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

[nismotigerwvu]

The hell. My CPU is utterly terrible in comparison. 965 BE @ 3.4Ghz stock took 4:05.

Spoiler

That just doesn't seem right. I'm running a 955 BE at 3.5Ghz with the NB at 2.6Ghz. I mean maybe I/O plays a role but the difference between our results makes little to no sense to me http://imgur.com/a/JR1G7

 

[R0H1T]

What spreadsheet?

This one, apparently it's still unconfirmed but shows 36s for 200 samples from (Blender) ryzen ~ https://docs.google.com/spreadsheets/d/1O2Or6XOETZr3a4gLAuCuYEx8m5X_lv23LdwZiCZpBcM/edit

 

[Hi-Fi Man]

i7 4790k @ 4.6GHz (4GHz uncore) with DDR3 1866MT/s CAS9 completes the render in 55 seconds with 150 samples.

 

[MajinCry]

That just doesn't seem right. I'm running a 955 BE at 3.5Ghz with the NB at 2.6Ghz. I mean maybe I/O plays a role but the difference between our results makes little to no sense to me http://imgur.com/a/JR1G7

My NB is either @ 2.4 or 2.6 as well. Had the thing on my hard drive. I wonder if having it on my HDD somehow slowed things down. I'll gi'es it a run on my SSD.

Edit: No difference. The run took 4:05.36

Bloody odd. Maybe it's a python thing? I've got several versions installed for legacy support.

 

[nismotigerwvu]

Even then, I ran it from my HDD and not my SSD with a youtube video on Firefox and Thunderbird and whatnot running in the background. I mean maybe it can keep everything in RAM (I'm running 12GB) but that difference just is really odd.

 

[Fanatical Meat]

What are you guys using to open this file?

 

[nismotigerwvu]

Dropping the background tasks brought my time down to 2:55

You need to install Blender to run this. You'll find everything you need here

 

[nismotigerwvu]

My NB is either @ 2.4 or 2.6 as well. Had the thing on my hard drive. I wonder if having it on my HDD somehow slowed things down. I'll gi'es it a run on my SSD.

Edit: No difference. The run took 4:05.36

Bloody odd. Maybe it's a python thing? I've got several versions installed for legacy support.

What OS are you running? I'm in W10 X64

 

[linaaslt]

Xeon 5660@4.2 windows 10 64bit, Blender 2.78a 64bit
100 samples 36.54sec
150 samples 55.73sec

Not too bad.

 

[MajinCry]

What OS are you running? I'm in W10 X64

Windows 7 x64, fully updated. Full system is:

CPU: AMD Phenom II x4 965 BE @ 3.4ghz
GPU: AMD Radeon HD 7850 2GB
RAM: 4x4GB DDR3 @ 1333MHz
Mobo: AM3 ASRock M3N78D
Soundcard: Creative Soundblaster X-Fi Titanium Fatal1ty Pro

I've seen it mentioned that Windows 10's threading benefits older processors, but I don't think there's been any tests to confirm that. Like running Fallout 4 with draw distances maxed at the top of the Corvega plant. That drives me down to the low teens in FPS.

 

[Dygaza]

This one, apparently it's still unconfirmed but shows 36s for 200 samples from (Blender) ryzen ~ https://docs.google.com/spreadsheets/d/1O2Or6XOETZr3a4gLAuCuYEx8m5X_lv23LdwZiCZpBcM/edit

AFAIK that spreadsheet isn't made by AMD, but some user in Reddit. AMD James just replied to that tread and corrected that their Ryzen and 6900k results were done with 150 samples. Whoever maintains that spreadsheet should fix it.

 

[jhu]

Throughput wise when i compare the results with my 48 core Opteron setup... it looks like each Zen core (with SMT) is pushing the same as 2.72 K10 cores worth. Basically making this perform like almost 22 K10 cores at the same Ghz. If that proportionate relationship holds we are talking about a chip that does about 19.5 on Cinebench 11.5 and around 1716 in Cinebench R15. Doggone impressive if you ask me.

This isn't quite right because IPC/core decreases as core count increases. This is especially true for multi-socket systems because data resides in memory local to one CPU. When other CPUs access that data, they have to fetch from non-local memory, and that's slower. Having large caches helps but it's not exactly 1:1 scaling.

i'm not a blender guy, but wouldn't this benchmark be stressing the floating point units more than integer? What says folks here?

FP is the bottleneck, but the renderer is written in mixed C/python (if I recall correctly), so there is still going to be a fair amount of integer in there.

 

[nismotigerwvu]

Windows 7 x64, fully updated. Full system is:

CPU: AMD Phenom II x4 965 BE @ 3.4ghz
GPU: AMD Radeon HD 7850 2GB
RAM: 4x4GB DDR3 @ 1333MHz
Mobo: AM3 ASRock M3N78D
Soundcard: Creative Soundblaster X-Fi Titanium Fatal1ty Pro

I've seen it mentioned that Windows 10's threading benefits older processors, but I don't think there's been any tests to confirm that. Like running Fallout 4 with draw distances maxed at the top of the Corvega plant. That drives me down to the low teens in FPS.

That's a possibility. What are your usage percentages looking like on your cores during the run? On W10 all 4 of mine are pegged at 99~100%. My Llano based HTPC is still on W7 so I'll give it a go tonight when I get home and see if the OS makes that much of a difference.

 

[MajinCry]

That's a possibility. What are your usage percentages looking like on your cores during the run? On W10 all 4 of mine are pegged at 99~100%. My Llano based HTPC is still on W7 so I'll give it a go tonight when I get home and see if the OS makes that much of a difference.

Task manager (as bad as it is) says 100% usage across all four of my cores.

