Apple M1 CPU & GPU speed is very disappointing

[Appletoni]

The original post was comparing the quad-core, entry level M1 chips to high-end server CPUs with 16 or more cores and declaring this as "disappointing". Some other posters joined in, explaining how "slow" this processor is. This is obviously not a rational position given the fact that M1 performs extremely well in a variety of real world workloads as well as industry standard benchmarks.

In retrospect, it does seem like M1 underperforms in Stockfish relative to expectations. My guess as a programmer with some exposure to low-level optimization is that Stockfish lacks optimizations for the ARM architecture. It is also possible that Stockfish code has inherently low intrusion-level parallelism (it could be the case if Stockfish relies on long chains of interdependent computations), which would be the worst possible match for M1 with it's slow-running very wide cores. If I understood correctly, a Stockfish developer read this thread and promised to investigate. But according to the benchmarks, the quad-core M1 is still way faster than any quad-core Intel or AMD chip and on par with 6-core SMT x86 chips.

The original post was comparing:

M1 CPU = 13000 kn/s

i7 3930k overclocked (from 2011 = 10 years old) = 13000 kn/s

I didn’t know that the i7 3930k is a high-end server CPU with 16 or more cores? and we are still declaring the results as „disappointing“.

 

[leman]

The original post was comparing:

M1 CPU = 13000 kn/s

i7 3930k overclocked (from 2011 = 10 years old) = 13000 kn/s

I didn’t know that the i7 3930k is a high-end server CPU with 16 or more cores? and we are still declaring the results as „disappointing“.

The i7 3930k is an enthusiast-class workstation CPU when it was released. Its power consumption is over 130 watts. It is still faster than many new laptops sold today.

That's what 10 years of progress give you: either same performance at dramatically reduced power usage or 2-3 better performance at the same power usage. If you find it dissatisfactory, you might need to reevaluate your expectations.

Besides, the logic in your argument is very questionable to begin with. A modern AMD 5800U is also slower than a 10-year old Xeon. Does it mean that AMD makes bad CPUs? Of course it doesn't.

 

[Appletoni]

all this for....chess...

Yes and we are also disappointed to see that Apple recognized to late after reading this discussion how slow the cpu is for running the Stockfish chess engine and they decided not to release the new MacBook Pro with also a to slow M1X chip at WWDC21.??

 

[leman]

They said it’s well optimized for ARM Neon: https://groups.google.com/g/fishcooking

Look friend, this is obviously very important to you and I would like to help. Here is what we can do. For a modest consultant fee of $20'000 I am prepared to look at the Stockfish code and submit a series of patches that will improve its performance on M1. If I can't do it by the end of summer, I give you the money back. What do you think?

 

[Appletoni]

Someone asked about Homebrew, see:

stockfish

Homebrew’s package index

formulae.brew.sh

 

[JMacHack]

Yes and we are also disappointed to see that Apple recognized to late after reading this discussion how slow the cpu is for running the Stockfish chess engine and they decided not to release the new MacBook Pro with also a to slow M1X chip at WWDC21.??

Yes, Apple saw the overwhelming praise of the M1 and anticipation of the M1x and said “hold up, this one chess benchmark that nobody’s ever heard of is bad, let’s delay the release.”

 

[leman]

Yes, Apple saw the overwhelming praise of the M1 and anticipation of the M1x and said “hold up, this one chess benchmark that nobody’s ever heard of is bad, let’s delay the release.”

Not delay, no. Cancel. I think I’ve heard they plan to move the next MacBook Air to a 10 year old Intel CPU. After all, it’s just as fast as M1, so why bother?

 

[johnhackworth]

Not sure that I care about the M1's ability to play chess - I regularly got spanked at Chess by the 1Kb version of ZX Chess on the Sinclair ZX81.

 

[dmccloud]

One of the cheapest Intel i5-11400 11 Gen CPU easily outperforms M1 chip .Here is my benchmark ; this same in Windows or MacOS. About 40% faster than M1 chip. Cinebench R23 benchmark.

