Apple M1 Pro and M1 Max

[4K8KW10]

This just goes to show how much we need Apple to step up and start producing discrete graphics cards

And how poor work both AMD and Nvidia have with their legacy graphics architectures.

 

[CAT-THE-FIFTH]

The Apple GPU IP is licensed from UK based Imagination Technologies. However,Apple,a few years afo tried to bankrupt them,and hire away their engineers. Then a Chinese back entity bought them up,and Apple then went back to licensing their GPU technology.

 

[HACO]

The Apple GPU IP is licensed from UK based Imagination Technologies. However,Apple,a few years afo tried to bankrupt them,and hire away their engineers. Then a Chinese back entity bought them up,and Apple then went back to licensing their GPU technology.

What happened to Imagination is a sad story. They used to make the best mobile GPUs, they were in iPhones, PS Vita and even Intel used their GPUs for their higher end integrated ones. But Apple wanted something different, they didn't play ball and Apple decided to just open up a new chip design centre in St Albans and hire all their key engineers. Then progress halted, they were almost bankrupt and were sold off to Chinese investors.

Apple never stopped licensing Imagination patents though, their original agreement ended at the end of 2019 and they signed a broader one which came into effect at the beginning of 2020, with the understanding that Apple just needs their parents, not the microarchitecture which has been done in-house for several years now.

Imagination are focusing their designs towards high performance computing, primarily for the auto industry. Their IMG-B architecture is geared towards multi-GPU implementations to maximise compute density. This is very different to how NVIDIA and AMD do multi-GPU, and apparently solves software compatibility issues by going away from alternate frame rendering towards a more compute-focused workload sharing where one big GPU decides how to distribute work among many smaller GPUs, similar to how execution engines parallelise workloads across a big cluster of workers.

Their C-series which is going to be announced soon is going to support ray tracing.

 

[Rroff]

Most GFXBench tasks are CPU-bottlenecked on high-end GPUs, so wouldn't read too much into this. The better one of the rest is Aztec High, Metal vs DirectX, which is consistent with everything else we've seen.

Having a play with my laptop (10870H w/ full power 3070) even Aztec High scales linear with CPU performance and GPU clocks/power barely affect results - not really surprising on stuff pumping out 200+ FPS though.

 

[Sean473]

There's no point in the Apple GPU having this power and that power - If they don't support the APIs which games use, they are good as useless for them..

 

[HACO]

There's no point in the Apple GPU having this power and that power - If they don't support the APIs which games use, they are good as useless for them..

Maybe... wait for it... they're not for gaming ¯\_(ツ)_/¯

 

[Deleted member 209350]

Cinebench R23 scores roughly:

M1 (4 performance, 4 efficiency)
SC: 1512
MC:7758

M1 Pro (8 core variant, 6 performance, 2 efficiency)
SC: 1527
MC: 8773

M1 Pro/Max (10 core variant, 8 performance, 2 efficiency)
SC: 1521
MC: 12380

Pretty decent looking numbers! Is a bit strange that the difference between 8 core Pro and regular M1 isnt that big in the CPU front though, single core seems the same for all of them.


Other R23 comparison numbers just for reference:

AMD 5600U (10-25w)
SC: 1372
MC: 7582

AMD 5800U (10-25w)
SC: 1478
MC: 11203

AMD 5800H (Upto 45w)
SC: 1445
MC: 12788


The very bleeding edge of apple's tech, performance about on par with AMD laptop chips, but i presume at a lot better efficiency. Will need to wait and see on how much power they draw roughly

 

[Deleted member 209350]

There's no point in the Apple GPU having this power and that power - If they don't support the APIs which games use, they are good as useless for them..

The 32 core variant seems about on par with the 3060 mobile chip from nvidia.. So if you were a hardcore gamer (who for some reason uses a laptop?) you wouldnt go with apple regardless

 

[HACO]

Anandtech review is out:

CPU:

Single threaded unchanged from M1 (as expected).
Multithreaded scaled as per extra cores, except for FP which went insane with M1 Max.

