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

361–370 of 448 posts

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

#361

Post by grecy »

jsheard wrote: post #299
torginus wrote: post #286

Earlier quoted context omitted.

GPUs won't work at all? Can't you even use them for ML/compute stuff?

They don't work at all, there are no discrete GPU drivers for ARM macOS. The cooling solution and 1250W(!) power supply they carried over from the 2019 model are kind of hilariously overkill now, I don't think there's anything you still can install into the PCIe slots which comes even remotely close to needing them.

> The cooling solution and 1250W(!) power supply they carried over from the 2019 model are kind of hilariously overkill now

I wonder if that's a small hint or giveaway of something coming in the future that will make use of it.... One can hope.

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

#362

Post by sarsaparilyptus »

threeseed wrote: post #339
tomnipotent wrote: post #336

Earlier quoted context omitted.

> cool the CPU you really need to be running AIO or dedicated water coolers There is no shortage of silent air coolers, such as the NH-D15, that are 20-30 dBA and undetectable unless the only noise in the room. It's usually stock case fans that are the culprit.

If you're running a PC with a dedicated GPU then you will have minimum 6 extra fans. 3 for the GPU and 3 for the case. The latter are pretty noisy and constantly run.

[dead]

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

#363

Post by MBCook »

florakel wrote: post #325
MBCook wrote: post #201

Earlier quoted context omitted.

The took the giant overkill design meant to hold a hot Intel chip and possibly multiple big GPUs, and put a M2 Ultra in it. It’s overcooled . There is no way it’s thermally limited. It’s got to have headroom for days. I’m very curious where this idea I keep seeing in the comments that Apple refuses to run chips at full speed for cooling reasons comes from.

The current Mac Pro is kind of pointless. Before you got a powerful workstation (power consumption was irrelevant), now you get a Studio Mac in a new package. I feel they just launched it for “compliance” not because it is a great product. They hit a wall with scaling the M chip architecture. Building an even larger chip than the Ultra would be insane and going multi socket is not possible.

They need a product for people who need expansion slots for non-GPU things, and it fits that.

I agree that it’s likely they wanted to use the quad M2 chip (jade 4 C die, if I remember the term correctly), which would have given them far more PCIE lanes and made it more desirable.

But that doesn’t seem to have worked out. And the lineup had a whole they needed to fill for some of their top end customers. So here we are.

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

#364

Post by MBCook »

markemer wrote: post #334
MBCook wrote: post #198

Earlier quoted context omitted.

> The i9-13900T and i7-13700T come in at a max turbo power of just 106W Now add in the GPU. Because the Apple number combined both.

Yep - the total TDP of the comparison is like 600W. The Pro is also for people with strange PCIe card needs - the Mac Studio delivers this performance in a tall MacMini form factor and nearly silent operation. Not that there aren't things to complain about, but Apple Silicon architecture is impressive for what it is. The GPU suffers more for its uncommon architecture that is similar to a phones but is harder to optim…

I think Epic could get Unreal to run just fine. Unfortunately they don’t have a great relationship with Apple (I can’t remember why but I think it was something Apple did).

The Mac Pro makes sense for only a few true professionals who need those PCI slots for non-GPU things. But those people are have BIG budgets to spend and are good customers.

Maybe the future Pros will improve, this one is a bit odd. But it has a purpose for those who truly need it.

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

#365

Post by Retric »

sudosysgen wrote: post #333
Retric wrote: post #297

Earlier quoted context omitted.

If the pipeline stalls for 70ms it’s a problem even if in theory you don’t “miss” a frame you still aren’t getting new frames until rendering finishes. All adaptive refresh rates do is avoiding unwanted delays after the frame finishes.

That's not how it works. A 70ms pipeline stall means a ~90ms frame which means that your 1% framerate low is going to be reduced. Adaptive refresh rate means that only frame rate matters. You can't miss a frame, you can only have a frame take longer, and that's measured perfectly well by 1% or 0.1% low frame rates. Also I don't see why an M series CPU is going to be any better at feeding the (asynchrous, multiple fra…

Yes occasional 90ms stall reduces 1% frame rates, but less than how annoying it is.

A 10 minute test at 120 FPS is 10 * 60 * 120 ~= 72,000 frames. Your 1% lows is an average of ~720 frames but so what if of 300 them are at 60 FPS you’re going to have trouble noticing.