Just stop comparing apples to oranges. Neither you nor Mi7chy seem to grasp that you're not comparing equivalent processors from Apple and AMD, but that also fits into your collective desire to bash Apple at every turn. Comparing the M1 to my Windows machine (9th gen i7 with 6 cores and HT), the Mac blows the Intel system out of the water on any CPU-based benchmark. The Windows machine does win on GPU-intensive benchmarks because it has a 1660Ti videocard, but on the CPU side it isn't even close. Here's the testing I did: https://forums.macrumors.com/thread...chmarks-comparisons-to-i7-gaming-rig.2273847/

Even if You use ONLY safari webrowser all cores are active ! Look at CPU history. I cant find any my apps use only one core cpu.

That's not even close to true. I have Firefox running on my M1 right now (with multiple tabs open), and I am not showing more than 2-3 cores at a time actually doing work. The rest are idling.

 

[Colstan]

For a modest consultant fee of $20'000 I am prepared to look at the Stockfish code and submit a series of patches that will improve its performance on M1. If I can't do it by the end of summer, I give you the money back.

I think I found a way to compile Stockfish in a much cheaper and efficient manner.

 

[mi7chy]

I want to see 3090 graphics performance on my MBA M1 that some are touting but are there only 3 native games? Anything more recent than 2 decade old WoW like maybe Warzone, Cyberpunk, BF6, etc? WoW runs on a potato anyway. On my old PC it gets between 250fps and 350fps (looking towards sky) with maxed out level 10 graphics settings.

https://www.macgamerhq.com/apple-m1/native-mac-m1-games/

 

[JMacHack]

I want to see 3090 graphics performance on my MBA M1 that some are touting

not a single soul has ever said that.

 

[thedocbwarren]

<big sigh /> Feels like this topic has jumped track and off into the absurd. unsubscribe...

 

[mi7chy]

not a single soul has ever said that.

Yes they have along with 8GB on M1 == 16GB on other architectures and other stuff. Might as well see it for myself since my MBA M1 is sitting idle.

 

[leman]

<big sigh /> Feels like this topic has jumped track and off into the absurd. unsubscribe...

Yeah, I agree. It was fun for a while, but some people seem to have completely lost any and all touch with reality over the last couple of pages. I’m calling it quits too.

 

[jeanlain]

Yes they have

Source?

 

[JMacHack]

Source?

Source: his ass.

 

[Jorbanead]

Yes they have along with 8GB on M1 == 16GB

People said this when it first came out but this has relatively stopped. Is this your favorite thing to comment on every single thread? Not a single person in this thread has said this, and I haven’t seen anyone recently make these claims.

Do you have a source on people who claim their M1 gets 3090 performance?

 

[Fawkesguyy]

People said this when it first came out but this has relatively stopped. Is this your favorite thing to comment on every single thread? Not a single person in this thread has said this, and I haven’t seen anyone recently make these claims.

Do you have a source on people who claim their M1 gets 3090 performance?

See post #367

 

[Appletoni]

It's hilarious that so many people are expending so much time with all this metaphorical (ahem) measuring contest, they all seem to forget the fact that this is a v1 release.

This is Apple's first foray into this market and, by all accounts, they've done exceptionally well.

Most people with M1s will probably not give a flying fing over benchmarks: all they care about is real world usage. And if it succeeds there with all the other benefits the new M1 gives, then it's a win.

You can get into "my benchmark's better than your benchmark" all day long - it ultimatly means nothing of the overall result is success.

It's all rather funny really. Don't y'all have better things to do then see who's got the biggest whatever?

The overwhelming number of people buy devices without giving a tinkers pot over benchmark scores, and I've yet to see a benchmark score that has sold me on anything.

Sure the true single core (when one takes SMT & saturation into account) shows many other CPUs are faster than the M1.

Again, who cares? Really?

