AMD new CPU RYZEN using blender to render.

Way slower, but on an old Intel 4930K ( stock speed ): 6cores / 12threads.

Ofc the benchmark in itself was just there for show cpu performances, most of Blender user will prefer gpu’s CUDA or OPenCL for it as it is way faster than any cpu’s out… but thats a bit offtopic when you show cpu performances.

Attachments


Before everybody get wrong impressions about performances of different platforms:
Blender got many optimisations in the last days.

  • 2.78a renders in 41s43 on Linux with i7 6700k at stock speed (4GHz) vs 34sec on latest master from buildbot and 31sec on optimized build.
    So compare using 2.78a, to get comparable results with AMD.
  • OpenCL card should be used with one tile. You can set it to something like 4096x4096 to ensure it always render one big tile. Cycles will then define the best tile size internally for optimal speed.
  • on 2.78a with RX480 I get 21sec28 using one big tile (Linux with AMD GPU PRO 16.50). With optimized build it goes down to 15sec. So build and tile size are very important.

Offcourse. AMD have use the 2.78a standard x64 windows build. So comparaison need to been made on similar softwares. with cycles CPU ( no openCL CPU ), same tiles size, samples, etc.

Well, comparing gpu’s vs cpu was not really the purpose.

Sounds like a better chip than the 8 core i7 already. I believe OpenCL can leverage both CPU and GPU at the same time. Hopefully that means we get faster rendering on a Ryzen + Vega Combo.

I do wonder if CPU and GPU can be used together, what settings it will use. for example Cycles render on GPU loves higher tiles, where as on CPU much much smaller.

on my Xeon 2687w (gen1) I get 1m04s… that is … bad :frowning: second cpu that I got was DOA.

Still … 1m… something is off on the threading side of things (could be windows 10) Ideas?

Linux is much faster than windows. But only with cpu. Try liveusb linux and test render. I bet it will go down with 15%.

As much as I like Blender, and applaud AMD’s recent involvement with open source communities, Doesn’t anyone find it odd for AMD to put out a demo like this on their main product launch page?

Thats kind of interesting point of view mpan3, they did nothing to blender for so many years. Their line of radeon cards still not comatibile with cycles but they hade time to create radeon pro renderer and even integrate with blender. I do not like how they use blender for their marketing hype, and they hire 1 developer for blender. 1? really? Million dollar corporation?

AMD can use Blender for commercial use. It’s in the GPL.

Also. Everyone can download blender and the demo file to test it on their computer. People that would not consider to download blender.

If Blender had a build with benchmark for cpu and gpu where you just click a button then there will be 100x more downloads. People love benchmarks.

Yes, bad AMD, they did the split kernel, paid 3 devs to work on Blender and then promote it on their website and advert videos. All of that because they know the Blender community is rich and like to give money and will make them rich.
NVidia did soooooo much more and they like you and will do always better card for cheap because they like you. NVidia doesn’t care about money. They did so much for Blender and for you. Can’t give link to their commits because they did so much it would take an insane amount of space on this forum. BUY NVIDIA.

It’s nice to finaly see some competition for Intel!
But for 3D rendering i sill prefer GPU. With 2.78a on my GTX1070 this scene render in 9.8 seconds.

Yeah the good old RISK architecture - too bad the consortium at that time did not push it better.

That’s why it would be more interesting to see this kind of CPU benchmarks in production scenes, using Hair, SSS and volumetrics for example (which are usually slow in GPU)

I have been thinking about your comment a bit. Honestly can’t see how can you feel applaud by this. AMD is supporting the open source community as OpenCL is suppose to drive openness. Nivida’s CUDA is closed language, and it has given Blender foundation plenty of hardware to ensure they code for it. So this is just second side of the coin.

From my perspective the reason they released the file is to simply create hype about its product. If you have a quad core with HT CPU, you’d get significantly slow performance compared to AMD’s latest/greatest. Honestly that is good marketing move for AMD, and a surprisingly positive one from AMD in the first place (seeing how they did other products in the past)

The second positive is that this is free advertising for Blender. Seeing so many techsites talking about blender.

Third, AMD now feels more involved in providing support to Blender by even funding a full time developer to ensure OpenCL is properly implemented. This again is a big plus.

I would really like to hear more details from you, maybe I’m missing something.

Their benchmark approach for Ryzen is far more open than what they did in the past, now people can actually verify the claims of their new chip being able to beat Intel.

I think there has been some marked changes in AMD’s business practices since the days of their Bulldozer chip (which effectively knocked them out of the CPU race and signaled the beginning of the end for major performance leaps between chip generations). Their CEO has changed for one thing and they have since been more invested in open source technologies.

If you’re going to bench yourselves download it anew or dial back the samples to 150 as AMD managed to mess it up.

that does not look good for amd


looks like another faildozer

Reading the article, the only benchmark where Ryzen didn’t match or beat Intel was Cinebench (look further down and you find it doing well with Blender and other programs). Even then, someone in the comments suggested that they actually used the wrong image for the Cinebench test.

Though there’s also the issue in the form of these benchmark rumors coming from Chinese sources (which can often bring sketchy information), the same is true for Wccftech itself according to some here.

I find it likely that there’s no mistake here and that this explains why they went for Cycles instead of the widely-used Cinebench - they need a cherry-pick to show that the chip is good at rendering.

Let’s say that this Zen eight-core costs 300$ and really does match with a 1000$ eight-core from Intel, that doesn’t mean there’s a big market for it. Gamers will still be better off with a 4+Ghz quad-core for 300$. Workstation users that really benefit from multi-core will probably go with a dual-cpu rig. In between, there’s a bunch of “pro-sumers” who do stuff like 3D rendering and video encoding - that’s where this product needs to fit in.

Also, having AMD developers contributing code to blender is nice, but having them then use that code to benchmark their own processors is somewhat ethically questionable. It’s pretty easy to come on top with optimizations when you have commit access to the code base. I bet if intel put 3 developers on the cycles code, they could make their processors come out on top too.