Apple M1 CPU & GPU speed is very disappointing

[Leifi]

Given a massive number of unknowns, I'd think that anyone with the skills and experience in software performance optimization wouldn't make an absolute statements.

You would have a point if there was a small advantage between lets say a 5800 and a M1...

But it is like 200%-300% performance diff... The performace diff i HUGE.. Bigger than what you could "solve" by tweaking pipelining SIMD etc..

Please Keep in mind for reference, that the diff in these bencmarks is larger than ANY benchmarked diff between M1 and the latest M1 Max!! (only 40-70% jump) and those have even dubbled the high performance cores of the M1!!

Would you honestly argue that the M1 could be "optimized" to beat the M1 Max as well.. Of course not. this 4-copre (full speed) CPU cannot beat neither Apples 8-core versions, nor AMDs 8-core versions.. Its NOT a "coding" issue. lets get real.

 

[pshufd]

You would have a point if there was a small advantage between lets say a 5800 and a M1...

But it is like 200%-300& performance diff... The performace diff i HUGE.. Biggar than what you could "solve" by tweaking pipelining SIMD etc..

Please Keep in mind for reference, that the diff in these bencmarks is larger than ANY benchmarked diff between M1 and the latest M1 Max!! (only 40-70% jump) and those have even dubbled the high performance cores of the M1!!

Would you honestly argue that the M1 could be "optimized" to beat the M1 Max as well.. Of course not.

Again, you don't know and I don't know.

Do you do software performance optimization on large software products? I've been doing that since the 1980s.

I came up with a process to improve Think or Swim performance on M1 by two-thirds. That is it now takes 33% of the time to startup and run and far less CPU than before. A good compiler can make some pretty impressive performance improvements if they have the proper optimizations.

A good example would be a loop that adds the first 100 million integers. Compiling with no optimization would result in generated code that looped through the digits and added them to a register. Good optimization would simply compute the end result of the loop and store the result as data in the program. So a million to one times performance improvement.

 

[leman]

So basically you even get a higher bench for your old MacBook, taking advantage of AVX2 instruction-set, than what people get from an M1 with NEON tweaks :-(

Which again strongly suggests that Stockbench Neon code is suboptimal.

The "benches" where M1 shines, mostly seem to be when running highly tweaked M1 specific code compared to generic, older un. optimized code for Intel CPUs (most commonly single-threaded type workloads).

Yes, "sightly tweaked M1 specific code", like you know, the SPEC benchmark suite, or Java, STAN, or browsers...

About cinebench r23:

Ah, right cibenench, a test where we know for sure that M1 is running a suboptimal path (it uses an Intel library optimized for x86 CPUs with a set of macros that translate SSE intrinsics to NEON). A beautiful example, really. Makes for an interesting story too. So far we have M1 underperform in 1) a benchmark that we know is flawed and b) a chess engine where apparently none of the devs have access to an M1 machine.

But it is like 200%-300& performance diff... The performace diff i HUGE.. Biggar than what you could "solve" by tweaking pipelining SIMD etc..

Which is exactly the point. We know for certain that M1 outperforms Zen3 in a wide variety of tests. So if I you show me a single benchmark where M1 is this much slower, there are two options: 1) this test is using a feature that is slow on M1 or 2) this test does not properly using the M1 hardware. Since we know that M1 is great at SIMD, option 2 is more likely.

Anyway, I am getting bored, so I think I will bid you all nice folks a farewell. Give me a call if there are any new interesting developments.

 

[Taz Mangus]

I don't. I just note that M1 is more expensive and a snail on demanding workloads compared to most cheap modern gaming laptops these days..

I have no desire to waste my time on the M1 if there is no tangible evidence that it can perform even closely on par with its asking pricepoint.

I guess in your world where one obscure chess benchmark is the only benchmark that counts, but in the real world where people actually do real workloads for a living, the M1 Pro/Max handle those workloads one par with or better than CPUs that use far more power, and not requiring that the 14" or 16" laptop be plugged into the power outlet to achieve maximum possible performance.