I‘m not interested in single core performance, because it’s always (fast) enough.
Only multicore is important, because more and more programs starts to support more and more cores and the number of programs which are running only on one core get smaller and smaller.
The distance between Stockfish on M1 and Stockfish on Threadripper is much higher than the little distance in this benchmark:

Cinebench R23 (Multi-Core)​

Cinebench R23 is the successor of Cinebench R20 and is also based on the Cinema 4 Suite. Cinema 4 is a worldwide used software to create 3D forms. The multi-core test involves all CPU cores and taks a big advantage of hyperthreading.

	Apple M1 8x 3.20 GHz	7760 (10%)
	AMD Ryzen Threadripper 3990X 64x 2.90 GHz (4.30 GHz) HT	74422 (100%)

 

[Appletoni]

One core benchmark doesn't really matter. Only few apps use only one core CPU.
We don't usually need powerful processors for these apps. I need only 250W power supply
for this CPU. So case is larger than mac mini but still small.

Some people also said that it’s the fault of Stockfish developers that Stockfish is slow on M1.
Yes yes and they will also say it’s the fault of all these developers, while I’m looking at these 437 Chess Engines running on CPU:

CCRL 40/15 - Pure list

 

[AdamNC]

I‘m not interested in single core performance, because it’s always (fast) enough.
Only multicore is important, because more and more programs starts to support more and more cores and the number of programs which are running only on one core get smaller and smaller.
The distance between Stockfish on M1 and Stockfish on Threadripper is much higher than the little distance in this benchmark:

Cinebench R23 (Multi-Core)​

Cinebench R23 is the successor of Cinebench R20 and is also based on the Cinema 4 Suite. Cinema 4 is a worldwide used software to create 3D forms. The multi-core test involves all CPU cores and taks a big advantage of hyperthreading.

	Apple M1 8x 3.20 GHz	7760 (10%)
	AMD Ryzen Threadripper 3990X 64x 2.90 GHz (4.30 GHz) HT	74422 (100%)

Really? Can you just go play chess and stop "trolling?"

 

[Jorbanead]

I‘m not interested in single core performance, because it’s always (fast) enough.
Only multicore is important, because more and more programs starts to support more and more cores and the number of programs which are running only on one core get smaller and smaller.
The distance between Stockfish on M1 and Stockfish on Threadripper is much higher than the little distance in this benchmark:

Cinebench R23 (Multi-Core)​

Cinebench R23 is the successor of Cinebench R20 and is also based on the Cinema 4 Suite. Cinema 4 is a worldwide used software to create 3D forms. The multi-core test involves all CPU cores and taks a big advantage of hyperthreading.

	Apple M1 8x 3.20 GHz	7760 (10%)
	AMD Ryzen Threadripper 3990X 64x 2.90 GHz (4.30 GHz) HT	74422 (100%)

Yeah I keep seeing people say this about single thread speeds, but it very much does matter.

As an audio engineer, single thread speeds greatly effect DAW workflows, as only a single track can be processed by a single thread. Higher single core speeds means more plugins, more processing on one track. Thus we can use more demanding plugins without needing to compromise our process.

That’s just one example. But for me and thousands of other audio engineers, having great single thread speed is super important.

 

[Appletoni]

Yeah I keep seeing people say this about single thread speeds, but it very much does matter.

As an audio engineer, single thread speeds greatly effect DAW workflows, as only a single track can be processed by a single thread. Higher single core speeds means more plugins, more processing on one track. Thus we can use more demanding plugins without needing to compromise our process.

That’s just one example. But for me and thousands of other audio engineers, having great single thread speed is super important.

The Apple-fanboys would answer: Your developers are bad and you need better developers.

 

[AgentMcGeek]

What’s the point exactly of comparing a 10W mobile SoC with a 280W workstation CPU?

M1 is actually doing pretty good. 10% of the perf for 3% of the power consumption is an impressive ratio. Imagine if you scale that in a Mac Pro.