Re:The Mac Pro’s biggest problem is the MacBook
by DrJokepu on Wednesday June 28, 2023 @10:16PM
(#36515542)
Would something like this be fundamentally incompatible with Apple’s shared memory architecture though?
Re:The Mac Pro’s biggest problem is the MacBook
by 7e on Wednesday June 28, 2023 @10:24PM
(#36515598)
Domain-specific accelerators have always had their own memory. Look at GPUs. I don't see why a power user who needs speed would be put off by some DMA.
Maybe Apple ends up implementing something like CXL down the line, anyway.
Re:The Mac Pro’s biggest problem is the MacBook
by wmf on Wednesday June 28, 2023 @10:19PM
(#36515563)
Yeah, it makes no sense.
Re:The Mac Pro’s biggest problem is the MacBook
by rgbrenner on Wednesday June 28, 2023 @11:36PM
(#36516025)
I think it would be compatible, in the same way a GPU would be compatible. You wouldn't be able to use it for your display though (for the same reason you cant use integrated graphics with your GPU).. so it would only be useful for accelerating ML or graphics rendering workloads for example.
Re:The Mac Pro’s biggest problem is the MacBook
by brucethemoose2 on Wednesday June 28, 2023 @10:21PM
(#36515575)
That sounds more like a server. Having an M1 ultra to "host" them makes no sense.
Re:The Mac Pro’s biggest problem is the MacBook
by 7e on Wednesday June 28, 2023 @10:36PM
(#36515673)
You get an interconnect which is faster than Infiniband. I could see some people using this.
They should implement ECC memory, though.
Re:The Mac Pro’s biggest problem is the MacBook
by brucethemoose2 on Wednesday June 28, 2023 @10:52PM
(#36515762)
AFAIK Intel and IBM sell 8P servers exactly like this. Physically the Intel ones look like big addon cards on a mobo.
They are very niche and very expensive, hence almost everyone buys 2P servers instead.
Re:The Mac Pro’s biggest problem is the MacBook
by datavirtue on Wednesday June 28, 2023 @10:36PM
(#36515670)
Yeah, these would make a nice server for people with decent internet service.
Re:The Mac Pro’s biggest problem is the MacBook
by themitigating on Wednesday June 28, 2023 @10:14PM
(#36515524)
You're right and the people who will pay need PCIe slots for high end stuff like
NICs for high speed networks that are used in share video editing
SDI cards for video
Re:The Mac Pro’s biggest problem is the MacBook
by numpad0 on Wednesday June 28, 2023 @10:55PM
(#36515789)
Without the Apple proprietary MPX slots, such cards will be limited to 75W bus power. PCIe bus on M2 Ultra is also limited to something like PCIe 4.0 16+8 lanes, or 47.28Gbps(5.91GB/s) total. This will be much slower than competing(in price) product such as Intel Xeon W/AMD EPYC, which has up to 112/128 lanes of PCIe 5.0 or 441.28/504.32Gbps(55.16/63.04GB/s) max bandwidth.
This means a M2 Ultra PCIe card/backplane desktop supercomputer takes 192GB/5.9GB/s = 32.54s per node minimum to fill the whole 192GB of RAM - I don't know how to interpret this, is this going to mean the cluster will be I/O limited, or will it be actually limited by CPU?
Re:The Mac Pro’s biggest problem is the MacBook
by wmf on Wednesday June 28, 2023 @11:25PM
(#36515964)
I think your math is off by a factor of eight.
Re:The Mac Pro’s biggest problem is the MacBook
by visionscaper on Friday June 30, 2023 @06:28PM
(#36543354)
Just read your comment, I made a similar comment here:
https://news.ycombinator.com/item?id=36543226
So here's my completely baseless theory: M2 Ultra PCIE coprocessor cards.
These cards could be upgradable, and they would include CPU, GPU, and memory. It would transform the Mac Pro into an incredibly powerful machine; restore the top end price near $35-50k; and re-enforce apple's ecosystem of metal, pro-res, etc.
This is a type of product that already exists. And Apple can definitely do this, since they control the entire product.