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

Apple M2 Ultra SoC isn’t faster than AMD and Intel last year desktop CPUs

wccftech.com

Post reply on HN
Print view View original

221–230 of 448 posts

Re: Apple M2 Ultra SoC isn’t faster than AMD and Intel last year desktop CPUs

#221

Post by microtonal »

samwillis wrote: post #148
microtonal wrote: post #121

Earlier quoted context omitted.

Maybe, but most of these comparisons are based on GPU performance (ie. games). For other workloads like machine learning, Tensor cores on NVIDIA GPUs will blow Apple Silicon GPUs out of the water. The M2 Ultra is 27 TFLOPS. The 4080't Tensor Cores are 48.7 TFLOPS in FP32, 194.9 TFLOPS in FP16, 389.9 TFLOPS in FP8 with FP16 accumulate. (IIRC on Apple Silicon GPUs FP16 performance is roughly the same as FP32.) (There i…

I think the exciting thing with Apple silicone and "AI" is inference, not training. Due to their unified memory, you can potentially have enormous local model inference. That's potentially much more expensive on a PC with a GPU having its own memory. Apple have an opportunity, if they 2-4x the memory on the entry level devices (not beyond the realms of possibility), to make local inference a thing available to all.

Sure, but for inference it’s much easier to work in lower precision. Training with very low precision is also possible, but is often more tricky due to numerical stability.

A lot of work is going on in 8-bit inference and even 4 bit inference. So, models that need 64 GB in FP32 can do with 16GB VRAM in FP8 or INT8, which is well within the realm of consumer NVIDIA cards. And the latest NVIDIA tensor cores will absolutely destroy Apple Silicon GPUs or the Neural Engine in 8 bit.

So, I don’t think it’s really a strong argument. And as someone who is a Mac user and a ML practitioner, I’d be very happy if they started supporting eGPUs again.

Apple Silicon has many strengths and the GPU core are fine for many ends, from games to graphics apps.

But let’s not pretend that Apple is beating NVIDIA at their own game (yet). That day might come, but currently it only leads to disappointed users in ML forums who were hyped into thinking that their vanilla M2 MacBook Airs can almost compete with a 4090 in training a deep transformer model. (Yes, that happens.)

Re: Apple M2 Ultra SoC isn’t faster than AMD and Intel last year desktop CPUs

#222

Post by visionscaper »

I'm wondering to what extend these benchmarks, especially the GPU benchmarks, are really optimised for the specific Apple M2 architecture? And to what extent the high GPU-GPU bandwidth is benchmarked, this also influences real-world performance.

Re: Apple M2 Ultra SoC isn’t faster than AMD and Intel last year desktop CPUs

#223

Post by jml78 »

littlestymaar wrote: post #214
achandlerwhite wrote: post #146

Earlier quoted context omitted.

They announced MetalFX for their upscaling tech last year and by all accounts it works well if developers take advantage of it.

Not familiar with DLSS at all, does it requires developers to do something in order to take avantage of it too? I had imagined it was automatic but then again I know nothing about it beyond the marketing pitch to consumers.

Yes, games have to do work to support DLSS.

I am not knowledgeable enough to know how much work it is but I have played games that didn’t initially support it but eventually released an updated that added support.

There are also multiple “levels” for DLSS in games that support it, eg. Quality, performance, etc

Re: Apple M2 Ultra SoC isn’t faster than AMD and Intel last year desktop CPUs

#224

Post by gumby »

MBCook wrote: post #195
KerrAvon wrote: post #31

Earlier quoted context omitted.

haven’t read the article, but there are no single threaded use cases on Apple platforms that Apple cares about, so I’d take any claim that Apple is optimizing for single threaded performance with an enormous bag of salt

> but there are no single threaded use cases on Apple platforms that Apple cares about I don’t see how that could be true. A huge amount of software tasks are basically single threaded. Remember since Apple does everything soup-to-nuts they have a ton of performance data from their computers to know what real user workloads look like so they can optimize the hardware + the software for them.

I made this claim (single threaded performance matters) but in a parallel comment to yours noted that I was unable to find substantiating statements from apple that I believe led to my assertion.

So in that spirit I will point out that Apple's support code / framworks etc do a bunch of multithreaded UI and network stuff even when and app's code is putatively single threaded.

Now that stuff IMHO is pretty high latency (e.g. waiting on user action) so as a developer I still think my statement, and your impression, are correct. But I'd like to see something from Apple on the topic.

