yes it is quite amazing in some cases animation playback back is even faster then my 3080ti 😮 and that with an M1 Pro.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
3D Rendering on Apple Silicon, CPU&GPU
- Thread starter vel0city
- Start date
- Sort by reaction score
That could be the effect of UMA and shared caches — even the fastest GPU in the world can't do its job if it has to fetch the data over the slow PCIe bus first.
Perhaps we will see for sure when they switch to Vulkan.
There are some other nice tricks coming for Apple however like this one.
Blender Archive - developer.blender.org
The progress and level of optimization for blender is extraordinary and very encouraging. In a way the M* SoC is very beautiful and “the way forward “ since sooner or later energy efficiency will be the limiting factor for all form factors. And obviously we will get RT hw for mac some time if rumors are to be believed. It is just that we right now are in limbo and stuff that matters is not in place. In a year it is quite possible we are in a very good position. Here’s hoping!
Nice to hear that the Ultra is a step up from the M1 Max, I’ll be upgrading in a years time and giving the M1 Max MBP to my son - so I’ll be interested to see if the Apple Studio gets an upgrade before then.
Not that i need more than I have performance wise, 50+ million polygons, 8k documents and fast BPR renders are ok for what I currently do.
The report about Apple having to scrap their plans of making a beefier A16 is very intriguing. If the report is correct, Apple planned on integrating hardware raytracing amongst other big updates to the A16 GPU.
This definitely adds to the possibility of seeing hardware raytracing come to the Mac chipset with the M2 Pro/max or introduced with the M3 generation. We may yet see a day where Apple closes the GPU performance gap with Nvidia.
This definitely adds to the possibility of seeing hardware raytracing come to the Mac chipset with the M2 Pro/max or introduced with the M3 generation. We may yet see a day where Apple closes the GPU performance gap with Nvidia.
My guess is if that’s the case it only fits on the N3(whatever modifier here) process, which is why the schedules have slipped.
That’s in the scenario that Apple is actually behind instead of analyst speculation.
M2 adds the Media engine, thereby making the M2 die larger than the M1 die...
M2 also adds two more GPU cores, also contributing to increased die size...
M2 Pro / Max is rumored to add two efficiency CPU cores, and have a higher GPU core count than the M1 Pro / Max SoCs; M1 Pro = 16 GPU cores, M1 Max = 32 GPU cores, M2 Pro rumored for 19 GPU cores, and M2 Max rumored for 38 GPU cores...
So throwing hardware ray-tracing into the mix would also increase die size; with the M2 Max, all of that may exceed the size available from the reticle...?
Apple needs to start producing dies with a 3nm process to "move the needle" on the Mn-series performance gauge...
M2 also adds two more GPU cores, also contributing to increased die size...
M2 Pro / Max is rumored to add two efficiency CPU cores, and have a higher GPU core count than the M1 Pro / Max SoCs; M1 Pro = 16 GPU cores, M1 Max = 32 GPU cores, M2 Pro rumored for 19 GPU cores, and M2 Max rumored for 38 GPU cores...
So throwing hardware ray-tracing into the mix would also increase die size; with the M2 Max, all of that may exceed the size available from the reticle...?
Apple needs to start producing dies with a 3nm process to "move the needle" on the Mn-series performance gauge...
Reticle limit is currently 832mm² and the M1 Max is 432mm² (57B transistors) at 5nm. Something like the NVIDIA A100 is 826mm² (54.2B transistors) on 7nm. That's nearly a 50% reduction in size by going one node down just to give you some perspective of what can be fit inside the reticle limit imposed by current technology.
Last edited:
The rumors and leaked benchmarks make it quite clear that these are indeed two E-cores. Apple CPUs are organized in clusters of four cores anyway.
I'd say clusters of up to four instead. iPhone A-series chips have 2-core P clusters, and some M-series chips (T600x aka M1 Pro / Max /Ultra) have 2-core E clusters.
Die photos of the latter do suggest that it's just the 4-core E cluster from base M1 with two cores chopped off. It still has a full-size L2 cache, and the layout appears identical other than the two missing E cores.
I didn't realize that Qualcomm's Snapdragon 8 gen 2 GPU is both faster and more efficient than the A16's and has hardware raytracing, but that makes me incredibly excited for the future of Apple Silicon GPUs.
I'd say this almost guarantees that HWRT is just a matter of time, and while I don't think Apple care too much about losing to nVidia on their high end Macs, but the iPhone getting beaten by an Android phone is going to have Apple's GPU team rushing to battle-stations.
I'd say this almost guarantees that HWRT is just a matter of time, and while I don't think Apple care too much about losing to nVidia on their high end Macs, but the iPhone getting beaten by an Android phone is going to have Apple's GPU team rushing to battle-stations.
Qualcomm did some impressive work in this area, absolutely! Nice to have more competition. From what I’ve seen, A16 is still a small bit faster in actual games, but the difference is negligible now.
Im fairly certain Apples HWRT is done and ready to go, but probably designed for the 3nm node and therefore only likely to ship sometime next year. But who know, maybe we will get a backported version in the M2 Pro/Max?
Can someone here with a M1 Ultra run the VFXArabia Houdini benchmark tests? If you really want to be through you should run with and without viewport being displayed. This is a important benchmark with few samples and AFAIK no Ultra results yet.
