Question x86 and ARM architectures comparison thread.

[poke01]

Exactly, but all they are comparing is DC MT which is irrelevant to individual core performance for architecture comparisons

thats not the point @OneEng2 raised, I made MT comparisons because he told me that a Zen5 core with SMT will beat the M4 in MT.

I showed him with heavy MT tests thats not case, Zen5 even with 32 threads it does poorly against a 16 thread M4. Now for things lke AVX-512, AMD will beat M4 by a large magin, but hey he said "any highly threaded app a wipe out in favor of Zen 5 IMO."

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.

 

[poke01]

And no one is comparing N3E Zen 5,which does exist in compromised form, to M4 either.

because N3E Zen5 is neither in a laptop/desktop nor is M4 P cores in a DC config.

 

[gdansk]

because N3E Zen5 is neither in a laptop/desktop nor is M4 P cores in a DC config.

And yet we have

AMD arguably looks better in client than they do in server in comparison to Apple silicon, from the CPU side.

I am not arguing against strawmen.

 

[MS_AT]

That David Huang SPEC chart is all very much correct.

You mean this one?

 

[Schmide]

The tests I referenced are all CPU blender tests.


What? It’s not the same power level. M4 Max is using 60 watts meanwhile the rest of the Zen5 mobile processors are above or at 80 watts.

The initial point you made was that Zen5 can beat the M4 with SMT by a large margin. Uh, no it can’t unless you feed ton of power but then the x86 cores lose laughably in perf/w.
The node isn’t the deciding factor here, it’s the microarchitecture being a LOT better.

It’s funny isn’t when AMD beats Intel using a older node, it goes “wow, AMD is beating Intel using an older node”, but when other companies have better microarchitectures on a better node and mind you N3E isn’t all that different from N3B, the argument becomes, “well AMD is on a an older node”.

I don’t need to, a Zen5 core is losing pathetically in a laptop environment.

There are no M4 DC products because Apple doesn’t care for that market just as there are no Zen5 powered smartphones because AMD doesn’t care. But where these architectures converge is laptop so we can do some comparisons here.


Because you can. Why can’t you compare client products to Apple's N3E but can for Intel's N3B?



Sure go ahead, but Apple will be on M6 around the time Zen6 is released.

No, this doesn't make ARM better than x86. It makes Apple's microarch M4 P core better than AMD's Zen5 core.

Yet you refuse to acknowledge the class of processor the M4 max is? It's like $3k-5k. It has a 546GB/s bandwidth. They basically took a workstation class processor, downclocked and optimized its power. It has its niche, but it's not in the same class as zen5.

The closest comparison would be strix halo to a M4 Pro. Otherwise you might as well start comparing to the lower core count threadripper wrx90 processors

 

[igor_kavinski]

Apple fangirls

Oh man, that's COLD

 

[poke01]

You mean this one?
View attachment 128126

Yep now account for perf/w.

 

[poke01]

At the end of the day no one else in the ARM world is doing anything close to Apple yet either. These threads always become useless because it's just Apple fangirls running in circles about how good their architecture is when made on always the most bleeding edge process and extrapolating that to ARM in general.

lol, let’s not pretend there is no short of AMD fanatics here.

Also I clearly mean the microarchitecture. Qualcomm will come close or even beat Apple this summer.

 

[igor_kavinski]

Qualcomm will come close or even beat Apple this summer.

That would be nice if it happens. At least, something to run Windows/Linux on and QC at least doing some Linux enablement at a snail's pace.

 

[gdansk]

lol, let’s not pretend there is no short of AMD fanatics here.

Also I clearly mean the microarchitecture. Qualcomm will come close or even beat Apple this summer.

It's hard to compare microarchitecture when AMD is behind on process by their own profiteering choice. But people will always do it anyway because that's what we can.

