Looks like they’ll be around 21,000 multicore and 3000 single core?
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.
First M3 Max Benchmark!
- Thread starter macduke
- Start date
- Sort by reaction score
Looks like they’ll be around 21,000 multicore and 3000 single core?
Analysis: This seems like a pretty good jump on the multicore, roughly equivalent to M2 Ultra! But for OpenCL, it’s an improvement, but not significant. It’s a higher score than OpenCL for any previous Max chip, but I would think it would be higher? Need some Metal scores.
M3 Max 40 core GPU 48 GB GB6 Metal Score
Metal score is around %15 higher than my M2 Max (38 core). For apps utilizing hardware RT and mesh shading, difference could go up.
Metal score is around %15 higher than my M2 Max (38 core). For apps utilizing hardware RT and mesh shading, difference could go up.
Blender 3.6 Benchmark
M3 Max between M2 Ultra 60 and 76 GPU cores. Blender 4.0 should give additional performance gains (let's wait and see...)
A few percent points below 3070 Laptop GPU so lets see what kind of improvements RT support in version 4.0 will bring
M3 Max between M2 Ultra 60 and 76 GPU cores. Blender 4.0 should give additional performance gains (let's wait and see...)
A few percent points below 3070 Laptop GPU so lets see what kind of improvements RT support in version 4.0 will bring
I am curious whether this is with or without Metal RT enabled. If this is using cycles-internal compute-based RT, then the improvement is absolutely staggering. If this is with Apple's hardware RT, the improvement is merely good (lower than I would expect to be honest).
Blender has supported a Metal RT backend for a while, the 4.0 is merely enabling it by default on M3 and later. Anyway, Geekerwan has done some quick raytracing tests and the M3 Max seems to be on par with OPTIX 4060 (desktop)/4070 (laptop), which is a very nice result (and would be 2x faster than M2 Max!). Looking forward to official Blender benchmark results!
P.S. Based on Geekerwan's tests, the M3 Max result in the 3.6 database is without hardware RT. This is a very impressive improvement just from Dynamic Caching and the improved GPU scheduling, and it illustrates how inefficient GPUs are when running complex shaders.
How did you come to this conclusion, for us mere mortals (who can't watch a YouTube video right now)?
In games, do you think dynamic caching would improve the lowest frame rates in particular? I suppose the most complex scenes are those that run at lower frame rates.
Last edited:
I doubt that many games have shaders that can benefit much from the new technology. Although Geekerwan did record some nice improvements in BG3.
His tests with hardware RT enabled show a bigger improvement. Max with RT on should be comparable to desktop 4060 with OPTIX
Game Mode activates automatically when a game launches. Any gaming benchmark would utilize it 👍🏼
BG3 and Shadow of the TR tests are probably fullscreen unless the reviewer specifically missed this detail.
Apple should sell the M3 MAX with 40 GPU cores and with at least 20 CPU cores and not only 16 cores.
The benchmark results would be much better.
Apple also forgot to sell the M3 MAX with ARMv9.2.
They're probably thinking about power consumption on the 14" and 16" devices. The full M3 Max already throttles a bit on the 14" due to thermal/power constraints.
The ultra is going to be a massive step ahead of anything we've seen before from Apple though.
Full Max has always done this so I doubt they care. I guess it's better than limiting it to the 16" for them.
That's why I bought the 16-inch MacBook Pro M3 MAX maxed out.
Yes we all are still waiting for the MacBook Pro M3 ULTRA, probably 18-, 19- or 20-inch.
In 1080p (and CPU limited) the performance in Baldur's Gate 3 on the M3 Max has increased by 19%, at 1440p the performance increase is 33% and at 2160p the increase is 35% compared to the M2 Max.