Rendering on the Apple M1 Max Chip

The home of Code & Visuals

blog.yiningkarlli.com

MacBook Pro m1 max for photographers — Travel Photographer - Austin Mann

austinmann.com

Apple M1 Max Catches up to RX 6800M, RTX 3080 Mobile in GFXBench 5

The M1 Max is flexing its graphics muscle in new benchmarks.

www.tomshardware.com

Tests Showed Premiere Pro Almost 80% Faster on M1 Compared to Intel - Y.M.Cinema Magazine

A comprehensive benchmark showed that Premiere Pro was almost 80% faster on M1 than on Intel hardware. The parameters checked were: Importing a variety of video formats in different frame rates; encoding video to commonly used codecs (such as H265 or ProRes), Effects performance, and finally...

ymcinema.com

Twitter Buys Fully-Loaded M1 Max MacBook Pros for App Engineers • iPhone in Canada Blog

"We're seeing improvements in both top line performance and thermal throttling."

www.iphoneincanada.ca

 

[pshufd]

I guess in your world where one obscure chess benchmark is the only benchmark that counts, but in the real world where people actually do real workloads for a living, the M1 Pro/Max handle those workloads one par with or better than CPUs that use far more power, and not requiring that the 14" or 16" laptop be plugged into the power outlet to achieve maximum possible performance.

Rendering on the Apple M1 Max Chip

The home of Code & Visuals

blog.yiningkarlli.com

MacBook Pro m1 max for photographers — Travel Photographer - Austin Mann

austinmann.com

Apple M1 Max Catches up to RX 6800M, RTX 3080 Mobile in GFXBench 5

The M1 Max is flexing its graphics muscle in new benchmarks.

www.tomshardware.com

Tests Showed Premiere Pro Almost 80% Faster on M1 Compared to Intel - Y.M.Cinema Magazine

A comprehensive benchmark showed that Premiere Pro was almost 80% faster on M1 than on Intel hardware. The parameters checked were: Importing a variety of video formats in different frame rates; encoding video to commonly used codecs (such as H265 or ProRes), Effects performance, and finally...

ymcinema.com

Twitter Buys Fully-Loaded M1 Max MacBook Pros for App Engineers • iPhone in Canada Blog

"We're seeing improvements in both top line performance and thermal throttling."

www.iphoneincanada.ca

My son is getting the first Apple Silicon system in his oncogenomics lab. He's going to have to wait a few weeks because of the backlog. The cost of a laptop is tiny compared to the labor cost of a scientist or engineer.

 

[Taz Mangus]

You would have a point if there was a small advantage between lets say a 5800 and a M1...

But it is like 200%-300% performance diff... The performace diff i HUGE.. Bigger than what you could "solve" by tweaking pipelining SIMD etc..

Please Keep in mind for reference, that the diff in these bencmarks is larger than ANY benchmarked diff between M1 and the latest M1 Max!! (only 40-70% jump) and those have even dubbled the high performance cores of the M1!!

Would you honestly argue that the M1 could be "optimized" to beat the M1 Max as well.. Of course not. this 4-copre (full speed) CPU cannot beat neither Apples 8-core versions, nor AMDs 8-core versions.. Its NOT a "coding" issue. lets get real.

And you keeping proving the point that the chess benchmark is not optimized for Apple Silicon and that is why it runs so poorly on the Apple Silicon. Yes, let's get real. You think just simply recompiling the code is enough. It requires having knowledge of the target hardware and how best to utilize that in the code. Has anyone at StockFish ever used Apple Silicon and examined the source code to see what needs to be changed in order for the code to take advantage of said hardware? Since you seem to be the expert on this, you tell us as you are the one who needs to prove that the benchmark code is correct for Apple Silicon. Somehow, I think your response will be to deflect away from this question and not answer it.

 

[pshufd]

And you keeping proving the point that the chess benchmark is not optimized for Apple Silicon and that is why it runs so poorly on the Apple Silicon. Yes, let's get real. You think just simply recompiling the code is enough. It requires having knowledge of the target hardware and how best to utilize that in the code. Has anyone at StockFish ever used Apple Silicon and examined the source code to see what needs to be changed in order for the code to take advantage of said hardware? Since you seem to be the expert on this, you tell us as you are the one who needs to prove that the benchmark code is correct for Apple Silicon. Somehow, I think you're response will be to deflect away from this question and not answer it.

Sometimes you can get source code with code profiles and sometimes you can't. It depends on your tools and setup. So when you don't have source code in your code profile reports, you get to read a lot of assembler code. Hopefully you're familiar with the architecture so that you have instructions and their costs and behaviors.

This troll doesn't understand this because it looks like he's never done this professionally. There are so many people that read a little on Wikipedia and think that they are knowledge domain experts. They don't know that they don't know.

 

[ddhhddhh2]

You would have a point if there was a small advantage between lets say a 5800 and a M1...

But it is like 200%-300% performance diff... The performace diff i HUGE.. Bigger than what you could "solve" by tweaking pipelining SIMD etc..

Please Keep in mind for reference, that the diff in these bencmarks is larger than ANY benchmarked diff between M1 and the latest M1 Max!! (only 40-70% jump) and those have even dubbled the high performance cores of the M1!!

