Hacker News Reader

Comment by sirwhinesalot | Hacker News Reader

Do I really need to be the one pointing out that M1 Ultra has an "integrated" GPU and that these tests are disadvantaging it due to Rosetta and weak engine support for Metal?

The results are nuts.

View on HN

What does that really mean at this point? A big part of what made integrated cards suck so much is how they had to share memory with the CPU and with a unified memory SiP design that doesn't seem like a problem. Plus the M1 Ultra has 114 billion transistors whereas the 3090 is 28 billion. Even if the M1 Ultra is only using a 1/4 for the GPU that's still the same number of transistors. Even a 6th is still huge.

If Apple wants people to take their claims seriously and if weak engine support for Metal and Rosetta support is a problem they should pay someone to port a popular graphics intensively game properly. Of course it works in their favor not to because fans can point to it and say we'll it's just not optimized for it! Like we all did back in the PPC days until they admitted how crap those chips were when switching to Intel.

Apple doesn't care about gaming, but they do care about 3D rendering. Blender benchmarks (which is just now getting official Metal support, sponsored by Apple), should be the real deal.

Blender is still 3/4 times slower on ultra compared to rtx 3090 according to people testing it

34s on ultra (64 core gpu) vs 11s on rtx 3090

So 3 times slower but taking ~4 times less power (350W 3090 70W ultra [on 48 core gpu so about 90 on 64 cores])

Sounds about right, it's not magic. AFAIK the Metal support in Blender is not very optimized yet (it's brand new) but even optimized I don't expect it to come close to an RTX 3090. I still find the results amazing, for the amount of power this thing needs if it reaches 2x slower after optimizations, that's massive. Don't forget that the Ultra also includes the CPU and RAM in the power consumption.

Its wall power without cpu encoding with cpu its ~140W.

But i can agree results are good but we need also to remember Samsung 8nm [rtx 3090 node] is 4 years old so 2 generations different from TSMC 5nm 4090 will be probably on 5nm node in few months and presumably will have 3x teraflops (probably 3x tf32 precision but it could be fp32 using 2/1.5x(?) more power).

(also I'm not sure about optimization it was write by apple employees and apple likes to drop open source support and focus on proprietary software)

The RAM packaging doesn't really matter much, Apple runs JEDEC spec memory timings anyway.

But it is impressive that they get high GPU performance with LPDDR. Consoles also use one big SoC and unified memory – but with GDDR memory!

The weird insistence on supporting only Metal and nothing outside of Apple's own bubble is entirely Apple's own doing.

I agree, but that is still something we need to take into account when comparing the hardware. Imagine if Arm Windows gets popular and it somehow has great drivers for this GPU (unlikely, but humor me here). Then the actual power of the hardware would be a lot more relevant for gaming related benchmarks.

Who cares about hardware you can't utilize? Its just a piece of silicon with potentially any performance imaginable. What you can do with it, now, matters. If this changes in some theoretical future, well we re-evaluate performance claims and stats.

Nobody bothers to guess what's the potential of some new graphic card once all drivers, games, OS etc are fully tweaked and optimized. You decide on performance now.

You can already use it now if you're targeting the platform. It's not the platform provider's problem if you're using a compatibility layer, that was your choice. Well, from a development POV at least, from a user POV, if your tools are using the compatibility layer, there is not much you can do.

We knew this would happen for everything using Rosetta, that's just part of transitioning to a new CPU architecture. For Vulkan vs Metal that's self-inflicted on Apple's part.

You mean that we should take into account how actually complicated is to get software that fully support the hardware into account for HW comparisons where a type of HW can only run with a specific OS?

No, I mean that's not relevant to compare what the hardware can do. It's like how the PS3 was super hard to program but in the end it was producing more impressive results than the Xbox360. This was not a surprising result, it was to be expected from the hardware. It's not that different from what is happening here, benchmarks running on x86 emulation on engines running compatibility layers for Metal are not representative of the hardware.

With the difference that Apple HW performance might be "good enough" for 3rd party software vendors if they take into account the extra development burden for the platform. Anyway some of those benchmarks were already using Metal and are still behind.

>The weird insistence

Well it worked for Microsoft

To be fair, while they promoted DirectX, Microsoft have provided support for using OpenGL in Windows[1] (while the docs here are dated 2019, the porting guide from IRIS GL, OpenGL's early 90s predecessor, indicates they're actually much older)

1: https://docs.microsoft.com/en-us/windows/win32/opengl/portin...

Microsoft supported OpenGL prior to developing Direct3D. As soon as that got into good enough shape to supplant OpenGL, OpenGL got dropped.

These days you're better off using OpenGL via ANGLE, built on top of Direct3D. That will probably also happen analogously on Apple's platform with Metal, so I don't quite get the kvetching about Apple deprecating OpenGL.

As if it matters whether Apple ships an OpenGL implementation as part of the OS or purely as a normal user space library.

The discussion over on blender devtalk is that they now have cycles working on Metal (including AMD GPUs) and now they can start to optimise.

https://devtalk.blender.org/t/cycles-apple-metal-device-feed...

I’ve already seen a significant CPU improvement on my 2019 16” and I can now use my 5500M to render. My PC’s RTX 2080 Super is significantly faster but it draws more than 200W.

This indeed is great progress!

An integrated GPU with 64 cores? (Compare this to the 112 "cores" of the 3090) At that point the distinction makes no damn difference. The rumor mills of the speculated that AMD is going to release high performance server APUs with 32 GB HBM years ago.

Really, if you can fit the memory on top of the GPU, then there is no meaningful distinction between a discrete and integrated GPU because the fundamental difference on the PC platform is whether the iGPU uses the system DDR RAM or whether it uses dedicated GDDR RAM. That distinction really doesn't exist on Apple's M1 platform because the memory is placed directly on top of/next to the SoC.

At this point, Apple Silicon is going on 18 months. If there still isn't software available to run GPU benchmarks natively, that's a problem for that GPU. And it's a problem for validating marketing claims.

Integrated is an advantage not a disadvantage on a technical level, it's just worse for the consumer as you cannot upgrade them separately. What you should be comparing is the number of transistors to see whether the architecture is impressive or if the speed is due to a size brute force. the M1 Ultra has 4x the number of transistors as the 3090 (probably about half is just the GPU)! It's a freaking humongous chip.

What difference does it make if Apple can achieve performance through more transistors than the competition?

It's more TSMC achieving it than apple achieving it, and that technology will get to competitors. https://www.technopixel.org/nvidia-hopper-architecture-comes...

It's just due to apple being the biggest customer they have access 1 year earlier.

>weak engine support for Metal

And who's fault is that?

Reply on news.ycombinator.com