How good is Ryzen? see for yourself !

[AlienALX]

OK, so after a few weeks of letting the Ryzen launch digest I decided to do some digging this morning to see just how good it is.

The method of this is simple. First you need to go and download Blender, making sure you get the correct version for your OS (X86 32 bit or X64).

https://www.blender.org/download/

It is completely free.

Then you need the AMD Ryzen test file.

http://download.amd.com/demo/RyzenGraphic_27.blend

OK, so go ahead and install Blender. Then open the file (the .blend one) Then click on render, and Render Image (or F12).

Once the benchmark has concluded look for your time.

And a close up of mine, I got 50.25 seconds.

Now when you compare with Ryzen the magic number is between 36 and 37 seconds for a full render.

Let's see how other CPUs compare.

I am running a Xeon E5 2680 V2 8c 16t @ 3.3ghz with 16gb RAM and a Fury X.

 

[Wraith]

I don't mean to open a can of worms here, but Blender!? Really... I've been using it for years now and as far as unfair goes this takes the biscuit, AMD CPUs always and I mean always win out over Intel in Blender while rendering, the same way cuda does over Radeon. I said the same thing when we watced the launch of Ryzen and they showed the test.

 

[g0ggles1994]

Going by what Wraith said, I would like to see how long it takes a Bulldozer to render this, would be good to see some AMD vs AMD numbers instead of just AMD vs Intel

Also, it took my 4790K 01:02:55 to render it.

 

[NickOmega]

I was bored so...

02:48.44

 

[AlienALX]

5820k @ 4.5ghz

Not as good as I had hoped. Oddly going up to 4.6ghz made no difference, so it obviously wants the threads.

 

[Wraith]

By default Compute is disabled for Intel as it is experimental, but turn that on and et voila you get magical numbers.

 

[NeverBackDown]

By default Compute is disabled for Intel as it is experimental, but turn that on and et voila you get magical numbers.

Now once we get Ryzen, turn on compute, and see how it performs. Otherwise it's comparing apples to oranges

 

[Wraith]

Now once we get Ryzen, turn on compute, and see how it performs. Otherwise it's comparing apples to oranges

AMD CPUs already utilise compute when in Blender, it's only experimental and disabled for Intel. This is why I said the Blender demo is horsepoop..

 

[Kilbane]

This right here is the problem with blender, you make 1 small tweak that nobody would notice and my 4670K is performing pretty much the same as a 5820K

 

[NickOmega]

Regardless of which CPU's Blender favours, from looking at everyone elses results I think it's plain to see that I'm in dire need of an upgrade

 

[NeverBackDown]

Regardless of which CPU's Blender favours, from looking at everyone elses results I think it's plain to see that I'm in dire need of an upgrade

Well at least you have good timing.. since Intel just released new stuff and AMD is soon

 

[SPS]

By default Compute is disabled for Intel as it is experimental, but turn that on and et voila you get magical numbers.

What do you mean turn on compute?

 

[AlienALX]

What do you mean turn on compute?

This. Come on, blab !

And Kilbane, you too. Stop speaking in riddles and tell us the problem.

Right just doing some reading. Wraith, you mean compute as in Open CL or CUDA? so that would enable the GPU to jump in on the job, right?

I'm pretty sure AMD left it disabled during the test. IE - what you get when you run the bench file I linked to was exactly what they ran. Which seems about right, as a 5960x @ 3.5ghz renders the file in just over 37 seconds.

Edit 1. There could be a few. Handbrake was also ran on both CPUs and Ryzen still won.

 

[Wraith]

What do you mean turn on compute?

OpenCL 2.0 has been supported since Haswell it just isn't enabled.

This. Come on, blab !

And Kilbane, you too. Stop speaking in riddles and tell us the problem.

Right just doing some reading. Wraith, you mean compute as in Open CL or CUDA? so that would enable the GPU to jump in on the job, right?

I'm pretty sure AMD left it disabled during the test. IE - what you get when you run the bench file I linked to was exactly what they ran. Which seems about right, as a 5960x @ 3.5ghz renders the file in just over 37 seconds.

Edit 1. There could be a few. Handbrake was also ran on both CPUs and Ryzen still won.

Yes I mean OpenCL, but it's doesn't use the GPU. CUDA renders that scene in roughly 15 seconds. My argument is that we were never shown the settings used my AMD in their test, there are multiple ways Blender can be "tuned" to render faster and visa versa.

 

[SPS]

OpenCL 2.0 has been supported since Haswell it just isn't enabled.

Yes I mean OpenCL, but it's doesn't use the GPU. CUDA renders that scene in roughly 15 seconds. My argument is that we were never shown the settings used my AMD in their test, there are multiple ways Blender can be "tuned" to render faster and visa versa.

OpenCL does use the GPU. It is its primary purpose.

 

[AlienALX]

OpenCL does use the GPU. It is its primary purpose.

Yup that is what I have been reading. That is why people build rendering farms with up to 7 GPUs. Some dude on Linus' forums cuts up the back brackets, removes the DVI connector and crams in 7 GPUs on water.

 

[Wraith]

OpenCL does use the GPU. It is its primary purpose.

Aye but it doesn't

create a batch file with this

set CYCLES_OPENCL_SPLIT_KERNEL_TEST=1
start blender
exit

place it in the Blender directory.

OpenCL on Intel CPU only.

 

[AlienALX]

Have you checked in something like Afterburner that the GPU is doing nothing?

 

[Wraith]

OK here it is with PresX.

I have Firestorm too if you'd like me to run that.

 

[AlienALX]

Hmm well I never. Might have to dig out the handbrake test and try that !