Would you honestly argue that the M1 could be "optimized" to beat the M1 Max as well.. Of course not. this 4-copre (full speed) CPU cannot beat neither Apples 8-core versions, nor AMDs 8-core versions.. Its NOT a "coding" issue. lets get real.

Well, I think you have a rich imagination, the scientific attitude is to make a hypothesis and then prove it, obviously you forget to prove it yourself, you even ask others to work and then prove it.

I'm reminded of another example, when Apple first started to promote OS X, adobe launched InDesign to counter Quark, and that year, my boss bought the OS X pre-installed G4.

At the beginning, out of curiosity, I tried InDesign for the first time on a magazine design, and it was a disaster, with its slow speed and many bugs.

But adobe kept trying to optimize InDesign (to prevent us from going back to Quark), and adobe worked diligently, and eventually InDesign was able to run on the G4 in a stable and ideal way. (Although the general opinion is that Quark is still better than InDesign), but if compared with InDesign before optimization, it is almost the difference between heaven and earth, the comparison conditions are the same, the same G4.

Although we can't say for sure that the new G4 is twice, or three times faster than the old G4, but to claim that the optimized software can't be twice or three times faster than the hardware specification is a completely unfounded guess!

Your biggest problem is that you are using a positive answer to declare something that you don't know.

But that's not everything about this discussion, you don't believe after all that optimized chess is better on M1.

You have already made your conclusion -
Because the chess is slow on M1, and you almost think that the game is optimized, M1 is garbage, and that's your conclusion.

Talking is almost meaningless, even debating.

If today, your title is: Chess does not seem to be optimized on M1, has any way to make it faster? As good as the other applications.

If that were the case, then it would be a very meaningful argument, but unfortunately that is not your attitude, and I was wrong, I thought you had missed the return M1, and it turns out you didn't.

Since you didn't, and I believe you should have a computer that performs well enough to run chess, then you should have enough judgment to decide if the M1 is worth buying, so now what? Are you waiting for the results of someone else's work?

 

[pshufd]

Well, I think you have a rich imagination, the scientific attitude is to make a hypothesis and then prove it, obviously you forget to prove it yourself, you even ask others to work and then prove it.

I'm reminded of another example, when Apple first started to promote OS X, adobe launched InDesign to counter Quark, and that year, my boss bought the OS X pre-installed G4.

At the beginning, out of curiosity, I tried InDesign for the first time on a magazine design, and it was a disaster, with its slow speed and many bugs.

But adobe kept trying to optimize InDesign (to prevent us from going back to Quark), and adobe worked diligently, and eventually InDesign was able to run on the G4 in a stable and ideal way. (Although the general opinion is that Quark is still better than InDesign), but if compared with InDesign before optimization, it is almost the difference between heaven and earth, the comparison conditions are the same, the same G4.

Although we can't say for sure that the new G4 is twice, or three times faster than the old G4, but to claim that the optimized software can't be twice or three times faster than the hardware specification is a completely unfounded guess!

Your biggest problem is that you are using a positive answer to declare something that you don't know.

But that's not everything about this discussion, you don't believe after all that optimized chess is better on M1.

You have already made your conclusion -
Because the chess is slow on M1, and you almost think that the game is optimized, M1 is garbage, and that's your conclusion.

Talking is almost meaningless, even debating.

If today, your title is: Chess does not seem to be optimized on M1, has any way to make it faster? As good as the other applications.

If that were the case, then it would be a very meaningful argument, but unfortunately that is not your attitude, and I was wrong, I thought you had missed the return M1, and it turns out you didn't.

Since you didn't, and I believe you should have a computer that performs well enough to run chess, then you should have enough judgment to decide if the M1 is worth buying, so now what? Are you waiting for the results of someone else's work?

The guy claims to have a computer science degrees, is very interested in Chess and knows about architectural and optimization issues. He just doesn't seem to be able to perform.

 

[JMacHack]

The guy claims to have a computer science degrees, is very interested in Chess and knows about architectural and optimization issues. He just doesn't seem to be able to perform.

I’d take Terry Davis’ opinion over this guy.

 

[bcortens]

It is Apple who is ignoring SPEC.. No offical tests published according to SPEC controlled tests.