The 8P alone is roughly equal to the desktop 5800X in SPECint, beats it in SPECfp, basically core for core compared to Zen 3 (which again is what we expected). The full 10-core beats 5800X in almost every SPEC test. In fact M1 Max is basically at the same level as 5950X in SPECfp, despite having half the cores. This is because of the huge memory bandwidth.

This is insane, insane performance levels.

The GPU is also interesting:

Premier Pro results are scaled to desktop RTX 3080, i.e. an RTX 3080 will get a score of 1000. M1 Max gets 85-95% of the way there. CPU plays a role here too, but the baseline 1000 score is with an AMD 5800X anyway.

Davici Resolve is more GPU-bound. This is roughly on par with desktop AMD RX 6700XT.

Gaming performance, as expected, isn't as good and doesn't compare as well to higher end AMD/NVIDIA GPUs after emulation and trickery to run them, but it was never the point of this machine anyway.

 

[Deleted member 209350]

Anandtech review is out:

CPU:

Single threaded unchanged from M1 (as expected).
Multithreaded scaled as per extra cores, except for FP which went insane with M1 Max.

The 8P alone is roughly equal to the desktop 5800X in SPECint, beats it in SPECfp, basically core for core compared to Zen 3 (which again is what we expected). The full 10-core beats 5800X in almost every SPEC test. In fact M1 Max is basically at the same level as 5950X in SPECfp, despite having half the cores. This is because of the huge memory bandwidth.

This is insane, insane performance levels.

The GPU is also interesting:

Premier Pro results are scaled to desktop RTX 3080, i.e. an RTX 3080 will get a score of 1000. M1 Max gets 85-95% of the way there. CPU plays a role here too, but the baseline 1000 score is with an AMD 5800X anyway.

Davici Resolve is more GPU-bound. This is roughly on par with desktop AMD RX 6700XT.

Gaming performance, as expected, isn't as good and doesn't compare as well to higher end AMD/NVIDIA GPUs after emulation and trickery to run them, but it was never the point of this machine anyway.

These cherry picked benchmarks are pretty wild, they are not near desktop 5000 series ryzen in performance and still behind the top end mobile GPU's as well lol.

https://www.anandtech.com/show/17024/apple-m1-max-performance-review

This anandtech article shows the rough level its at. in some cases it will will outperform bigger more powerful chips, but general case its not as powerful as people were claiming.

3080 mobile and desktop ryzen performance its nowhere near really

 

[HACO]

These cherry picked benchmarks are pretty wild, they are not near desktop 5000 series ryzen in performance and still behind the top end mobile GPU's as well lol.

https://www.anandtech.com/show/17024/apple-m1-max-performance-review

This anandtech article shows the rough level its at. in some cases it will will outperform bigger more powerful chips, but general case its not as powerful as people were claiming.

3080 mobile and desktop ryzen performance its nowhere near really

What are you talking about? I posted the average multithreaded SPEC benchmarks, by definition, the average is not a cherry picked. I also posted 100% of the compute benchmarks that Anandtech did. Again, posting 100% of the benchmarks is not cherry picked.

It's actually more powerful than people were expecting, because multithreaded FP scaling was significantly better than expected. You're claiming literally the opposite of what Anandtech concluded in the review you linked:

>> The M1 Pro and M1 Max change the narrative completely – these designs feel like truly SoCs that have been made with power users in mind, with Apple increasing the performance metrics in all vectors. We expected large performance jumps, but we didn’t expect the some of the monstrous increases that the new chips are able to achieve.

>> On the CPU side, doubling up on the performance cores is an evident way to increase performance – the competition also does so with some of their designs. How Apple does it differently, is that it not only scaled the CPU cores, but everything surrounding them. It’s not just 4 additional performance cores, it’s a whole new performance cluster with its own L2. On the memory side, Apple has scaled its memory subsystem to never before seen dimensions, and this allows the M1 Pro & Max to achieve performance figures that simply weren’t even considered possible in a laptop chip. The chips here aren’t only able to outclass any competitor laptop design, but also competes against the best desktop systems out there, you’d have to bring out server-class hardware to get ahead of the M1 Max – it’s just generally absurd.​

^^^ That is what Anandtech says.

You seem to have not read the article.

