News Apple M4 benchmarks suggest it is the new single-core performance champ, beating Intel's Core i9-14900KS — incredible results of 3,800+ posted

[TheHerald]

Ok, someones gonna hafta give me an ELI5 on this one. A tablet CPU beating a 14900K? There has to be something I'm missing. Is this one unique cherry-picked benchmark, specifically tailored in favour of the M4, or is this some massive revolution on apples part, and where does the 7950X3D fit in? Can someone tell me whats going on here? I am very confused

GB6 is not the best indicator of performance, they are only testing the ST portion and what they are not telling you is that the m4 core is HUGE. The whole chip is 28b transistors divined into 10 cores, the 14900k is 25b transistors divined into 24 cores. We don't have a die shot yet but with very raw estimates the m4 P core should be 280% larger than the 14900k P core. It should be beating the crap out of the 14900k in both MT and ST, but it doesn't.

 

[TheHerald]

Yes you should. A battery-powered tablet beating a 400w desktop processor.

Does it? This is my 12900k, average power draw through the whole test was 19w, it beats the m4 in MT performance.

"400w", yeah, right

 

[YouFilthyHippo]

GB6 is not the best indicator of performance, they are only testing the ST portion and what they are not telling you is that the m4 core is HUGE. The whole chip is 28b transistors divined into 10 cores, the 14900k is 25b transistors divined into 24 cores. We don't have a die shot yet but with very raw estimates the m4 P core should be 280% larger than the 14900k P core. It should be beating the crap out of the 14900k in both MT and ST, but it doesn't.

Ok. But what I am not understanding is that.... why should a tablet that consumes significantly less power then a desktop CPU destroy the desktop CPU in performance? I thought it would be the other way around. Did apple discover something revolutionary? Or is it a cherry-picked benchmark? Or is it straight up lying? Or what is going on here?

 

[TheHerald]

Ok. But what I am not understanding is that.... why should a tablet that consumes significantly less power then a desktop CPU destroy the desktop CPU in performance? I thought it would be the other way around. Did apple discover something revolutionary? Or is it a cherry-picked benchmark? Or is it straight up lying? Or what is going on here?

Α tablet cpu doesn't destroy anything. The m4, a cpu that has a substantially larger transistor count than my 12900k, loses in the mt gb6 test by over 40%. My 12900k averaged a power draw of 19.1w during the test.

 

[YouFilthyHippo]

Α tablet cpu doesn't destroy anything. The m4, a cpu that has a substantially larger transistor count than my 12900k, loses in the mt gb6 test by over 40%. My 12900k averaged a power draw of 19.1w during the test.

So how is it the "single core champ" as the article says? I am very confused. How can it win in any category against a desktop 14900K..... Well .. besides power efficiency I guess. But in performance, the 14900K should crush it on every bench. Or am I missing something?

 

[TheHerald]

So how is it the "single core champ" as the article says? I am very confused. How can it win in any category against a desktop 14900K..... Well .. besides power efficiency I guess. But in performance, the 14900K should crush it on every bench. Or am I missing something?

Why would a desktop cpu crash it on every bench? The m4 has a much bigger core, so it makes sense that it is faster in single core performance.

A 10 core cpu with a 28b transistor count should be faster than a 24 core cpu with a 25b transistor count in single thread since each core is much bigger.

With that said gb6 ain't the most reliable benchmark

 

[YouFilthyHippo]

Why would a desktop cpu crash it on every bench? The m4 has a much bigger core, so it makes sense that it is faster in single core performance.

A 10 core cpu with a 28b transistor count should be faster than a 24 core cpu with a 25b transistor count in single thread since each core is much bigger.

With that said gb6 ain't the most reliable benchmark

So why is Intel making such an inferior product? Or is apple really 10 years ahead of everyone? Or what's the story here? That's just the base m4, nevermind the ultra. That's even more scary

 

[TheHerald]

So why is Intel making such an inferior product? Or is apple really 10 years ahead of everyone? Or what's the story here? That's just the base m4, nevermind the ultra. That's even more scary

What do you mean inferior? The m4 is a bigger chip than the 14900k yet it gets blasted in performance by 2021s much smaller 12900k. At least in gb6.

If I'm not mistaken, the m4 is 28b transistors, the 12900k is 10.4b, yet the 12900k almost 40% faster in gb6 multithreaded.

 

[YouFilthyHippo]

What do you mean inferior? The m4 is a bigger chip than the 14900k yet it gets blasted in performance by 2021s much smaller 12900k. At least in gb6.

If I'm not mistaken, the m4 is 28b transistors, the 12900k is 10.4b, yet the 12900k almost 40% faster in gb6 multithreaded.

