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

M1 Pro First Impressions: Core Management and CPU Performance

eclecticlight.co

Post reply on HN
Print view View original

221–230 of 234 posts

Re: M1 Pro First Impressions: Core Management and CPU Performance

#221

Post by floatboth »

weatherlight wrote: post #194
floatboth wrote: post #149

Earlier quoted context omitted.

Relatively obscure fun thing: CRC32 checksum instructions. https://github.com/srijs/rust-crc32fast/pull/6 (Get rust nightly via https://rustup.rs , pull the repo, run: cargo bench --features nightly)

(16-inch, 2019) - 2.6 GHz 6-Core Intel Core i7 - 16 GB 2667 MHz DDR4 test result: ok. 0 passed; 0 failed; 4 ignored; 0 measured; 0 filtered out; finished in 0.00s Running unittests (target/release/deps/bench-b9bcbb53a5ef2c28) running 4 tests test bench_kilobyte_baseline ... bench: 310 ns/iter (+/- 70) = 3303 MB/s test bench_kilobyte_specialized ... bench: 48 ns/iter (+/- 5) = 21333 MB/s test bench_megabyte_baseline .…

Thanks! Interesting result, baseline performance is impressive (pretty close to my R9 5950X which does 4853/5069 MB/s) but the CRC32 hardware is surprisingly mediocre (Cortex-A72 on Graviton1 does 14840/14483, Neoverse N1 on Graviton2 does 17066/19560)

Re: M1 Pro First Impressions: Core Management and CPU Performance

#223

Post by fold3 »

zenron wrote: post #213
blitzar wrote: post #92

Earlier quoted context omitted.

Out of curiosity which laptop should I be looking at if I dont want to be robbed by the apple on a M1?

I just got a Lenovo Blue Phantom w/ Ryzen 5800H and 3050TI for 999 USD. 3.2ghz 8c/16t Ryzen 3. Feels amazing even with Win11 on it and I am going from a Dell G3 i5 9300H with 1660TI QMax and its a very tangible feeling upgrade. (even if the OS is a downgrade lol). I did have to do 40 USD memory upgrade to get it to 16g Dual channel so 1039 USD + tax so there is a hidden extra cost there. Or you could do a 32gb upgrad…

Correct me if I'm wrong but the screen is not nearly as good,1080p vs 3K, worse color spectrum, worse contrast. The fan noise is typical of gaming laptop while under load although M1 pro stays mostly inaudible. I haven't checked but I doubt the battery life can be compared as well.

Re: M1 Pro First Impressions: Core Management and CPU Performance

#224

Post by fold3 »

sydthrowaway wrote: post #18

These laptops are too expensive. What the hell is the markup on these things

I couldn't find a monitor with the same spec for less than $4000. The performance is comparable to high ens oc laptop but with better thermal efficiency. If you works with visual, theses are unbeatable value

Re: M1 Pro First Impressions: Core Management and CPU Performance

#225

Post by jjcon »

aroman wrote: post #38
jjcon wrote: post #32

Earlier quoted context omitted.

I’m still personally seeing quite a bit better performance on my old laptop with a dedicated gpu in Unity 21.2 - but the power draw is much much lower (consequently better battery life) on the Mac side which definitely has its place.

Are you comparing Windows vs Mac? If so I think that might have even more to do with perf than the hardware. I'm sure the overwhelming majority of Unity developers use it on Windows, and so it probably gets more love from Unity as a result.

Thats possible but baking lights etc should come down to raw power unless theyve implemented something oddly

Re: M1 Pro First Impressions: Core Management and CPU Performance

#226

Post by Zababa »

Filligree wrote: post #167
Zababa wrote: post #103

Earlier quoted context omitted.

> With that said, there is a huge M1 hype on HN. I wonder how much of that is due to a "finally, a good laptop" sentiment. I have used a lot of laptops over the years, and each and every one of them have been terrible in lots of ways, especially recently. If M1s "solve" the laptop problem, that would be great.

They pretty much do. The M1 macbook is, unironically, the first laptop I've had that doesn't have any glitches. It's less powerful than a modern desktop, but not by much. More importantly, none of the power management, suspend, wifi, whatever... ever seems to fail.

> More importantly, none of the power management, suspend, wifi, whatever... ever seems to fail.

UI scaling is also a big one. I have a Dell XPS 15 in 2014/2015 with Windows 8.1 and the UI scaling wasn't great.

Re: M1 Pro First Impressions: Core Management and CPU Performance

#227

Post by Zababa »

BackBlast wrote: post #198
Zababa wrote: post #102

Earlier quoted context omitted.

I know that AWS already have their own ARM chips: https://aws.amazon.com/ec2/graviton/ . I don't know about chips enough to understand how large is the gap between AWS and Apple when it comes to designing chips though.

I run arm servers in production. M1 performance core performance is about 3x graviton 2 and 2.5x ampere ultra in our stack. The efficiency core is about 0.6 graviton and a bit less than half ampere. Would love to have reasonably costed m1 instances but the value play for us is ampere.

Thanks for sharing that.

Re: M1 Pro First Impressions: Core Management and CPU Performance

#228

Post by bullen »

brrrrrm wrote: post #209
bullen wrote: post #185

Earlier quoted context omitted.

Is that CPU or GPU? Sounds a bit high when 2Gflops/watt (but I guess at 40W for all cores =2,5Gflops/W it could make sense?!) is the industry standard! Can you turn off the acceleration to get CPU only figures? What is FMA/s? Edit: Ok that is some sort of SIMD/GPU... I need figures for the raw CPU that can also do general purpose calculations and branches! Maybe MIPS is a better "apples to apples" measurement?

Ah, it's actually 100gflops (not 100Gfmas) because the fmla instruction is fused. I just ran it on CPU, single core (no pinning), just using the aarch64 fmla instruction. Using pure BLAS (multithreaded, with the MM ISA extension), I get like 1.2TFlops in fp32 on the whole device. Code for the blas test: https://jott.live/code/blas_test.cc

So a friend that has a M1 did this test: https://github.com/brianolson/flops/blob/master/flops.c

And the 5nm M1 has ~2.5Gflops/W which is not a huge increase compared to the 28nm Pi 4 at 2Gflops/W.

No-moores law in effect. Game Over!

Re: M1 Pro First Impressions: Core Management and CPU Performance

#229

Post by bullen »

ReaLNero wrote: post #202
bullen wrote: post #130

Earlier quoted context omitted.

I'm having a hard time finding mips/flops for the M1 CPU, can you measure one of them?

Could you give me a benchmark in particular? Or maybe this one works: https://github.com/mmperf/mmperf . I'll run it in an hour.

So a friend that has a M1 did this test: https://github.com/brianolson/flops/blob/master/flops.c

And the 5nm M1 has ~2.5Gflops/W which is not a huge increase compared to the 28nm Pi 4 at 2Gflops/W.

No-moores law in effect. Game Over!

Re: M1 Pro First Impressions: Core Management and CPU Performance

#230

Post by asdff »

fouc wrote: post #49
asdff wrote: post #46

Earlier quoted context omitted.

They must be getting their bread by charging $400 for the ram upgrade instead of the traditional $200 apple price for what should be $80 of memory.

> traditional $200 apple price That might've been for when it was only an 8GB increase in RAM.

That's also what it was when it was a 2gb increase in RAM. RAM has gotten a lot cheaper per GB than what Apple continues to charge per GB and thats honestly never not the case. Only now its worse, especially when you have developers trapped in docker centric workflows where they need to have more memory, no way around it, and apple decides to raise prices on their already high markup.
Post reply on HN