Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Apple M1 Ultra

apple.com

Post reply on HN
Print view View original

601–610 of 855 posts

Re: Apple M1 Ultra

#601

Post by scarface74 »

Melatonic wrote: post #589
kllrnohj wrote: post #360

Earlier quoted context omitted.

The GPU claims on the M1 Pro & Max were, let's say, cherry picked to put it nicely. The M1 Ultra claims already look suspicious since the GPU graph tops out at ~120W & the CPU graph tops out at ~60W yet the M1 Studio is rated for 370W continuous power draw. Since you mention ML specifically, looking at some benchmarks out there (like https://tlkh.dev/benchmarking-the-apple-m1-max#heading-gpu & https://wandb.ai/tcapel…

I am sure it has great GPU performance for what it is but comparing it to a top Nvidia chip just seems ridiculous on Apples part. Apple I think is going to have trouble conquering back the semi pro workstation market they abandoned not that many years ago if they do not start offering M1 chips along with Nvidia GPUs. Once again we get a Mac Semi-Pro Mini (seems like the Studio is more like a replacement for the Trash…

Just a note: they explicitly said the Mac Pro is coming.

Re: Apple M1 Ultra

#602

Post by hedgehog »

TuringNYC wrote: post #491
polyrand wrote: post #258

I think the GPU claims are interesting. According to the graph's footer, the M1 Ultra was compared to an RTX 3090. If the performance/wattage claims are correct, I'm wondering if the Mac Studio could become an "affordable" personal machine learning workstation (which also won't make the electricity bill skyrocket). If Pytorch becomes stable and easy to use on Apple Silicon [0][1], it could be an appealing choice. [0]…

It shocks me how much payroll and cap-ex is spent on the M1 and how little is invested in getting TensorFlow/Pytorch to work on it. I could 10x my M1 purchases for our business if we could reliably run TensorFlow on it. Seems pretty shortsighted. The GPU claims wouldnt even need to be on parity with NVIDIA, it would just need to offer a vertically integrated alternative to having to use EC2.

Having beaten my head on this for a while (and shipped the first reasonably complete ML framework that runs on Metal) Apple's opinion as expressed by their priorities is that it's just not important.

Re: Apple M1 Ultra

#603

Post by sytelus »

polyrand wrote: post #258

I think the GPU claims are interesting. According to the graph's footer, the M1 Ultra was compared to an RTX 3090. If the performance/wattage claims are correct, I'm wondering if the Mac Studio could become an "affordable" personal machine learning workstation (which also won't make the electricity bill skyrocket). If Pytorch becomes stable and easy to use on Apple Silicon [0][1], it could be an appealing choice. [0]…

The most important detail here is 128GB ram for GPU computation! This allows to train monster models, ex 1B params GPT series, on single M1 Ultra. This is quite unprecedented. Unfortunately, it is also about 3.5x slower than 3090.

Re: Apple M1 Ultra

#604

Post by perfopt »

Macha wrote: post #538
momothereal wrote: post #479

Earlier quoted context omitted.

I believed that until I realized I couldn't individually upgrade my CPU or RAM because I have a mobo with LGA1150 socket and only supports DDR3 (and it's only 6 years old). So eventually you still have to "replace everything" to upgrade a PC.

You were unlucky to buy ddr3 near its end of life then (like someone buying ddr4 now), but you could still upgrade stuff like your GPU or drives independently. My first SSD (a 240gb Samsung 840) is still in service after 9 years with its smart metrics indicating only 50% of its expected lifetime cycles have been used, for example. You could also put a 4790k, 16gb of ddr3 and a modern gpu in that system to get a perfe…

Right now not many DDR5 boards. In fact none for AMD

Re: Apple M1 Ultra

#605

Post by Geee »

needle0 wrote: post #558
Geee wrote: post #554

Earlier quoted context omitted.

I didn't see display resolution (or the model of HMD) mentioned in the video. I'm using Varjo Aero, which has 2880x2720 per eye, which is almost quapruple (per eye) compared to Valve Index. I think this resolution is enough for a good VR experience; pixels are almost invisible, and even small text is readable. However, HL Alyx doesn't run at 90 fps on full resolution.

Uh, Varjo Aero is a clear outlier in terms of resolution and definitely does not represent the current typical VR headset experience (which in early 2022 would be around Quest 2 or Valve Index). If you went out of your way to get an expensive niche high-end headset, it struggling on a 3090 is your problem, not HL:A's.