All I ask is that the ARM fans try to understand that AMD has more or less slapped all ARM server attempts back to the safety of the hypervisors big enough to exploit ARM's discount IP. And why that might be. One contributing factor is that Apple is the exception to ARM implementations so far. Maybe Nuvia/Qualcomm can also pull it off but they haven't yet. It seems just as much a matter of time as it did five years ago, which is odd for something inevitable.

 

[poke01]

Yet you refuse to acknowledge the class of processor the M4 max is? It's like $3k-5k. It has a 546GB/s bandwidth. They basically took a workstation class processor, downclocked and optimized its power. It has its niche, but it's not in the same class as zen5.

The closest comparison would be strix halo to a M4 Pro. Otherwise you might as well start comparing to the lower core count threadripper wrx90 processors

I DID acknowledge it in post #98. The 12 core/12 thread (8P + 4E) M4 Pro is 64% slower than 16/32 thread strix halo. But the 12 core version uses less than 40 watts.

 

[poke01]

It's hard to compare microarchitecture when AMD is behind on process by their own profiteering choice. But people will always do it anyway because that's what we can.

but its not hard to compare Intel uarch is it? This strawman argument makes no sense.

 

[gdansk]

but its not hard to compare Intel uarch is it? This strawman argument makes no sense.

No, because they are ahead in process yet behind in result. That's categorical.
AMD is behind in process and behind in result. That is not categorical. One might almost call it expected.

 

[gdansk]

You mean this one?
View attachment 128126

Unfortunately the highest AMD shipped that CCD is 5.5GHz which with extrapolation would be only about about 13.3.

 

[poke01]

No, because they are ahead in process yet behind in result. That's categorical.
AMD is behind in process and behind in result. That is not categorical. One might almost call it expected.

nope, we still compared Zen4 to Raptor Lake. AMD was ahead on node and result

 

[gdansk]

nope, we still compared Zen4 to Raptor Lake. AMD was ahead on node and result

I'm not seeing the similarity. No one screamed that x86 was doomed because Raptor Lake sucked compared to Zen 4. Or that it was the inevitable consequence of AMD having more money than Intel. You think I'm defending AMD, but I am not. I am pretty sure I was vocal about how screwed they are in mobile, they won't have an M4 competitor until 2027. And by then one can assume Apple will be shipping much better chips.

 

[adroc_thurston]

they won't have an M4 competitor until 2027.

Ever?
AMD does not make tablet parts anymore.

 

[gdansk]

Ever?
AMD does not make tablet parts anymore.

Sorry, I mean M4 family. M4 Pro/Max level of performance in 1T and probably more in MT. AMD won't have that until 2027.

 

[poke01]

I'm not seeing the similarity. No one screamed that x86 was doomed because Raptor Lake sucked compared to Zen 4. Or that it was the inevitable consequence of AMD having more money than Intel.

Where did I say x86 was doomed or rather did anyone else say in this thread? You brought it up here. All I said was Apple has the better uArch, is it a crime to say it?

Or that it was the inevitable consequence of AMD having more money than Intel.

They can't because that would put AMD in a bad light. Why would they? But for Apple they have to come up with excuses

 

[gdansk]

. All I said was Apple has the better uArch, is it a crime to say it?

No, but it's questionable. In the same area as N3E Zen 5, how many M4 P cores do you fit? Do you know the answer? One would need to know this in order to state that.

They can't because that would AMD in a bad light. Why would they? But for Apple they have to come up with excuses

It's also absurd because still AMD doesn't have more money than Intel.

 

[adroc_thurston]

AMD won't have that until 2027.

2027?

 

[OneEng2]

The initial point you made was that Zen5 can beat the M4 with SMT by a large margin. Uh, no it can’t unless you feed ton of power but then the x86 cores lose laughably in perf/w.
The node isn’t the deciding factor here, it’s the microarchitecture being a LOT better.

... and I never stated within a power envelop. Zen 5 can pound M4 in MT by a large margin .... even in Laptop .... but ESPECIALLY everywhere else where Zen 5's infrastructure is simply light years ahead of M4.

