Apple M1 ARM Performance With A 2020 Mac Mini

[PerformanceExpert]

Or Rosetta is better in static translating x86->RISC than AMD's and Intel's CPUs built-in translator (since both provide x86 interface to their internal RISC cores).
That actually is easier to believe as Rosetta is not restricted to real-time translation.

All modern CPUs decode their instructions into internal micro-ops. RISC vs CISC is about the ISA, not about the internals.

But what really matters is the incredible single-threaded performance of M1 - since it beats the fastest desktop x86 cores (even Zen 3!), it can still beat most cores while running translated code.

[BillBroadley]

So, the main question is - Will it run Linux?
As nice as OSX UI is, I'm not sure I'm up to running OSX. Well, that, and the first generation is sorely lacking on RAM department and general extensibility.

Sadly unlikely. Locked bootloader, signed images, and zero docs on the GPU, neural GPUs, image processing, security enclave etc. So a port would be MUCH harder than it would be where it's running on nearly commodity hardware like an intel i5+IGPU.

[BillBroadley]

Mixed bag when comes to performance and terrible software. No, thanks.

Keep in mind that many of the benchmark losses were not against the m1, but against the m1 emulating an x86-64. Give it a few weeks and there will be MUCH more native support.

[Michael]

Have SoC package power monitoring working with PTS now so future articles will include raw power and perf-per-Watt.

[ultimA]

Man,

go to bed and sleep over it. You are clearly tired
You are talking non-sense

Some numbers?

• Processor Number i7-6700HQ
• Launch Date Q3'15
• Lithography 14 nm
• # of Cores 4
• # of Threads 8

https://ark.intel.com/content/www/us...-3-50-ghz.html

• Processor Number i7-8700B
• Launch Date Q2'18
• Lithography 14 nm
• # of Cores 6
• # of Threads 12

https://ark.intel.com/content/www/us...-4-60-ghz.html

The M1 is an 8 cores big.little (4 big 4 little) -- not sure how many threads per core, let's assume 1
lithography 5nm -- Intel may not get there soon, AMD will

W.r.t to Rosetta: it is NOT an emulator -- if it was Michael would still be there crunching numbers

The only conclusion is :

The new Mac (arm based and bla bla bla) is faster often but not always than a 2years old mac -- that is underpowered and thermally throtled

Thanks Apple, 42 minutes standing ovation

Yes these Intel CPUs are older ones but IMHO that doesn't make these benchmarks invalid or apples-to-oranges. Simply because the performance advantage of the M1 compared to the intels in many of these benchmarks is so huge (40-50%, in some even more), that the age doesn't justify the difference. For example, the i7-6700HQ is now 5 yrs old, but current Intel CPUs are not 40% faster (not even close) than the i7-6700HQ at similar core counts. And if that wasn't enough, the M1 manages this performance with much lower clocks, less heat, and lower power consumption. So while I agree that the perf advantage of the M1 is not as big as these benchmarks suggest when compared to current CPUs of today, but I do think they are still a clear winner. A certain and sure wakeup call to Intel and even AMD. Which is only good for us

[BillBroadley]

Here's hoping the next benchmarks measure bandwidth and latency for both sequential and random accesses. From what I can one of the most unique thing about the apple M1 is having 8 channels to rather quick LP-DDR4X running somewhere north of 4200 GT/sec. Is someone wants to grant a user shell on one I could quantify this myself, just require a working gcc compiler.

[PerformanceExpert]

The only conclusion is :

The new Mac (arm based and bla bla bla) is faster often but not always than a 2years old mac -- that is underpowered and thermally throtled

Thanks Apple, 42 minutes standing ovation

It outperforms the fastest x86 desktops including Zen 3 at a fraction of the power. The only conclusion: game over.

[Grinness]

Yes these Intel CPUs are older ones but IMHO that doesn't make these benchmarks invalid or apples-to-oranges. Simply because the performance advantage of the M1 compared to the intels in many of these benchmarks is so huge (40-50%, in some even more), that the age doesn't justify the difference. For example, the i7-6700HQ is now 5 yrs old, but current Intel CPUs are not 40% faster (not even close) than the i7-6700HQ at similar core counts. And if that wasn't enough, the M1 manages this performance with much lower clocks, less heat, and lower power consumption. So while I agree that the perf advantage of the M1 is not as big as these benchmarks suggest when compared to current CPUs of today, but I do think they are still a clear winner. A certain and sure wakeup call to Intel and even AMD. Which is only good for us

The results on Openbenchmarks with modern CPUs tell a different story

https://openbenchmarking.org/result/2011202-PTS-MERGE68890&sor

Again, the M1 performance are good -- this is not a weak CPU
But it is not (yet?) a revolution

And yes, I agree with you that this is a wake up call for AMD and Intel.
But I do not want that the wake up call translates into RAM-on-chip for all.

I personally believe that the major advantage of he M1 is the RAM in the same package, but this has serious limitation for i.e. ppl that build their own PC from individual parts, ect

[carewolf]

So, the main question is - Will it run Linux?
As nice as OSX UI is, I'm not sure I'm up to running OSX. Well, that, and the first generation is sorely lacking on RAM department and general extensibility.

Well a lot of the performance probably comes from integrating the memory directly on to the chip. It will performance worse if they make it flexible.

[PerformanceExpert]

Here's hoping the next benchmarks measure bandwidth and latency for both sequential and random accesses. From what I can one of the most unique thing about the apple M1 is having 8 channels to rather quick LP-DDR4X running somewhere north of 4200 GT/sec. Is someone wants to grant a user shell on one I could quantify this myself, just require a working gcc compiler.

AnandTech has some memory system results, it's 68GB/s max. But with those humongous caches you can run most code without relying on fast DRAM.