About cinebench r23:

Mobile CPU	Score
Apple M1	9,569
Apple M1 Pro	12,390
Apple M1 Max	12,402
AMD Ryzen 5800H	12,788
AMD Ryzen 9 5900HS	12,794
Intel Core i9-11950H	12,836
AMD Ryzen 9 5900H	12,892
AMD Ryzen 9 5900HS Creator Edition	13,875
AMD Ryzen 9 5980HX	14,363

Two things:

1. Apple has never submitted to SPEC, in the absence of this we have only the independently run SPEC tests which show excellent M1 performance, do you believe that Anand tech is fudging the tests? If so, how? Evidence of Ananandtech cheating in favour of Apple would be rather explosive.

2. Cinebench is known to be unable to fully utilize the M1 cores and yet contrary to the claim that the M1 is actually very underpowered compared to x86-64 it is instead able to essentially match its direct competitors.

 

[Leifi]

Which is exactly the point. We know for certain that M1 outperforms Zen3 in a wide variety of tests. So if I you show me a single benchmark where M1 is this much slower, there are two options: 1) this test is using a feature that is slow on M1 or 2) this test does not properly using the M1 hardware. Since we know that M1 is great at SIMD, option 2 is more likely.

Anyway, I am getting bored talking to a broken record player, so I think I will bid you all nice folks a farewell. Give me a call if there are any new interesting developments.

If you bother to read this thread you will see examples of benchmarks like the LC0 (GPU bound engine) where M1's GPU part gets even more trounced. M1 soc is more than 58000% slower compared to discrete SoCs.

Many CPU-only benches show similar weakness or even worse for M1).

Just some examples.

Appleseed 2.0 - 233% performance advantage on a 5900..
Asmfish 2018 - 5900 225% faster than M1
Blender 2.92 - 590 220% faster than M1
Chaos Group V-ray 5 +200% for 5900.
Kvazaar - +500% for a 5900 over M1

etc.

Do you require more examples?

 

[robco74]

Oh Apple in their hubris has underestimated the chess community. Who knew that these brave souls would be the David who slew the mighty Goliath. It's only a matter of time before Apple's sinister deception is uncovered. Soon everyone will wake up and realize that Apple has been scamming everyone. All those sheeple praising Apple Silicon must have been paid off.

Thank you for your service. Have a gluten-free cookie and a glass of almond milk.

 

[Leifi]

Evidence of Ananandtech cheating in favour of Apple would be rather explosive.

Don't need to accuse anyone of cheating, but not a very incomplete set of results there.

And I would rather have an offical "real" verified specCPU rating than som "maybe" "honest" tests from a site launced by someone currently employed by Apple (that now is not involved at least officially).

 

[Homy]

1080p?
That’s a joke.
I only run games and movies with at least 4k.

Stockfish is running on cpu.
But look also at the gpu. LC0 is slow.
Ceres is using cpu and gpu and I’m sure it is slow too.

Yeah, I agree. That Alienware with 140W 3070 is really a joke running slower than iGPU.

 

[bcortens]

Don't need to accuse anyone of cheating, but not a very incomplete set of results there.

And I would rather have an offical "real" verified specCPU ratting than som "maybe" "honest" tests from a site launced by someone currently employed by Apple (that now is not involved at least officially).

So lets take this standard then ... every single one of the benchmarks you have provided are not submitted by apple nor are they verified by some standards body... so should we ignore all of yours too?

 

[Leifi]

Oh Apple in their hubris has underestimated the chess community. Who knew that these brave souls would be the David who slew the mighty Goliath. It's only a matter of time before Apple's sinister deception is uncovered. Soon everyone will wake up and realize that Apple has been scamming everyone. All those sheeple praising Apple Silicon must have been paid off.

Thank you for your service. Have a gluten-free cookie and a glass of almond milk.

I think Apple will do just fine with sub par HW.. Just as it has done on the GPU side of laptops for years.. But I frankly don't see the point of your post if you are not just out to troll..

How Many Chess Players Are There in The World? (Finally Solved)

It’s impossible to determine an exact number of people globally that play Chess. However, it is estimated that the number of Chess players is about 800 million globally. No one knows how many people enjoy playing Chess. FIDE (Fédération Internationale des Échecs) proposed this estimate of the...

www.chessjournal.com

 

[Wolff Weber]

If you bother to read this thread you will see examples of benchmarks like the LC0 (GPU bound engine) where M1's GPU part gets even more trounced. M1 soc is more than 58000% slower compared to discrete SoCs.

Many CPU-only benches show similar weakness or even worse for M1).

Just some examples.

Appleseed 2.0 - 233% performance advantage on a 5900..
Asmfish 2018 - 5900 225% faster than M1
Blender 2.92 - 590 220% faster than M1
Chaos Group V-ray 5 +200% for 5900.
Kvazaar - +500% for a 5900 over M1

etc.

