Apple M1 CPU & GPU speed is very disappointing

[pshufd]

Here is a different picture in SPEC2006 and 2017 beating Ryzen 9 5950X, i7-1185G7 and i9-10900K in several tests.

"Apple’s performance is extremely balanced across the board, but what stands out is the excellent 502.gcc_r performance where it takes a considerable leap ahead of the competition, meaning that the new Apple core does extremely well on very complex code and code compiling."

The 2020 Mac Mini Unleashed: Putting Apple Silicon M1 To The Test

www.anandtech.com

There's performance per watt and nobody bothers to compare on this.

The scary things are:

- Apple's one-year cadence which I don't think that AMD or Intel can match
- Apple can optimize any specific application that they care to by analyzing it, figuring out the hotspots, and building specific hardware for the hotspots and then putting in an API to access it. Intel and AMD can't do this alone as they don't own the operating systems.

 

[Homy]

There's performance per watt and nobody bothers to compare on this.

The scary things are:

- Apple's one-year cadence which I don't think that AMD or Intel can match
- Apple can optimize any specific application that they care to by analyzing it, figuring out the hotspots, and building specific hardware for the hotspots and then putting in an API to access it. Intel and AMD can't do this alone as they don't own the operating systems.

About why Intel and AMD can't do like Apple: https://debugger.medium.com/why-is-apples-m1-chip-so-fast-3262b158cba2

"Here we get a big problem with the Intel and AMD business model. Their business models are based on selling general-purpose CPUs, which people just slot onto a large PC motherboard. Thus computer-makers can simply buy motherboards, memory, CPUs, and graphics cards from different vendors and integrate them into one solution.

But we are quickly moving away from that world. In the new SoC world, you don’t assemble physical components from different vendors. Instead, you assemble IP (intellectual property) from different vendors. You buy the design for graphics cards, CPUs, modems, IO controllers, and other things from different vendors and use that to design an SoC in-house. Then you get a foundry to manufacture this.

Now you got a big problem, because neither Intel, AMD, or Nvidia are going to license their intellectual property to Dell or HP for them to make an SoC for their machines.

Sure Intel and AMD may simply begin to sell whole finished SoCs. But what are these to contain? PC-makers may have different ideas of what they should contain. You potentially get a conflict between Intel, AMD, Microsoft, and PC-makers about what sort of specialized chips should be included because these will need software support.

For Apple this is simple. They control the whole widget. They give you, for example, the Core ML library for developers to write machine learning stuff. Whether Core ML runs on Apple’s CPU or the Neural Engine is an implementation detail developers don’t have to care about."

 

[argon2020]

- Apple can optimize any specific application that they care to by analyzing it, figuring out the hotspots, and building specific hardware for the hotspots and then putting in an API to access it. Intel and AMD can't do this alone as they don't own the operating systems.

In theory, yes. In practice, this would take *years* to implement. Even with the chip, OS, and apps all in-house.

 

[pshufd]

In theory, yes. In practice, this would take *years* to implement. Even with the chip, OS, and apps all in-house.

Yet they are actually doing this. Their fast cadence may be helping here.

There are some standard algorithms going back to the 1990s that are widely used by billions of people every day but I don't see specific hardware optimizations for them in x86 CPUs. I'd love to be able to optimize Huffman Decoding in hardware but it's a very difficult nut to crack without specific hardware. And it's been around since the 1950s.

 

[argon2020]

About why Intel and AMD can't do like Apple: https://debugger.medium.com/why-is-apples-m1-chip-so-fast-3262b158cba2

The funny thing is, there is nothing new or novel about Apple's approach here. Those of us who have been in the industry a while will recognize this model from the UNIX workstations of the 1990's. DEC Alpha and Tru64, PA-RISC and HP-UX, SPARC and Solaris, MIPS and SGI Irix, etc. When a single vendor designs and builds the CPU, system architecture, OS, and applications, it opens the door for custom ASICs and extreme performance tuning. The UNIX workstation market was built on exactly this concept. SGI was especially successful in this regard, with their various image and video processing chips.

The Wintel era of the late 90's and early 2000's was built on exactly the opposite - a rejection of the monolithic system design in favor of inexpensive commodity components that could be assembled in infinite variations like Lego bricks. Cost was the ultimate driver of this trend, and the result was a race to the bottom. Anyone remember the silly deals of the time? A low-dollar junk PC for $299 if you sign up for 3 years of dial-up MSN Internet service? Lots of folks fell for that one.

Enter Apple in the 2000+ era, bringing back the concept of monolithic design and premium pricing. Apple was striking deals with Intel for lightly customized versions of their x86 chips that were not available to others. The M1 is really just a continuation of this concept. What's old is new again!

 

[argon2020]

Yet they are actually doing this. Their fast cadence may be helping here.

There are some standard algorithms going back to the 1990s that are widely used by billions of people every day but I don't see specific hardware optimizations for them in x86 CPUs. I'd love to be able to optimize Huffman Decoding in hardware but it's a very difficult nut to crack without specific hardware. And it's been around since the 1950s.

I'm not disagreeing. But I think you will be sorely disappointed if you think there will be an annual cadence of M1 chips with new hardware optimizations, and API's and apps that leverage them. The timeline for this stuff is measured in multiple years, I predict this is true even for Apple.

 

[pshufd]

I'm not disagreeing. But I think you will be sorely disappointed if you think there will be an annual cadence of M1 chips with new hardware optimizations, and API's and apps that leverage them. The timeline for this stuff is measured in multiple years, I predict this is true even for Apple.

