- Why is the 32-core gpu said by apple to be only 60% faster than the 16-core gpu? (See slide where they are respectively 400% and 250% speed of the 5600M)
- What are the cores? (Most people seem to think they're the same as the M1s, but from GPU perf alone, those cores are likely the same gen as the A15. No clue about the CPU cores, but you'd think they'd be same-gen.)
Not all benchmarks fully use the capability of the hardware. On Apple's page there is a section on GPU tests and the Photoshop test shows M1 Max at 2.1x the 5600M and M1 Pro at 2.0x, which means Max is just 5% faster in that test:
Only benchmarks that fully use the hardware show the true capability of the GPU. The best ones are usually compute benchmarks like Redshift. The Redshift test showed 60% improvement between M1 Pro and Max but not all of that process is the render itself. If the render test was around 1 minute but it takes 20 seconds to prepare assets for the render and 40 seconds to render on M1 Pro then if M1 Max is twice as fast, it will only affect the latter part of the process. M1 Pro would take 20s prepare + 40s render, M1 Max would take 20s + 20s, this makes it 50% faster overall despite the compute part being 100% faster.
Apple listed the theoretical maximum performance of each GPU as 5.2TFLOPs for M1 Pro and 10.4TFLOPs for M1 Max so the raw hardware is double but this only shows in software that uses it all. The same can be seen in other GPUs, the 3080 is theoretically 50% faster than the 3070 but in most tests, it doesn't show this much difference, it's usually around 30%:
The highest difference Apple showed was 70% in the Final Cut render test. Once people get the machines to test, there will be some tests that come closer to the theoretical maximum.
Some benchmarks also show much faster performance than expected. The 5600M is over 5TFLOPs so a 4x speedup would suggest over 20TFLOPs but it's not just compute hardware that affects performance, there's memory capacity, bandwidth, time to sync between system memory and GPU memory. The unified memory system will give a large performance gain vs the 5600M in some tests. In media encoding tasks, the special hardware in Apple's chips helps boost this.
The cores seem like they are the same as M1 as the theoretical performance number match up with M1. M1 = 2.6TFLOPs, M1 Pro = 5.2TFLOP, M1 Max = 10.4TFLOP. It doubles each time as they doubled the core count so I'd assume they are the same M1 cores.
As noted in other threads, while the performance gains are massive (and larger than widely appreciated - AI got it wrong at least once already), there's a lot we really don't know yet.
- Clock speed: how much higher is it? (My guess: minimal)
- Why is the 32-core gpu said by apple to be only 60% faster than the 16-core gpu? (See slide where they are respectively 400% and 250% speed of the 5600M)
- What are the cores? (Most people seem to think they're the same as the M1s, but from GPU perf alone, those cores are likely the same gen as the A15. No clue about the CPU cores, but you'd think they'd be same-gen.)
- How big is the SLC? (My guess: 24/48MB for Pro/Max)
- How fast is the LPDDR5? (Probably LPDDR5-6400)
- >7GB/s reads of the SSD are great, but what about writes?
Also some non-performance questions, like can we run 4x 4k displays on the Pro? (Probably not.)
Apple generally does not release this information, and to mostbuyers it doesn't matter.
Regarding the displays - it was answered during the presentation & on the MacBook Pro spec page:
- Why is the 32-core gpu said by apple to be only 60% faster than the 16-core gpu? (See slide where they are respectively 400% and 250% speed of the 5600M)
- What are the cores? (Most people seem to think they're the same as the M1s, but from GPU perf alone, those cores are likely the same gen as the A15. No clue about the CPU cores, but you'd think they'd be same-gen.)
Not all benchmarks fully use the capability of the hardware. On Apple's page there is a section on GPU tests and the Photoshop test shows M1 Max at 2.1x the 5600M and M1 Pro at 2.0x, which means Max is just 5% faster in that test:
[good stuff for a benchmarking 101 class]
The cores seem like they are the same as M1 as the theoretical performance number match up with M1. M1 = 2.6TFLOPs, M1 Pro = 5.2TFLOP, M1 Max = 10.4TFLOP. It doubles each time as they doubled the core count so I'd assume they are the same M1 cores.
That's an extremely strong argument.
Your discussion of benchmarking is a lot less relevant because for the most part, Apple hasn't cherry-picked benchmarks, but rather used broad ones that tend to align with other broad ones like SPEC's. That despite the Photoshop example. I'm talking specifically about the numbers and graphs they used in the M1 introduction, which are very similar to the ones they used for the M1P/M intro (though this week, they gave more specifics than they did with the M1).
Despite that, I am regretfully coming to the conclusion that you (and mjtomlin in the other thread, if I'm remembering him right) are probably right, and the numbers they provided in the comparison to the 5600M were either totally wrong, or else horribly skewed by the nature of whatever oddball benchmark they used for that graph. It's really unfortunate.
This is all also *really* strong support for the notion that these chips were ready to go much earlier in the year, and that everything got held up.
As noted in other threads, while the performance gains are massive (and larger than widely appreciated - AI got it wrong at least once already), there's a lot we really don't know yet.
- Clock speed: how much higher is it? (My guess: minimal)
- Why is the 32-core gpu said by apple to be only 60% faster than the 16-core gpu? (See slide where they are respectively 400% and 250% speed of the 5600M)
- What are the cores? (Most people seem to think they're the same as the M1s, but from GPU perf alone, those cores are likely the same gen as the A15. No clue about the CPU cores, but you'd think they'd be same-gen.)
