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

[itsmydamnation]

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.

Yes i get that but without knowing the clock it makes it hard to judge given the very wide clock window that chip has. You can try https://software.intel.com/en-us/articles/intel-power-gadget-20

 

[superstition]

Yes i get that but without knowing the clock it makes it hard to judge given the very wide clock window that chip has. You can try https://software.intel.com/en-us/articles/intel-power-gadget-20

The CPU was performing as it does normally at stock. I don't see why you're worried about this. It is what it is. I don't influence its specs.

Anyway, I was able to get a better score by closing a ton of stuff that was open.

37.73

Not bad.

 

[itsmydamnation]

The CPU was performing as it does normally at stock. I don't see why you're worried about this. It is what it is. I don't influence its specs.

Anyway, I was able to get a better score by closing a ton of stuff that was open.

37.73

Not bad.

then im guessing its running @ 3.8 otherwise your haswell has near 100% IPC over my IVB

Also are we sure the demo was @ 100 samples those images at the start aren't from the live demo?

 

[Valantar]

then im guessing its running @ 3.8 otherwise your haswell has near 100% IPC over my IVB

Also are we sure the demo was @ 100 samples those images at the start aren't from the live demo?

OS and a whole bunch of other variables seem to affect scores too. I wouldn't be hugely surprised if stock rendering settings were different for different OSes.

 

[superstition]

then im guessing its running @ 3.8 otherwise your haswell has near 100% IPC over my IVB

It does have EDRAM. Perhaps it uses it as a cache some like Broadwell C? I'll check.

update:

Appears so. https://en.wikichip.org/wiki/intel/core_i7/i7-4760hq

Looks like 128 MB of L4 from the EDRAM like Broadwell C.

Kinda fun to have had Ryzen's next-generation performance in my laptop since late 2013.

 

[superstition]

then im guessing its running @ 3.8 otherwise your haswell has near 100% IPC over my IVB?

Since when is Blender single-threaded?

Each core offers a base speed of 2.6 GHz, but can dynamically increase clock rates with Turbo Boost up to 3.6 GHz (for 4 active cores), 3.7 GHz (for 2 active cores) and 3.8 GHz (for 1 active core).

 

[CatMerc]

Looks like it wasn't 100 samples in the live demo after all. The press demo completes in 25 seconds, which matches up with our 6900K's at 100 samples.
Some other setting was used with the live demo to reach 35~ seconds.

 

[cytg111]

Looks like it wasn't 100 samples in the live demo after all. The press demo completes in 25 seconds, which matches up with our 6900K's at 100 samples.
Some other setting was used with the live demo to reach 35~ seconds.

And that would put it inline with gusklines 5960x again, maybe a little faster.
This is a mess.

 

[Head1985]

6700k 4.5Ghz
1:07

6700k 4.5ghz 100 samples
34s

 

[CatMerc]

And that would put it inline with gusklines 5960x again, maybe a little faster.
This is a mess.

Actually looks fairly reasonable looking at his numbers and halving them.

 

[itsmydamnation]

And that would put it inline with gusklines 5960x again, maybe a little faster.
This is a mess.

maybe the 5960 is turboing better?

 

[Atari2600]

Hopefully this episode makes people appreciate the effort that numerous journos etc have to put in to perform proper comparable benchmarking!

 

[itsmydamnation]

6700k 4.5ghz 100 samples
34s

thats a bigger gap to my IVB then i would have expected (53 seconds @ 4.3) a solid 33/50 (depending which way you count) performance per clock better!

 

[Valantar]

Hopefully this episode makes people appreciate the effort that numerous journos etc have to put in to perform proper comparable benchmarking!

Indeed. Hopefully we'll see more not too long after CES.

 

[cytg111]

Actually looks fairly reasonable looking at his numbers and halving them.

Are you disagreeing with my post? That was what I was getting at; 5960x ~ Zen .. with a little advantage to Zen.
Still a mess though.

 

[CatMerc]

Are you disagreeing with my post? That was what I was getting at; 5960x ~ Zen .. with a little advantage to Zen.
Still a mess though.

I'm trying to understand what you meant by "this is a mess". Like the 5960x isn't supposed to do that?

 

[Atari2600]

I'm trying to understand what you meant by "this is a mess". Like the 5960x isn't supposed to do that?

No, the ambiguity over what was run.

 

[CatMerc]

No, the ambiguity over what was run.

Ah, gotchya.

 

[cytg111]

I'm trying to understand what you meant by "this is a mess". Like the 5960x isn't supposed to do that?

No, the 200->100 sample rate, 36 vs 25 seconds, different results from different operating systems.. We need a clinical description of the test envoriment and all parameters involved so we can reproduce the results.

 

[superstition]

Hopefully this episode makes people appreciate the effort that numerous journos etc have to put in to perform proper comparable benchmarking!

They don't always do a good job, though. In fact, it's somewhat common to see red flags.

One example was that at least two sites, if not three, overclocked E FX chips to high levels without turning off APM. Oops.

Another was a certain site's decision to include like 5 AMD APUs but not a single 8 core FX (particularly at a readily-achievable overclock of 4.4 GHz or so) when comparing AMD's performance with Intel's. When asked about this the response was "Our readers don't care about the FX". Apparently they care about a bunch of APUs, though.

My favorite thing, though, is when the conclusion doesn't match the data. That is a lot more common. But, yeah, some benchmarkers put a huge amount of effort into their work. I particularly like one game reviewer that uses a wide variety of CPUs to test games on.

 

[jihe]

Single E5-2670 Win10 2.78a:
100 samples: 37.35s
200 samples: 72.83s

All core turbo is 3Ghz.

 

[superstition]

Lynnfield i5 750, 3.8 GHz, 16 GB 1600, Windows 7

100: 1:22
125: 1:42
200: 2:43

CB R15 multi: 451
CB R15 single: 116

LinX 0.7 GFlops (14336 RAM): 56.27

 

[superstition]

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.

I wonder if there is something amiss. My 4 thread i5 from 2009 at 3.8 shouldn't outperform a 4 GHz 8 core CPU in a well-multithreaded test unless it's primarily FPU-oriented.

 

[Nebojsha]

My result is
200 Samples 2:42.86
100 samples 1:19.60
AMD FX8350@4.6GHz

 

[superstition]

My result is
200 Samples 2:42.86
100 samples 1:19.60
AMD FX8350@4.6GHz

This result really seems to suggest that this test is favoring single-thread performance a lot more than multithread.

Your 200 samples score is about the same as my 4 core/thread Lynnfield at 3.8. When I did a fast overclock to 4 GHz I got the same 100 sample score you did. 4 cores/threads at 4 GHz should not post the same result as a 4.6 GHz 8 core/thread processor unless it's not only favoring single-threaded performance more but leaning primarily on the FPU (since the FX chips have only 4 FPU units).