Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
…for LLMs… you can put up to 512GB of unified memory in the Mac Studio Ultra and destroy the Nvidia 5090. Watched a few YT comparison videos. Even when running LLMs locally that are small enough to work on the 5090, it got crushed by the M3 Ultra. Anything that requires VRAM gets crushed except video games. In addition, the power draw was almost comically different. Spend at least 4x as much on the power to run the 5090 when running but in idle mode, spend even more than the M3 Ultra. Almost everything was faster for those two purposes as well as advanced maths and anything science or work would be done on would use a Mac Studio…

Not intended to nitpick but I think the Youtuber was probably running some specific workloads that favor Apple Silicon.

For large models and to some extent fine tuning (when linking multiple Mac Studios) Apple Silicon is in a league of its own because it can fit everything into memory but for inference speed they aren’t, they need much better matrix hardware on the GPU side which hopefully we’ll get with M5.
 
It's not easy to fairly compare GPU performance between Apple and other manufacturers.
  • Apple prioritizes efficiency, and they likely will never allow their computers to have a huge power draw. Like why do you think they never push the single core performance on the higher tiers even with extra power room?
  • Comparing Metal-based software to OpenCL, Vulkan, or DirectX-based software is just hard. Other platforms have a software advantage because most applications were initially designed for those APIs and later ported to Metal. And why wouldn’t Apple just support other APIs instead of just using proprietary Metal? Because they want to control their stacks so they can control the efficiency of the whole process.
  • Especially games. I have yet to see a Mac game truly designed from the ground up for Metal. Why? Because it's not financially reasonable for game studios to do so. All those Tomb Riders, Resident Evils, Death Stranding are just ”ports”. They were not initially designed for Metal in mind.
  • The same can be said for software. For example, like Blender. Its Metal integration seems like an afterthought, likely because Apple only recently began supporting them. The only 3D rendering software that seem to effectively utilize Metal and Apple Silicon are Redshift, Octane, and other "financially funded" applications, where the M4 Max can actually rival the 4090.
  • Nvidia and Windows will always be in the lead in this area. Because they won the ecosystem war. And that’s mean majority of software and hardware will prioritize them first.
 
  • The same can be said for software. For example, like Blender. Its Metal integration seems like an afterthought, likely because Apple only recently began supporting them. The only 3D rendering software that seem to effectively utilize Metal and Apple Silicon are Redshift, Octane, and other "financially funded" applications, where the M4 Max can actually rival the 4090.
Odd my experience has been the opposite, in fact looking at GPU rendering benchmarks Blender 4.3 is probably one of the best GPU renderers, largely because Apple themselves have contributed code. The CB R24 CPU renderer is pretty clearly better than Blender's though.

E.g. unfortunately no M4s here, but M3 GPUs perform worse compared to Nvidia hardware in CB R24 than in Blender 4.3 according to this:


as compared to in Blender:


Examples:

CB R24Blender 4.3
409034,77210,944
407018,6655,112
M3 Max (40-core)12,9804,146
A550016,7024,216
M3 Max (30-core)10,3133,374

Oddly the 4090 in CB R24 appears to have proportionally lower performance relative to everyone else including other Nvidia chips than in Blender 4.3 and thus it's proportion to 40 core M3 Max is the same as in Blender, but all other Nvidia GPU hardware does better in CB R24 than in Blender compared to Apple Silicon*. The same is even more true for other GPU renderers as far as I can tell. However, if you have data that shows otherwise, I'd love to see it! Non-Blender GPU benchmarks are harder to find for Macs.

*Side note: user submitted benchmarks mean many Nvidia GPUs, especially desktop ones, will not be at stock (they are often factory or user overclocked).
 
  • Like
Reactions: novagamer and leman
I have yet to see a Mac game truly designed from the ground up for Metal. Why? Because it's not financially reasonable for game studios to do so.

There is one "game". It's a flight simulator called X-Plane which is primarily developed on Mac and then ported to Windows.

Skärmavbild 2025-04-06 kl. 07.51.04.png
 
It's not easy to fairly compare GPU performance between Apple and other manufacturers.
  • Apple prioritizes efficiency, and they likely will never allow their computers to have a huge power draw. Like why do you think they never push the single core performance on the higher tiers even with extra power room?
  • Comparing Metal-based software to OpenCL, Vulkan, or DirectX-based software is just hard. Other platforms have a software advantage because most applications were initially designed for those APIs and later ported to Metal. And why wouldn’t Apple just support other APIs instead of just using proprietary Metal? Because they want to control their stacks so they can control the efficiency of the whole process.
  • Especially games. I have yet to see a Mac game truly designed from the ground up for Metal. Why? Because it's not financially reasonable for game studios to do so. All those Tomb Riders, Resident Evils, Death Stranding are just ”ports”. They were not initially designed for Metal in mind.
  • The same can be said for software. For example, like Blender. Its Metal integration seems like an afterthought, likely because Apple only recently began supporting them. The only 3D rendering software that seem to effectively utilize Metal and Apple Silicon are Redshift, Octane, and other "financially funded" applications, where the M4 Max can actually rival the 4090.
  • Nvidia and Windows will always be in the lead in this area. Because they won the ecosystem war. And that’s mean majority of software and hardware will prioritize them first.

These are all the choices Apple has made, right? So, in order to maximize their hardware performance by tightly integrating it with their own proprietary instruction sets they have alienated the majority of GPU dependent apps. No different than iOS/Android developers not wanting to develop their apps for Windows Mobile, webOS, BB10 etc.

But even when there is native Metal support, M4 Max and M2 Ultra GPUs are not on par with dedicated GPUs. M4 Max/M2 Ultra are about equal to 3060 when you compare Octane X to OctaneRender. 4060 is about 20% faster while 4070 is more than twice as fast.

Besides all of that, the only real problem I have with Apple M strategy is to limit power draw on their desktop computers. That doesn't make any sense. Unless scalability is not that great or there is poor stability or perhaps it requires a lot of cooling. We will never know.
 
  • Like
Reactions: nathansz
  • The same can be said for software. For example, like Blender. Its Metal integration seems like an afterthought, likely because Apple only recently began supporting them. The only 3D rendering software that seem to effectively utilize Metal and Apple Silicon are Redshift, Octane, and other "financially funded" applications, where the M4 Max can actually rival the 4090.
I’m not sure that’s fair. Blender is actually optimized for Metal pretty well, and has had several Apple employees dedicated to improvements for a few years now.
 
  • Like
Reactions: crazy dave
Has anyone actually used a laptop with an RTX in it for any period of time. I've got one. Apart from the battery life being absolutely unusably terrible, there is the problem of the inevitable sausage barbecue. I have to plug an external keyboard in to prevent my sausages getting cooked and keep it away from my other sausage.

1743971227691.png
 
But even when there is native Metal support, M4 Max and M2 Ultra GPUs are not on par with dedicated GPUs. M4 Max/M2 Ultra are about equal to 3060 when you compare Octane X to OctaneRender. 4060 is about 20% faster while 4070 is more than twice as fast.

And in Blender M3 Ultra is almost as fast as dekstop 5070 Ti which has a TGP of 300W. M4 Max 40c is as fast as a desktop 4070 which has a TGP of 200W. M4 Max 32c is faster than desktop 7900 XTX which has a TGP of 355W. Many times it’s a matter of optimization, not the HW.
 
When the M1 Max launched, Apple said that it rivaled the flagship NVIDIA GPU at the time, the RTX 3080.

However, in 2025, when comparing the M4 Max MacBook Pro to PC laptops equipped with the flagship NVIDIA GPU, the RTX 5090, which costs the same as a M4 Max MacBook Pro, the MacBook Pro gets destroyed, it is not even close.

And this is true even on battery power.

View attachment 2498773

View attachment 2498774
Now compare performance per watt
 
Especially games. I have yet to see a Mac game truly designed from the ground up for Metal. Why? Because it's not financially reasonable for game studios to do so. All those Tomb Riders, Resident Evils, Death Stranding are just ”ports”. They were not initially designed for Metal in mind.
Windows has been an after thought as well, so if something as popular as Windows gaming gets a secondary concern, of course Macs will get even less of a priority.
 
"In 2023, Apple Mac shipments reached 21.9 million units, less than 10 percent of the total number of PC shipments observed across the entire market. In the same year, global PC shipments..."
--------------
Too bad Apples Mac mini M4 for $600 doesn't beat a $600 Playstation 5 or $600 Xbox series for gaming.
Thats hard to understand. They could get a few more dollars with better GPUs in the cheaper Mac minis made for gaming.
The processors power is there.
https://nanoreview.net/en/gpu-compare/geforce-rtx-4060-mobile-vs-apple-m4-gpu
 
  • Haha
Reactions: eltoslightfoot
Trying to compare S/C high fidelity macOS computers with mono Windows systems is a fools errand. Mac is not a PC and never will be.
Please tell that to MacRumors.
Because when it comes to comparisons in which the Mac performs better in benchmark results, the editors no longer know that.
 
Thats hard to understand. They could get a few more dollars with better GPUs in the cheaper Mac minis made for gaming.
The processors power is there.

Even if Apple did that, nobody would buy it. Because Apple has no interest in gamers.

Gaming means long-lasting compatibility, simple hardware upgrades and individual optimization of performance.

The exact opposite of Apple.
 
  • Like
Reactions: IJustWannaTalk
"In 2023, Apple Mac shipments reached 21.9 million units, less than 10 percent of the total number of PC shipments observed across the entire market. In the same year, global PC shipments..."
--------------
Too bad Apples Mac mini M4 for $600 doesn't beat a $600 Playstation 5 or $600 Xbox series for gaming.
Thats hard to understand. They could get a few more dollars with better GPUs in the cheaper Mac minis made for gaming.
The processors power is there.
https://nanoreview.net/en/gpu-compare/geforce-rtx-4060-mobile-vs-apple-m4-gpu
?? Apple is the 4th largest personal computer vendor worldwide and AFAIK their market share is not shrinking. And they are making good profit.
 
If Apple would just fix the macOS to allow External GPU's plugged in by Thunderbolt 5 it would be Awesome.
Apple has been pushing hard for the unified memory approach. My assumption that things are moving away from external GPUs. Once macOS can rely on all Macs having Apple Silicon chips then Metal can be simplified by removing all code that requires copying data between CPU and GPU. They'll have achieved their pointer-passing-only nirvana. And if that is where they're going, then external GPUs will never come back again. "Fixing" this would be like asking Ford to "fix" the F150 so that it can swim, too.
 
Unified memory is the way things will go eventually for all vendors, because the power consumption for data transfer is excessive and getting worse. Transfer from one pool to another is wasted time, energy and heat.

First it will hit the low end (in terms of throughput / cost - e.g., mobile then laptop, then desktop), eventually it will eat the high end market (datacenter) too.
 
  • Like
Reactions: Unregistered 4U
Big picture, I think Apple does a good job with its laptop offerings relative to NVIDIA, but falls short in comparison to their desktop line.

Yes, the 5090 laptop GPU is significantly more powerful than the top M4 Max GPU in the MBP. But the former requires significant compromises, namely a hot, noisy laptop with limited battery life. Most people want a more balanced package, and AFAIK only the MBP offers that with a top-end GPU.

Where Apple continues to fall short is in desktop GPU's. You should be able to build a 5090 desktop PC with much less significant compromises. Yes, it's going to be bigger and hotter than the M3 Ultra Studio. But with sufficient cooling it should be reasonably quiet (maybe not quite a quiet as the Studio), and battery life becomes a non-issue. And it will have significantly more GPU processing power (losing out only* in the amount of RAM available to the GPU: 32 GB for the 5070 vs. ≈0.5 TB for the M3 Ultra).

[*Here I'm referring to the video side only; I believe the M3 Ultra's CPU side equals or exceeds the other top workstation CPU's, except in max RAM, where the latter can be >512 GB.]
 
Last edited:
Big picture, I think Apple does a good job with its laptop offerings relative to NVIDIA, but falls short in comparison to their desktop line.

Yep, this is where their mobile-first design shows some weakness. The hardware is quite clearly unable to sustain higher clocks, even if the thermal headroom is there, and the tightly integrated solution is less flexible and more expensive than that of other vendors.

I wonder whether they have a solution to this going forward, or whether they will focus on improving the baseline tech hoping that it will be good enough. They have the potential to significantly improve the GPU performance with only minimal increase of die area — dual FP units and matrix accelerators would significantly boost performance in several key areas where Apple is lagging behind.
 
Yep, this is where their mobile-first design shows some weakness. The hardware is quite clearly unable to sustain higher clocks, even if the thermal headroom is there, and the tightly integrated solution is less flexible and more expensive than that of other vendors.

I wonder whether they have a solution to this going forward, or whether they will focus on improving the baseline tech hoping that it will be good enough. They have the potential to significantly improve the GPU performance with only minimal increase of die area — dual FP units and matrix accelerators would significantly boost performance in several key areas where Apple is lagging behind.

 

I am not convinced it means what people think it means. In particular, moving to 2.5D or 3D designs is not guaranteed to make the die arrangement more flexible. What I can see however is a larger GPU complex enabled by stacking.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.