Question x86 and ARM architectures comparison thread.

[johnsonwax]

I haven't benched a multicore Excel workload yet (keep forgetting) but using overclock.net's Excel benchmark sheet, the M1 was almost as fast as a 12700K. This sheet is mostly single thread heavy and seeing the M1 crunch through it without making any noise was insane.

I could be confusing Excel with Librecalc though. I just remember that I was blown away by the benchmark sheet giving the final result so quickly.

The real lesson with Excel is that if you are hitting performance bottlenecks with Excel, your problem isn't the architecture you're running on, your problem is that you're using Excel when you shouldn't be.

 

[igor_kavinski]

MP4 and MKV are container formats, not video formats. You are demanding technical explanations when you don't understand what you are talking about.

Those are common formats and they would normally be using some commonly used codec. TV shows aren't distributed over the internet using odd codec choices.

 

[igor_kavinski]

The real lesson with Excel is that if you are hitting performance bottlenecks with Excel, your problem isn't the architecture you're running on, your problem is that you're using Excel when you shouldn't be.

Not true. My 14 core 245KF easily beat the crap out of my workplace's 6248R 48 thread Xeon CPU doing a simple sumif() over a million rows.

 

[johnsonwax]

That's an incredible claim and I have to wonder if it was a browser game

I don't play browser games. You can hit 100% single core CPU and 100% GPU in Minecraft btw without too much effort.

Since I dislike using MacOS for anything other than watching movies (because it's the only usable 13.3 inch laptop I have), I can't think of any proper way of analyzing my laptop's real world battery life. I don't want to forcefully browse or watch Youtube on that laptop just to test the battery life. And even if it delivers on that promise, useless to me since that's not my use case for it.

Why do you need a proper way to analyze it? You can just use the laptop and if the battery lasts long enough, then it's good enough. There's a battery health tool in settings. What's the point of a synthetic benchmark to measure battery life for something you aren't going to do. That's like measuring how fast my car can drive backward on the moon. Who gives a sh*t?

 

[johnsonwax]

Those are common formats and they would normally be using some commonly used codec. TV shows aren't distributed over the internet using odd codec choices.

Of course they are. Again, you're describing containers formats, not video formats. MKV can have, among others, HEVC/H.265, H.264/AVC, H.263, VP3, Theora, VP8, VP9, AV1, MPEG-2, MPEG-4, AVCHD, DivX, XviD. MP4 can have most of those as well. Apple Silicon doesn't have hardware codecs for some of those. For instance AV1 hardware decode wasn't added until M3. Depending on what GPU you have on your Windows machine you might have AV1 hardware decode on it.

 

[johnsonwax]

Not true. My 14 core 245KF easily beat the crap out of my workplace's 6248R 48 thread Xeon CPU doing a simple sumif() over a million rows.

You're missing my point. If you have a million record dataset, don't use Excel. That's the wrong tool. Use R, python, etc.

 

[igor_kavinski]

You can just use the laptop and if the battery lasts long enough, then it's good enough.

Well, it doesn't fare so well with VLC player. And that's my only use case. I personally would like the legendary battery life please since I paid for this laptop with a Geforce 3080 10GB and a 3060 Ti 8GB swap and currently I feel like I should've gotten it for free.

 

[igor_kavinski]

You're missing my point. If you have a million record dataset, don't use Excel. That's the wrong tool. Use R, python, etc.

If I'm paid like a drone, I will work like a drone. No need to activate any extra neurons for my company

 

[poke01]

If I'm paid like a drone, I will work like a drone. No need to activate any extra neurons for my company

100% agree, use what the company gives you. Do not do anything that speeds up your work cause then they give you more work

 

[igor_kavinski]

Do not do anything that speeds up your work cause then they give you more work

Yeah. 100% my experience. And they hire dumbasses just because they were paid more than me in the previous company and then the dumbasses wake me up from sleep and ask me stupid questions which they should've known had they actually done any work in the previous company and then I have to give them a dose of reality but it doesn't change the fact that I'm still poorer than them.

 

[johnsonwax]

This thread reminds me why I had a reputation for firing people.

 

[igor_kavinski]

This thread reminds me why I had a reputation for firing people.

Elaborate please

 

[Doug S]

Finally a thread where I can openly call Apple absolute cowards. Because if they aren't, why don't they support at least one Linux distro officially so we can have a proper comparison with other architectures? They are sitting on a ton of cash. Surely they have enthusiast engineers who wouldn't mind doing open source in their free time.

