fully available to work but no threads assigned or too few because proc 0's Logical I/O can't assign fast enough. Of course those are some VERY large systems.....
 
Version 27.8c gives different results, been using 27.8a before and topped at 44 seconds @ 5.2/3866, now this setting renders the 150er setting in just 31.5 seconds. How come ?

edit* run at 5.2GHz 3866MHz, clean
 

Attachments

  • Blender 5.2-3866.JPG
    Blender 5.2-3866.JPG
    90.3 KB · Views: 58
Version 27.8c gives different results, been using 27.8a before and topped at 44 seconds @ 5.2/3866, now this setting renders the 150er setting in just 31.5 seconds. How come ?

edit* run at 5.2GHz 3866MHz, clean

BitMaster . Blender 27.8a scores different than Blender 27.8c because the blender developers failed to create a good FINAL product. They release alphas and betas versions for the public instead of waiting in order to fix the performance issues, the bugs, etc.

Dont worry, 27.8a scores bad on AMD platforms also. If you dont trust me, you cant test it yourself, or ask a friend who has AMD and run the test to see for yourself.
Here is my test on AMD FX 6350 3.9Ghz stock [same hardware]

Blender 27.8a - 3m 27s on Win7
Blender 27.8a - 3m 05s on Win10
Blender 27.8a - 2m 47s on Linux

Blender 27.8b - 1m 50s on Win7
Blender 27.8b - 1m 47s on Win10
Blender 27.8b - 1m 28s on Linux

Blender 27.8c - 1m 49s on Win7
Blender 27.8c - 1m 47s on Win10
Blender 27.8c - 1m 27s on Linux

If you dont believe me, you should try to test on Linux, boot up a Live Distro such as Linux Mint, Linux LXLE, Ubuntu etc and download the zip blender, no need to install anything, just unzip, run Blender, load the demo file and hit F12. You will have the surprise to see that your CPU will score better under Linux, Why? ask the Blender developers. i guess Blender is optimized for Linux, maybe they dont like Windows...maybe the developers team working for Windows platform made some mistakes, that cripples the performance on Windows Platform.

AMD has nothing to do with the crippled performance of the INTEL or other AMD cpus , running Blender 27.8a version. They chose to use Blender because they thought Blender it is a FREE Opensource app and therefore it shouldnt be manipulated by.....competition or...by them [amd]. You must be stupid and retarded to be AMD, and taking a risk to pay a company in order to optimize Blender for your CPUs, and to cripple the Intel performance. So i think, it is out of the question. If there are big differences in scores between blender versions, is because....blender developers.
 
buzzbomb , you get 1m 47sec with your overclocked FX 8300 4.8Ghz. Ouch! The old FX 6350 3.9Ghz[stock,Turbo Boost Disabled] scores 1m 49sec in Win7 and Win10. Did you try the latest v27.8c ?

OttoD , your 4x Xeon E7 v4 88cores is extreemly slow. Yes, 11seconds is less than 23seconds [ ryzen score posted by JimmiG ], but be aware, you have 88 cores vs 8 cores. I dont know if you "see" this. With your intel monster, you should have scored under 1 second. [the entire blender fits inside your L3 cache]
 
It's strange that on the same system that I would get 34s on Debian, 45s on Windows 10, and 48s on Windows 7.
 
It's strange that on the same system that I would get 34s on Debian, 45s on Windows 10, and 48s on Windows 7.

KarsusTG, i am surprized that nobody have seen it so far, that proves that most of the users are Windows users.
I am glad you could see the difference, this is happening for both Intel and AMD cpus. Some people were saying that "AMD made Blender to cripple intel cpus" which i dont believe.
So Blender needs a lot of patches until they make a good final product.
ps: nice times you have there, you are not far from the 4x Xeon E7 v4 88cores score.
 
Back
Top