Live data from Hacker News

Apple M1 Ultra

apple.com

111–120 of 855 posts

Re: Apple M1 Ultra

#111
I wonder how long it will be until a CPU as capable as this one will be the required baseline for ordinary apps. Is there any hope that the upgrade treadmill will stop any time soon?

Re: Apple M1 Ultra

#112

Haha, bloody hell, what a monster of a chip. I find my M1 Max already remarkably fast. The change is so huge. It's like in the old days when you'd get a new computer and it felt like it could do things before you could think of doing them. But surely the GPU things can't be real? The GPU in the M1 Ultra beats the top-of-the-line Nvidia? That's nuts.

> The GPU in the M1 Ultra beats the top-of-the-line Nvidia? That's nuts.

We don't know yet. Apple is benchmarking against Workstation graphics cards

"production 2.5GHz 28-core Intel Xeon W-based Mac Pro systems with 384GB of RAM and AMD Radeon Pro W6900X graphics with 32GB of GDDR6"

Re: Apple M1 Ultra

#113
post #13

Haha, bloody hell, what a monster of a chip. I find my M1 Max already remarkably fast. The change is so huge. It's like in the old days when you'd get a new computer and it felt like it could do things before you could think of doing them. But surely the GPU things can't be real? The GPU in the M1 Ultra beats the top-of-the-line Nvidia? That's nuts.

That thing has four times as many transistors as a 3090.

Although true, transistor count is only tangentially related to performance

Re: Apple M1 Ultra

#114

Is it actually more powerful than a top of the line threadripper[0] or is that not a "personal computer" CPU by this definition? I feel like 64 cores would beat 20 on some workloads even if the 20 were way faster in single core performance. [0] https://www.amd.com/en/products/cpu/amd-ryzen-threadripper-3...

Apple's ARM chips can process a metric ton of ops per cycle due to the architecture of the chip: https://news.ycombinator.com/item?id=25257932

Re: Apple M1 Ultra

#115
post #76

At what point Apple will put those chips in their servers or sell server chips? It only makes sense for them to take this architecture to the cloud deployments

Unfortunately I still don't think the market has much interest in energy-efficient servers. But maybe the energy-sector crunch created by Putin's war will precipitate some change here...

Re: Apple M1 Ultra

#116
> This enables M1 Ultra to behave and be recognized by software as one chip, so developers don’t need to rewrite code to take advantage of its performance. There’s never been anything like it.

Since when did the average developer care about how many sockets a mobo has...?

Surely you still have to carefully pin processes and reason about memory access patterns if you want maximum performance.

Re: Apple M1 Ultra

#118

Earlier quoted context omitted.

I think some of this can be guessed from the SoC codenames https://en.wikipedia.org/wiki/List_of_Apple_codenames M1 Max is Jade C-Die => 64GB M1 Ultra is Jade 2C-Die => 128GB There is a still unreleased SoC called Jade 4C-Die =>256GB So I think that's the most we'll see this generation, unless they somehow add (much slower) slotted RAM If they were to double the max RAM on M2 Pro/Max (Rhodes Chop / Rhodes 1C), which…

How is slotted RAM slower? 6400Mhz DIMM exists. This would match the specs of the RAM on the M1 Max. Even octa-channel has been done before so the memory bus would have the exact same width, latency and clock frequency.

The memory bandwidth of the M1 Max is 400 GB/s with 64GB of RAM, where as the memory bandwidth of Corsair's 6400MHz DDR5 32GB RAM module is 51GB/s per stick, or 102GB/s for the M1 Max equivalent.

Re: Apple M1 Ultra

#119
> M1 Ultra can be configured with up to 128GB of high-bandwidth, low-latency unified memory

Nice! Good enough to run a Solana node!

I was slightly annoyed that the M1 Max’s 64gb RAM puts it just under the system requirements, at that premium price

But I don't have any other theoretical use case for that much resources

Re: Apple M1 Ultra

#120
post #36

Earlier quoted context omitted.

A few points to make... - the shared CPU+GPU RAM doesn't necessarily mean the GPU has to eat up system RAM when in use, because it can share addressing. So whereas the current Mac pro would require two copies of data (CPU+GPU) the new Mac studio can have one. Theoretically. - they do have very significant video decoder blocks. That means that you may use less RAM than without since you can keep frames compressed in f…

Also, the memory model is quite different - with the ultra-fast SSD and ultra-fast on-die RAM. You can get away with significantly less RAM for the same tasks, not just because of de-duplication but because data comes in so quickly from the SSD that paging isn't nearly the hit it is on say an Intel based Mac. I'd expect it to work more like a game console, streaming in content from the SSD to working memory on the fl…

Another thing to consider is memory compression. If Apple added dedicated hardware for that, it can effectively double the total memory with minimal performance hit.
Post reply on HN