Ok but that's multithreaded. Why is Intel building an inferior single threaded chip in 2024 ? Or is apple just 10 years ahead?

 

[TheHerald]

Ok but that's multithreaded. Why is Intel building an inferior single threaded chip in 2024 ? Or is apple just 10 years ahead?

Let me reverse that, why is apple building an inferior mt chip in 2024? Or is Intel 10 years ahead?

Going by transistor count alone intel is actually light years ahead of apple. Both in st and mt. We don't yet have a die shot but going by the current numbers, the transistor count alone, the m4 should be 280% faster than the 14900k in ST. Yet it's barely 10%. Think about that.

 

[YouFilthyHippo]

Let me reverse that, why is apple building an inferior mt chip in 2024? Or is Intel 10 years ahead?

Going by transistor count alone intel is actually light years ahead of apple. Both in st and mt. We don't yet have a die shot but going by the current numbers, the transistor count alone, the m4 should be 280% faster than the 14900k in ST. Yet it's barely 10%. Think about that.

I guess my next question is.... How does apple have so many more transistors in a much smaller low power chip?

 

[TheHerald]

I guess my next question is.... How does apple have so many more transistors in a much smaller low power chip?

The second part of the question is meaningless. Power draw is an arbitrary number that each company can decide to run their cpus at. The 14900 exists in 3 configurations, at 35w, 65w and 253w. If Intel wanted they could release it also in 50w, 100w, 120w etc. It's not a metric that actually shows anything.

Apple packs more transistors cause it uses tsmcs node which packs more transistors per die size. But that's also meaningless since Intel has cpus much bigger than the 14900k anyways. So apple doesn't have more transistors, their specific cpu (m4) has more transistors than Intel's 14900k. Intel has cpus with 100b+ transistors.

But you didnt answer my question. Is apple 10 years behind intel or why else are they releasing so slow cpus in MT?

 

[YouFilthyHippo]

The second part of the question is meaningless. Power draw is an arbitrary number that each company can decide to run their cpus at. The 14900 exists in 3 configurations, at 35w, 65w and 253w. If Intel wanted they could release it also in 50w, 100w, 120w etc. It's not a metric that actually shows anything.

Apple packs more transistors cause it uses tsmcs node which packs more transistors per die size. But that's also meaningless since Intel has cpus much bigger than the 14900k anyways. So apple doesn't have more transistors, their specific cpu (m4) has more transistors than Intel's 14900k. Intel has cpus with 100b+ transistors.

But you didnt answer my question. Is apple 10 years behind intel or why else are they releasing so slow cpus in MT?

I don't know the answer to that question. I'm the one who's confused trying to figure out what's going on here

 

[TheHerald]

I don't know the answer to that question. I'm the one who's confused trying to figure out what's going on here

Well, let me simplify it for you. M chips are much much much larger than Intel's desktop chips. That's why a Mac with M2 ultra costs a whooping 3.999$. Way more transistors, way more expensive.

For comparison, the M2 ultra has over 110b transistors and it barely beats the 3 year old 12900k in gb6. The 12900k is literally 1/10th the size.

Now with that said the M2 ultra is much more than just a cpu, since it has accelerators and a gpu, but so does the 12900k, since it has avx and a gpu.

So direct comparisons are silly, but if you wanna make them regardless - the ultra variation of the m4 will be a lot slower than any intel or amd cpu that packs the same amount of transistors in pure cpu performance. It really won't even be competitive, a 52 core xeon or the equivalent threadripper will dwarf the m4 ultra in cpu performance.

 

[YouFilthyHippo]

Well, let me simplify it for you. M chips are much much much larger than Intel's desktop chips. That's why a Mac with M2 ultra costs a whooping 3.999$. Way more transistors, way more expensive.

For comparison, the M2 ultra has over 110b transistors and it barely beats the 3 year old 12900k in gb6. The 12900k is literally 1/10th the size.

Now with that said the M2 ultra is much more than just a cpu, since it has accelerators and a gpu, but so does the 12900k, since it has avx and a gpu.

So direct comparisons are silly, but if you wanna make them regardless - the ultra variation of the m4 will be a lot slower than any intel or amd cpu that packs the same amount of transistors in pure cpu performance. It really won't even be competitive, a 52 core xeon or the equivalent threadripper will dwarf the m4 ultra in cpu performance.

So why are Intel and AMD not using bigger CPUs to get more out of em?

 

[TheHerald]

So why are Intel and AMD not using bigger CPUs to get more out of em?

Huh? Both intel and amd do have much bigger cpus,they are called threadripper and xeons.

 

[YouFilthyHippo]

Huh? Both intel and amd do have much bigger cpus,they are called threadripper and xeons.

