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

[Ajay]

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

OK, thanks - that was bloody confusing

 

[jhu]

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

Cinebench is NUMA aware. Blender isn't. Scaling is nearly linear for low core counts. For much higher core counts, it drops ofd fast.

 

[nismotigerwvu]

Just finished up testing on my A8 3850 based HTPC (W7 12 GB RAM) and saw 3:32. Not a bad showing for old Llano, 18.75% lower clocks 19.12% slower render. Basically lines up with the core being tweaked just enough to reach equal performance per clock even without the L3 cache.

 

[bystander36]

i7 5820K @ 4.4Ghz, 150 samples, on 2.78a: 39.3 seconds.

 

[Ranulf]

2:05.48 on the 8350 (stock), win7 64bit, 16gb ram DDR3 @1600, ran it from hard drive.

 

[unseenmorbidity]

I saw something interesting on reddit the other day. Someone pointed out that AMD said that these CPUs were using stock coolers.

The stock intel cooler, which correct me if I am wrong, doesn't even come with the 6900k. But could they have used that tiny cooler anyway? Could this explain some discrepancies that people are seeing?!

 

[sirmo]

I saw something interesting on reddit the other day. Someone pointed out that AMD said that these CPUs were using stock coolers.

The stock intel cooler, which correct me if I am wrong, doesn't even come with the 6900k. But could they have used that tiny cooler anyway? Could this explain some discrepancies that people are seeing?!

One of AMD's employees chimed in and confirmed they were using the Haswell-E cooler, forget the part number. Don't think it mattered much though since the runs were pretty short.

 

[BTRY B 529th FA BN]

X58 6 core Hyper Threading off 4.4GHz core
3.6GHz Uncore
3.6GHz QPI
2000MHz cl8.8.8.24

150 samples 01m:18s:95


X58 6 core Hyper Threading on 4.4GHz core
3.6GHz Uncore
3.6GHz QPI
2000MHz cl8.8.8.24

150 samples 01m:18s:74
This above text is wrong with HT on, the bios never enabled it

 

[Ajay]

X58 6 core Hyper Threading off 4.4GHz core
3.6GHz Uncore
3.6GHz QPI
2000MHz cl8.8.8.24

150 samples 01m:18s:95


X58 6 core Hyper Threading on 4.4GHz core
3.6GHz Uncore
3.6GHz QPI
2000MHz cl8.8.8.24

150 samples 01m:18s:74

That odd that both are the same - and they are lower than my rig (4.0 GHz core, don't remember Uncore and QPI).

 

[Deders]

i7 6700K @4.6GHz

150 - 50.07
100 - 33.90

Windows 10

 

[guskline]

One of AMD's employees chimed in and confirmed they were using the Haswell-E cooler, forget the part number. Don't think it mattered much though since the runs were pretty short.

HMM? I have a 5960x which is Haswell E and it did not come with a cooler.

Perhaps he meant a cooler designed to handle Haswell E heat output?

Intel did market an upscale heatsink but most Haswell E users bought a separate non intel cooler or like me, water cooled.

 

[unseenmorbidity]

HMM? I have a 5960x which is Haswell E and it did not come with a cooler.

Perhaps he meant a cooler designed to handle Haswell E heat output?

Intel did market an upscale heatsink but most Haswell E users bought a separate non intel cooler or like me, water cooled.

There is an AIO intel water cooler.

 

[BTRY B 529th FA BN]

That odd that both are the same - and they are lower than my rig (4.0 GHz core, don't remember Uncore and QPI).

Specially when my daily driver does this:

post # 96

Daily driver:
X58 - 6 core 12 threads
X5690 stock + Turbo (3.46GHz / 3.73GHz 1-2 cores)
Uncore - 2.66GHz
QPI - 3.2GHz
DDR3 - 1333MHz cl9

Blender @ 150 samples = 01:02:80


EDIT: Re ran Blender on my 4.4GHz Machine with HT on at 150 samples and it did 00:51:56. I think my bios didn't enable HT. I thought I enabled it in bios. After the results I turned it back off and booted up, checked device manager and clicked 'show hidden devices' and the other 6 cores weren't there in light grey (how a device usually shows when it's not in use) so the bios most not have enabled HT when I thought it did.