I'll give it to you that M4 gets better battery life, but once you get "all day" battery life in my book, everything else is just slightly incrementally nicer.

I would rather take 13 hr battery life with the ability to have very high performance over 32 hour battery life and much less performance. I am not alone.

This becomes especially true when most laptops running this kind of software are usually just plugged in anyway.

It’s funny isn’t when AMD beats Intel using a older node, it goes “wow, AMD is beating Intel using an older node”, but when other companies have better microarchitectures on a better node and mind you N3E isn’t all that different from N3B, the argument becomes, “well AMD is on a an older node”.

No, I made EXACTLY this argument when 285K using N3B was bested by Zen 5 on N4P. N3E is actually LESS impressive than N3B if you want to add insult to injury to poor limping Intel.

But where these architectures converge is laptop so we can do some comparisons here.

By all means. Compare away! I agree... BUT if we are evaluating an ARCHITECTURE, it seems like a very THIN argument to say .... "M4 is good at laptop therefore it is fundamentally better everywhere" .... and that is even more true since my arguement is that M4 isn't even fundamentally better than Zen 5 in laptop! It is better at some things, but not most things. Not sure how that makes it "better". Furthermore, it is more expensive AND has a process advantage.

It's awful hard for me to conclude that M4 has a better architecture than Zen 5 with any of this evidence.

Because you can. Why can’t you compare client products to Apple's N3E but can for Intel's N3B?

We all did. It is shameful that Intel can't best Zen 5 with a full die shrink and ~25% higher transistor budget. I don't think anyone here pulled their punches when ARL was released.

Exactly, but all they are comparing is DC MT which is irrelevant to individual core performance for architecture comparisons

I think it is quite fair to determine the breadth of applicability of an architecture when having a discussion on what architecture is "best". In my book, an architecture that is ONLY "best" for a wrist watch can't be compared with an architecture that works "really well" for a bunch of markets .... many of which are much higher margin markets than the wrist watch market.

You mean this one?
View attachment 128126

So Zen 5 is equaling the more expensive, M4 that is built on a superior process node.

Again, I am having a hard time with the argument that M4 has a superior architecture based on the evidence in this thread.

No, because they are ahead in process yet behind in result. That's categorical.
AMD is behind in process and behind in result. That is not categorical. One might almost call it expected.

Absolutely!

 

[gdansk]

2027?

Q1. Or cancelled. It isn't clear at this time.

 

[MS_AT]

Yep now account for perf/w.

That is problematic, I don't know what power the CPUs consumed in his tests, he did not put it on the sldies. Then I don't know how to account for various platform differences that also influence the outcome.

Unfortunately the highest AMD shipped that CCD is 5.5GHz which with extrapolation would be only about about 13.3.

There is also memory BW and latency difference to account M4 will have much better BW, but higher latency. We could nitpick and also say that two or three odd subtests would have benefited from AVX512, while M4 doesn't have that problem as Neon will be enabled.

Also if one translates the comments on his spec page:

• Apple Silicon devices use macOS to test large cores and Debian Linux ARM64 (Asahi Linux kernel) to test small cores. Under macOS, although you can use taskpolicy to test the test on the small core, it is suspected that the performance cannot be played properly because of the energy-saving mechanism, and in the end, only about half of the performance of Linux.
• About the performance of macOS: Due to the difference in the operating environment (especially macOS libc/malloc), various types of processors, including x86_64/ARM64, run 523.xalancbmk on macOS, have significant advantages over the default Linux/glibcc, and other subkeys are mutually beneficial. In the end, macOS will lead Linux by about 3%-4%, such as the data of the MacOS and Linux test M2 Pro provided

 

[poke01]

That is problematic, I don't know what power the CPUs consumed in his tests, he did not put it on the sldies. Then I don't know how to account for various platform differences that also influence the outcome.

good, then why did you bring it up?