Re: Apple M2 Ultra SoC isn’t faster than AMD and Intel last year desktop CPUs

#226

Post by Lapha »

BaculumMeumEst wrote: post #130
spiderfarmer wrote: post #126

Earlier quoted context omitted.

Desktop pcs requiring a ludicrous amount of power and cooling is absolutely problematic. The amount of people willing to put up with big honking machines is dwindling.

AMD and Intel are jacking up the default power settings to absurd levels because it gives marketing bigger numbers to to throw around. You can drastically reduce the power you supply to desktop chips with BIOS settings. You'll generate far less heat, can use a smaller power supply and form factor, while still getting great performance. If Intel/AMD get to a level where their chips rival M1/M2 while power throttled, t…

Both companies are juicing up the default power limits which absolutely matters if you're concerned about total Wh for sustained loads, thermal considerations, PSU rating, noise, etc, as you mentioned, but to be fair to AMD here they've released an eco mode on the 7000 series which does lower the default power limit quite substantially. You could do this before with manual adjustments to voltage and clock speeds, but now it's a simple toggle.

Another point is that in terms of power efficiency AMD are absolutely mopping the floor with Intel, where even older high power R9s are consistently outperforming Intel's lower power i5 and i7s by a wide margin. This sort of discussion is often left out of reviews which only look at TDP or peak wall power. It's still not ARM levels of efficiency, but x86 vs ARM or even desktop vs laptop efficiency is an entirely separate conversation.

Re: Apple M2 Ultra SoC isn’t faster than AMD and Intel last year desktop CPUs

#227

Post by pram »

mnd999 wrote: post #108
lopkeny12ko wrote: post #94

One of my biggest gripes about Macs is that their thermal engineers refuse to get off their high horses and build real cooling solutions that don't artifically limit the capabilities of an objectively performance-competitive CPU. I don't care if it consumes 200 W of power. I don't care if the chassis needs to be thicker to accommodate a larger heatsink. I don't care if you need to run the fans at 100%. Just let me us…

They used to, the G5 Macs had impressive cooling.

The G5s had to have impressive cooling lol

Re: Apple M2 Ultra SoC isn’t faster than AMD and Intel last year desktop CPUs

#228

Post by charrondev »

kemayo wrote: post #42
jacooper wrote: post #21

Earlier quoted context omitted.

Well, the article is speaking about the M2 ultra, which only going to be used in the Mac pro under Mac studio, both desktop computers.

The Mac Pro does have a rack-mounted configuration for the non-desktop data centers case. (I have no idea whether people will actually use it that way, but it exists.) https://www.apple.com/shop/buy-mac/mac-pro/rack

I can see those being bought up for Datcenters and CI use. There have been companies hosting huge racks of Mac minis for ages to do CI for MacOS and iOS software.

Re: Apple M2 Ultra SoC isn’t faster than AMD and Intel last year desktop CPUs

#229

Post by pram »

hulitu wrote: post #114
nickdothutton wrote: post #110

Earlier quoted context omitted.

I wasnt a G5 owner, waited for the Intel switch, but I gather the G5 water cooling had a very high failure rate and this was at least a large part of why Jobs dropped IBM/MOT/PPC int he end. They promised him a high clock, air cooled G5 but couldn't deliver.

> They promised him a high clock, air cooled G5 but couldn't deliver. Then how were IBM processors cooled ? IBM used also PowerPC

If you've never been around an AIX Power server: they're extraordinarily loud. Usually these 3/4U rack servers that sound like a plane taking off.

Re: Apple M2 Ultra SoC isn’t faster than AMD and Intel last year desktop CPUs

#230

Post by waboremo »

burmanm wrote: post #220
waboremo wrote: post #204

Earlier quoted context omitted.

You should care because you're paying for that energy use lol

In the end that sum it takes energy during gaming is meaningless. I spend more on coffee while playing than to electricity.

Right, and you could have afforded better/more coffee with the savings costs to consuming less power. Which is the point: energy costs are not 0. These costs vary depending on which devices you play on, regardless of games. Someone playing the same games as you on PS5 rather than a gaming PC, is going to save enough money to then buy another indie game per month or more coffees in that month than you.

Note: this isn't about debating which platform is best, rather understanding power consumption impact and the cost to your budget.

Are these costs major? Life changing? Absolutely not. But you should care about them and make decisions incorporating them, not ignoring it as if power consumption is irrelevant even if doubled!

Post reply on HN