M1 PassMark promotion over Intel Core i7-11700K clarifies the damage Apple Silicon has done to Rocket Lake

[Redaktion]

The Apple M1 8-core chip has finally managed to leapfrog the Intel Core i7-11700K in PassMark's CPU Mark chart for single-thread performance in desktop processors. This leaves only the top-end Rocket Lake i9-11900K ahead of the Apple Silicon. While the i9-11900K has an MSRP of US$539, an M1 Mac mini with 8 GB RAM and 256 GB storage costs from US$699.

https://www.notebookcheck.net/M1-PassMark-promotion-over-Intel-Core-i7-11700K-clarifies-the-damage-Apple-Silicon-has-done-to-Rocket-Lake.529209.0.html

[Ali Jradeh]

M1 is a heavily customized CPU, M1 is certainly faster in some tasks but it will be slower while processing other tasks , sure Apple did a great job here. but I never trust all bench marks and especially geekbench, I remember when MSFT used some SD chips on windows and they were getting much lower scores than their Android counter parts, I think reviewing the mechanism that these processors calculate their results  is a must(and why it is so OS depended) . never to mention the upgrade in geekbench 4 gave apple a much higher hand on it's android rivals where it was the opposite in geekbench 3.

I think such things must be reviewed carefully before getting it to people hands.
thanks!

[_MT_]

M1 is a heavily customized CPU, M1 is certainly faster in some tasks but it will be slower while processing other tasks , sure Apple did a great job here. but I never trust all bench marks and especially geekbench, I remember when MSFT used some SD chips on windows and they were getting much lower scores than their Android counter parts, I think reviewing the mechanism that these processors calculate their results  is a must(and why it is so OS depended) ...

There is no denying that Firestorm cores are formidable in single threaded generic computing. You can look up results from SPEC benchmarks. That's a very respected suite. And isn't consumer oriented.

Single threaded performance was always a weakness for ARM processors. So, this is remarkable. But, x86 designs typically offer 2-way SMT. While Firestorm is very strong when only one thread is at play, both AMD and Intel pull ahead when second thread per core is added. And x86 processors can offer more than just four high performance cores. While Firestorm cores are remarkable and there is no doubt Apple could build a first class workstation processor with them, M1 is a ultrabook level processor. A very good one, but that's still where it belongs. It competes with processors like 5400U. Not 5800X.

[ribirto]

But, x86 designs typically offer 2-way SMT. While Firestorm is very strong when only one thread is at play, both AMD and Intel pull ahead when second thread per core is added. And x86 processors can offer more than just four high performance cores.

M1 still comfortably outpaces comparable quad-core offerings from both Intel and AMD, SMT or not, by a healthy margin and while consuming half as much power. If you want, you can consider the Icestorm cores as an SMT equalizer (they offer around 20% of performance compared to Firestorm, which is more or less on par with improvement from SMT on a typical x86 CPU). With their deep OOE capability, Firestorm simply does not need SMT to achieve good backend utilization.

While Firestorm cores are remarkable and there is no doubt Apple could build a first class workstation processor with them, M1 is a ultrabook level processor. A very good one, but that's still where it belongs. It competes with processors like 5400U. Not 5800X.

Very true! And that is why it's so remarkable: the entry-level ultrabook M1 with its meter 4 CPU cores and 10-25W TDP is trading blows with large desktop class chips. Imagine what the rest of the lineup will be capable of.

[actionjksn]

What Intel and AMD should be concerned about is these are very power efficient ships that apple is making.

If they design a really good liquid cooling system for the next Mac Pro and tell their chip designers don't worry about power efficiency at all, make this thing as powerful as possible, what kind of performance could they end up getting?

I don't think this thing is even close to its full potential, if they increase the power consumption, core count and size who knows what they can get out of it.

Maybe it won't scale up that well but then again maybe it will, I don't see why it wouldn't at least by double digit percentages.

[expresspotato]

Remember the M1 is reaching these scores at 10x-15x less power... What's the point of X86/X64 again?

