AMD Ryzen (Summit Ridge) Benchmarks Thread (use new thread)

Page 26 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
I

iuno

Junior Member
Sep 2, 2016
2
0
1
I did some similar testing for the same reason yesterday, and now found your results.

The Stilt said:
MSVC 2013 is old and generally ICC provides superior performance to any other compiler.
Click to expand...
I tested your builds and experience the very opposite. The performance in general is shockingly low on windows. It takes 1/3 longer to render compared to my gcc builds on Linux.
On Windows, the performance of all builds is about the same, and especially slow. On Linux, the differences between avx, avx2 and sse are clearly visible.
Haven't tested gcc on windows or icc on Linux, though...

The Stilt said:
In the previous builds I posted, I only disabled AVX2 (AVX was enabled). Now I made an additional build (57.5MB) with only SSE2, SSE3 and SSE4.1 kernels enabled (CCX_HAS_AVX & CCX_HAS_AVX2 = False).
Click to expand...
I can't download that one, b/c onedrive says it was unable to scan this file for viruses.
 
The Stilt

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
iuno said:
I did some similar testing for the same reason yesterday, and now found your results.


I tested your builds and experience the very opposite. The performance in general is shockingly low on windows. It takes 1/3 longer to render compared to my gcc builds on Linux.
On Windows, the performance of all builds is about the same, and especially slow. On Linux, the differences between avx, avx2 and sse are clearly visible.
Haven't tested gcc on windows or icc on Linux, though...


I can't download that one, b/c onedrive says it was unable to scan this file for viruses.
Click to expand...

I had the completely opposite results in Linux, while I was testing Carrizo with Blender. The performance in Windows was significantly better than in Linux (precompiled official binaries used in both cases).

So what kind of differences are you seeing under Linux, between SSE2, SSE4.1, AVX and AVX2? And how exactly did you test the different options (compile time parameters)?

Here is a new link for the non-AVX/AVX2 build: https://1drv.ms/u/s!Ag6oE4SOsCmDhEiqxBAsBmTj3OTh
Add z (7 -> 7z) to the file suffix and use "blender" as a password.
 
I

iuno

Junior Member
Sep 2, 2016
2
0
1
Here are all the results I got. I made at least two runs in each config but on Windows the values were fluctuating heavily. I selected the two lowest values then.

Apparently, I did it the exact same way like you. Set CXX_HAS_AVX and CXX_HAS_AVX2 to FALSE in the Cycles CMakeLists file.

Code: 
build                    run 1    run 2
Linux                         
gcc (avx2)               2:41.73  2:42.67
gcc (avx2, prefer 128b)  2:46.37  2:44.15
gcc (avx)                2:49.58  2:49.71
gcc (sse)                2:58.08  2:57.59
Windows                       
icc (avx2)               3:38.22  3:38.02
icc (avx)                3:40.49  3:37.89
icc (sse)                     
official 2.77a           3:39.19  3:37.44

With "prefer 128b" I added this flag, which is supported by gcc and should help bulldozer CPUs:
Code: 
-mprefer-avx128
Just added it to the CYCLES_AVX2_KERNEL_FLAGS

BTW: This is Skylake, 4C/8T at 4.0 GHz.

SMT scaling:
Code: 
        4C/4T   4C/8T
Linux   3:37    2:41
Windows 5:13    3:37

Thanks for the reupload, I'll have a look the next time I boot Windows (which could take a while ;) )
 
Last edited:
SpaceBeer

SpaceBeer

Senior member
Apr 2, 2016
307
100
116
Here are results of some tests I’ve made on different systems during last year. Though I’ve been using different versions of Blender, but it’s still good enough for overview. And conclusion could be – Between Zen and Broadwell-E, I would choose GCN :d


 
inf64

inf64

Diamond Member
Mar 11, 2011
3,884
4,692
136
jpiniero said:
The scores are so bad I don't think it's legit, even accounting for the clock speed at 1.44 Ghz.
Click to expand...
Well ST score if it was achieved at 1.4Ghz is OK I suppose (~16% behind Broadwell EP per clock). MT score is just awful for some reason.
 
L

leoneazzurro

Golden Member
Jul 26, 2016
1,114
1,867
136
It is written in the header data that there is no L3 cache, this could probably impact MT (and also ST, of course, in a lesser extent)
 
S

Sweepr

Diamond Member
May 12, 2006
5,148
1,143
136
Last edited:
inf64

inf64

Diamond Member
Mar 11, 2011
3,884
4,692
136
It is obvious the score is way off for some reason. Just look at the MT portion, it is crazy low.
 
FIVR

FIVR

Diamond Member
Jun 1, 2016
3,753
911
106
That MT score just about matches my Haswell 4-core i7 laptop from 2013. How is a 32C/64T processor scoring the same as 2-generation old 4C/8T notebook processor? In MT no less?


I do not believe this is a legit score. The people getting exited about this score tell you exactly how believable it is...
 
B

blublub

Member
Jul 19, 2016
135
61
101
Well this CPU won't be out for some time so it is likely that some stuff is disabled.

ST performance isn't to bad for 1.4GHZ. Maybe SMT is also disabled hence MT is so low
 
The Stilt

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
The SKU for this CPU has not been leaked before (AFAIK) and it is definitely legit.
The score itself obviously doesn't represent the performance with a default configuration.

The SKU is supposed to have 1450MHz base clock, and 2.9GHz MSCB.
 
D

deasd

Senior member
Dec 31, 2013
606
1,038
136
Looks like L3 cache as well as turbo core have been disabled on that ES SKUs.
 
A

Abwx

Lifer
Apr 2, 2011
11,937
4,910
136
Could a Model 0 and Stepping 0 exist..?..

Besides the ST/clock is barely the level of the X6 since this latter is handicapped by the absence of AES instructions that boost the score of such equipped CPUs..
 
J

jpiniero

Lifer
Oct 1, 2010
17,015
7,409
136
The Stilt said:
The SKU for this CPU has not been leaked before (AFAIK) and it is definitely legit.
The score itself obviously doesn't represent the performance with a default configuration.

The SKU is supposed to have 1450MHz base clock, and 2.9GHz MSCB.
Click to expand...

Think maybe it's GMI wreaking havoc?
 
FIVR

FIVR

Diamond Member
Jun 1, 2016
3,753
911
106
So I guess the process is:
1. Acquire AMD ES CPU
2. Short AMD
3. Cripple L3 Cache, lock at base clock
4. Post terrible benchmark
5. Profit
 
sirmo

sirmo

Golden Member
Oct 10, 2011
1,014
391
136
The Stilt said:
The SKU for this CPU has not been leaked before (AFAIK) and it is definitely legit.
Click to expand...
Is that our new criteria for legitimate benchmarks? An SKU we haven't seen. Or do you have some other source supporting this conviction?
 
The Stilt

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
sirmo said:
Is that our new criteria for legitimate benchmarks? An SKU we haven't seen. Or do you have some other source supporting this conviction?
Click to expand...

:rolleyes:

It is pretty hard for someone to pull a fake result out of their arse, when there is valid information included that hasn't been previously leaked.
 
  • Like
Reactions: Phynaz and Arachnotronic
Status
Not open for further replies.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom