I did and couldn’t find any, certainly by anyone credible. I’m sure you wouldn’t be using some claim by a random person to make a case against the majority of apple users, just as you wouldn’t want any old claim about windows pcs to be held against you. In that spirit I’d appreciate some more direction to these claims.
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.
So M1 Max with 32 Core GPU matches RTX 3080 Mobile.
- Thread starter iBug2
- Start date
- Sort by reaction score
I remember claims, legitimate ones, that Apple was running Windows for Arm apps under emulation faster than WoA laptops because of how slow the Qualcomm chips were. But to beat native Windows performance with Rosetta, for games no less, would require … uhhh … quite a slow PC.
Yep, this is exactly what we are waiting for to truly evaluate the GPU performance.
It would have been nice for Andre to show the scores for different quality settings. He says the M1 Max is cpu limited with Rosetta but doesn’t show it with numbers.
They did show the 4K numbers GPU limited scenario with one other laptop for reference. The AMD GPU advantage gets bigger, probably again, the limitations of the port wrt to the TBDR architecture.
Taken from Ars Technica:
https://twitter.com/andysomerfield/stat ... 0721448963
"The #M1Max is the fastest GPU we have ever measured in the @affinitybyserif Photo benchmark. It outperforms the W6900X - a $6000, 300W desktop part - because it has immense compute performance, immense on-chip bandwidth and immediate transfer of data on and off the GPU (UMA).""
https://twitter.com/andysomerfield/stat ... 0721448963
"The #M1Max is the fastest GPU we have ever measured in the @affinitybyserif Photo benchmark. It outperforms the W6900X - a $6000, 300W desktop part - because it has immense compute performance, immense on-chip bandwidth and immediate transfer of data on and off the GPU (UMA).""
Last edited:
He would have needed to show 720p and 1440p with same settings to show the scaling. Unless you guys honestly believe 85 fps is as fast as the M1 Max can go for Shadow of the Tomb Raider.
Ah okay to truly measure the effect, yes. But you can say there is one with two points: not GPU limited and GPU limited and see how the FPS numbers scale between the pro and max in each.
The M1 Max cannot be more than 4x faster than the M1. And it is about exactly 4x faster at this game.
Sure, but it’s unoptimised on all M1 chips. Performance could increase on all of them, while maintaining the 4x difference.
Yo this is ****ing HUGE. I didn't see this so awesome find. I'm just REALLY hoping that more game companies will do native ports. If so, we could really see the mac come back in STRONG for gaming.Further to the questions about gaming performance on the M1/Pro/Max, Brad Oliver, who works for Feral, posted this on Ars:
ARMageddon
Andrei from Anandtech has endorsed the theory that the Max die shot is doctored. Along the bottom seem to be duplicated parts, including the neural cores. The assumption is that this hides interconnects for Mac Pro purposes. This seems so ridiculous. The lifetime of that supposed secret is...arstechnica.com
I don’t want to speak for @jeanlain but we’re discussing how to test performance in cpu limited scenarios and measuring what the effect of Rosetta is on performance. @diamond.g was wishing we got more data in those scenarios and @jeanlain and I are pointing out that at 4K we know we’re saturating the GPU with work but aren’t at 1080p and the difference can give you some information about Rosetta and cpu limited scenarios - though not the complete picture. Which is a little different than just is the game optimized for the M1.
Don't get your hopes too high. Who is going to spend time optimising AAA games for Apple GPUs, which represent a tiny fraction of the market? So far, we only have two optimized AAA games, both from Larian Studios, which I suppose happen to like the Mac.
Maybe Feral will spend time tuning their code for Apple GPUs, but I don't think that's a given.
As for the other developers, they'll use moltenVK at best.
I'm not sure how much it'll matter for native ports. I'm rather doubtful that the industry will suddenly switch to Apple and what not. What I am more excited for is something like Crossover/Parallels. I really would like to see how those work. The M1 with either of those works incredibly well for what the M1 is.
The fact that the M1 Max is almost exactly 4x faster than the M1 at 1080p suggests that the GPU is limiting performance even at this resolution.
Nah it’s off by 20%. It’s 1.8 scaling whereas at 4K it’s perfect 2. You can also tell because the gap grows between them and the 6800 as it should. So at 1080p it’s CPU limited and Rosetta is having some effect. To know how much @diamond.g is right you’d have to test other settings.
I’m guessing Andrei didn’t have time to run all that.
Rosetta will be better as it is only translating from x86 to arm64 for native Mac apps. Emulation and crossover/parallels has to do that and translate or emulate Windows and if it is Windows doing the translating then performance will take another hit as their version of Rosetta is not (yet) as performant.
What I personally find impressive that Overwatch is getting 120FPS while inside of a VM AND going through Microsoft's bad translation layer lmao. Imagine if we could bring this down one layer and/or just get native.
The M1 yields 22 fps and the M1 Max 85 fps. It's >96% of quadrupling the M1 frame rate. I'm not sure why the M1 Pro is more than 2x faster than the M1, but even compared to the M1 Pro, the M1 Max is only 10% short of double the performance.
Huh interesting … you know… if I remember right Aztec ruins high does something similar. The jump from OG M1 to Pro is bigger than 2x while from Pro to Max is less than 2x, while the combination is exactly 4x. Most interesting. There’s still a lot to figure out about this chip.
In the absence of any actual AS graphics programmer picking up the gauntlet, I'm going to say that there are a couple of things to keep in mind. Even though the M1PM is well endowed in terms of bandwidth, writing to and from main RAM is still going to be slower than keeping data in on chip tile caches. And there are steps you can just avoid.
The other aspect that springs to mind is perhaps trickier to evaluate the effect of, it's that programmers code to a target. This affects all cross platform benchmarking. Outside of mobile, IMRs have been the only game in town, and code and hardware has co-evolved. While you can run that code on a TBDR, it was originally coded taking the strengths and weaknesses of the target (industry standard) hardware in mind. Not only doesn't it take advantage of the benefits of TBDR, it may choose techniques that you simply wouldn't have gone for at all, if you knew that the code would run on a TBDR. You would have done things differently to get similar results with better performance. Indeed the game itself might have looked different!*
(*Ancient trivial example - the hidden surface removal made overdraw really cheap on TBDRs, so having villages and towns with lots of structures was no problem, whereas foliage (at the time handled via transparency) was a kick in the TBDR nuts. The TBDRs of the time was benchmarked using game environments that would simply have looked different, had it been know that they would be produced by that kind of architecture. But they were created based on the strengths and weaknesses of the dominant approach.)
Apples 3D developer videos are relatively easy to follow, and this sort of thing crops up in them - not only the more common case of "if you have been doing things like this, you'd gain a lot by modifying your code like that instead", but also "since this is really cheap, it's a good way to...". But it would take an active 3D-programmer to say just how much of a difference using an Apple TBDR as an original target for your code would make. I can only say that yes, it would make a difference, but don't have the real world experience to express that difference in numbers.
@Andropov had a pretty good post with similar points
So M1 Max with 32 Core GPU matches RTX 3080 Mobile.
The only way to know for sure is to run custom benchmarks and I don't have a M1 Pro MacBook (yet). Apple even provides sample code of modern rendering techniques on the same scene, from WWDC videos, so it wouldn't be too difficult to try. But I'd expect that there's still a lot to be gained...
forums.macrumors.com
Interesting. But there is also this from WCCFTECH:
MacBook Pro With M1 Max Is Barely Faster Than Microsoft’s Surface Laptop Studio in 8K Video Render Test
That's about the 8K video rendering test which was performed on Adobe Premiere Pro (it's available as AS native).