Post your Ryzen Blender Demo Scores! (AMD clarifies Blender Benchmark Confusion, Run @ 150 Samples)

[polyzp]

Post your Ryzen Blender Demo Scores with your setup!

Ryzen SR7 - 3.4 Ghz (No Turbo) : 25 seconds - Samples 100
Ryzen Sr7 - 3.4 Ghz (No Turbo) : 36 Seconds- Samples 150

EDIT: Make sure to set to 100 or 150 samples to compare !




EDIT: To Compare to the presentation score of 36 seconds, set samples to 150!

AMD clarified on Reddit

"AMD_jamesAMD Employee 6 points 11 minutes ago

Set render samples to 150. A new file with the sample level set correctly will be uploaded shortly. Apologies for the confusion.

Blender 2.78a x64 is what we used, binary download from blender.org."


The press benchmark is different than the one they did on stage. Press reports 25s at 100 samples (done independantly), the one on stage is 36 seconds and the settings are a mystery.


How do you compare?

Link is here : http://download.amd.com/demo/RyzenGraphic_27.blend

 

[guskline]

My OC'd 5960x @4.4Ghz did it in 41.54 seconds. At stock (3.3) it was 54.04 seconds. These were at 200 sample.

I set the sample to 100 and got the following results on the 5960X. At stock, 27.62 seconds and OC'd to 4.4, 20.16 seconds.

Finally at 150 sample size, which now apparently is what the AMD demo was running at, my 5960x, at stock had a render time of 40.87 seconds and at my 4.4 ghz speed, 29.83 seconds.

I've got to give AMD credit. My 5960x stock is 3.3ghz but turboed. AMD's non-turbo 3.4Ghz RYZEN 8c/16t yields at 150 sample size 36 seconds.

 

[Sven_eng]

At 100 samples my stock i7 6700k took 38 seconds.

 

[Saylick]

Some people over on OCN were posting up their Blender render times as well; sounds like if you just hit the "GO" button, you won't achieve anywhere near what AMD was reporting during the event, but if you change the settings to 100 Samples in lieu of 200 Samples (which is apparently the default parameter), you get a much closer result. It's going to be hard to know exactly what settings AMD used during their event, so an apples-to-apples comparison isn't going to be likely until we get the actual product in reviewers' hands.

 

[bigboxes]

Rig in sig: 1:17.54

 

[jhu]

34 seconds

2x Xeon E5-2670
FreeBSD 11.0
Blender 2.76b

 

[Head1985]

6700k 4.5Ghz
1:07

 

[jhu]

Some people over on OCN were posting up their Blender render times as well; sounds like if you just hit the "GO" button, you won't achieve anywhere near what AMD was reporting during the event, but if you change the settings to 100 Samples in lieu of 200 Samples (which is apparently the default parameter), you get a much closer result. It's going to be hard to know exactly what settings AMD used during their event, so an apples-to-apples comparison isn't going to be likely until we get the actual product in reviewers' hands.

Of course if you halve the number of samples it would be faster. Not so obvious, however, is that

1) the Blender version affects render performance
2) Operating system affects performance. In my experience, the speed hierarchy is: Linux > FreeBSD > Windows.
3) Tile size affects render speed slightly (fastest is 16 x 16).

 

[moonbogg]

My OC'd 5960x @4.4Ghz did it in 41.54 seconds. At stock (3.3) it was 54.04 seconds.

Then something is off because no way a stock Zen lays waste like that to your chip @ 4.4 across all cores.

I just did it and got 1:03

 

[cytg111]

https://forums.anandtech.com/thread...zen-benchmarks.2482739/page-116#post-38629970

Two options:
A. AMD used a build with AVX256, which apparently lines up with the scores
B. AMD used a lower sample rate (from 200 to 100), which again lines up with the scores

We should defn get first hand confirmation of these setting. Loving the thread otherwise though.

 

[CatMerc]

https://forums.anandtech.com/thread...zen-benchmarks.2482739/page-116#post-38629977

Mystery solved.

 

[cytg111]

https://forums.anandtech.com/thread...zen-benchmarks.2482739/page-116#post-38629977

Mystery solved.

Op .. perhaps update your post with this?

"clearly see samples -> render set to 100"

and everyone run again? .

 

[cytg111]

My OC'd 5960x @4.4Ghz did it in 41.54 seconds. At stock (3.3) it was 54.04 seconds.

A rerun with samples, render set to 100 perhaps? At stock? At haswell 8c/16t 3.3 your chip is the closest "apples to apples" we have.

 

[Valantar]

C2D Q9450 @3,52GHz: 3:54.34

I dare anyone here to beat me.



That's OOTB, though, didn't touch a single setting.

100 samples, same CPU: 1:56.66. Pretty much exactly half the time. The image is really grainy, though.

 

[cytg111]

If it finishes in ½ the time then gusklines 5960 will lay the smack down @stock vs zen...

 

[SpaceBeer]

i7-4790 @ Stock (3.6/4 GHz)
During rendering, all cores are @3.6 GHz

Has anyone tride to use GPU to render this image? It's strange, it takes much more time than the CPU, at least in my case (R9 380)

 

[Kiska]

Celeron 857: 0:58:12 at 200 samples
Celeron 857: 0:47:12 at 100 samples

Results may be skewed, as I am using the laptop

 

[dacostafilipe]

I'm running 38.62 on my 2010 Mac Pro (2x Xeon Quad E5620 @2.4/2.6)

 

[lixlax]

Seems that it was indeed 100 samples. My results:
200 samples 2:49 (169 seconds);
100 samples 1:25 (85 seconds).
Cpu is stock FX 8350. The demo Rysen was still more than 2x faster which isn't a bad result.

If we roughly halve gusklines scores from the the second post for 6950X we get ~27s at stock and 21s when overclocked. In that case 35-36s on a stock 6900K make a lot more sense as well.

Edit: Misread the gusklines post, its indeed 5960 not 6950. But that also makes the presentation 6900K noticeable amount slower than 5960X. Any other settings may have been changed?

 

[richaron]

Seems OS makes a huge difference...

1:43.67 ~ 104 seconds for an a8-7600 in Linux.

 

[itsmydamnation]

Windows 10 3770k @4.3 with 100 samples = 56.30

 

[superstition]

38.97, OS X Mavericks, 2.6 GHz i7-4960HQ, Blender 2.78a

100 samples.

update:

Closed lots of open stuff and cleared off my cluttered desktop. Got the time to 37.73.

https://s24.postimg.org/5f32adf11/proof.jpg

 

[Valantar]

Celeron 857: 0:58:12 at 200 samples
Celeron 857: 0:47:12 at 100 samples

Results may be skewed, as I am using the laptop

Uhm, did you run the Ryzen workload linked in the OP? There's no way a 17W Celeron 857 can do that.

 

[itsmydamnation]

38.97, OS X Mavericks, 2.6 GHz i7-4960HQ, Blender 2.78a

100 samples.

what were you clocked at while it ran? big difference between 2.6 and 3.8 ghz......

 

[superstition]

what were you clocked at while it ran? big difference between 2.6 and 3.8 ghz......

I don't control the turbo. It's set up in the default configuration. There is no way I know of to manually adjust Macbook Pro BIOS settings for clock rate and such. It's a "Late 2013" Macbook Pro running OS X 10.9.5.