Hey @leman I think you made a mistake here. The M5 is indeed around a stock 4050 mobile in 3DMark's benchmarks (slightly better actually). However in Blender the M5 scores around 1733 but the 4060 (no way to control for boosted GPUs in the blender database sadly) laptop/desktop chips are nearly double at ~3200. Were you thinking of the estimations for the M5 Pro? Those should easily blow past the 4060 and may even rival the 5060 - including the full M5 Pro and the 5060 ti.
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.
Do you think a theoretical M5 Extreme's GPU could match the RTX 5090?
- Thread starter ProQuiz
- Start date
- Sort by reaction score
Thank you for pointing this out! I think I must have been looking at CUDA scores, apologies!
M4 Max is no way near RTX 3090. RTX 3090 is 35.6 TFLOPS while M4 Max is 19 ~ 20 TFLOPS. To match RTX 5090's performance, you will need 190 M4 GPU cores. But then, Ultrafusion wont gonna help since doubling the core does not mean doubling the performance.
Also, dont forget that a lot of software are Nvidia based which boost the performance just like Apple Silicon works great with Blender due to unified memory.
M4 Max is closer to 16 TFLOPS.
Of course, peak FP32 throughput is increasingly less useful for comparison. On paper the M4 Max should be two times slower, but it is within 15% of 3090 score in Wild Life and within 5% on Blender benchmark suite.
You could make the same argument for pretty much any GPU, Nvidia included. Apple’s performance scaling has been on par or better than the rest of the industry.
I wouldn’t say that Blender performance advantage has anything to do with unified memory. The Blender benchmark suite uses fairly small scenes that should fit within the RAM budget of most consumer GPUs. Apples secret sauce for Blender are dynamic caching and superscalar execution, which allow them to achieve much better hardware utilization than anyone else.
At least Nvidia graphic cards has PCIe slots which allow them to add and expand unlike Ultrafusion which is a huge difference.
Still and so far, Blender is the only software that Apple Silicon chip is close to RTX 40/50 series from benchmarks tho not proven in realistic scenario.
Cant say M4 Max is as good as RTX 3090 in real life cause those tests are too short and games proves differently which requires RAW GPU performance.
Not quite sure what you are talking about. I’ve never seen an expandable NVIDIA GPU. How would that even work?
Where do you get this from? And what does “raw performance” mean?
You literally insert GPU on the motherboard and expand.
Games are great example to compare such as Cyberpunk 2077.
diamond.g
macrumors G5
Games are a horrible example because as some say when the performance isn't what is expected "they aren't optimized for Apple Silicon".
maflynn
macrumors Broadwell
PCIe slots allow you to upgrade you're systems GPU performance by replacing the GPU card, i.e., going from a RTX 3070 to a RTX 5080. You're post was awkwardly written to imply you could upgrade an existing GPU's performance - not replace it
To further add to your point, depending on the game, the amount of optimization will differ. There's been some games and at the extreme level there are some games that have AMD GPUs heavily optimized where as they hadn't put as much work on the nvidia side.
Alameda
macrumors 68000
thenewperson
macrumors 65816
Pressure
macrumors 603
I don't think we should expect a 25-30% GPU architecture improvement from M4 to M5 AND a 20% increase in GPU cores on practically the same process node (albeit third iteration - N3E to N3P). Much of the improvement is due to the higher memory bandwidth anyway.
That would balloon the die size and the M4 to M5 base chip retained it's transistor count at around 28B transistors.
That would balloon the die size and the M4 to M5 base chip retained it's transistor count at around 28B transistors.
I hope that M5 Max will be good enough to play WOW at ultra settings on 16 inch Macbook Pro 🙂
(Even though it's a CPU dependent game.)
(Even though it's a CPU dependent game.)
Last edited:
Sounds unlikely. The 5090 RTX offers just over 100 TFLOPs of FP32 compute. A single M5 GPU core running at 1.8 Ghz (which is a very generous estimate for M5 Max/Ultra) is just under 0.5 TFLOPs. You'd need to double either the number of cores or the frequency to get to 5090 levels.