That's fine. There are lots of algorithms out there to optimize that have been out there for many years. Intel asked us if we wanted any special instructions many years ago and they did add something for us. It took a while but it's something useful. I don't recall what it was though. I just thought that it was cool that they offered.

I was on a conference call with Intel to look at a problem we had at Mozilla and an Intel engineer came on the call to look at it. That was something else that I was impressed with - that they made engineers available to look at problems. Intel also has their software libraries and ICC that they sell and that's very good performance stuff. So companies can do this. But it would be easier for Apple as they own everything and they don't have to serve all of the different market segments that x86 addresses.

 

[v0n]

Is that your experience? I've just updated my MBP16 from Catalina to Big Sur 11.4 and think it runs noticeably faster...not by much, but it feels a little bit "snappier".

I think the general experience is that Bug Sir is a bit of a 'Vista' of MacOSes? No?

 

[JMacHack]

I think the general experience is that Bug Sir is a bit of a 'Vista' of MacOSes? No?

not in my experience, Catalina was more like Vista, Big Sur is like 7, which is to say “fixed Vista”

 

[pshufd]

I like the Big Sur GUI but my old hardware doesn't. My impression is that it makes a good reason to upgrade your hardware which is why I've been looking at the 2020 iMac 27. I tried it out at the Apple Store with Big Sur and the combination is very snappy in terms of UI responsiveness. Perhaps it's the GPU that really helps out with responsiveness. The M1 has a much higher single-core Geekbench 5 score but I found the 2020 iMac 27 to be more responsive than the M1.

I am really looking forward to Monterey. I like the idea of macOS as an AirPlay target. I will be able to stream video from my phone or iPad to my Macs which is something that I've wanted for a long time.

 

[v0n]

I like the Big Sur GUI but my old hardware doesn't.

Is your hardware 2015/2016?

 

[pshufd]

Is your hardware 2015/2016?

2014/2015.

 

[v0n]

2014/2015.

Try regaining control over your fans. You'll probably find that your 2014/2015 hardware and i5 or i7 is perfectly fine, but SMC controls installed with Bug Sir are literally cooking your machine alive - your MBP might be throttling at 90-100 deg. C. while doing nothing more than browsing the net or watching YouTube video. Get some software to set your own defaults, start with about 2000 rpm up to 60 deg C and see if your old MBP come alive and become its old agile and snappier self with new MacOS.

 

[pshufd]

Try regaining control over your fans. You'll probably find that your 2014/2015 hardware and i5 or i7 is perfectly fine, but SMC controls installed with Bug Sir are literally cooking your machine alive - your MBP might be throttling at 90-100 deg. C. while doing nothing more than browsing the net or watching YouTube video. Get some software to set your own defaults, start with about 2000 rpm up to 60 deg C and see if your old MBP come alive and become it's old agile and snappier self with new MacOS.

I'm perfectly happy using Mojave.

I have a big system for production work that runs really cool.

 

[thedocbwarren]

I think the general experience is that Bug Sir is a bit of a 'Vista' of MacOSes? No?

This kind of thing seemed to be repeated every single release.

 

[Dammit Cubs]

all this for....chess...

 

[Joelist]

And an incorrectly configured chess benchmark no less...

 

[Colstan]

Believe it or not, at one point, the Mac's included Chess program was diagnostically useful. It was the first program to go 64-bit. When I switched out the 32-bit Core Solo in my Mac mini of that era, for a 64-bit Core 2 Duo, I pulled up Chess to make sure that it was functioning and supported 64-bit binaries. I assume Apple used Chess as some sort of early test before compiling the other programs to support 64-bit.

So, Chess has been useful on the Mac in the past. I can't say the same thing for this thread.

 

[Appletoni]

I’m sure you have seen WWDC21 https://www.apple.com/apple-events/...d_124763416682_&mtid=20925fey39950&aosid=p238

What improvements can be used to speed up Stockfish on cpu or LC0 on gpu?

 

[leman]

What improvements can be used to speed up Stockfish on cpu

Profile the software on the target platform, figure out what the inefficiencies are, fix the code.

or LC0 on gpu?

Update your software to actually use the GPU properly. Since LC0 uses machine learning, they should probably look into implementing an ML compute backend if they want their stuff to work well on a Mac.

This has nothing to do with WWDC. Apple can’t magically “fix“ software.

 

[NotTooLate]

Didn't Apple either acqire Exponential outright or bring in some of their talent? I know Srouji came from Intel, so they know how to put together a competent chip design team.

He came from IBM after being at Intel before that.

 

[NotTooLate]

On topic , engines like engines are heavily optimized SW on target HW, be it chess or game engines , when the code is not using the HW in an optimized way and uses it as a general purpose CPU then it is no surprise (unless your the OP) that he lags behind , the matter of fact is that most of the SW is optimized for X86 and Nvidia/AMD GPU`s , even porting a SW application from X86-> ARM wont mean its optimize , it can just mean it wont use Rosetta , but would the engine use the MLkit ? would it use the GPU ? what optimization would it have ?

Having native SW doesnt mean optimized SW , but its better then Rosetta.

 

[Joelist]

Or……run Stockfish on the correct settings. It is mentioned somewhere upthread and the stockfish numbers on the correct setting are rather a lot higher.

 

[JMacHack]

Or……run Stockfish on the correct settings. It is mentioned somewhere upthread and the stockfish numbers on the correct setting are rather a lot higher.

That would defeat the purpose of the troll thread

 

[dmccloud]

Or……run Stockfish on the correct settings. It is mentioned somewhere upthread and the stockfish numbers on the correct setting are rather a lot higher.

They can't do that - it will upset their poorly constructed and logically unsubstantiated agendas!!