Apple M1 Max Geekbench Score • Hacker News Board

Re: Apple M1 Max Geekbench Score

#521

Post by ihattendorf » Thu, Oct 21, 2021, 1:30 AM UTC

kbenson wrote: post #513

caycep wrote: post #476

Earlier quoted context omitted.

well....M1 Max single core also beating a 5950x But tbh, it doesn't seem the new hotness in chips is single core CPU, it's about how fancy you spend the die space in custom processors, in which case the M1 will always be tailored to Apple (and presumably Mac users') specific use-cases...

You actually get better single core performance out of a 5900x than a 5950x. The more cores the AMD CPU has, generally the more they constrain the speed it can perform at on the top end. In this case, the 5950x is 3.5GHz and the 5900x is 3.7GHz. The 5800x is even slightly faster than that, and there's some Geekbench results that show single core performance faster than the score listed here, but at that point the mul…

That's not necessarily true, single core boost on a 5950x is higher than the 5900x (4.9GHz vs 4.8GHz).

Re: Apple M1 Max Geekbench Score

#522

Post by ChrisMarshallNY » Thu, Oct 21, 2021, 1:32 AM UTC

blakesterz wrote: post #100

I just can't figure out what I'm missing on the "M1 is so fast" side of things. For years I worked* on an Ubuntu desktop machine I built myself. Early this year I switched to a brand new M1 mini and this this is slower and less reliable than the thing I built myself that runs Ubuntu. My Ubuntu machine had a few little issues every no and then. My Mini has weird bugs all the time. e.g. Green Screen Crashes when I have…

I thought native M1 Linux was still a work in progress (disclaimer: I’m not a Linux geek).

Even if it’s out there, I suspect it’s not optimized.

I think standard distros are running on Rosetta2, which is an (excellent, from all reports) X86 emulator.

Re: Apple M1 Max Geekbench Score

#523

Post by perryizgr8 » Thu, Oct 21, 2021, 1:36 AM UTC

blakesterz wrote: post #100

I just can't figure out what I'm missing on the "M1 is so fast" side of things. For years I worked* on an Ubuntu desktop machine I built myself. Early this year I switched to a brand new M1 mini and this this is slower and less reliable than the thing I built myself that runs Ubuntu. My Ubuntu machine had a few little issues every no and then. My Mini has weird bugs all the time. e.g. Green Screen Crashes when I have…

The Intel macs were extremely slow. Even a core i9 16 inch ran hot and noisy. A small xps 13 felt much faster in general use. I don't know if this was deliberate by Apple or just a side effect of the extremely anemic cooling.

So all those people who had been using macbooks felt the speed when they used a decently fast laptop. Everybody else knows that M1 laptops are fast, like non-apple laptops have been for quite some time.

Re: Apple M1 Max Geekbench Score

#524

Post by throwawaylinux » Thu, Oct 21, 2021, 1:37 AM UTC

delfinom wrote: post #443

smoldesu wrote: post #431

Earlier quoted context omitted.

This definitely is a factor. Another thing that people frequently overlook is how competitive Zen 2 is with M1: the 4800u stands toe-to-toe with the M1 in a lot of benchmarks, and consistently beats it in multicore performance. Make no mistake, the M1 is a truly solid processor. It has seriously stiff competition though, and I get the feeling x86 won't be dead for another half decade or so. By then, Apple will be com…

AMD is already planning not only 20% YoY performance improvements for x86 but now has 30x efficiency plan for 2025. I think x86 is in it for much longer than a decade. Intel otoh, depends on if they can gut all the MBAs infesting them.

30x efficiency is specifically for 16 bit FP DGEMM operations, and it is in the context of an HPC compute node including GPUs and any other accelerators or fixed function units.

For general purpose compute, so such luck unfortunately. Performance and efficiency follows process technology to a first order approximation.

https://www.tomshardware.com/news/amd-increase-efficiency-of...

Re: Apple M1 Max Geekbench Score

#525

Post by josephg » Thu, Oct 21, 2021, 1:43 AM UTC

smoldesu wrote: post #61

mhh__ wrote: post #39

Earlier quoted context omitted.

I'm waiting to be corrected by someone who knows GPU architecture better than me but as far as I can tell the synthetic benchmarks can trade blows with a 3070 or 80 (mobile), but the actual gaming performance isn't going to be as rosy. Also recall that very few games needing that performance actually work on MacOS

The gaming performance will be CPU-bottlenecked. Without proper Wine/DXVK support, they have to settle for interpreted HLE or dynamic recompilation, neither of which are very feasible on modern CPUs, much less ARM chips.

From all reports, rosetta2 performs remarkably well. Apparently they added special hardware support for some x86 features to improve the performance of dynamically recompiled code.

Re: Apple M1 Max Geekbench Score

#526

Post by sunflowerdeath » Thu, Oct 21, 2021, 1:46 AM UTC

keymone wrote: post #297

munro wrote: post #274

Here's the link to the MacBookPro18,2 OpenCL benchmark: * M1 Max OpenCL https://browser.geekbench.com/v5/compute/3551790 [60,167 OpenCL Score] Comparing to my current MacBook Pro (16-inch Late 2019) & my Hetzner AX101 server: * MacBook Pro (16-inch Late 2019) vs M1 Max - CPU https://browser.geekbench.com/v5/cpu/compare/10496766?baseli... [single 163.6%, multi 188.8%] * MacBook Pro (16-inch Late 2019) vs M1 Max - Open…

> * NVIDIA GeForce RTX 3090 vs M1 Max - OpenCL https://browser.geekbench.com/v5/compute/compare/3551790?bas ... [29.0%] 30% performance for ~15% of power use in laptop form factor? that's not boo, that seems like a clear win for apple.