Thread reply # 185 Deder's 6700K @ 4.6Ghz scoring 00:50:07 vs my 4.4GHz X58 6 core score of 00:51:56 is right on target with CineBench15 results. Around 4.4GHz the 6c 12t X58 rig normally pulls around 1000 - 1031 in CB15


EDIT2: If those scores between the SR7 vs the 6900K were both at the same sample rate then I personally think it's safe to say Zen is on par Broadwell-E. Time to check AMD stock pricing.

 

[sirmo]

HMM? I have a 5960x which is Haswell E and it did not come with a cooler.

Perhaps he meant a cooler designed to handle Haswell E heat output?

Intel did market an upscale heatsink but most Haswell E users bought a separate non intel cooler or like me, water cooled.

I don't know dude, here is the post: https://www.reddit.com/r/Amd/comments/5ihole/amd_convinced_me_to_wait_for_zen/db8izzw/

Maybe it didn't come with all the CPUs.

 

[Haserath]

X5650 @ 2.93
Uncore: 2.666
Memory: 1.066 8-8-8-20

HT on: 1:16:99
HT off: 1:50:14

Good HT gains for this.

 

[PG]

The AMD guy posted the Intel heatsink part number. It's this one:

http://www.newegg.com/Product/Product.aspx?Item=9SIA1UH3Z96291

The BX at the front of the part number just means it's boxed, as in retail packaged.

It does not come with any 2011-3 cpus as far as I know. It's a separate purchase.

 

[nathanddrews]

Is anyone compiling these scores for posterity?

 

[SlickR12345]

Got 1.22 on my I5 3330 at 100 samples. Running 2.78, win10, with background applications running like chrome and stuff got 1.44 seconds at 100 samples.

 

[Yongsta]

i7 2600K (stock), 32GB DDR3 ram @ 1600, Samsung 850 Evo SSD, Windows 10 Pro, 100 samples, on 2.78a: 1:02:87.

 

[rvborgh]

Just for giggles... on my quad Opteron system i set 16 of the K10 cores to 3.4 GHz (to match Ryzen), set threads in Blender to 16, and set thread affinity for Blender to them and reran (150 samples). Result: 46.99 seconds.

Basically AMD must have been lying their rear ends off as far as throughput (40% my rear). Each Zen core is indeed pushing between 2.5 and 2.7 K10 cores worth on blender... and one Piledriver core at same clock is only worth around .85 of a K10 core (at least on Cinebench).

AMD might as well name the thing Phenom x20+ because that is what this thing is the equivalent of (conservatively speaking).

Kudos to AMD for creating the biggest sandbag job in x86 CPU history )))))

 

[.vodka]

Well, Lisa did say they exceeded their 40% target over XV. By how much, well, that's the question. So far it looks like one of the best cases of underpromise and overdeliver we've seen in the past few years, I think we could all get used to this approach to stuff in general. Not a Netburst -> Core jump, but yeah, pretty impressive nonetheless.


It's safe to say that AMD has left K10 behind.. at last!

 

[Makaveli]

i7-970 at 4Ghz HT on
Windows 10

100 samples = 38.25
150 samples = 58.11

 

[majord]

Just for giggles... on my quad Opteron system i set 16 of the K10 cores to 3.4 GHz (to match Ryzen), set threads in Blender to 16, and set thread affinity for Blender to them and reran (150 samples). Result: 46.99 seconds.

Basically AMD must have been lying their rear ends off as far as throughput (40% my rear). Each Zen core is indeed pushing between 2.5 and 2.7 K10 cores worth on blender... and one Piledriver core at same clock is only worth around .85 of a K10 core (at least on Cinebench).

AMD might as well name the thing Phenom x20+ because that is what this thing is the equivalent of (conservatively speaking).

Kudos to AMD for creating the biggest sandbag job in x86 CPU history )))))

I doubt this will be case for other applications. these Blender versions clearly run like shit on Excavator, and seemingly K10 also , not really representative, unless one believes Broadwell also is 2.7x a K10 core

 

[leper84]

Ran again with 150 samples with and without hyper-threading.

4790k, stock clocks. Windows 10

HT on- 58.54s
HT off- 1:26.23

 

[rvborgh]

Broadwell core is about 2.2x a K10 core in Cinebench 11.5 at same GHz... it seems to do even better with Blender (better utilization of SMT likely - so around 2.5). For older code SSE2 etc... K10 does just fine... as its got an FPU for each core unlike BD arch.

I doubt this will be case for other applications. these Blender versions clearly run like shit on Excavator, and seemingly K10 also , not really representative, unless one believes Broadwell also is 2.7x a K10 core