- How big is the SLC? (My guess: 24/48MB for Pro/Max)
- How fast is the LPDDR5? (Probably LPDDR5-6400)
- >7GB/s reads of the SSD are great, but what about writes?
Also some non-performance questions, like can we run 4x 4k displays on the Pro? (Probably not.)
Apple generally does not release this information, and to mostbuyers it doesn't matter.
Regarding the displays - it was answered during the presentation & on the MacBook Pro spec page:
About the displays - That in no way answers the question. The reason those limits exist is not made clear, and depending on that reason, more 4k displays might or might not be possible. It will depend on implementation details that could go either way. Basically, can the chip drive multiple 4x streams over a single thunderbolt link. The M1 couldn't. But that doesn't mean the M1P/M can't. They already have more support for displays than the M1. So anything's possible (within the bandwidth limits of Thunderbolt, anyway).
As for the other questions... Of course it doesn't matter to buyers, my order's in already. It matters to those of us interested in technology for its own sake. And the answers will be forthcoming even without Apple's cooperation. I'll figure out at least some of it when mine comes, if I haven't already learned it by reading Andrei's analysis.
As noted in other threads, while the performance gains are massive (and larger than widely appreciated - AI got it wrong at least once already), there's a lot we really don't know yet.
- Clock speed: how much higher is it? (My guess: minimal)
- Why is the 32-core gpu said by apple to be only 60% faster than the 16-core gpu? (See slide where they are respectively 400% and 250% speed of the 5600M)
- What are the cores? (Most people seem to think they're the same as the M1s, but from GPU perf alone, those cores are likely the same gen as the A15. No clue about the CPU cores, but you'd think they'd be same-gen.)
- How big is the SLC? (My guess: 24/48MB for Pro/Max)
- How fast is the LPDDR5? (Probably LPDDR5-6400)
- >7GB/s reads of the SSD are great, but what about writes?
Also some non-performance questions, like can we run 4x 4k displays on the Pro? (Probably not.)
Apple generally does not release this information, and to mostbuyers it doesn't matter.
Regarding the displays - it was answered during the presentation & on the MacBook Pro spec page:
About the displays - That in no way answers the question. The reason those limits exist is not made clear, and depending on that reason, more 4k displays might or might not be possible. It will depend on implementation details that could go either way. Basically, can the chip drive multiple 4x streams over a single thunderbolt link. The M1 couldn't. But that doesn't mean the M1P/M can't. They already have more support for displays than the M1. So anything's possible (within the bandwidth limits of Thunderbolt, anyway).
As for the other questions... Of course it doesn't matter to buyers, my order's in already. It matters to those of us interested in technology for its own sake. And the answers will be forthcoming even without Apple's cooperation. I'll figure out at least some of it when mine comes, if I haven't already learned it by reading Andrei's analysis.
It was clear in the presentation - M1 Pro can drive 2 x external displays, up to 6K resolution. The M1 Max can drive 4 x external displays, with 3 x external displays up to 6K & 1 x external display up to 4K. There was no doubt about the number of external displays possible - it was crystal clear. And for me the tech specs are also crystal clear. You seem to be looking for something else, but that does not make it possible.
Comments
https://www.apple.com/macbook-pro-14-and-16/
Only benchmarks that fully use the hardware show the true capability of the GPU. The best ones are usually compute benchmarks like Redshift. The Redshift test showed 60% improvement between M1 Pro and Max but not all of that process is the render itself. If the render test was around 1 minute but it takes 20 seconds to prepare assets for the render and 40 seconds to render on M1 Pro then if M1 Max is twice as fast, it will only affect the latter part of the process. M1 Pro would take 20s prepare + 40s render, M1 Max would take 20s + 20s, this makes it 50% faster overall despite the compute part being 100% faster.
Apple listed the theoretical maximum performance of each GPU as 5.2TFLOPs for M1 Pro and 10.4TFLOPs for M1 Max so the raw hardware is double but this only shows in software that uses it all. The same can be seen in other GPUs, the 3080 is theoretically 50% faster than the 3070 but in most tests, it doesn't show this much difference, it's usually around 30%:
https://www.notebookcheck.net/NVIDIA-GeForce-RTX-3070-vs-NVIDIA-GeForce-RTX-3080_10559_10487.247598.0.html
The highest difference Apple showed was 70% in the Final Cut render test. Once people get the machines to test, there will be some tests that come closer to the theoretical maximum.
Some benchmarks also show much faster performance than expected. The 5600M is over 5TFLOPs so a 4x speedup would suggest over 20TFLOPs but it's not just compute hardware that affects performance, there's memory capacity, bandwidth, time to sync between system memory and GPU memory. The unified memory system will give a large performance gain vs the 5600M in some tests. In media encoding tasks, the special hardware in Apple's chips helps boost this.
The cores seem like they are the same as M1 as the theoretical performance number match up with M1. M1 = 2.6TFLOPs, M1 Pro = 5.2TFLOP, M1 Max = 10.4TFLOP. It doubles each time as they doubled the core count so I'd assume they are the same M1 cores.
Regarding the displays - it was answered during the presentation & on the MacBook Pro spec page:
Simultaneously supports full native resolution on the built-in display at 1 billion colors and:
Up to two external displays with up to 6K resolution at 60Hz at over a billion colors (M1 Pro) or
Up to three external displays with up to 6K resolution and one external display with up to 4K resolution at 60Hz at over a billion colors (M1 Max)