Also, the 3090 costs almost like the whole macbook

Re: Apple M1 Max Geekbench Score

#527

Post by touristtam » Thu, Oct 21, 2021, 1:46 AM UTC

flatiron wrote: post #444

jandrese wrote: post #336

Earlier quoted context omitted.

Downside of RHEL is the package repo is anemic and out of date. Sometimes horribly out of date. It's hardly uncommon to run into some issue with an application and then look it up online and find out that the fix was applied 8 versions after the one that's in the repo. Worse is when you start grabbing code off of Git and the configure script bombs out because it wants a library two versions ahead of the one in the re…

This use case begs for arch. Not bloated and up to date upstream and AUR is very robust for the “one off” apps. Rhel on desktop makes sense for development for rhel on servers. Can’t think really of any other use case. At least they could try out centos stream? For all its hate it’s really nice for certain use cases. Although people are religiously against it (rh needs to really work on that before it dies on the vin…

Why go all the way to Arch when Fedora is perfectly serviceable? Used it for a few years before being imposed a MBP.

Re: Apple M1 Max Geekbench Score

#528

Post by inkyoto » Thu, Oct 21, 2021, 1:53 AM UTC

jermaustin1 wrote: post #128

hazeii wrote: post #108

Earlier quoted context omitted.

Over 6 years, 2x the performance in single-core and 4x (for 2.5x the cores, depending how you count) seems surprisingly low compared to stuff I've read (not experienced yet) with the M1.

I think a lot of the perceivable speedup in M1 Macs is less to do with these benchmarks and more to do with optimizing the OS and Apple software for the M1 and the various accelerators on the SoC (video encoding and ML).

> perceivable speedup in M1 Macs

Perception is key here.

Of course M1's are perceived as being faster because:

- video camera encoding/decoding is offloaded to a dedicated processor;

- disk encryption is offloaded to a dedicated processor;

- rendering is offloaded to dedicated GPU's;

- matrix multiplication is offloaded to dedicated matrix multiplication processors (neural engines);

- no wasteful memory transfers (even though the memory bus is 2048 bit wide anyway) across dedicated processors.

Kind of like channel processors running channel programs in IBM mainframes, brought to a mainstream ultraportable laptop.

And, on top of that, the screen that drops the refresh rate by a factor of 6 when there is no need to refresh the picture too often.

Ironically, the CPU is left with pretty much one job to do: to compute. It is possible to get fast compilation times whilst doing all of the above at once. Or have an app that does all of the above at once, too (no, it won't save Microsoft Teams). And be power efficient, too; hence a long battery life and an Apple marketing team brouhaha: «We can do this all day». They are right – they can actually do all of that, and they can brag about it now because they have a full list from the above fulfilled.

System performance is a holistic matter, not just single/multicore comparisons oftetimes taken out of the context.

Re: Apple M1 Max Geekbench Score

#529

Post by paulmd » Thu, Oct 21, 2021, 1:54 AM UTC

throwawaylinux wrote: post #524

delfinom wrote: post #443

Earlier quoted context omitted.

AMD is already planning not only 20% YoY performance improvements for x86 but now has 30x efficiency plan for 2025. I think x86 is in it for much longer than a decade. Intel otoh, depends on if they can gut all the MBAs infesting them.

30x efficiency is specifically for 16 bit FP DGEMM operations, and it is in the context of an HPC compute node including GPUs and any other accelerators or fixed function units. For general purpose compute, so such luck unfortunately. Performance and efficiency follows process technology to a first order approximation. https://www.tomshardware.com/news/amd-increase-efficiency-of...

also bear in mind that AMD's standards for these "challenges" have always involved some "funny math", like their previous 25x20 goal, where they considered a 5.02x average gain in performance (10x in CB R15 and 2.5x in 3D Mark 11) at iso-power (same TDP) to be a "32x efficiency gain" because they divided it by idle power or some shit like that.

But 5x average performance gain at the same TDP doesn't mean you do 32x as much computation for the same amount of power. Except in AMD marketing world. But it sounds good!

https://www.anandtech.com/show/15881/amd-succeeds-in-its-25x...

Like, even bearing in mind that that's coming from a Bulldozer derivative on GF 32nm (which is probably more like Intel 40nm) 5x gain in actual computation efficiency is still a lot, and it's actually even more in CPU-based workloads, but AMD marketing can't help but stretch the truth with these "challenges".

Re: Apple M1 Max Geekbench Score

#530

Post by _kbh_ » Thu, Oct 21, 2021, 1:56 AM UTC

kllrnohj wrote: post #520

spitfire wrote: post #397

What am I missing here? The marketing presentations spoke about 200GB/sec and 400GB/sec parts. Existing CPU's generally have 10s of GB/sec. But I see these parts beating out existing parts by small margins - 100% at best. Where is all that bandwidth going? Are none of these synthetic benchmarks stressing memory IO that much? Surely compression should benefit from 400GB/sec bandwidth? This also raises the question how…

The SoC having 400GB/s of memory bandwidth doesn't mean that any individual CPU core can saturate 400GB/s of memory bandwidth (or even if all the CPU cores combined can achieve that). The SoC's memory bus is also feeding the GPU, and GPUs tend to be _very_ memory bandwidth hungry (see other discreet GPUs pushing over 1TB/s of memory bandwidth). The CPU performance, even where memory IO limited, is more likely limited…

A single firestorm (performance) m1 core in the m1 Mac mini can sustain ~60Gb/s read from ram. I have no doubt that 8 of them could come close to if not entirely saturate 400GB/s.

https://www.anandtech.com/show/16252/mac-mini-apple-m1-teste...