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

No, I'm not really very well versed in Blender TBH. I'll try it later, lunch is about over.

And I don't appear to have that option. I have the "Full Screen", the "Supported", but nothing under that.

I finally got it to run with GPU Compute or whatever it was called, but SLI TXM @ 1500 was still way slower than 5960X @ 4.6.
 
Upgraded my workstation - dual E5-2690 v4 @ 2.60GHz with Quadro M2000 takes 0:13.80
 
Rig in sig. 58.96 seconds.

AceGoober_Blender_Test_1.JPG
 
The only thing this benchmark is making me want is more! Oh, and that I wish the Zen release day would get here quickly.
 
Forget my password for photo bucket. My time was 01:15.83 on my rig in my sig.
 
Re-ran it with the new file.
Blender 2.78a 64bit
All my H/W set to stock clocks.
No settings changed except for sample size.

200 samples: 1:54.67
150 samples: 1:33.61
128 samples: 1:13.47
100 samples: 0:57.60
 
FX--8300 @ 3.3Ghz (Boost Disabled)
DDR3 1600

I don't know anything about blender, but the "render" setting was set to 150 in the samples section. I guess that means samples were set to 150.

3.09....ouch.

Btw, I disabled boost because I own the worlds cheapest motherboard (ASRock N68-VS3 FX) that likes to throttle my cpu to 1400mhz whenever boost is enabled. Yeah, No thanks.
 
I tested my I7 920 @3.8 again with the recommended 150 sample setting.

I7 920 @3.8
Corsair Dominator DDR3 1600
EVGA X58 sli x3 mobo

Test1 - 1:30:52
Test2 - 1:29:21
Test3 - 1:29:89
Test4 - 1:30:25
Test5 - 1:30:10

Very interested in Ryzen :happy:
 
Wanted to see how my old workhorse did here and what effect an older version of HT had on things:

X5650 (6c) @ 3.6GHz no turbo 18x200, 18GB (2x3 + 4x3)
HT off - 1:28.10
HT on - 1:03.29

I usually keep HT off because 6 threads is more than enough for most jobs and it makes my temps go up a bit, but now that it's winter that's just helping heat the house so why not. My x5650 still seems up to the task in anything I use it for, it's really the connectivity side of things that the x58 platform is showing its age. Looks like there are some really nice gains to be had with Ryzen if blender performance manages to translate into other applications as well. Will likely skip out on the first gen but keep my eyes open for Zen+ when the platform has any initial bugs and kinks worked out. At that point my motherboard and 3 sticks of RAM I have in use will be going on 8 years :oldman:
 
Just for giggles... on my quad Opteron system i set 16 of the K10 cores to 3.4 GHz (to match Ryzen), set threads in Blender to 16, and set thread affinity for Blender to them and reran (150 samples). Result: 46.99 seconds.

Basically AMD must have been lying their rear ends off as far as throughput (40% my rear). Each Zen core is indeed pushing between 2.5 and 2.7 K10 cores worth on blender... and one Piledriver core at same clock is only worth around .85 of a K10 core (at least on Cinebench).

Ryzen = total and compleat sandbag job on the part of AMD.
 
Just for giggles... on my quad Opteron system i set 16 of the K10 cores to 3.4 GHz (to match Ryzen), set threads in Blender to 16, and set thread affinity for Blender to them and reran (150 samples). Result: 46.99 seconds.

Basically AMD must have been lying their rear ends off as far as throughput (40% my rear). Each Zen core is indeed pushing between 2.5 and 2.7 K10 cores worth on blender... and one Piledriver core at same clock is only worth around .85 of a K10 core (at least on Cinebench).

Ryzen = total and compleat sandbag job on the part of AMD.

IPC != performance increases

IPC is a part of the function of performance increase. Blender is a heavy multi threaded test, Bulldozer had major issues fully utilizing all of its cores at the same time when doing FP operations and yeah rendering on a CPU will use the FP units greatly.

So you have three problems on Bulldozer that Ryzen would have solved, IPC, threading, and core utilization in this specific benchmark.
 
Stock 6700K using Blender with 256bit AVX2. 100 samples. 27.17seconds!

blend256.png
 
Two of my plugged in rigs (just moved to Austin a few weeks ago so the rest of the rack is in boxes):

HP ProLiant DL165 G7 (2x6172 Opteron) - 53.91 seconds
Main rig in my sig at stock speeds (6700k) - 54.94 seconds

Will try on my Sun SunFire X4600 M2 when I find it in my pile of server boxes. That machine has 8x8384 Opterons (32 total cores), but has DDR2.
 
Did you get the new file? Amd used 150 samples.

They used 150 for the live and 100 for the press. I get 39.30 with 150.

With 200 samples it takes 53 seconds. With the other version it took 77 second.
 
They used 150 for the live and 100 for the press. I get 39.30 with 150.

With 200 samples it takes 53 seconds. With the other version it took 77 second.

What other version are you referring to if I may ask? I'm aware of the 100, 150 and 200 sample tests. Was there another that people were testing?
 
Sure could use those Ryzen 8 cores today, I'm transcoding my entire DVD movie library to .m4v on a 2012 MacBook Pro i7. Takes about 8 to 10 minutes per movie. :-/

Yeah, I know, it won't be available for at least a month and the Old Build in my sig would be faster than the MBP but its packed up ready for a household move.
 
Followed the instructions in OP as of 12/18/16.

27.55 on an SR2 w/2x X5690 @ 4.3 Render 150. System has been on continuously since last win 10 update reboot on Tuesday. So, no tweaks.
upload_2016-12-18_12-48-15.png


18.32 @ Render 100:
upload_2016-12-18_12-59-58.png
 
Last edited:
2.78a vs 2.78.4. 128bit vs 256bit.
https://builder.blender.org/download/

The IPC load is also low with 1.14, hence why SMT does so well. Easy branch prediction too.
Interesting but we don't have a Ryzen test using that version could be 16sec or less. Who knows?

Using the exact same scene and software you can compare between the cpu's. I also get 50sec for I7 6700K at 4.6ghz and the same configuration for software as AMD Ryzen test.
 
Interesting but we don't have a Ryzen test using that version could be 16sec or less. Who knows?

Ryzen doesn't support single cycle 256bit execution. Its the same for SB/IB/FX/Jaguar etc.
 
Back
Top