Date: Thursday , December 15, 2016
On Tuesday of this week AMD presented to us its "New Horizon" event. This event was a soft launch of its upcoming desktop CPU that carries the architecture codename "Zen." This desktop Zen-based processor will be marketed under the name of "Ryzen." Undoubtedly a lot of us computer and gaming enthusiasts are excited about this, I know I certainly am.
While we can debate the worth of the content presented to us at New Horizon, as we know for sure that tech companies work hard to show new products in the best possible light, AMD did throw us guys that actual understand performance metrics a bone.
The video below shows that moment in the presentation. And AMD goes on to invite us to run the benchmark at home to compare and shares with us the file that is used in its Blender rendering benchmark....or did it?
The test is simply the new Ryzen CPU at 3.4GHz compared to an Intel Core i7-6900K processor. Both these are 8 core / 16 thread processors, which makes it a good comparison. The Ryzen CPU is not using any form of boost, but rather its locked base clock of 3.4GHz. The Intel 6900K is being run at stock with Turbo Boost, which makes things a bit fuzzy seeing the clocks are dynamic. Also, in notes that were shared by AMD on the Intel system being used, it notes that it is running 2400MHz dual-channel memory with a 16GB footprint. A bit odd since the system supports quad-channel RAM, however in the testing I have done in the last couple of days here, it seems that this particular Blender scene is not tremendously sensitive to memory bandwidth. I have run tests with RAM up to 3400MHz with etc. Toggling back and forth between dual and quad channel memories at cited 2400MHz seems to make no difference at all in this particular test, so if you see someone complaining about that, their point is moot for this discussion.
Blender is an open source 3D rendering program. It is a good program to use for looking at CPU Instructions per clock, and for reaching into multiple processor cores and threads. AMD suggests that we use version 2.78a 64bit so we downloaded that. AMD shared with us the file that it used for the demo as well and is downloadable from its New Horizon site and we moved forward with trying to replicate its Intel results, which showed a render time of a bit under 36 seconds.
So assuredly, me and many others started looking into the results. My primary reason was to validate what AMD was telling us in that the new Ryzen at 3.4GHz is pretty much on par with the performance of a Core i7-6900K. Now I do not have a 6900K on hand, but I do have a 6950X which is a 10C/20T processor, both of which are Broadwell-E CPUs.. Thankfully, in the BIOS of the motherboard I have using, I can disable cores, so that is what I did. Then I went on about my way to match AMD's 6900K results.
Long story short, I could not replicate the results, nor could anyone else for that matter. I tried till the wee hours of the morning to make sense of the results. Surely a lot of tinfoil hats were put on, but I was firmly in the camp of just being confused rather than looking to conspiracies. While AMD has made a lot of marketing claims in the GPU arena in the last year that are simply horseshit, this was CEO Lisa Su presenting and I simply give more value to her statements than I do to AMD's Roy Taylor's and Raja Koduri's.
We sent emails to AMD which were never answered. AMD did however finally come forward to confirm that it did not share the Blender file with the correct rendering variables.
"AMD_james281 - Set render samples to 150. A new file with the sample level set correctly will be uploaded shortly. Apologies for the confusion.
Blender 2.78a x64 is what we used, binary download from blender.org."
So this morning when we got time, we loaded everything up again with the new file from AMD's New Horizon page which seems to load the correct Blender rendering variables.
With our 6950X CPU cut down to 8C/16T in the BIOS, with the same core and thread count (and core architecture) as the 6900K processor that AMD used in its Blender demo, and all our cores clock locked to 3.5GHz, we were able to "match" the results shown at AMD's event in relation to Ryzen Blender performance. This result shows us coming in at just under 36 seconds render time.
First and foremost, WTF AMD? I love the fact that you shared the benchmark files and system specs with us, but if you are going to do this, at least hire some people that will make sure to get this right the first time. The last thing AMD needs is a self-manufactured conspiracy theory jump-off point. And spellcheck the damn slides? Yes, I am surely a bit of hypocrite on this last point, but HardOCP is far and away from a multi-million dollar tech company with its resources. On to what is important.
AMD has allowed us at least one data point to run with during its Ryzen soft-launch, and for that we are thankful. From what we can "prove" here today is that AMD's Blender demo was not smoke and mirrors and the Ryzen at 3.4GHz is possibly a little bit faster than Intel's Broadwell-E CPU that was launched earlier this year. This is a pretty big deal! This actually shows us that AMD and its Ryzen CPU have a shot at being competitive in the desktop CPU market almost assuredly.
Now IPC is not everything. We all know that clock speeds are a big part of that equation. Given that AMD noted Ryzen's TDP at being under 100 watts, it is very conceivable that we might see Ryzen scale to 4GHz and possibly a bit above that. But assuredly, I think in the mind of the enthusiast, if AMD can push clocks to 4GHz, Ryzen will very likely be a win. Obviously pricing is another lynchpin that will come into play.