What's wrong with Asahi? Yes it doesn't fully support every bit of hardware on ARM Macs, or full GPU acceleration, but it is more than good enough for running stuff on the command line if you're just looking for "proper comparison" (though I don't think you understand the meaning of that phrase if you think VLC is a reasonable app to use for comparison)

I don't get your obsession with constantly being so mad at Apple that they don't sell stuff for the price you think they should charge, or offer the configurations you think you they should offer, or support other operating systems because you wish they would.

Did you grow up filthy rich or something? Only someone with a seriously spoiled upbringing could believe the entire world should bend to their will as much as you seem to.

If anyone thinks that Apple computers will become as popular as iPhones, keep dreaming.

Who has ever suggested that? Even Apple doesn't suggest it. Heck, I remember Steve Jobs suggesting at launch that the iPhone might take 1% of the cellular market within a couple years and people laughed and laughed. If he was still alive today even he would admit it greatly exceed his wildest expectations. They just sold their 3 billionth iPhone. I don't know how many Macs Apple has sold since the 80s, but I'd be surprised if it is even 300 million. There's no way it could ever match the iPhone's success, even if Microsoft announced tomorrow that they were discontinuing Windows at the end of the month.

 

[Doug S]

M4 is on N3E and is about 80mm2 and has 10 cores (and a bunch of other stuff). Zen 5c in DC is 73mm2 and has 16 cores.

Zen 5 pretty much wipes the floor with M4 in highly threaded workloads. Of course, M4 is the performance per power king as its roots are from this standpoint.

Wow you compare something with 10 cores (many of them E cores) with something that has 16 cores and we're supposed to be impressed that the one with 16 cores beats the one with half as many P equivalent cores? That "bunch of other stuff" is doing a LOT of heavy lifting in your comparison. Zen 5c is useless without a "chipset", which the M4 doesn't need. So if you want to do a 10 core vs 16 core comparison I say let's do just the 80 mm^2 chip versus the 73 mm^2 chip comparison next. I predict Zen 5c's performance will be 0, because it can't do anything at all on its own without a whole other chip you decided not to include when comparing mm^2.

 

[igor_kavinski]

What's wrong with Asahi?

Any Asahi M4 benchmark figures you would like to share?

I don't get your obsession with constantly being so mad at Apple that they don't sell stuff for the price you think they should charge, or offer the configurations you think you they should offer, or support other operating systems because you wish they would.

They are operating a monopoly and due to their size, that's unacceptable.

Who has ever suggested that?

Implied, if not outright suggested, by everyone who feels it is their duty to point out that Apple Silicon is the best thing to happen to the world.

 

[Doug S]

Any Asahi M4 benchmark figures you would like to share?

I don't own a Mac. So not only do you want Apple to bend to your will now you want me to bend to your will too? You're gonna have an awfully long wait, son.

 

[igor_kavinski]

Wow you compare something with 10 cores (many of them E cores) with something that has 16 cores and we're supposed to be impressed that the one with 16 cores beats the one with half as many P equivalent cores?

M4 Max has 16 cores. Zen 5 offers 32 threads. Arrow Lake 24 threads.

You know what that tells me? It's very expensive for Apple to scale up the core count.

Sorry. At least with respect to core count scaling, Apple Silicon is currently at a disadvantage and unable to satisfy anyone who needs more cores.

 

[igor_kavinski]

I don't own a Mac. So not only do you want Apple to bend to your will now you want me to bend to your will too? You're gonna have an awfully long wait, son.

I thought maybe you had seen them somewhere.

 

[Doug S]

M4 Max has 16 cores. Zen 5 offers 32 threads. Arrow Lake 24 threads.

You know what that tells me? It's very expensive for Apple to scale up the core count.

Sorry. At least with respect to core count scaling, Apple Silicon is currently at a disadvantage and unable to satisfy anyone who needs more cores.

Its not expensive for APPLE to scale up the core count. It is expensive for Apple customers to scale up the core count. That's not the same thing.

Apple is not trying to compete in the market for "cheapest MT performance". They aren't trying to compete in the market for "cheapest" anything. I know you don't like that, but that's their prerogative to charge what they want. And your prerogative to choose not to buy their products.

 

[johnsonwax]

M4 Max has 16 cores. Zen 5 offers 32 threads. Arrow Lake 24 threads.

You know what that tells me? It's very expensive for Apple to scale up the core count.