 

[CAT-THE-FIFTH]

AT answers the gaming question - its not quite there yet even in the best case scenario. Realise also AT is not testing a lot of Zen3 based laptops either - you can get 8C Zen3 based laptops with an RTX3060 for as low as £900 if you look on HUKD.

https://images.anandtech.com/graphs/graph17024/126685.png

https://images.anandtech.com/graphs/graph17024/126683.png

https://images.anandtech.com/graphs/graph17024/126681.png

https://images.anandtech.com/graphs/graph17024/126682.png

** Hotlinked Images **

Also,some tests in a YT video today:
https://www.youtube.com/watch?v=IhqCC70ZfDM
https://imgur.com/a/YdErGtK

Its solid,but as usual PCMR has gotten way too overhyped on these things and people have to realise the media gets in on the hype too. The Asus G15 has a Ryzen 9 5900HS,which is made on an old 7NM process and AMD is already moving over to newer designs over the next 12 months.

So as usual the Macs work well if you are already in that ecosystem and TBH,you would be already using a Mac irrespective of what CPU/GPU would be in there because of the software you can't get on PC.

Cinebench R23 scores roughly:

M1 (4 performance, 4 efficiency)
SC: 1512
MC:7758

M1 Pro (8 core variant, 6 performance, 2 efficiency)
SC: 1527
MC: 8773

M1 Pro/Max (10 core variant, 8 performance, 2 efficiency)
SC: 1521
MC: 12380

Pretty decent looking numbers! Is a bit strange that the difference between 8 core Pro and regular M1 isnt that big in the CPU front though, single core seems the same for all of them.


Other R23 comparison numbers just for reference:

AMD 5600U (10-25w)
SC: 1372
MC: 7582

AMD 5800U (10-25w)
SC: 1478
MC: 11203

AMD 5800H (Upto 45w)
SC: 1445
MC: 12788


The very bleeding edge of apple's tech, performance about on par with AMD laptop chips, but i presume at a lot better efficiency. Will need to wait and see on how much power they draw roughly

You can get laptops with a Ryzen 5 5600H and an RTX3060 for under £1000 now.

What happened to Imagination is a sad story. They used to make the best mobile GPUs, they were in iPhones, PS Vita and even Intel used their GPUs for their higher end integrated ones. But Apple wanted something different, they didn't play ball and Apple decided to just open up a new chip design centre in St Albans and hire all their key engineers. Then progress halted, they were almost bankrupt and were sold off to Chinese investors.

Apple never stopped licensing Imagination patents though, their original agreement ended at the end of 2019 and they signed a broader one which came into effect at the beginning of 2020, with the understanding that Apple just needs their parents, not the microarchitecture which has been done in-house for several years now.

Imagination are focusing their designs towards high performance computing, primarily for the auto industry. Their IMG-B architecture is geared towards multi-GPU implementations to maximise compute density. This is very different to how NVIDIA and AMD do multi-GPU, and apparently solves software compatibility issues by going away from alternate frame rendering towards a more compute-focused workload sharing where one big GPU decides how to distribute work among many smaller GPUs, similar to how execution engines parallelise workloads across a big cluster of workers.

Their C-series which is going to be announced soon is going to support ray tracing.

The thing is Intel and Nvidia get called out for doing crappy things,but Apple doing this to a British company put me off them to a degree. They blatantly have enough money,but instead went the underhanded way.

 

[HACO]

AT answers the gaming question - its not quite there yet even in the best case scenario. Realise also AT is not testing a lot of Zen3 based laptops either - you can get 8C Zen3 based laptops with an RTX3060 for as low as £900 if you look on HUKD.

https://images.anandtech.com/graphs/graph17024/126685.png

https://images.anandtech.com/graphs/graph17024/126683.png

[IMG]https://images.anandtech.com/graphs/graph17024/126681.png

[IMG]https://images.anandtech.com/graphs/graph17024/126682.png

Also,some tests in a YT video today:
[URL]https://www.youtube.com/watch?v=IhqCC70ZfDM[/URL]
[URL]https://imgur.com/a/YdErGtK[/URL]

[IMG]https://i.imgur.com/kqKBEjl.png

Its solid,but as usual PCMR has gotten way too overhyped on these things and people have to realise the media gets in on the hype too. The Asus G15 has a Ryzen 9 5900HS,which is made on an old 7NM process and AMD is already moving over to newer designs over the next 12 months.