 

[Lebdnil]

Ok, my 2 systems (using Blender 2.78a 64bits, Win10 x64 fully updated, new Ryzengraphic_27.blend with default 150 samples):
1- 1m37.46s
Skylake Core i5-6600 (running at 3.63GHz with max power package 37W). Asrock Z170Extreme4 + 8Gb DDR4-2133 Kingston dualchannel

2- 2m17.36s
Sandy Bridge Core i7-2630QM, Acer Aspire 5750G also with 8Gb DDR3-1333 dualchannel. It starts at 2.6GHz (2.594GHz to be exact) using 46W (throttlestop monitoring, same as Skylake system), then drops to 2.4GHz using 40W, so 2.45GHz@41W average or so.

Using Experimental version dated Dec 11, 2016 (also from blender website):
1- 57.86s, Skylake i5-6600 this time using 42W
2- 1m42.81s Core i7-2630QM, using 44W. Again, starts at 2.6GHz using 48W, then drops to 2.4GHz using 44W, and last drops to 2.3GHz using 42W. 2.4GHz@44W average approx.

Screenshots here: https://1drv.ms/i/s!AjqAFTn4xM0Zga9z4n8bxyoHpalT4Q
(yep, I know, I'm using different Throttlestop versions, but 6.00 doesn't work with Skylake and 8.30 works worse than 6.00 in my old Sandy)

Byes

 

[IEC]

57.03 seconds using 150 samples on a i7-6700K stock (4C/8T, 4.2 GHz Turbo). Win 10 Pro, Blender 2.78a x64

 

[[DHT]Osiris]

Xeon 5660@4.2 windows 10 64bit, Blender 2.78a 64bit
100 samples 36.54sec
150 samples 55.73sec

Not too bad.

On a x5660? Crushing it dude. Is this with the same settings as everyone else (are there other settings?).

 

[nismotigerwvu]

Just for fun, I ran this on my Dell Venue 11 Pro (i5 4300Y 2C/4T 9 watt boost/6 watt sustained, base clock 1.6 GHz, 8GB ram, W10 X64). Honestly, it didn't embarrass itself but it isn't setting any speed records. Considering it is running at about 1/10th the wattage per core, the fact that it performed in the neighborhood of my Phenom II on a per core level is really quite impressive. It boosted for the first minute or so and then settled in at ~1.45 Ghz for the rest of the run, which I'm assuming is due to some minor thermal throttling.

 

[William Gaatjes]

57.03 seconds using 150 samples on a i7-6700K stock (4C/8T, 4.2 GHz Turbo). Win 10 Pro, Blender 2.78a x64

That fits with the i7-6700(non K) i tested with.

Blender 2.78a x64.
Skylake 65W. I7-6700@3.4GHz, 4GHz turbo)
DDR4 memory @2133Mhz.
HT enabled : 8 cpu cores in W10 pro resource monitor
set at 150 samples.

Time = 1:04:42.

My A10-6700 Richland (piledriver) (3,7GHz, 4,2GHz turbo)
DDR3@1600MHz.
But it has only 2 modules with 2x FPU total.

Time = 4:18:42.
There is only one strange thing, before when i tested the a10-6700 under windows 7 it used to go up to multiplier 43 with prime95.
Now under windows 10 it does not go higher than multiplier 42.
I tested it with 1 core and prime95 as well to keep the core constantly at multiplier 42.

edit:
Well, i just got it to go up 43 occasionally for a second or 2 after a minute at 1 thread with prime95.

 

[Abwx]

This isn't quite right because IPC/core decreases as core count increases. .

Scaling is something like 99% with CB 11.5, dunno for Blender but it seems to be in the 95%+ range..

 

[KTE]

2- 2m17.36s
Sandy Bridge Core i7-2630QM, Acer Aspire 5750G also with 8Gb DDR3-1333 dualchannel. It starts at 2.6GHz (2.594GHz to be exact) using 46W (throttlestop monitoring, same as Skylake system), then drops to 2.4GHz using 40W, so 2.45GHz@41W average or so.

2.50GHz i5-2520M 2C/4T gets 3m43s

Doubling cores (2->4) improves 38-9%?

For ref, 2C/2T gets 5m12s.

Which is a 28.5% improvement 2T->4T.

Sent from HTC 10
(opinions are own)

 

[R0H1T]

AFAIK that spreadsheet isn't made by AMD, but some user in Reddit. AMD James just replied to that tread and corrected that their Ryzen and 6900k results were done with 150 samples. Whoever maintains that spreadsheet should fix it.

Yes I apparently did missed that, sorry for the confusion. The point's still valid though wrt official communication from AMD, they somehow manage to just mess up that last step.

 

[Lebdnil]

2.50GHz i5-2520M 2C/4T gets 3m43s

Doubling cores (2->4) improves 38-9%?

For ref, 2C/2T gets 5m12s.

Which is a 28.5% improvement 2T->4T.

Sent from HTC 10
(opinions are own)

I tried a second round without Throttlestop running: https://1drv.ms/i/s!AjqAFTn4xM0Zga945jkcLxFKFsfmjQ
2m15.10s almost the same. I don't really know why, because HT get a substantial boost (those Core i7-6700 get 30+% from HT vs my i5-6600).

Byes

 

[Ajay]

Don't know if I missed this, but SweClockers posted this image (zoomed in on the demo).
Apparently Blender v2.77 was used

 

[BeepBeep2]

Don't know if I missed this, but SweClockers posted this image (zoomed in on the demo).
Apparently Blender v2.77 was used
*img snip*

2.77 for the press demo and 100 samples that ran 25 seconds, 2.78a for New Horizon 150 samples 36 seconds