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.
M3 Max GPU vs NVIDIA RTX 4090 Laptop
- Thread starter AnonFr13n45
- Start date
- Sort by reaction score
It will probably come close in some rasterization/gaming benchmarks. But for many tasks the 4090 will be massively ahead.
Why not? The OP wants to compare the most powerful GPU in a notebook in the PC and Apple world. Whereas your comparison is between the most powerful GPU in a desktop in the PC and Apple world.
On the question, depends what you want to do with it, and what your expectations are. M3 max has more RAM and can be upgraded to enormous RAM amounts than no VRAM of any laptop can match. For games and 2d rendering they should be close. But for 3d rendering stuff I assume 4090 to be ahead still, assuming you are ok with the jet engine noise in your room.
"As fast" for what?
For gaming or workloads that are heavily optimized around CUDA? Doubtful, but one can dream.
For areas Apple has traditionally done well in? It will probably depend on the app and what exactly is being done. Time will tell.
For workloads that require massive amounts of ram? Well the 4090 just can't run those so that's a win for the M3 Max!
Either way we'll know more in the coming weeks.
OP asked IF m3 max will be as fast to 4090 laptop....he was not asking what will be the difference between those two
So based on history...the closest Apple gpu to nvidia 4090 laptop will be the desktop M ultra
Last edited:
To be fair, given the power consumption of the Nvidia GPU, Apple could as well put an M3 ultra in a laptop. And the power consumption may end up being lower.
I'm sure that @leman and @MayaUser are correct.
The GeekBench OpenCL benchmarks are currently showing the M2 Max at around 80,000, the 4090 Laptop just under 182,000. There's surely no way that the M3 Max will even come close to bridging that gap.
(Note: I don't know which version of the M2 Max that figure represents. It may be a blend of the two different numbers of GPU cores.)
As one might expect, the M2 Max scores much better on the Metal benchmark, just under 132,000, although I don't know just how much of an apples-to-apples comparison (no pun intended) that really is. (I don't believe there would be any Nvidia benchmark numbers for Metal .... 😬)
Of course, for some workloads -- ones that suit the M3's hardware optimisations, to state the obvious -- the M3 Max may be much closer to the 4090 Laptop but, equally, the 4090 may have advantages in other scenarios (CUDA being an obvious possibility).
Incidentally, the GeekBench OpenCL list has the M2 Ultra scoring just under 119,000, but comparing that with the 4090 is, I think, comparing a laptop GPU with a non-laptop one. (Again, I don't know which version of the M2 Ultra that represents.)
Andrew
I wouldn't look at OpenCL results for Apple Silicon. They are known to run much slower than Metal (for reasons I don't know).
Apple deprecated OpenCL way back in macOS Mojave, years before the first AS Mac was released. It's just not optimized for the new hardware.
These are perfectly valid points, but the GB OpenCL numbers are the only ones I know of that give a direct comparison between Nvidia and Apple GPUs, albeit one that certainly isn't optimised for the Apple GPUs (it may well be that running OpenCL on Nvidia GPUs isn't ideal either).
So, yes, absolutely, the Metal numbers for Apple GPUs are considerably higher -- which is why I mentioned the one for the M2 Max (although it's still much lower than the 4090 Laptop OpenCL number) -- just as one might expect that Nvidia GPUs might have an advantage for CUDA-based workloads.
So looking at a combination of the OpenCL and Metal numbers does give some feel for relative performance. But, in practice, the best comparison is always going to be based on rigorous benchmarking of whatever specific workloads are of interest -- which I presume doesn't yet exist for the M3 GPUs. It seems reasonable to assume that there will be a huge variation for different workloads (e.g. 2D vs. 3D). Also a huge variation in power consumption (and noise!).
(Note: I don't suppose there are any CUDA benchmark figures for Apple GPUs. I can't say I've ever spent any time looking for any, but if any of you know of any, I'd be very interested!)
The best thing to do is compare the highest numbers, which in this case would be Metal vs. Vulkan. All GB6 GPU benchmarks perform the same work and use the same algorithms, so it boils down to the maturity of the respective code and the frameworks. Both Metal and Vulkan codebases should be reasonably mature (in fact, I wouldn't be surprised if they use the same code under the hood). As OpenCL is neglected by both Nvidia and Apple I wouldn't use it for anything.
With apologies to the OP for getting into what is probably too much of a digression ....
Comparing Metal and Vulkan is certainly an option but, in the case of the 4090, the OpenCL numbers are quite a bit higher (for the Laptop version: approx. 182,000 vs. 156,000). I don't have an explanation of that, although I would expect that for at least some workloads on Nvidia (compute rather than graphical, for instance), both OpenCL and CUDA would be more efficient. And given the CUDA-oriented optimisations in the nvcc compiler, I would also expect CUDA to be (far?) superior to OpenCL for Nvidia GPUs. So, while Metal is obviously the best option for Apple GPUs, I wouldn't feel comfortable making that claim for Nvidia.
The bottom line is, I think, that there isn't a good, simple, apples-to-apples comparison (and it may be unrealistic to expect that there should be one, given the huge range of workloads that GPUs can be used for).
If one wishes to make use of GeekBench, I agree that the best -- or, at least, the least bad -- option may be to compare the best numbers for each GPU that is of interest which, in this instance (M2 Max and 4090 Laptop) appear to be Metal and OpenCL. It's just a shame that GB no longer seems to have CUDA numbers.
Yes, but in that case, performance on battery would suffer just like Windows laptops, which is not how Apple designs laptops. Fan noise would be higher as well. It would be a nice desktop replacement bu certainly not a portable device.
I think highest numbers for each platforms should be compared. Metal for Mac and CUDA/Vulkan/OpenCL (whichever is higher)...
Last edited:
For gaming? Snowballs chance in hell.
Video editing, probably trades places depending on software.
3D rendering, I’d imagine it gets close, but I’d bet on NVidia still being ahead on this one.
Abstract stuff like tflops and such? Doubt that’s even relevant.
Video editing, probably trades places depending on software.
3D rendering, I’d imagine it gets close, but I’d bet on NVidia still being ahead on this one.
Abstract stuff like tflops and such? Doubt that’s even relevant.
Good rule of thumb is about 20-40% performance gain just from good CUDA code vs OpenCL last time i wrote smth there. And even 50%+ using cublas/cudnn and other nvidia stuff baked in NV ecosystem for more complex workloads.
Im not sure about whole GeekBench situation but 15-20% should be minimum gain from same quality CUDA code compared to OpenCL, so score could be around 200 000 - 240 000 for mobile RTX 4090.
Last edited:
Don't be so smug on Apple's GPU power usage when gaming on laptops. My M1Max MBP 16 gets about the same battery life as my Legion Lenovo 7 3080 playing the same game BG3. They're both dead in about a hour. Now browing and video editing, the MBP wins everytime on battery life.
I suppose the Lenovo reduces its GPU performance by a lot when unplugged.