4 ms·

Yes, but you need to setup in the bios how much memory you want to dedicate to the gpu, while apples approach afaik doesn't gpu/cpu can access the same memory,

Yes, but you need to setup in the bios how much memory you want to dedicate to the gpu, while apples approach afaik doesn't gpu/cpu can access the same memory, so if you for example want to game and the game requires more memory than the cpu you can get as much as ram you have. I'm on having the cpu/gpu integrated, but more powerful, if apple can do it, then amd/intel can do it too, but I suspect amd will lose their dedicated gpu market, while intel could gain more with this kind of move, but I suspect that intel doesn't care too much about the enthusiast or end users as apple/amd (I mean that apple market are end user and not server/enterprise like intel).

sgjohnson 5y ago
There are bigger problems with that. Bundling CPU and GPU is not feasible in the present day for high end systems. There are use cases where you need say HBM2 memory. Or you need 4 outputs (most motherboards have just HDMI+VGA). It’s easy for Apple to do, because they have 0 skin in the game in gaming or AI or even GPU compute markets. I don’t think it would make sense for AMD to do it. It would make sense for Intel, to compete in entry level gaming market, but that’s it. IMHO it’s just not feasible if your product is a general purpose GPU, because there’s a lot more to it than just the processing unit.
- runnerup 5y ago
  The M1 Ultra gets fully half of the performance of a desktop RTX 3090 in both synthetic and real world benchmarks. That’s extremely impressive. Sure, no one has CUDA except NVidia but it looks like Apples GPU architecture really truly does work quite well at this stage. I wouldn’t be shocked to see them catch up to the highest performance GPUs in the next 5 years or so. I mean I also would t be shocked if they didn’t, but it seems like a reasonable possibility at this point. Which is crazy for a SOC, especially one borne from mobile phone architectures. I also feel like M1 rackmount servers could be amazing if Linux ran on them.
  - novok 5y ago
    So the m1 ultra is like a 2070S or a little bit slower than a 3060ti? A ryzen 5950 also has a similar amount of cores, and if most of the power efficiency is coming from being a process node ahead due to special tsmc contracts and covid, its not a durable advantage either because it is not due to technical innovation from apple. The m1 ultra and max basically targeted for video editing as their target market. Same with the m1 max pro with it’s xdr screen to become a great hdr editing laptop. M1 studio similarly is targeted for the video editors that want a mac mini made for them.
    - josephg 5y ago
      You won’t take advantage of the GPU cores, but the M1 is also a fantastic chip for compiling code. The core count and memory bandwidth are delightful.
      - novok 5y ago
        Your paying a lot for all of that GPU compute with the amount of die space it takes up. Apple's Intel CPU options were also pretty old and low power, so it feels like a much bigger difference if you are stuck on apple platforms to do your work, like iOS. If apple kept up with x86 CPUs the difference on macOS wouldn't have been as wide. When you compare a Ryzen 5000 series with the same core count, they are fairly comparable in compile times AFAIK.
        josephg 5y ago
        > they are fairly comparable in compile times AFAIK. Yeah I can confirm that. I have an 8 core ryzen 5800, and its performance when compiling rust code is almost identical to my 8+2 core macbook pro. The macbook pulls ahead slightly in single core performance, but I think the ryzen takes longer before thermally throttling - which makes sense given its a desktop workstation.
    - runnerup 5y ago
      You’re looking at it from a perspective of “this instant in time”. I’m assuming that some of the phenomenal, market-beating growth of the Apple Silicon architecture over the past 5-10 years will continue. Also if Apple merely creates a quad M1 Pro, as some prognosticate, that would likely match the 3090 in performance.

sgjohnson 5y ago

There are bigger problems with that. Bundling CPU and GPU is not feasible in the present day for high end systems. There are use cases where you need say HBM2 memory. Or you need 4 outputs (most motherboards have just HDMI+VGA). It’s easy for Apple to do, because they have 0 skin in the game in gaming or AI or even GPU compute markets. I don’t think it would make sense for AMD to do it. It would make sense for Intel, to compete in entry level gaming market, but that’s it. IMHO it’s just not feasible if your product is a general purpose GPU, because there’s a lot more to it than just the processing unit.

runnerup 5y ago
The M1 Ultra gets fully half of the performance of a desktop RTX 3090 in both synthetic and real world benchmarks. That’s extremely impressive. Sure, no one has CUDA except NVidia but it looks like Apples GPU architecture really truly does work quite well at this stage. I wouldn’t be shocked to see them catch up to the highest performance GPUs in the next 5 years or so. I mean I also would t be shocked if they didn’t, but it seems like a reasonable possibility at this point. Which is crazy for a SOC, especially one borne from mobile phone architectures. I also feel like M1 rackmount servers could be amazing if Linux ran on them.
- novok 5y ago
  So the m1 ultra is like a 2070S or a little bit slower than a 3060ti? A ryzen 5950 also has a similar amount of cores, and if most of the power efficiency is coming from being a process node ahead due to special tsmc contracts and covid, its not a durable advantage either because it is not due to technical innovation from apple. The m1 ultra and max basically targeted for video editing as their target market. Same with the m1 max pro with it’s xdr screen to become a great hdr editing laptop. M1 studio similarly is targeted for the video editors that want a mac mini made for them.
  - josephg 5y ago
    You won’t take advantage of the GPU cores, but the M1 is also a fantastic chip for compiling code. The core count and memory bandwidth are delightful.
    - novok 5y ago
      Your paying a lot for all of that GPU compute with the amount of die space it takes up. Apple's Intel CPU options were also pretty old and low power, so it feels like a much bigger difference if you are stuck on apple platforms to do your work, like iOS. If apple kept up with x86 CPUs the difference on macOS wouldn't have been as wide. When you compare a Ryzen 5000 series with the same core count, they are fairly comparable in compile times AFAIK.
      - josephg 5y ago
        > they are fairly comparable in compile times AFAIK. Yeah I can confirm that. I have an 8 core ryzen 5800, and its performance when compiling rust code is almost identical to my 8+2 core macbook pro. The macbook pulls ahead slightly in single core performance, but I think the ryzen takes longer before thermally throttling - which makes sense given its a desktop workstation.
  - runnerup 5y ago
    You’re looking at it from a perspective of “this instant in time”. I’m assuming that some of the phenomenal, market-beating growth of the Apple Silicon architecture over the past 5-10 years will continue. Also if Apple merely creates a quad M1 Pro, as some prognosticate, that would likely match the 3090 in performance.