Mac: M3 - *Hardware accelerated RT (Part 1)

BMW:
RX 6800 XT – 0:22
MacBook Air M1 – 2:09

You can use RX 6800 with HIP drivers.

1 Like

How can I do that.

The Pro Driver said … if you are a content creator … install that driver …
(similar to Nvidia’s gaming vs content creator drivers I thought)
Supported hardware : AMD Pro/WX cards …
(I have the RX 6800 without any XT)

I tried to install it anyway but after nearly installation it skipped by saying,
no supported GPU found.
So I went back to standard Windows 11 optional
(not recommended WHQL driver)

Blender > System > says no supported device for Optix/…/HID …
RenderSettings let me choose GPU but its grayed out.

What am I doing wrong ?

But thanks for the Benchmark, what I would get if I would get something :slight_smile:

Just install the Radeon Pro Software for Enterprise. It should work.
You can try also beta version:
https://www.amd.com/en/support/kb/release-notes/rn-rad-win-21-40-beta-blender-3-0

Thanks, I will try to install again tomorrow …

Jebus!
Imagine what one of the current Mac Pros with dual W6900X 32GB DDR6 can do with Blender with AMD throwing some attention back towards this little DCC. Wonder if you can pool the VRAM to 64GB on those cards?

Big thanks for AMD staying with open source API. :metal:t2:

Here’s to hoping that there’s some future slide in dGPU function on future AS Mac Pros.

It’s great that AMD is back to high performance Blending. I expect similar performance on the Mac and Metal Cycles. Still, Nvidia Optix junk has much better performance, and it is unclear if we’ll see hardware raytracing on the Mac (it is announced for HIP). After all, AMD cards that support RT have been around for many months, but still don’t have full support in macOS. I think it depends on what plans Apple has for Apple Silicon and if they will continue to use AMD graphics cards in Mac Pro 2022

2 Likes

this just made my day, as i sent back my 16” and will get the binned 14”

bmw27_gpu.blend, M1 Pro 10/14 CPU+GPU render 1:11 so 71 sec.

16’’ MBP M1 Max with 10‑Core CPU, 32‑Core GPU and 32GB and Blender 3.1.0 (Dec-16-2021)

For bmw27_gpu.blend with GPU the render time is between 0:42 and 0:43.

The 32 core is only 65 % faster than the 14. ops did not see he was using the cpu too….

so For bmw27_gpu.blend with GPU+CPU the render time is between 0:41 and 0:43.

Still not much different.

so 73% if I am not mistaken for 2 cpu cores and 18 GPU cores more?

Anyway thought that was interesting. I would have expected more like 35 sec from the max when seeing the 14 core result.

Yah blender performance is not scaling linearly to GPU cores. I wonder if that’s just what we need to expect with this SoC design or if it’s a consequence of the lack of performance tuning … or something else maybe blender specific.

I other way, I personally still think the binned 14” is a great pick.

or if it is related to compute and geek bench was right.

it seem to scale fairly linear in rasterization but not compute? Would that make sense? I don’t know…

as for the binned yes me too, I am still not convinced of especially a max in the 14” and for a portable I want battery and no noise (call me picky). But that sure will be an audible difference.

They are all 14”, binned, pro, max

doh that is easier

It seems to me like this is still very early days. You guys seem to be drawing conclusions about all of this based on very early alpha software.

Just look at Blackmagic and all of the optimization that’s been happening with Resolve over the past year.

I feel that as both the Apple engineers gain more understanding of how Cycles works, and the Blender engineers gain more understanding of how Metal works, things will get progressively faster and more optimized.

I said from the beginning that what Apple needs is a reason why Macs are so compellingly better that it can’t be ignored – they’re not there yet, but almost.

3 Likes

I agree. Blender Cycles usually scales great. There is for sure still much to be done in terms of optimisation. Not to mention many other things such as accelerated ProRes textures and video clips.

1 Like

Speaking of Prores. :slightly_smiling_face:

Mac is video editing Performance King.

I sometimes use hi-res ProRes4444 textures with alpha in Blender and it sucks in term of the performance. I dream about hardware acceleration.

2 Likes

That’s right. MacUsers can now continue to use Blender in their projects. The future is safe here for now. That’s great. I have migrated all my software to Linux and Windows. If nothing else had happened with Blender and Mac, I would have switched completely as a hard core MacUser. Apple knows that for sure :smiley:

2 Likes

Dedicated hardware acceleration à la the media engine blocks on the SOC, but for ray tracing acceleration, would be the next motivator I can foresee pushing me to an upgrade.

But then again, the NE cores seem to just sit there… who knows though, maybe its on the blender/apple docket for things to utilize (c’mon denoising!)

5 Likes

Yeah. I also love Apple Motion’s performance. Way more responsive than the After Effects bloatware, and more accessible as well.

4 Likes

the MBP M1 Max is 4 times SLOWER than the PC with the 2080.

yes of course it is… has RT cores for one.

If Apple cares they will have RT cores in their next ones too?

The again is is only M1 Max CPU not GPU… cheeky Nvidia.

Same way the last LTT Macbook video smells very Intel sponsored this seems sponsored by Nvidia.

In other news shame this seems only to work with cinema 4d

Why are people comparing the M1 Max with desktops that consume hundreds of Watts and think that it is relevant to the discussion. :wink:
Having said that, for the alpha version of Metal support for Cycles and Apple’s first cut at their own GPU’s, being a quarter as fast as a desktop while consuming a fraction of the power, leads me to think that nVidia have a lot to worry about.
For me, I’m happy with the 4.5x increase in speed over CPU rendering.

4 Likes

I don’t think this has the effect you are thinking or hoping for. Lol :grin:

From your post “… is 4 times slower” but you left out ONLY against Optix task in Blender… cause of RT cores. But you do post to a link that clearly the folks know this and are talking about it.

Cuda renders are pretty much on par

Games where both Metal and Cuda are available they’re on par.

Video editing task the M1M seems to be a step ahead

Programs where both devices are optimized for Apple’s new SOC, are on even ground, just like Apple said. :metal:t2:

Like others have said, RT cores could very well be coming to the next iteration of Apple’s SOCs. So that’s even more scary/embarrassing to think about for these current dGPUs

And don’t forget, this is all while the GPU shares a space with the CPU on the same die.

I don’t think people realize they’re actually championing Apple’s SOCs when they post this stuff. :grimacing:

7 Likes