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

[rvborgh]

8 core/16 thread Ryzen at 3.4 GHz is equal to 24 K10 processors at 3.1 GHz on standard 2.78a build @150 samples.

 

[inf64]

8 core/16 thread Ryzen at 3.4 GHz is equal to 24 K10 processors at 3.1 GHz on standard 2.78a build @150 samples.

Or per core and per clock one Zen core(with SMT) is 2.75x faster than one K10 core in Blender .

 

[walk2k]

i7 980X (6 core Gulftown) @ 4ghz = 57s.
150 samples. Win10 x64

 

[lopri]

V2.78a
1045T @4.05 GHz (13.5 x 300)
150 Samples, Win7x64

1 min 39 sec

Edit: SS

 

[Otto-D]

V2.78a
4x Intel Xeon E7-8880v4 @ 2.20ghz 88c/176t
150 Samples, Server 2012R2

11.74 sec

 

[chuy409]

v2.78a
AMD Phenom 960t (unlocked to x6) @4.2ghz
150 samples Win 10 Pro


https://ibb.co/b8rdyv

1 min 36 sec

 

[nvgpu]

https://www.blender.org/download/

https://wiki.blender.org/index.php/Dev:Ref/Release_Notes/2.78/Performance_update

"Blender 2.78b is the latest stable release from the Blender Foundation, released on February 8, 2017."

 

[lopri]

V2.78a
4x Intel Xeon E7-8880v4 @ 2.20ghz 88c/176t
150 Samples, Server 2012R2

Talk about bringing a gun bazuka to a knife fight.

 

[FlyingNinja]

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

At 100 samples my stock 4790k took 27 seconds. You sure yours is right?

 

[Teizo]

Saw this thread and got curious, so decided to run the demo with the provided file in the OP.

i7 7700k @ 4.5ghz - 1.2v - Cpu Package Power = 65 watts
Gigabyte Auros Z270X Gaming 7 w/F4q bios
16gb G.Skill Trident Z 3200mhz - CL 16

Ryzen Graphic Blender Time = 35.81 seconds

 

[itsmydamnation]

Saw this thread and got curious, so decided to run the demo with the provided file in the OP.

i7 7700k @ 4.5ghz - 1.2v - Cpu Package Power = 65 watts
Gigabyte Auros Z270X Gaming 7 w/F4q bios
16gb G.Skill Trident Z 3200mhz - CL 16

Ryzen Graphic Blender Time = 35.81 seconds

You're not running the right version of blender, also make sure.your samples are 150.

 

[Teizo]

You're not running the right version of blender, also make sure.your samples are 150.

Eh.....I ran the actual file provided in the OP with the correct settings and ran the latest version of blender from their page. My results are legit. Just look at the pics and see.

 

[itsmydamnation]

Eh.....I ran the actual file provided in the OP with the correct settings and ran the latest version of blender from their page. My results are legit. Just look at the pics and see.

At the time 2.78a was the latest version
http://download.blender.org/release/Blender2.78/

 

[Teizo]

Got 52 seconds one time, and 53 seconds the other time. Perhaps the folks at AMD need to re-run the test with the new release and re-compare. Maybe there was a bug that was holding back Intel processors. Not sure, but if I got almost 15 seconds better time with the new release, Ryzen chips should get done in about 20 seconds now.

 

[itsmydamnation]

Got 52 seconds one time, and 53 seconds the other time. Perhaps the folks at AMD need to re-run the test with the new release and re-compare. Maybe there was a bug that was holding back Intel processors. Not sure, but if I got almost 15 seconds better time with the new release, Ryzen chips should get done in about 20 seconds now.

Nope there are big vector code handling improvements in b. Amd con core processesors see an even bigger improvement then Intel core do. In afew weeks I'm sure we will have Zen 2.78b data

 

[Teizo]

Nope there are big vector code handling improvements in b. Amd con core processesors see an even bigger improvement then Intel core do. In afew weeks I'm sure we will have Zen 2.78b data

That's good. I re-ran 2.78b again and got 35.51 seconds. Will be interesting to see what the new Ryzen scores are. I seriously doubt they'd intentionally show Intel in a negative light like that trying to run a benchmark that ran buggy on Intel hardware....so the new Ryzen score should really be impressive.

 

[walk2k]

if you're not using 150 samples you're literally just uttering gibberish here.

 

[The Stilt]

Nope there are big vector code handling improvements in b. Amd con core processesors see an even bigger improvement then Intel core do. In afew weeks I'm sure we will have Zen 2.78b data

Tested the version which is supposed to contain these improvements and the difference falls within the margin of error, at least on Haswell.
Both binaries were custom compilations (MSVC + ICL). The older version already has ~41% speedup from AVX/2 kernels.