Sorry. At least with respect to core count scaling, Apple Silicon is currently at a disadvantage and unable to satisfy anyone who needs more cores.

No, that just tells you Apple doesn't care about that market. I addressed this over in the Apple Silicon thread. Who's the market for needing more cores, and in what universe would they be buying from Apple?

 

[igor_kavinski]

Its not expensive for APPLE to scale up the core count. It is expensive for Apple customers to scale up the core count. That's not the same thing.

I'm not convinced. Why not core spam and offer even higher priced products?

They need to stitch two Max dies to get higher than 16 cores and that's expensive for them. That's why they are forced to offer previous gen Ultra chip. And then they price it so high that people with even great interest are forced to think twice about plopping down so much hard cash.

Apple is good at pretending to offer more value than they actually do. That's their secret.

 

[igor_kavinski]

Who's the market for needing more cores, and in what universe would they be buying from Apple?

Someone or some company that has written a custom Apple Silicon application that scales very well with more cores.

But that's just my guess. You give more cores. More use cases get invented. This is what developers do. Use hardware if it's available.

 

[poke01]

I'm not convinced. Why not core spam and offer even higher priced products?

They need to stitch two Max dies to get higher than 16 cores and that's expensive for them. That's why they are forced to offer previous gen Ultra chip. And then they price it so high that people with even great interest are forced to think twice about plopping down so much hard cash.

Apple's main market and focus in the computer space is laptops, not desktops or servers. They really don't care if the Ultra sells or not. Its there for people who want 512GB of unified memory for LLMs.

M4 Max has 16 cores. Zen 5 offers 32 threads. Arrow Lake 24 threads.

You know what that tells me? It's very expensive for Apple to scale up the core count.

Lets break it down flagship offerings in the laptop space.
Apple offers 12P+4E/16 threads with M4 Max
AMD 16P/32 threads for strix halo. 9955hx3d is also the same but with a higher cTDP
Intel offers 8P+16E/24 threads for 285HX and 6P+8E/14 threads for 285H.

Now lets take look at Blender because a) it multithreaded b) open source so no hidden tricks and c) its optimised for x86 and ARM.

https://www.notebookcheck.net/MSI-Titan-18-HX-AI-A2XWJG-Review-No-holds-barred-Core-Ultra-9-285HX-and-175-W-RTX-5090-Laptop-performance.1010632.0.html

Notebookcheck tested the x86 CPUs with no power limits.



Now for some 9955HX3D Blender 4.4.0 tests and Vray compared at 80 watts




M4 Max at 60 watts. (Yes the when all the P cores are loaded at 3.9GHz, the M4 Max consume a max peak of around 60 watts for the CPU.)

Apple Mac Studio (M4 Max) Review - More Performance, As Usual - Pokde.Net

The M4 Max-powered Apple Mac Studio is just as before on the outside, but it's the inside that get all the new upgrades. Here's how it performs.

pokde.net

Thread count is meaningless when comparing to a CPU that has superior 1t by a large margin but has a decent amount of P cores ie M4 Max. While SMT did help AMD when compared to Intel however when AMD is faced with Apple's M4 P core the lead vastly diminishes at 80 watts and this isn't just because Apple is on N3E, Intel is also on N3B and its not as good. Its because the M4 P core uArch is great.

I'll go indepth regarding these results in a few days, but just read the comparisons for now.

 

[poke01]

Does it matter? Core for core Zen 5 with SMT will outperform M4. Zen 5 also comes in variants having many many more cores than M4 making any highly threaded app a wipe out in favor of Zen 5 IMO.

this was proved wrong, see post #98. Despite having more P cores and double the threads AMD barely has any notable lead with SMT.

AMD and Intel need to feed their cores more power because of the lack of 1t improvements and only then do they beat Apple by a huge margin, but is it really a win if you feed nearly 2 times the power but only get get a small % of improvement.

 

[igor_kavinski]

Good comparison but Apple had to go through a lot more trouble for that performance at 60W whereas AMD is doing it AFTER they got done with their main market, i.e. server. It's pretty close and Apple seems to be at a disadvantage because their design cost a lot more money and man-hours than AMD's.

If it's just these two benchmarks that we use to compare x86 and ARM, yes, Apple has an edge at a much higher cost (design, manufacturing and end-user cost). There's also the question of whether Apple can continue getting IPC gains. Things will be pretty interesting to compare between Zen 6, M5 and Snapdragon Elite X Gen 2.