[_MT_]

If you want, you can consider the Icestorm cores as an SMT equalizer (they offer around 20% of performance compared to Firestorm, which is more or less on par with improvement from SMT on a typical x86 CPU). With their deep OOE capability, Firestorm simply does not need SMT to achieve good backend utilization.

Very true! And that is why it's so remarkable: the entry-level ultrabook M1 with its meter 4 CPU cores and 10-25W TDP is trading blows with large desktop class chips. Imagine what the rest of the lineup will be capable of.

The point was that a single x86 core can outpace single Firestorm core if you take advantage of the second hardware thread. It's good to keep this in mind when comparing core counts. It's really more of a 4 core processor counterpart. And yes, in the case of M1, Icestorm cores can do a decent job of compensating for this. Depending on workload. But I can't see a reason for adding more Icestorm cores into designs with more cores. 12+12 design wouldn't make sense to me. I would rather have 16+4. Or even 16+0 if it's a desktop.

The thing is that power limit isn't that big of a factor in single threaded loads. Single core takes relatively little. Even fully loaded. The problem is when you've got 64 of those cores. Then, even 10 W per core would be a big problem. And that's not even counting ancillaries (just the fabric connecting 64 cores is going to eat a load of power). No, the truly impressive part is the kind of MP scores it can produce while being passively cooled in MBA. That's impressive. There are ultrabooks that will trash it. But none of them come close to being passively cooled. Single threaded numbers are interesting because it's ARM. Not because of the power limit.

[Gggshdj]

Remember the M1 is reaching these scores at 10x-15x less power... What's the point of X86/X64 again?

Well luckily AMD or Intel can easily choose to make a high performance ARM or RISC-V CPU and sell those alongside x86/64 ;) Since like AMD and Intel has architecture license and that everyone on the whole planet can make and sell ARM CPU.

[Gggshdj]

Remember the M1 is reaching these scores at 10x-15x less power... What's the point of X86/X64 again?

Also somehow people keep forgetting that the M2 is made on 5nm, it's hilarious when people proclaim M2 to be the best CPU when AMD or Intel isn't even on the same node yet, oh before you guys say stuff like nodes don't matter, hell yeah it matters, wire length is considerably shorter in 5nm vs 7nm which means transmission latency is reduced, plus not to mention that AMD with zen 2 already has 15watt TDP option via their ultrabook CPU lineup and that is on 7nm imagine zen 4 on 5nm which is the proper comparison, yeah apple M2 currently has the fastest single core IPC, but for how long?

[Ben29]

The point was that a single x86 core can outpace single Firestorm core if you take advantage of the second hardware thread. It's good to keep this in mind when comparing core counts.

I do not understand why people keep posting that same nonsense. Single thread means the performance that the core is able to deliver in that single thread. Most of the programs that you use are coded single thread ( the most basic and easiest to program ).

AMD and Intel can perform HT/SMT but that is tested in MultiThread benchmarks. As this requires programs that are multithread aware and programmed as such. And even then you can have bottlenecks where programs are single thread limited ( games where the main logic is in the main thread and thus limited your performance when that main thread is 100% loaded, while your other threads are doing a lot less ).

The fact that AMD/Intel can do better on a core per core basis with MT programs is not relevant for programs that use one thread ( what is something like 80% of the programs that you daily use ).

Nobody cares that AMD/Intel can do more workload in MT for a "core" basis because that is already included in ST vs MT benchmarks. I do not understand why one article resulted in such misinformation of the actual practical limitations of our software and how it works / interacts.

The fact that a M1 with barely a 2.5W a 3W / core is the same as a 13W core, is painful. Even if we add that HT/SMT to the mix, then a M1 core + 30% ( SMT/HT max gain is 30%, most cases less ). So lets make a 3W core, a 4W core. It is still using only 1/3 the power, for the same performance. Ok, lets say its not 7nm vs 5nm, lets add 50% watt gain from a process shrink in reverse ( make that 5nm a 7nm ). That becomes a 6W core vs 13W core.

This is the issue that people do not see. Excuses are excuses, ARM is simply more efficient and has been a for a long time on a watt / performance basis ( note: Apple unified memory etc also help ). Now we are just seeing it in practice on a desktop level field.