 

[Otto-D]

if you're not using 150 samples you're literally just uttering gibberish here.

I did run it at "stock" settings 150 samples

 

[Teizo]

if you're not using 150 samples you're literally just uttering gibberish here.

Well, it's been confirmed there was a bug using the Ryzen blend with Intel processors that 2.78b fixed. I ran the exact file that was listed for DL in the OP. It's not my fault there was a bug in 2.78a slowing down Intel processors.

https://www.pcper.com/news/General-Tech/Blender-Foundation-Releases-278b-Performance

"Also, some of the optimizations solve bugs with Intel’s CPU implementation as well as increase the use of SSE 4.1+ and AVX2. Unfortunately for AMD, these were pushed up right before the launch of Ryzen, and Blender with Cycles has been one of their go-to benchmarks for multi-threaded performance. While this won’t hurt AMD any more than typical version-to-version variations, it should give a last-minute boost to their competitors on AMD’s home turf."

And don't get me wrong either....I am rooting for AMD and want to see them do well.

 

[legcramp]

i7 980X (6 core Gulftown) @ 4ghz = 57s.
150 samples. Win10 x64

x5650 @ 4ghz here, 56.69s. Pretty much spot-on with yours =)

 

[itsmydamnation]

Well, it's been confirmed there was a bug using the Ryzen blend with Intel processors that 2.78b fixed.

no it hasn't thats just crap. that build has been out for 4 years!, here are the release notes for b;

https://wiki.blender.org/index.php/Dev:Ref/Release_Notes/2.78/Performance_update

no bugs of any note, but look here:

Cycles: CPU

• Use more global SSE optimizations for SSE4.1+ kernels.
• Multiple improvements for the latest AVX2 CPUS:
  • Optimized various math utilities (cross-products, dot-products, min-axis-selection and others).
  • Faster version of triangle intersection function.
  • Optimization of various steps in BVH traversal algorithm (including both construction and traversal).

I ran the exact file that was listed for DL in the OP. It's not my fault there was a bug in 2.78a slowing down Intel processors.

There was no bug i already explained it to you, there where performance improvements from better vectoriszation, see the quote from the release notes.

https://www.pcper.com/news/General-Tech/Blender-Foundation-Releases-278b-Performance

Who cares what pcper think, they are hardware reviewers they aren't the bastion of all knowledge.

And don't get me wrong either....I am rooting for AMD and want to see them do well.

O RLY.jpeg? Maybe do your research first

Here is a start http://i.imgur.com/WjG3n3v.png this is data from this very thread posted by Majord which uses a build compiled by the stilt which uses most if lot all the optimizations found in 2.78b.

Fundamentally we are talk about 128bit vectors because renders work on XYZA data, excavator handles AVX2 fine so there is no reason Zen wont. Given the workload there is unlikely if at all to be any 256bit avx/2 hit to Zen. Excavators speed up is far greater then skylakes, i dont expect Zen to see the same improvements as Excavator ( CON cores are broken and im not going to detail it yet again) but i do expect Zen to see the same kind of scaling between A and B as Broadwell/Skylake.

 

[Otto-D]

had a run on my good old i7-4770k, in a Mini ITX case, running on air and Overclocked here are the settings:
BLCK: 100
Core multiplire: 49 49 49 46 <-- CPU-Z and other programs report the speed as 4600 but u can see in the multiplier part that is actually 4900
Cache multiplire: 44 42
vcore: 1.365

Blender time: 38.96

im changing to water because i hit my thermal comfort limit at 4.9ghz, and after that i will try to delid the cpu and i hope to hit 5ghz on the 3 cores, the last core wont go over 4.6

Best regards Otto

 

[Teizo]

No need to get so riled up man. No need at all. I didn't say AMD's Ryzen was not going to see improvements from 2.78b...but it is blatantly obvious Intel chips improved drastically from it. I look forward to seeing the Summit Ridge scores once released.

 

[itsmydamnation]

No need to get so riled up man. No need at all. I didn't say AMD's Ryzen was not going to see improvements from 2.78b...but it is blatantly obvious Intel chips improved drastically from it. I look forward to seeing the Summit Ridge scores once released.

Don't take it personally, the problem is people take wrong info and use it for their agenda's and then i have to read it over and over again as other people pick it up and run with it. All i want it people to use correct accurate data and knowledge . Like the original fake Zen slidedeck that kept getting reused by people and its not even close to accurate. Not even bothering to try and determine whats fake or not for current Zen craziness.

With the data we have 2.78b will likely be just as good for Zen as it is for skylake which is a very different conclusion then what otherwise might be painted.