I'd say that its just a couple of years ahead of the mainstream curve. My thinking is that Apple probably wants to make a very high resolution VR headset (similar or close to Varjo Aero), and it seems that with their new chips, they might be able to pull it off.

Re: Apple M1 Ultra

#606

Post by kiratp »

atq2119 wrote: post #462
kiratp wrote: post #230

Earlier quoted context omitted.

My workstation has a 3990x. Our "world" build is slightly faster on my M1 Max. https://twitter.com/kiratpandya/status/1457438725680480257 The 3990x runs a bit faster on the initial compile stage but the linking is single threaded and the M1 Max catches up at that point. I expect the M1 Ultra to crush the 3990x on compile time.

Are both cases compiling to the same target architecture? If not, you may well be comparing the relative performance of different compiler backends instead of comparing the performance of your CPUs. (+ now I see it's rust: how parallel is your build, really?)

The target arch doesn’t matter for the change-build-test loop d vs do. All that matters is how fast can you compile your code to test it.

If the final x86 production build takes longer it doesn’t matter - that happens on the cloud anyway.

Edit: Rust builds are very parallel until linking. No different than any other LLVM build.

Re: Apple M1 Ultra

#607

Post by kiratp »

zerd wrote: post #455
kiratp wrote: post #230

Earlier quoted context omitted.

My workstation has a 3990x. Our "world" build is slightly faster on my M1 Max. https://twitter.com/kiratpandya/status/1457438725680480257 The 3990x runs a bit faster on the initial compile stage but the linking is single threaded and the M1 Max catches up at that point. I expect the M1 Ultra to crush the 3990x on compile time.

From the tweet reply you used sccache with hot cache, which would probably be mostly single-threaded since it's just fetching and copying things from cache.

I’m not trying to compare the details of SOC performance.

Just that with the same hot caches, the average change-build-test loop that developers do 100+ times a day is just faster on the M1 Max.

Re: Apple M1 Ultra

#608

Post by hajile »

whoisburbansky wrote: post #301
polyrand wrote: post #258

I think the GPU claims are interesting. According to the graph's footer, the M1 Ultra was compared to an RTX 3090. If the performance/wattage claims are correct, I'm wondering if the Mac Studio could become an "affordable" personal machine learning workstation (which also won't make the electricity bill skyrocket). If Pytorch becomes stable and easy to use on Apple Silicon [0][1], it could be an appealing choice. [0]…

Cursory look gives you a ~$3500 price tag for a gaming PC with a 3090 [1], vs. at least $4k for a Mac Studio with an M1 Ultra. Roughly the same ballpark, but I wouldn't call the M1 Ultra more affordable given those numbers. 1. https://techguided.com/best-rtx-3090-gaming-pc/#:~:text=With... .

I can't put a 3090 into a 3.5 liter case. Even a 10 liter case is really pushing it. That's before mentioning power savings. 3090 real world power when in use is something like 4x as high.

Re: Apple M1 Ultra

#609

Post by vimy »

whoisburbansky wrote: post #301
polyrand wrote: post #258

I think the GPU claims are interesting. According to the graph's footer, the M1 Ultra was compared to an RTX 3090. If the performance/wattage claims are correct, I'm wondering if the Mac Studio could become an "affordable" personal machine learning workstation (which also won't make the electricity bill skyrocket). If Pytorch becomes stable and easy to use on Apple Silicon [0][1], it could be an appealing choice. [0]…

Cursory look gives you a ~$3500 price tag for a gaming PC with a 3090 [1], vs. at least $4k for a Mac Studio with an M1 Ultra. Roughly the same ballpark, but I wouldn't call the M1 Ultra more affordable given those numbers. 1. https://techguided.com/best-rtx-3090-gaming-pc/#:~:text=With... .

You also need to compare the right cpu. M1 Ultra cpu is the equivalent of the fastest threadripper. Which costs $3990. So a pc with similar performance would be $7500

Re: Apple M1 Ultra

#610

Post by afavour »

codyb wrote: post #568
derac wrote: post #528

Earlier quoted context omitted.

Seems unlikely that the general population would spend 2-4k on a console.

Don’t gamers spend tons of money on gaming PCs? Also, might be cheaper a couple years down the line.

PC gamers do. Console gamers don't. And there are a lot more of the latter than the former.
Post reply on HN