So as usual the Macs work well if you are already in that ecosystem and TBH,you would be already using a Mac irrespective of what CPU/GPU would be in there.

Honestly people who expected these to be gaming machines are going to be disappointed, for good reason. We spent a week telling people that these aren't gaming machines, and they didn't listen

 

[Poneros]

The sobering reality... and this is before Alder Lake even hits.

 

[Deleted member 209350]

The sobering reality... and this is before Alder Lake even hits.

Yup, not to mention that Apple is on 5nm and Intel/AMD are on older architecture.. Once they catch up and get on 5nm on an even playing field, the difference will be even greater

 

[Deleted member 209350]

AT answers the gaming question - its not quite there yet even in the best case scenario. Realise also AT is not testing a lot of Zen3 based laptops either - you can get 8C Zen3 based laptops with an RTX3060 for as low as £900 if you look on HUKD.

Its solid,but as usual PCMR has gotten way too overhyped on these things and people have to realise the media gets in on the hype too. The Asus G15 has a Ryzen 9 5900HS,which is made on an old 7NM process and AMD is already moving over to newer designs over the next 12 months.

So as usual the Macs work well if you are already in that ecosystem and TBH,you would be already using a Mac irrespective of what CPU/GPU would be in there because of the software you can't get on PC.


You can get laptops with a Ryzen 5 5600H and an RTX3060 for under £1000 now.

Yeah there are better value for money options out there, but im almost a little disappointed that Apple whilst being on 5nm still cant match upto AMD/Intel when they are on older architecture. I mean going from 7nm to 5nm is quite a big jump, and the fact that apple aren't easily beating all these older architecture chips is quite telling.. Going to be very interesting to see how they compare once AMD release their 5nm chips

 

[CAT-THE-FIFTH]

Honestly people who expected these to be gaming machines are going to be disappointed, for good reason. We spent a week telling people that these aren't gaming machines, and they didn't listen

Its no different than all those compute focussed GPUs AMD/Nvidia have made which never did that well in gaming! Although one could argue for an ARM based SOC these are the quickest so far. It does make you wonder what AMD could do if someone would commission them to do a laptop focussed Zen3 SOC.

Yeah there are better value for money options out there, but im almost a little disappointed that Apple whilst being on 5nm still cant match upto AMD/Intel when they are on older architecture. I mean going from 7nm to 5nm is quite a big jump, and the fact that apple aren't easily beating all these older architecture chips is quite telling.. Going to be very interesting to see how they compare once AMD release their 5nm chips

I do hope it means AMD/Intel actually get on an try and push their laptop CPUs a bit more. The Zen3 SOCs are only 180MM2 - I am hoping once DDR5 comes along,AMD can try and push more cores.

 

[Grim5]

The huge performance gains due to the memory system makes we wish we had that on PC.

 

[Grim5]

The sobering reality... and this is before Alder Lake even hits.

Now do performance per watt

 

[HACO]

Its no different than all those compute focussed GPUs AMD/Nvidia have made which never did that well in gaming! Although one could argue for an ARM based SOC these are the quickest so far. It does make you wonder what AMD could do if someone would commission them to do a laptop focussed Zen3 SOC.


I do hope it means AMD/Intel actually get on an try and push their laptop CPUs a bit more. The Zen3 SOCs are only 180MM2 - I am hoping once DDR5 comes along,AMD can try and push more cores.

There really isn't much they can do, they've pushed as far as they've been able to within power and thermal constraints. Adding more cores on a laptop means running them at lower clocks to keep within power and thermal limits, which diminishes returns while the chip gets bigger and more expensive, there always is a viability range and for Zen 3 the higher end chips are already at the top end of that.

The huge performance gains due to the memory system makes we wish we had that on PC.

8-channel DDR5 really isn't something we're going to see on PC anytime soon. That bandwidth only exists on server-grade products but those lack the single-threaded performance.