While if you’re almost rock solid at 120FPS a dozen 70ms stalls it’s really obvious that something is wrong but the calculated 1% low’s may actually look better. This is especially true as those stalls generally correlate with interning things happening.

The M series CPU comparison people are talking about isn’t average FPS or even 1% lows where dedicated graphics cars have an advantage, the comparison is what happens when things go very wrong. And it’s in that very specific case when they may have an advantage.

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

#366

Post by jsheard »

grecy wrote: post #361
jsheard wrote: post #299

Earlier quoted context omitted.

They don't work at all, there are no discrete GPU drivers for ARM macOS. The cooling solution and 1250W(!) power supply they carried over from the 2019 model are kind of hilariously overkill now, I don't think there's anything you still can install into the PCIe slots which comes even remotely close to needing them.

> The cooling solution and 1250W(!) power supply they carried over from the 2019 model are kind of hilariously overkill now I wonder if that's a small hint or giveaway of something coming in the future that will make use of it.... One can hope.

If they were going to make a dGPU or other big accelerator in the future then I think they would have maintained support for the MPX form factor cards. The new Mac Pro doesn't support MPX, despite being based on the same chassis that MPX debuted with.

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

#367

Post by zarzavat »

astrange wrote: post #343
zarzavat wrote: post #335

Earlier quoted context omitted.

Apple has beef with the worlds largest GPU maker and refuses to sign their drivers because they got screwed by that company over a laptop GPU a decade ago and still hold a grudge. Microsoft on the other hand, is run by adults.

Where can you get these drivers that exist but just aren't signed?

You can’t because Nvidia gave up and stopped developing drivers for macOS. The OS won’t let you install unsigned drivers.

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

#368

Post by Ygg2 »

threeseed wrote: post #324
LeoNatan25 wrote: post #322

Earlier quoted context omitted.

I don’t know what tower case you are using, or what cooling system you have set up, but in modern times, if you hear much noise from your cooling solution, you have chosen components poorly. From fans to pumps to dampened cases, you could have build a relatively silent system.

With a Mac Studio it isn't just relatively silent in almost cases it is silent. With a PC you still have case fans running because even with modern motherboards they don't support switching them off at low temperature. You need to be running special Windows software. And to cool the CPU you really need to be running AIO or dedicated water coolers which themselves will have 3 fans almost always running.

> With a PC you still have case fans running because even with modern motherboards they don't support switching them off at low temperature

What do you mean? Most motherboards support sending 0W to fans at given temps. You just install the motherboard software.

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

#369

Post by kristianp »

xattt wrote: post #313
kmeisthax wrote: post #275

Earlier quoted context omitted.

Don't waste your money on the Pro, you can get an M2 Ultra in a Mac Studio for $4000. The only reason to buy the Pro is if you have a bunch of really weird PCIe cards[0]. As for the desktop use case... sure, you aren't going to care about the $200 of saved power draw, but not having a very loud and hot machine at your desk has to count for something, right? [0] GPUs won't work, their memory access is locked out in ha…

> The only reason to buy the Pro is if you have a bunch of really weird PCIe cards They’re telegraphing the use case for future SOCs.

> They’re telegraphing the use case for future SOCs.

Can someone explain what that means?

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

#370

Post by wtallis »

astrange wrote: post #353
wtallis wrote: post #84

Earlier quoted context omitted.

> and also eliminates a lot of the need to copy data from RAM to VRAM, as both the GPU and the CPU can access the same memory. In return, this allows the GPU to match the performance of discrete GPUs that have a lot more cores. Plenty of PC hardware reviewers have done sensitivity analysis experiments to see how discrete GPU performance is affected by running with a slower or narrower PCIe link. The consensus is usua…

The only reason it wouldn't matter is that software is designed for it to not matter. It enables software to be designed differently ie by not having to ever copy things to VRAM.

That's true to an extent, but when discussing the current performance of GPU hardware the context is necessarily software that actually exists rather than hypothetical software that could exist in the future. The high CPU-to-GPU bandwidth is not currently enabling Apple's GPUs to punch above their weight class (GFLOPS or whatever) except in scenarios where competing GPUs simply don't have enough VRAM. In a future where applications are written with Apple's architecture in mind, any advantage this provides would pretty much boil down to using more VRAM because VRAM capacity isn't as limited anymore.
Post reply on HN