Personally I’m very interested in seeing how the high resolution FLIP tests performs.
Personally I’m very interested in seeing how the high resolution FLIP tests performs.
Article from August 2020 had TSMC in volume production of 3nm wafers for 2H22; that didn't happen and probably messed up Apple's plans a good bit, as we see now with the "cancellation of the M2 Extreme"...
So now we look towards M3 on 3nm for an Extreme SoC...?
If we use maximum acceptable reticle size for M3 Max SoCs and aim for a 16-core CPU (12P/4E) build, how many GPU cores might be able to fit on said die, and how do we take ray-tracing cores into account; are they separate cores from the GPU cores, or might every GPU core have RT built-in...?
I dunno how all this stuff works...
M3 Extreme on 3nm (N3/N3E/N3P/N3S/N3X...?)
64-core CPU (48P/16E)
??? GPU cores (w/hardware ray-tracing...?)
1TB LPDDR5X SDRAM (ECC...?)
2.13TB/s UMA bandwidth
So now we look towards M3 on 3nm for an Extreme SoC...?
If we use maximum acceptable reticle size for M3 Max SoCs and aim for a 16-core CPU (12P/4E) build, how many GPU cores might be able to fit on said die, and how do we take ray-tracing cores into account; are they separate cores from the GPU cores, or might every GPU core have RT built-in...?
I dunno how all this stuff works...
M3 Extreme on 3nm (N3/N3E/N3P/N3S/N3X...?)
64-core CPU (48P/16E)
??? GPU cores (w/hardware ray-tracing...?)
1TB LPDDR5X SDRAM (ECC...?)
2.13TB/s UMA bandwidth
Not that I can tell so I will say probably not.
Dont think Blender even uses any Metal 3 features.
I have run this and also gotten data from a friend that have an ultra. Basically great perf on the ultra, slightly faster than a 5950x. Flip fits AS very well and is quick already for the max chip. These are the values I have found just checking quickly using different builds so take it with a grain (hehe)!of salt. Max at 0.05 30 min, ultra at about 21 Min
Last edited:
That sounds great, besides rendering final frames FLIP can be the most time consuming process in my workflow. If Ultra is at 21 min that’s faster than my 3970x which is great news.
Houdini Benchmark Cores vs Clockspeed (updated)
Something that's gonna face every artist building a new rig, which is better a fast cpu or more cores. Well let's find out, I'll focus more on simulation tasks (Pyro,Grain,ocean). This chart is going to be up to date, so I'll really appreciate it if you ran the tests as well and shared your...
www.vfxarabia.co
Just read the 7900 xtx creators review on techgage. Would be good to do these tests for Macs as well. https://techgage.com/article/amd-radeon-rx-7900-xt-radeon-rx-7900-xtx-creator-review/
I tried out 2 of the tests since the blender scenes are just standard scenes that are freely available.
Maybe I did something wrong but just testing the first blender example scene renders in about 9 mins(canceled the render at 20% after 2 mins) on my 16" m1 Max (32). On a 7900xtx that render takes less than 2 minutes in total and a 4090 finishes in just 21 secs... I guess that is a scene that is extra well suited for RT HW with all the small details but using quite simple shading. However, the diff in speed between the M1 series and PC based solution is quite extreme and shows why there is a need for RT hardware on a pro machine if it is to be used for 3D beyond a few stills from time to time. I also quickly tested the eevee scene and got render times for 100 frames in around 290s (slightly slower than a nvidia 3050 but faster than a radeon 6500xt. 7900 xtx did the 100 frames in 65 sec)
Eevee tests done on Blender 3.5 alpha using metal viewport.
I know some of you like to say that render times doesn't matter and we should not compare to PC. I totally disagree and we must know what the alternative solution is at. All in all, we better hope that Apples m2 pro/max/ultra chips have HW RT or we will have to wait another 1.5 years hoping for the M3 series.
I tried out 2 of the tests since the blender scenes are just standard scenes that are freely available.
Maybe I did something wrong but just testing the first blender example scene renders in about 9 mins(canceled the render at 20% after 2 mins) on my 16" m1 Max (32). On a 7900xtx that render takes less than 2 minutes in total and a 4090 finishes in just 21 secs... I guess that is a scene that is extra well suited for RT HW with all the small details but using quite simple shading. However, the diff in speed between the M1 series and PC based solution is quite extreme and shows why there is a need for RT hardware on a pro machine if it is to be used for 3D beyond a few stills from time to time. I also quickly tested the eevee scene and got render times for 100 frames in around 290s (slightly slower than a nvidia 3050 but faster than a radeon 6500xt. 7900 xtx did the 100 frames in 65 sec)
Eevee tests done on Blender 3.5 alpha using metal viewport.
I know some of you like to say that render times doesn't matter and we should not compare to PC. I totally disagree and we must know what the alternative solution is at. All in all, we better hope that Apples m2 pro/max/ultra chips have HW RT or we will have to wait another 1.5 years hoping for the M3 series.
It could be a bug in Cycles because your laptop should be slightly faster. It should be about a little over 4 times slower than the 7900xtx.
Someone in the forum can help you find out if this is a bug in Cycles and if you should report it.
Cycles Apple Metal device feedback
Yes, I’m also having a lot of crashes in 3.4. Trying to render I get these Metal Integrator Errors on animations along with just regular crashes that I wasn’t getting in 3.3.