Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
True!

I have a Surface Laptop 7 with X Elite to play around with. Windows on Arm itself is ok - x86 translation works fine, the system is fast and stable and all that. The foundations are as solid as they can be.

The issue is that developers and users are apathetic towards Windows as a whole. It’s an ******tified legacy platform that’s tolerated rather than loved.

IMO, the Arm PC guys (Qualcomm, Nvidia, Mediatek etc.) should be focusing as much on Linux as Windows. Linux is where all the exciting developments in PC are happening (e.g. Steam OS for gaming). Focusing on Windows is shortsighted.
Which Linux?
 
Which Linux?
Not a specific distro, the ideal scenario would be:
1. Have support ready in the mainline kernel before launch
2. Provide an actively supported, feature complete open source graphics driver for Linux, just as Intel and AMD do. The burden of support shouldn’t fall on the community.
3. Support the PC BSA spec so any generic aarch64 image boots and runs on the hardware (the experience needs to be x86 like)

I doubt we’ll see this but one can hope 😅
 
Not a specific distro, the ideal scenario would be:
1. Have support ready in the mainline kernel before launch
2. Provide an actively supported, feature complete open source graphics driver for Linux, just as Intel and AMD do. The burden of support shouldn’t fall on the community.
3. Support the PC BSA spec so any generic aarch64 image boots and runs on the hardware (the experience needs to be x86 like)

I doubt we’ll see this but one can hope 😅
Weren't these laptops made in collaboration with Microsoft in the first place? I know that Microsoft was working with either Qualcomm or Snapdragon to make their own ARM development kits. I would assume the drivers were written by Microsoft then.

I certainly wouldn't trust Microsoft to support *any* open source Linux drivers, no matter what they say about "Microsoft Loves Linux"
 
First M5 results on the left Screenshot 2025-09-30 at 18.44.38.png
 
So for the M4Pro we can speculate the gpu will be around 140k ?!
Sure, but if it is SoIC, then all bets are off? Probably safe to assume the result will be better than what the monolithic SoC approach would have achieved, but I imagine it would be hard for even an expert to predict. I guess I'd study what AMD is doing and try to extrapolate from there?
 
15% faster vs 16 pro on ST (3500 vs 4000) on 6% higher freq (4.0 vs 4.26). 25% improvement on MT.

Nicely done.

Lets say M5 uses same freq boost and similar arch improvement, and keeps core count, could mean 4300 ST and 18700 MT for the 10 core.
4100/15400 for the 9-core iPad?

Since mac scores are usually higher, maybe I should just call Gurman and tell him he’s out and I’m in?
 
Does anyone anticipate an increase in core count on M5 generation? Probably not, given there's no process node shrink. So instead, just performance optimisations for potentially faster clocks?

Looking at the leaked benchmarks from that iPad, things are looking really good for M5 to knock it out the park on the GPU front. CPU is already king of the hill with M4, so a 10-13% improvement is only going to make that even better.

I'm mostly keen on the improvement to raster performance on the GPU front so that we can all be a little more competitive on the gaming/graphics/3d-content front. M4 Max is a wonderful GPU for sure, but it would be nice if it had a bit more juice.

If we end up with a ~40% boost in GPU, that would be a huge improvement year over year. I appreciate that the GPU cores haven't changed TOO much on the raster front, ray-tracing etc notwithstanding.

I wonder if the improvements to Metal benchmarks are coming from the neural improvements on the GPU or raw raster? I guess GFX benchmarks need to be done. Surely they will come very soon given the leaks from Russia.
 
  • Like
Reactions: Populus
Updated codenames - 09/30/2025

Those with "?" are my personal speculation

MarketingInternalCodenameCPIDP-CoreE-Core
A16H15PCreteT8120EverestSawtooth
M3H15GIbizaT8122EverestSawtooth
M3 ProH15JLobosT6030EverestSawtooth
M3 MaxH15J/H15SPalmaT6031 / T6034EverestSawtooth
M3 UltraPalma 2C ?T6032EverestSawtooth
A17 ProH16PCollT8130
M4H16GDonanT8132
M4 ProH16SBrava ChopT6040
M4 MaxH16CBravaT6041
A18H17ATupaiT8140a
A18 ProH17PTahitiT8140
M5H17G ?HidraT8142
M5 ProH17S ?Sotra_ST6050
M5 Max ?H17C ?Sotra_CT6051?
M5 Ultra ?H17C ?Sotra_DT6052?
A19H18ATilosT8150a
A19 ProH18PTheraT8150
M6 ?H18G ?KomodoT8152?
A20 ?H19ABandaT8160a
A20 Pro ?H19P ?BorneoT8160
M7 ?H19G ?DelosT8162


MarketingInternalCodenameCPIDE-Core
S6, S7, S8M10TurksT8301Thunder
S9, S10M11CaicosT8310Sawtooth
S11 ?M12NevisT8320

MarketingCodenameCPID
C1LedaC4010 ?
C2 ?PandiaC4020
C3 ?IsonoeC4030


MarketingCodenameCPID
R1BoraT6500
R2 ?CorfuT6502

sources:

 
Last edited:
Does anyone anticipate an increase in core count on M5 generation? Probably not, given there's no process node shrink. So instead, just performance optimisations for potentially faster clocks?

