Holy crap. (unexpectedly good way)…
Upgraded to new MacPro. (upgrading from trashcan, so literally 10x faster in some cases, very responsive, plays anything)
Have Vega II Duo 2x32GB card, apparently one card made out of 2 GPUs (option from Apple, not third party) This is outside of official spec for Flame, but workflow now includes Houdini, UE, etc.
2021.1.1.
As expected, Flame on Mac really only uses one of the GPUs
EXCEPT: MoVector Analysis node uses both GPU.
2021.2
BOTH GPUs seem to be used now. Can’t speak to stability/performance yet, but potentially a very helpful step for Mac folks.
Haven’t tried Octane. Using Solaris in Houdini, and free options like Karma, Renderman (non-commercial), Mantra is where I’m heading currently. Not sure if anyone has tried the AMD RT plugin on houdini and UE yet but likely going down that road too.
Is there a reason you’re not trying Octane? It’s one of the main reasons I’m considering upgrading to Big Sur but I haven’t used their shaders or anything before so I don’t know if it’s worth the drawbacks of updating.
Hi Greg. May be these links give you an idea of the current Octane X and Red Shift (beta) metal implementations in C4D also powered by a Radeon Pro Vega II Duo MPX Module:
I would love to hear from Autodesk if they are going to officially support multiple GPU’s in the future. Would also like to hear from Andy more details about the speed improvement in render times and interactivity.
Ok. Here’s what Ive found so far:
Waiting so long to upgrade has made the difference night/day.
Both GPU’s are engaged for UI but not for render.
Makes for snappy performance and interactivity.
(Both apparently DO work in render to compute precomp MoVector Analysis, so suggest you do this. Difference is significant vs computing/caching in Action)
No matter how you slice it, even w/o nVidia card, 2x 32GB VRAM is enormously helpful in making Flame feel like Flame.
I no longer worry about having scopes on during playback (ok 6-8k may still want to hide scopes).
Seems no more issues w large amount of gmask shapes. This used to feel like the masks got physically heavy to move around, and workarounds like working specifically in f12 solo mode were needed to keep working.
Action generally very interactive.
My OFOW setups from a couple years back were awful to open and render.
No longer the case.
Working in batch, no longer concerned w impact of 2nd GFX monitor, or needing it to display library instead of schematic or result to keep render times quick.
Same w my external HDR monitor thru the AJA.
These are all UI/interactivity only use of the GPUs, not rendering.
Where rendering would likely have a big impact from GPU is Action.(Tracer/Image likely too)
Games and 3d are using the GPUs for raytracing, etc. They have great implementation of leveraging all that VRAM, but Flame isn’t really designed for heavy 3D.
Would be helpful to get both GPUs useful during render on Mac.
Summary of experience so far:
Even tho I’m sure the render times will likely lag compared to the latest linux z8, the macPro is BY FAR the most interactive machine I’ve ever run Flame on. Not saying Linux is sluggish in any way, but this machine is fast enough that I don’t feel any lag at all.
Also if you wait so many years to upgrade, literally everything feels crazy fast.
It doesn’t sound like there’s enough of an improvement for me to upgrade to a 2nd GPU, unless Autodesk commits to greater utilization. As it stands this unannounced feature could disappear in the next version.
I don’t think it’s actually an unannounced feature as something perhaps done natively by Big Sur OS. And no, wouldn’t go out and buy a 2nd card for it, even if it does help a bunch w moVector stuff
What would you like to see @miles, I have a couple different machines to test, a 16core and a 28 core, with a single Radeon Pro II and a Radeon Pro II Duo.
First off:
the biggest reason my machine feels fast is that I’ve upgraded after 7 years on a trashcan.
A starving man is going to think whatever meal he’s given to be delicious.
The config I chose for MacPro may not be the most ideal for Flame, as it’s not the only app I’m leaning heavily upon it for.
Agree w it would be good to see if there is any science to it.
What I was pointing out is that it SEEMS to use both cards according to the GPU History in the Activity monitor. Should anyone go out and buy a second GPU for Flame on Mac based on this?
Definitely not.
Would any hardware upgrade be great after 7 years?
Certainly so.
Are the GPUs are being used usefully?
Perhaps the GPU are just passing Flame processes back/forth natively in Big Sur, but still getting same result as single GPU?
Honestly don’t know. It’s just something I noticed while testing.
Working in Flame in the UI seems to engage both.
This is where it seems different in 2021.2 vs earlier releases.
On renders, this seems to go to only one of the GPUs,
The exception seemed to be w pre-rendering the MoVectors vs caching directly in Action.
This seems to work even on a recent job using 2020.3.1 (still Big Sur).
If I get quantifiable results from this, I’ll post them.
It’d be great to see if it is reproducible by other folks and whether they see any quantifiable benefits.
I’d guess this has more to do w Apple than Autodesk, but don’t honestly know.
Agree w you Randy.
Can’t recreate the moVectors rendering using both GPUs currently.
Seemed repeatable on multiple occasions last week, but now uses just the one GPU at render.
Perhaps I had something on in the BG that I didn’t notice.
The UI still seems to use both GPUs, but not at render.
Scrubbing thru a particle setup in action seems to back this up.
If I find something further w the renders, I’ll post it.
I seldom see one GPU being fully utilized with the recommended Radeon Pro Vega II, and suspect that the Mac platform is not being optimized to keep the ‘systems’ group happy. Judging from Jon Hollis’ observations on the M1 Mac Mini on facebook, this is a situation that won’t be tenable for long.