pouët.net

Apple M1 (ARM) architecture demos

category: general [glöplog]
Quote:
Quote:
console can be "demo platform" if you hack it, install linux there, and made your own graphic driver and API for that GPU it has

Now you're just pushing the cost towards the end-user - rarely a good strategy.


(until someone puts it in a neat package à la letterbomb for the wii)
added on the 2020-11-26 01:44:09 by porocyon porocyon
I was about to make a comment about how convincing someone to google "Letterbomb" is surely gonna make them endear to your idea of "neat", but then I realized the fact that you have to somehow also convince them to find out their MAC address which is probably even worse when all you tried to convince them of is to watch a demo.

(But sure, Windows demos and DLLs.)
added on the 2020-11-26 10:27:54 by Gargaj Gargaj
I have no experience with any Apple products, but something tells me that these spectacular benchmark numbers wont translate into real world performance under real payload.

I have ported some of my software to ARM for a while, there was always a magnitude of difference in favor of x86 chips despite of similarities in synthetic benchmarks (at the same power consumption class).

I know Apple redesigned the whole core, but even with their well-pipelined superscalar design i have doubts about it.
added on the 2020-11-26 14:38:56 by Geri Geri
I know some people who got one already (and definitely not all apple fans, far from), and they definitely *are* fast, even in real-world scenarios.
added on the 2020-11-26 16:46:47 by porocyon porocyon
Quote:
but then I realized the fact that you have to somehow also convince them to find out their MAC address which is probably even worse


(it's printed on the sticker on the bottom of the console, and also can be found in the network settings menu)
added on the 2020-11-26 16:48:40 by porocyon porocyon
Again... Shadertoy on M1 how does it run?

I don't think there will be a big difference in GLSL code performance vs MLS (Metal Shading Language) they are very similar, both compilers should generate similar code to run on the M1.

But.. it's super simple to convert some shadertoy shaders from GLSL to METAL, just to check if using "All the Apple recomended pipeline" makes any difference in shaders
It's not going to be great for shadertoy... they've only released low end models with integrated graphics. For low end / integrated graphics, it'll be extremely good, but we're probably talking mid-range video card from a few years ago, not 3090. the tflops numbers put it around 1/4 the speed of my vega56.

The benchmarks really do translate to real world performance tho, I've been looking at people posting compilation times where a low end mac mini is a lot faster than a 12 core mac pro. While waiting for my code to compile. 🙄 Nearly ordered one today, but they're sold out and by the time it arrives, I'll probably be done with this slow-compiling work.
added on the 2020-11-27 21:12:55 by alia alia
@All

All of a sudden everyone now wants all and only ARM.
added on the 2020-11-28 14:28:40 by AlienTech AlienTech
https://twitter.com/ErrataRob/status/1331735383193903104

A guy explaining some of the black magic.
added on the 2020-11-28 16:22:33 by xernobyl xernobyl
Meanwhile Linux for the M1 seems to be on the way. Check out or even support the effort here.
added on the 2020-12-01 15:50:23 by Kuemmel Kuemmel
https://i.imgur.com/fXkvMJa.png

Linus tech tips reviewed the chip. Unless gpu acceleration/decoders/encoders or heavy dsp usage is involved, the M1 is 3-4 times slower than a regular modern x86 chip (but has half the power consumption), and about half as fast on the same power consumption levels.
added on the 2020-12-01 16:47:55 by Geri Geri
As measured on x265, which has a bazillion AVX optimizations and runs in pure C for 64-bit Arm? Yeah, that's a reasonable test… :-)
added on the 2020-12-01 18:00:04 by Sesse Sesse
yeah, totally mindblowing ;-)
you just need to wait 4 more months until there's a working Google Drive app, but maybe it runs with Rosetta :P
8800 GTX level look like, back to 2006
(Vega8 in AMD processors like 10x times faster than this)
added on the 2020-12-03 20:00:16 by Danilw Danilw
Quote:
8800 GTX level look like, back to 2006
(Vega8 in AMD processors like 10x times faster than this)


Is it?

https://technical.city/en/video/Apple-M1-8-Core-GPU-vs-Radeon-RX-Vega-8-Ryzen-4000
added on the 2020-12-03 20:21:22 by Rob Rob
I do not trust this link at all
youtube has a video of launching TeamFortress2 on AppleM1 and it works worse than 10 years ago on those old videocards

GPU that launch TF2 with 10-30FPS can not launch Fortnite with 140FPS, its literally impossible
added on the 2020-12-03 20:26:46 by Danilw Danilw
🆗
added on the 2020-12-03 20:30:20 by Rob Rob
We just wait and see when games are patched for native ARM-code. Feral does porting the middleware actually...and forget stupid comparisons like a 50 USD chip against a 650 USD-Chip... it's like to compare a nissan-micra to a porsche-boxter...
at the end of the day the M1 is a huge step forward for apple compared to intel igp before. there are better cpu's/gpu's on the market like the actual ryzen-chips but apple has no ryzen chips in their machines, so as an apple user I compare intel to m1 and I don't care about ryzen because I will stick to osx and the ecosystem.
that has nothing to do with fanboy or whatsoever, there are actual things I hate totally (itunes->music)...but the overall experiance is quite good after all. I ordered my M1-Mini and I will order the 32" iMac with dedicated gpu (lifuka) as well when it comes out in q3/4 2021.
Quote:
when games are patched for native ARM-code

Patched, never.

A few new games, maybe, but I'd predict mostly smallish Unity/Unreal-based ones, as no major studio will port their engine to a platform that has ~4% coverage.
added on the 2020-12-05 14:03:37 by Gargaj Gargaj

login