Right. I see what you mean. So if apple made such a large chip, how do they keep it cool in something thin like a tablet? Yet a smaller chip like the 14900K pretty much needs an AIO

 

[palladin9479]

Α tablet cpu doesn't destroy anything. The m4, a cpu that has a substantially larger transistor count than my 12900k, loses in the mt gb6 test by over 40%. My 12900k averaged a power draw of 19.1w during the test.

It's very ... disingenuous and clickbaity. The idea of "performance" really depends on what you are measuring and how you are measuring it. A cheap older nVidia card will smoke every single modern CPU on vector processing math, by orders of magnitude. Does that mean nVidia discovered some sort of quantum-blockchain-AI-aether technique? No it just means that nVidia processor design was specialized to that type of performance. There are ASIC's we can build that are several orders of magnitude faster in AES/SHA/MD5 processing then anything a general purpose processor can do.

So when evaluating performance we have to keep that in mind. Apple just builds typical ARM CPU's on whatever TSMC's latest node is, and usually pays out the nose for the right to do so. Since single thread processing is about maximizing the clock rate / IPC of a single core, basically allocating the entire TDP budget to a single task. The M4 is on TSMC's 2nd generation 3nm (kinda) node and the Intel raptor lake is based on Intel-7 (7nm), what we are seeing is just the process node advantage. Apple did the same thing with the M1 chip, bought first priority access to TSMC's latest node and shipped a bunch of CPU's before everyone else got access. Once other manufacturers got access to that node, the M1 faded into the distance.

On the whole size thing, it's because Apple makes a SoC design where that single chip is basically an APU, and not that great of one either. But because it's a different ISA, only Apple approved stuff makes it there and there isn't really anything to compare to.

 

[TheHerald]

Right. I see what you mean. So if apple made such a large chip, how do they keep it cool in something thin like a tablet? Yet a smaller chip like the 14900K pretty much needs an AIO

Bigger chips are easier to cool since they have a larger area for dissipation.

The 14900k doesn't need an aio. Doesn't need anything in particular. You can cool it passively. Obviously, it will thermal throttle, but so do apples cpus. In fact apple laptops are notorious for thermal throttling, constantly hitting 110c in heavy workloads.

 

[JamesJones44]

Completely and utterly wrong. Number of cores is irrelevant. The M4 is a bigger chip than their 14900k. It's made of 28 billion transistors vs 25 billion of the 14900k. It should be beating the crap out of the 14900k in both ST and MT performance (since each core is much bigger in transistor size) but it doesn't.

That's for the entire SoC. Let's not forget that the iGPU in Apple Silicon is far more capable than that found in the 14900k. For that comparison to work you need to compare the CPU to CPU transistor count and I don't believe there is anyway to obtain that information without hours and hours of digging.

 

[helper800]

That's for the entire SoC. Let's not forget that the iGPU in Apple Silicon is far more capable than that found in the 14900k. For that comparison to work you need to compare the CPU to CPU transistor count and I don't believe there is anyway to obtain that information without hours and hours of digging.

With a die shot you can approximate within a reasonable degree of accuracy how many transistors are where because we know the density of the process node and can calculate the area of specific parts of the chip, but until we get that shot its all a guessing game.

 

[TheHerald]

That's for the entire SoC. Let's not forget that the iGPU in Apple Silicon is far more capable than that found in the 14900k. For that comparison to work you need to compare the CPU to CPU transistor count and I don't believe there is anyway to obtain that information without hours and hours of digging.

Yes but on the same note 20% of the 12900k is taken by the igpu and the encoders.

 

[helper800]

Yes but on the same note 20% of the 12900k is taken by the igpu and the encoders.

I would be willing to bet even more of the M4 is, but that's besides the point.

 

[TheHerald]

I would be willing to bet even more of the M4 is, but that's besides the point.

Probably yeah, but the m4 is already 2.5 times the transistor count of the 12900k

 

[JamesJones44]

Yes but on the same note 20% of the 12900k is taken by the igpu and the encoders.

Just based of the 13900k

die shot and the M3 die shot

It's pretty obvious Intel has more space dedicated to the CPU than than Apple Silicon does. The 13900K P cores alone take up at least 30% of the die, with e cores it's about 50%. Apple Silicon of the base M line is about 20ish%.

Also, it looks like Intel actually has a larger die. From what I can find the 13900K has a die size of 257 mm2 and M3 has 146 mm2. If we want to argue node size just rough scaling gives 146 * 2.34 = 341 which would be 32% larger. 50% of 257 is 128 mm2 for the 13900K. 20% of 341 is 68.2 mm2 would be dedicated to CPU on an M3 (TSMC N3 to Intel 7 rough scaling). This is all speculation and density scaling guesses, but it's hard to make an argument on the evidence we have that Intel is using less transistors for their CPU clusters.