Do you require more examples?

I am still enjoing my Macbook Air M1 on the contrary to any x86 hardware I have to use at work with similar software suite. So.. be happy with x86 hardware and let us be happy with our Mxx crap.
BTW first time I heard the rumours about Apple transition I thought that this was horrible idea. Then, just because I was curious of that, nearly year ago I had bought M1 MBP. Today I consider this to be a perfect decision and I find my M1 MBA the best single piece of hardware I ever had.

 

[dgdosen]

You would have a point if there was a small advantage between lets say a 5800 and a M1...

But it is like 200%-300% performance diff... The performace diff i HUGE.. Bigger than what you could "solve" by tweaking pipelining SIMD etc..

Please Keep in mind for reference, that the diff in these bencmarks is larger than ANY benchmarked diff between M1 and the latest M1 Max!! (only 40-70% jump) and those have even dubbled the high performance cores of the M1!!

Would you honestly argue that the M1 could be "optimized" to beat the M1 Max as well.. Of course not. this 4-copre (full speed) CPU cannot beat neither Apples 8-core versions, nor AMDs 8-core versions.. Its NOT a "coding" issue. lets get real.

Just pointing out that looking at the `src/misc.cpp` file in Stockfish, a file holding optimizations for different architectures, has 250 commits to that file in the repo. Over time, there's quite a bit of optimizing, which IMO might make one question whether stockfish is a valid comparison to other benchmarks, or more precisely, to benchmarks that are designed in good faith to not give an advantage to any vendor.

I'm not necessarily discounting it, but this thread looks at areas where stockfish was bottlenecked on an M1. I'm all for anyone with the time and energy to try to make it better. That's certainly better use of time and energy than continuing to debate on this thread

 

[Leifi]

So lets take this standard then ... every single one of the benchmarks you have provided are not submitted by apple nor are they verified by some standards body... so should we ignore all of yours too?

Every single benchmark I have mentioned here is many different users of the hardware averaging the results. Not a single tester. And generally, I put more weight on open-source benches performed by programmers and developers that compile and run the benches rather than some commercial internet bloggers/influencers videos and tech-blogs "results" taking out the laptop and starting some apps to benchmark.

 

[bcortens]

If you bother to read this thread you will see examples of benchmarks like the LC0 (GPU bound engine) where M1's GPU part gets even more trounced. M1 soc is more than 58000% slower compared to discrete SoCs.

Many CPU-only benches show similar weakness or even worse for M1).

Just some examples.

Appleseed 2.0 - 233% performance advantage on a 5900..
Asmfish 2018 - 5900 225% faster than M1
Blender 2.92 - 590 220% faster than M1
Chaos Group V-ray 5 +200% for 5900.
Kvazaar - +500% for a 5900 over M1

etc.

Do you require more examples?

Wait seriously?

Do you mean this Asmfish ? If so you are comparing assembly code, for x87-64 running through rosetta to raw hardware and using that as a baseline for why the M1 hardware is weaker than we all think it is? Do I have that right?

 

[Leifi]

I'm not necessarily discounting it, but this thread looks at areas where stockfish was bottlenecked on an M1. I'm all for anyone with the time and energy to try to make it better. That's certainly better use of time and energy than continuing to debate on this thread

I would be more happy If Apple spent some time trying to improve their CPUs, GPUs, and stop being so d*mn overpriced, proprietary, and un-open .... More honesty and less dishonest sales-pitches, and less uninformative misleading graphs when presenting new stuff.. Thanks.... Please

 

[Leifi]

Wait seriously?

Do you mean this Asmfish ? If so you are comparing assembly code, for x87-64 running through rosetta to raw hardware and using that as a baseline for why the M1 hardware is weaker than we all think it is? Do I have that right?

Asmfish running native ARM assembly for the M1.

 

[dgdosen]

I would be more happy If Apple spent some time trying to improve their CPUs, GPUs, and stop being so d*mn overpriced, proprietary, and un-open .... More honesty and less dishonest sales-pitches, and less unfinformative misleading graphs when presenting new stuff.. Thanks.... Please

Well, we agree that Apple stuff is expensive...

But for all the rest you've written, how is Intel more open? Or how is your argument less sales-pitchy? misleading?

 

[Taz Mangus]

I would be more happy If Apple spent some time trying to improve their CPUs, GPUs, and stop being so d*mn overpriced, proprietary, and un-open .... More honesty and less dishonest sales-pitches, and less unfinformative misleading graphs when presenting new stuff.. Thanks.... Please

So sad that you are spouting wrong information based on a fault benchmark test.