Looking at the leaked benchmarks from that iPad, things are looking really good for M5 to knock it out the park on the GPU front. CPU is already king of the hill with M4, so a 10-13% improvement is only going to make that even better.

I'm mostly keen on the improvement to raster performance on the GPU front so that we can all be a little more competitive on the gaming/graphics/3d-content front. M4 Max is a wonderful GPU for sure, but it would be nice if it had a bit more juice.

If we end up with a ~40% boost in GPU, that would be a huge improvement year over year. I appreciate that the GPU cores haven't changed TOO much on the raster front, ray-tracing etc notwithstanding.

I wonder if the improvements to Metal benchmarks are coming from the neural improvements on the GPU or raw raster? I guess GFX benchmarks need to be done. Surely they will come very soon given the leaks from Russia.
So far, every M-generation has futzed with the core count and composition. M1 Pro was 8+2, M2 Pro was 8+4, M3 Pro was 6+6, M4 Pro was 10+4. It would be unusual for them to break this pattern and use the same number of cores for M5 Pro. If I had to guess, I would say maybe we see 10+6 this year, as Geekerwan has suggested the E cores saw a big performance leap this year. GPU has also been all over the place, going from 16 to 19 to 18 and 20 across M1 Pro->M4 Pro. I think we can count on some change in core count this generation.
 
  • Like
Reactions: scottrichardson
So far, every M-generation has futzed with the core count and composition. M1 Pro was 8+2, M2 Pro was 8+4, M3 Pro was 6+6, M4 Pro was 10+4. It would be unusual for them to break this pattern and use the same number of cores for M5 Pro. If I had to guess, I would say maybe we see 10+6 this year, as Geekerwan has suggested the E cores saw a big performance leap this year. GPU has also been all over the place, going from 16 to 19 to 18 and 20 across M1 Pro->M4 Pro. I think we can count on some change in core count this generation.
I don't think they have to make any changes but if they did I could see them going for 8+6 in the M5 Pro.
 
Considering the leaked iPad and iPhone GPU improvements, I really expect 35-40 % improvement for the Macbooks . What I really hope for besides that is tensor cores and fp8 support.
 
  • Like
Reactions: Populus
Considering the leaked iPad and iPhone GPU improvements, I really expect 35-40 % improvement for the Macbooks . What I really hope for besides that is tensor cores and fp8 support.
It's clear from the M5 iPad Pro leak that the GPU will feature the new features introduced with the A19 Pro, like neural accelerators in the GPU core, improved dynamic caching and unified image compression.
 
WHy are you repeating and quoting yourself a bit of narcissism ?
Gurman is not predicting benchmarks but devices
We also can say that based on existing A19 pro and M5 that M5 pro metal will be around 150k...but thats math not predictions
But if you really want to be a true leaker or a Messiah tell us from now what M6 will be in both cpu and gpu departments ( scores/ nr of cores/ frequencies/ cash)
 
Last edited:
WHy are you repeating and quoting yourself a bit of narcissism ?
Gurman is not predicting benchmarks but devices
We also can say that based on existing A19 pro and M5 that M5 pro metal will be around 150k...but thats math not predictions
But if you really want to be a true leaker or a Messiah tell us from now what M6 will be in both cpu and gpu departments ( scores/ nr of cores/ frequencies/ cash)
I guess he’s just using a bit of humor. Don’t take it so seriously.

As for predictions, some people in this thread already pointed towards GPU being the main target of improvements. This quote is from a year ago:

If they are serious about mainstream ML, memory bandwidth and GPU-specific ML features will be important. SME and the NPU are great, but the GPU has tremendous performance potential for ML applications. On the GPU side, I have speculated that Apple might be aiming for 2x FP32 pipes per scheduler in the near future. If they indeed go that route, they could implement GEMM features for various data formats, resulting in 512 BF16 and 512 or 1024 INT8 GEMM OPS per GPU core.

Now I am interested in the memory bandwidth and the speed of RAM on the M5. Also wether or not it supports Thunderbolt 5.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.