4 ms·

Something buried in this, Apple are starting their own server manufacturing.

by showmexyz 2y ago

reaperducer 2y ago
And in the United States, too. (Houston)
jermaustin1 2y ago
I would assume it is for their own use, not for people wanting the XServe to come back with M4 Ultra (i.e. me...).
- throwaway48476 2y ago
  Such a shame. I'd be interested in a personal server.
  - olyjohn 2y ago
    Why would Apple want you to have one though?
- btucker 2y ago
  Agreed, but I wonder--given investors demands for continued growth--if they're considering going up against NVIDIA.
  - smallmancontrov 2y ago
    Nah, surely 80% margins for matrix multiplication on the latest TSMC node will last forever.
- sampton 2y ago
  Nvidia has proven the space is incredibly lucrative and Apple is best equipped for high end chip designs. Remember 10 years ago it was unthinkable for an ARM chip to compete with x86.
  - nobankai 2y ago
    First Apple has to prove they have competitive designs. Apple Silicon GPUs simply do not compete with the efficiency of Nvidia's GPU compute architecture: https://browser.geekbench.com/opencl-benchmarks https://browser.geekbench.com/opencl-benchmarks Apple's obsessive focus with raster efficiency really shot their GPU designs in the foot. It will be interesting to see if they adopt Nvidia-style designs or spend more time trying to force NPU hardware to work.
    - jermaustin1 2y ago
      I think performance per watt is way in Apple's favor, but raw performance is not. That said, an M4 Ultra (extrapolating from Max and Pro) would likely compete with my 3090, and with 192GB of memory (for 10x the amount it should cost) will out perform my 3x3090 AI server. And honestly, cost less than my 3 3090s + rest of the computer + electricity. It won't outperform a bunch of A/H 100s (or even a single one, or any other cards in the enterprise realm) though, but it will cost an order of magnitude less than a single card.
      - jdsully 2y ago
        Careful when comparing performance and efficiency. As a rough factor power increases quadratically as you increase clocks on a design, so you can quite easily make a high performance design low power by under-clocking it. The same is not true for the reverse.
      - m463 2y ago
        I think you are comparing apples and oranges. inference is not the same as training.
        jermaustin1 2y ago
        Sorry, I was coming at this from the consumer side (since apple is a consumer product company). The majority of LLM use (by consumers) is in inference, not training, so I'd hazard to guess, the majority of people would rather have inference machines than training machine.
barkerja 2y ago
That's not necessarily news, unless I am missing something. Craig made an indirect mention of this during last year's WWDC regarding the private cloud compute. https://security.apple.com/blog/private-cloud-compute/ https://security.apple.com/blog/private-cloud-compute/
layer8 2y ago
That’s not completely news: https://www.macrumors.com/2024/05/06/apple-building-m2-ultra-and-m4-servers/ https://www.macrumors.com/2024/05/06/apple-building-m2-ultra...
dang 2y ago
(This comment was originally posted to https://news.ycombinator.com/item?id=43158187 https://news.ycombinator.com/item?id=43158187, so "this" was a different article. We merged that thread hither.)

Something buried in this, Apple are starting their own server manufacturing.

reaperducer 2y ago

And in the United States, too. (Houston)

jermaustin1 2y ago

I would assume it is for their own use, not for people wanting the XServe to come back with M4 Ultra (i.e. me...).

throwaway48476 2y ago
Such a shame. I'd be interested in a personal server.
- olyjohn 2y ago
  Why would Apple want you to have one though?
btucker 2y ago
Agreed, but I wonder--given investors demands for continued growth--if they're considering going up against NVIDIA.
- smallmancontrov 2y ago
  Nah, surely 80% margins for matrix multiplication on the latest TSMC node will last forever.
sampton 2y ago
Nvidia has proven the space is incredibly lucrative and Apple is best equipped for high end chip designs. Remember 10 years ago it was unthinkable for an ARM chip to compete with x86.
- nobankai 2y ago
  First Apple has to prove they have competitive designs. Apple Silicon GPUs simply do not compete with the efficiency of Nvidia's GPU compute architecture: https://browser.geekbench.com/opencl-benchmarks https://browser.geekbench.com/opencl-benchmarks Apple's obsessive focus with raster efficiency really shot their GPU designs in the foot. It will be interesting to see if they adopt Nvidia-style designs or spend more time trying to force NPU hardware to work.
  - jermaustin1 2y ago
    I think performance per watt is way in Apple's favor, but raw performance is not. That said, an M4 Ultra (extrapolating from Max and Pro) would likely compete with my 3090, and with 192GB of memory (for 10x the amount it should cost) will out perform my 3x3090 AI server. And honestly, cost less than my 3 3090s + rest of the computer + electricity. It won't outperform a bunch of A/H 100s (or even a single one, or any other cards in the enterprise realm) though, but it will cost an order of magnitude less than a single card.
    - jdsully 2y ago
      Careful when comparing performance and efficiency. As a rough factor power increases quadratically as you increase clocks on a design, so you can quite easily make a high performance design low power by under-clocking it. The same is not true for the reverse.
    - m463 2y ago
      I think you are comparing apples and oranges. inference is not the same as training.
      - jermaustin1 2y ago
        Sorry, I was coming at this from the consumer side (since apple is a consumer product company). The majority of LLM use (by consumers) is in inference, not training, so I'd hazard to guess, the majority of people would rather have inference machines than training machine.

barkerja 2y ago

That's not necessarily news, unless I am missing something. Craig made an indirect mention of this during last year's WWDC regarding the private cloud compute. https://security.apple.com/blog/private-cloud-compute/ https://security.apple.com/blog/private-cloud-compute/

layer8 2y ago

That’s not completely news: https://www.macrumors.com/2024/05/06/apple-building-m2-ultra-and-m4-servers/ https://www.macrumors.com/2024/05/06/apple-building-m2-ultra...

dang 2y ago

(This comment was originally posted to https://news.ycombinator.com/item?id=43158187 https://news.ycombinator.com/item?id=43158187, so "this" was a different article. We merged that thread hither.)