Of course, TFLOPs is barely a helpful metric nowadays. It all depends on the workload. Apple has better compute efficiency, which plays a larger role, especially for complex shaders.
There are plenty of improvements besides the memory bandwidth. But the architectural differences make it very hard to provide ballpark estimates.
Pressure
macrumors 603
Yeah I agree, it also varies wildly by workload type but an increase from 120GB/s to 153GB/s (27.5% increase) in memory bandwidth certainly helps overall.
diamond.g
macrumors G5
Why doesn't the dynamic caching and superscalar execution help the gaming side as much as it does the rendering side?
Because Apple isn’t working on the games you’re comparing to the same extent as Blender. Also the workload is different I imagine.
diamond.g
macrumors G5
Is TBDR hindering Apple with respect to how say Assasins Creed Shadows performs? We know that Apple has the CPU advantage, so running at lower resolutions aren't as capped as in x86 land, but the GPU performance doesn't punch up in games like it does practically everywhere else.
I don’t think TBDR hinders Apple, but the reality is that the priority for these game studios will be PC and console. They aren’t going to make the changes necessary to take advantage of Apple silicon. That being said, Nat Brown stated that CP2077 uses many features of Apple Silicon.
A couple of things to note is firstly, the degree to which driver optimisation plays a part in game performance. I personally wouldn’t want that on Apple’s platforms. Secondly, many of these gpus have a brute force approach to graphics. Both Nvidia and Amd have many more alus and more power to throw at the problem.
SDAVE
macrumors 601
Geekbench Metal score will be 400,000 for 96 core M5 Ultra
Geekbench Metal score for 5090 is 380,000
Of course it won't have the raw power of the 5090 but it should be close in performance for irl performance
SDAVE
macrumors 601
There were some measurements done by some folks on X based on how M5 is performing and scaling
diamond.g
macrumors G5
I assume driver optimizations are needed because API's and the code compilation is too generic. Apple doesn't have that crutch, therefore they don't have to do
Sometimes I miss the days of dedicated graphics api per vendor.
Pre-DX12, Vulkan, and Metal (so OpenGL and DX<=11), the drivers were extremely complicated and filled with loopholes and game specific optimizations. Sometimes in fact you can read people who do porting of older games in WINE talking about all the things they have to do to get specific games working because those games relied on a particular behavior that wasn't exactly in-spec shall we say.
The new APIs were meant to simplify the drivers down and more of the performance control was being passed to the developers for good and bad. My impression is that promise wasn't fully realized. Some custom game engines still deliberately break spec to get a little extra performance boost and day 0 driver patches are still required to get them to work at all never mind be performant. Further's Intel's first forays into dGPUs were thought to be hobbled by poor driver performance (though I'm less certain if that was specifically DX<11 games or if DX12 games likewise performed badly). So the drivers may not be quite as salient as they once were, but still tricky and yet necessary to get right.
Geekbench OpenCL for the 5090, no Metal for the Nvidia chip 😉 (and sadly no CUDA on the latest version of Geekebench). I know Primate Labs advertises GB GPU as being crossplatform, but you can see the discrepancies between API scores and I would be very careful comparing Geekbench Metal on the Apple GPU with Geekbench OpenCL on the Nvidia GPU. That said, a theoretical M5 Ultra should be very competitive in Blender.
I think the question is why 96 cores (the current Ultra has 80) ... Apple tends to double core counts going from the Pro to the Ultra (M3 Pro very slight exception, but based on the M1, M2, and M4 Pro/Max/Ultra die design the GPU core counts have to be exactly doubled from tier to tier) - so you think the M5 Max is getting a bump to 48 cores from 40 and is the Pro getting a core count bump as well? What is your basis for that? I don't recall any rumor that the chips are getting a core count increase this generation though of course it's possible.
Last edited: