【Maxon】Cinebench R20 Benchmark Thread

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

csbin

Senior member
Feb 4, 2013
816
8
136
#26
tweet1.png


Went from full sse/sse2 128wide code to AVX2+FMA 256bit.Power consumption nearly the same.I think Maxon has added only a few parts of embree on the renderer(looking at the leftover sse stuff).Embree as a whole is way more demanding than this.






Profane Tweet edited and image put back.

Iron Woode
Super Moderator
 
Last edited by a moderator:
Mar 7, 2019
60
15
16
#28
Intel i9 9900K @ 5.0 GHz (No AVX Offset)

Result.jpg


Living Room PC (Ryzen 2700X Stock)


LivingRoom.jpg
 
Last edited:

Abwx

Diamond Member
Apr 2, 2011
8,939
298
126
#29
There s a summary at Computerbase.de using datas from their community.

https://www.computerbase.de/2019-03/cinebench-r20-community-benchmarks/

What is to be noticed is not only SMT scaling being lower than in R15 but also that the tables have been turned , with Intel suddenly scaling better, this has nothing to do with instructions but with specific optimisation of the code to suit better a uarch than the other, at least in this specific scene.
 

ondma

Senior member
Mar 18, 2018
434
122
61
#30
The whole point was that as the latest gen AMD CPUs were likely to take the ST crown, the goalposts get moved in a manner that would favour Intel.
Whilst it's true that AMD will be improving on the AVX front themselves, they are still likely to lag in any workload that utilizes it.
Just dont give up do you? Current Ryzen certainly would not take the single threaded crown at 4.3 ghz or so compared to 5+ ghz on Coffee Lake. If you are talking about next gen, well, lets wait and see.
 

Gideon

Senior member
Nov 27, 2007
475
128
136
#31
What is to be noticed is not only SMT scaling being lower than in R15 but also that the tables have been turned , with Intel suddenly scaling better, this has nothing to do with instructions but with specific optimisation of the code to suit better a uarch than the other, at least in this specific scene.
I still don't see the point of Maxon taking the time NOW to deliberatly favoring Intel in this benchmark (compared to,say 2017). Zen 2 will certainly do excellently here, and AVX2 will just increase performance even more compared to Zen+
 

TheELF

Platinum Member
Dec 22, 2012
2,864
128
126
#32
Just dont give up do you? Current Ryzen certainly would not take the single threaded crown at 4.3 ghz or so compared to 5+ ghz on Coffee Lake. If you are talking about next gen, well, lets wait and see.
People where talking bout IPC at the same clocks where everybody (that is a AMD fanboy) believed that Ryzen would overtake intel in cinebench,but cinebench leaves about 30% + of current CPU cores idle so any CPU that can reach this threshold of IPC will have the exact same score in cinebench.
There is just no way to run something faster then what it let's you to.
 

Abwx

Diamond Member
Apr 2, 2011
8,939
298
126
#33
I still don't see the point of Maxon taking the time NOW to deliberatly favoring Intel in this benchmark (compared to,say 2017). Zen 2 will certainly do excellently here, and AVX2 will just increase performance even more compared to Zen+
This has nothing to do with instructions, rememeber, we are talking of the MT score comparatively to the ST score...

In Cinebench R15 SMT scaling of Zen was 1.42x, with Intel being a little lower, now AMD is at 1.23x and Intel at 1.26x, so something has changed in this respect.

If R15 was hugely underusing the CPU then this can be measured through power comsumption comparison when using one thread per core.

Since scaling went from 1.42x to 1.23x this imply that the CPU should consume about 1.42/1.23 = 1.15x more in R20 when loaded at 1T/Core.
 

jpiniero

Diamond Member
Oct 1, 2010
6,500
295
126
#34
MT Clocks should be in general lower due to using AVX. The Coffee Lake K models configured with an unrestricted TDP are going to maintain their clocks obviously.
 

naukkis

Senior member
Jun 5, 2002
213
39
101
#35
Zen 128-bit vector units are busy calculating 256-bit AVX vectors so obviously SMT scaling is lower as there's less free FPU-time for second thread. Intel's 256-bit vector units have more free time to do same job with 256 bit vectors than with 128 bit so there's more free resources to other thread to use.

That SMT scaling happens just like it should.
 

Abwx

Diamond Member
Apr 2, 2011
8,939
298
126
#36
Zen 128-bit vector units are busy calculating 256-bit AVX vectors so obviously SMT scaling is lower as there's less free FPU-time for second thread. Intel's 256-bit vector units have more free time to do same job with 256 bit vectors than with 128 bit so there's more free resources to other thread to use.

That SMT scaling happens just like it should.
That s one of the most uninformed post i ever had to read...
 

naukkis

Senior member
Jun 5, 2002
213
39
101
#37
That s one of the most uninformed post i ever had to read...
Yeah sorry about that :D

Try again, using 256bit AVX2 ops Zen FPU utilisation goes up per thread as calculating 256bit ops takes twice the resources vs 128 bit ops. There's less FPU-resources left to have SMT gains. Intel's 256 bit vector units uses same amount of cpu resources whether using 128 or 256 bit ops but with 256 bit ops there's high possibility that same code uses less instructions with 256 bit ops -> more fpu resources available for SMT gains.
 

Fir

Senior member
Jan 15, 2010
396
23
116
#38
Do they have a version that does not require the MS Store?
 

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
18,006
1,835
136
#39
Do they have a version that does not require the MS Store?
This version does not require that. Just download, and run.
 

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
18,006
1,835
136
#40
WOW, I just tried this on my 2695v3 Xeon, and it was painful watvching it run. And task manager said it was at 54% utilization@1.1 or 1.2 ghz !
Only 1732 ?? pathectic And I stopped BOINC, so nothing else was running.

Edit: Odd, after running this, I could not get BOINC to run@2.5 ghz/100% either. I had to reboot to get BOINC to run properly.
 
Last edited:
Apr 27, 2000
12,376
1,325
126
#41
There s a summary at Computerbase.de using datas from their community.

https://www.computerbase.de/2019-03/cinebench-r20-community-benchmarks/

What is to be noticed is not only SMT scaling being lower than in R15 but also that the tables have been turned , with Intel suddenly scaling better, this has nothing to do with instructions but with specific optimisation of the code to suit better a uarch than the other, at least in this specific scene.
Probably due to 256-bit SIMD and better pipeline optimization. We've seen this before with software like Linpack. Keep that pipeline stuffed and you have fewer execution resources available for SMT. It's like the polar opposite effect of SMT scaling (Intel only) on 3DPM v1. 3DPM v1 scales almost perfectly with SMT on Intel implementations.

I still don't see the point of Maxon taking the time NOW to deliberatly favoring Intel in this benchmark (compared to,say 2017). Zen 2 will certainly do excellently here, and AVX2 will just increase performance even more compared to Zen+
Depends on how conspiracy-minded you are. Intel could easily have put the money hat on Maxon. Hell we know they're using one of Intel's own software anyway, so why not? Intel wasn't as threatened in 2017.

Or it could just be that Maxon was tired of being stuck on SSE4 and wanted an upgrade. Behold the natural consequences.
 

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
18,006
1,835
136
#42
Do you have any why, after running cinebench 2.0 as well as during, my 2695v3 is stuck at 1.2 ghz and a little over 54% utilization ? I quit the program altogether, and started BOINC back up, and then BOINC does the same thing ! Then I reboot and BOINC is fine. Run cinebench and its back to 1.2 ghz and 54%, reboot again and all is fine !!!*^&$&^%$#&%
 
Apr 27, 2000
12,376
1,325
126
#43
@Markfw

Sounds like a processor bug to be perfectly honest. And not something unique to your sample. But you would really need to find someone else with the same CPU to confirm that.
 

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
18,006
1,835
136
#44
@Markfw

Sounds like a processor bug to be perfectly honest. And not something unique to your sample. But you would really need to find someone else with the same CPU to confirm that.
Well, it is a ES, so I suppose thats it. I have a 2683v3 thats almost identical, but retail, let me try on that one.

But VERY odd bug if thats it.
 

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
18,006
1,835
136
#46
Well, this is interesting... 14 core/28 thread Xeon 2683v3 works fine, does not have the "bug" the 2695v3 ES has, but gets beat by an 8 core 2700x.
My math (while it has no real bearing, so don't yell at me) is 16*4.0 (ghz*threads) = 64 while 28*2.5 = 70, so the Xeon should still win, but gets trunced soundly by the 2700x Ryzen.3678 for the Xeon and 3932 for my 2700x:

 
Last edited:

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
18,006
1,835
136
#47
See the above post, means that as soon as the new 3000 series comes out, I will be replacing all of my 26xx Xeons. They are down to under $200 on ebay,
 
Jan 31, 2019
143
19
51
#48
I have e5 2686v3 18cores
It gets 2800cb in cinebench R15 (stays at all core 3.15ghz)
And around 5709cb in cinebench r20 (for some reason it downclocks to 2.6ghz)

My e5 2686v3 score 4728cb with only 14 cores enabled. (All core downclocks to 2.6ghz again)

Without this xeon v3 turbo bios mod, cpu always downclocks to 2.2ghz for some reason and scores low on r20.
Hope this helps you.
 
Last edited:
Jan 31, 2019
143
19
51
#49
See the above post, means that as soon as the new 3000 series comes out, I will be replacing all of my 26xx Xeons. They are down to under $200 on ebay,
@Markfw
Try xeon v3 turbo overclock bios mod on this forum itself, i have provided guide few days ago.
I have e5 2686v3 18cores
It gets 2800cb in cinebench R15 (stays at all core 3.15ghz)
And around 5709cb in cinebench r20 (for some reason it downclocks to 2.6ghz)

My e5 2686v3 score 4728cb with only 14 cores enabled. (All core downclocks to 2.6ghz again)

Without this xeon v3 turbo bios mod, cpu always downclocks to 2.2ghz for some reason and scores low on r20.
Hope this helps you.


Sent from my ONEPLUS A3003 using Tapatalk
 
Jan 31, 2019
143
19
51
#50
Do you have any why, after running cinebench 2.0 as well as during, my 2695v3 is stuck at 1.2 ghz and a little over 54% utilization ? I quit the program altogether, and started BOINC back up, and then BOINC does the same thing ! Then I reboot and BOINC is fine. Run cinebench and its back to 1.2 ghz and 54%, reboot again and all is fine !!!*^&$&^%$#&%
The bug you faced is known as LFM bug, "lowest frequency mode" bug.
 
Last edited:


ASK THE COMMUNITY