apple M1 Pro

[bob]

Back in 2020, I bought a new apple pro with the 2.3ghz intel i7 processor, and used it a lot. But earlier this year, I decided that I wanted to play around with the ARM architecture, which is a cousin of the Sun Sparc, MIPS, etc (the so-called RISC computers that I prefer to call "load store" architectures.)

So, last week I went and traded in my 2020 MacBook pro for a new 2022 MacBook pro with the new M1 Pro chip (8 performance cores, 2 efficiency cores, 16 GPU cores.). I was not really sure what to expect but I wanted to fiddle with ARM.

So, first thing I noticed was it is _really_ fast. IE running Crafty with 8 cores produces similar numbers compared to the Intel 20 core Nehlam machine I used right before I retired. Crafty is reaching beyond 80M NPS which I didn't expect.

So, in summary, it is really a good development platform that is faster than any laptop ought to be. Bit pricey, but works flawlessly. I've removed ALL asm code from crafty and am now using built-in functions for msb() and such. About the only quirk is that using a built in spin lock requires you to include a builtin barrier() as the ARM / M1 architecture is beginning to play fast and loose with memory write ordering to improve speed. It can compile the entire crafty source (one big file) in about 2.5 seconds, for example.)

highly recommended. And it appears that since the load/store architecture is the future, it's worth consideration.

[Modern Times]

The really impressive feature of these CPUs is the outstanding energy efficiency and performance per watt. The very latest Intel and AMD mobile CPUs are fantastic performers now too, but they can't match the low power consumption and heat of the M1.

[Milos]

Back in 2020, I bought a new apple pro with the 2.3ghz intel i7 processor, and used it a lot. But earlier this year, I decided that I wanted to play around with the ARM architecture, which is a cousin of the Sun Sparc, MIPS, etc (the so-called RISC computers that I prefer to call "load store" architectures.)

So, last week I went and traded in my 2020 MacBook pro for a new 2022 MacBook pro with the new M1 Pro chip (8 performance cores, 2 efficiency cores, 16 GPU cores.). I was not really sure what to expect but I wanted to fiddle with ARM.

So, first thing I noticed was it is _really_ fast. IE running Crafty with 8 cores produces similar numbers compared to the Intel 20 core Nehlam machine I used right before I retired. Crafty is reaching beyond 80M NPS which I didn't expect.

So, in summary, it is really a good development platform that is faster than any laptop ought to be. Bit pricey, but works flawlessly. I've removed ALL asm code from crafty and am now using built-in functions for msb() and such. About the only quirk is that using a built in spin lock requires you to include a builtin barrier() as the ARM / M1 architecture is beginning to play fast and loose with memory write ordering to improve speed. It can compile the entire crafty source (one big file) in about 2.5 seconds, for example.)

highly recommended. And it appears that since the load/store architecture is the future, it's worth consideration.

Good to have you back on the forum but you seems to have been living under the rock for a past few years.
20 core Intel Nehlam is an absolute crap that is slower than an 8 core 5800H Zen3. M1 Pro is ok but still not on par with much cheaper laptops with top of the line Zen 3 or Intel Gen 12 mobile CPUs.
Performance per W is useless for computer chess. For light workloads it's great with battery life though.

[brianr]

Another Welcome Back!

The monthly tourney happens to be this Sat, if you might want to enter Crafty with your new rig.
Joost Buijs has been running it for quite a while now and there are typically has from 20-40 engines, often with their authors.
Crafty is a pretty stable engine, so you can leave it unattended too, although the games are fast and things wrap up around 5ish, IIRC.
It is a rare weekend off for me, and I plan to play with Lc0 on what is now a meh 2080ti GPU (sigh, time marches on).

Link: https://talkchess.com/forum3/viewtopic.php?f=7&t=79685

[bob]

Back in 2020, I bought a new apple pro with the 2.3ghz intel i7 processor, and used it a lot. But earlier this year, I decided that I wanted to play around with the ARM architecture, which is a cousin of the Sun Sparc, MIPS, etc (the so-called RISC computers that I prefer to call "load store" architectures.)

So, last week I went and traded in my 2020 MacBook pro for a new 2022 MacBook pro with the new M1 Pro chip (8 performance cores, 2 efficiency cores, 16 GPU cores.). I was not really sure what to expect but I wanted to fiddle with ARM.

So, first thing I noticed was it is _really_ fast. IE running Crafty with 8 cores produces similar numbers compared to the Intel 20 core Nehlam machine I used right before I retired. Crafty is reaching beyond 80M NPS which I didn't expect.

So, in summary, it is really a good development platform that is faster than any laptop ought to be. Bit pricey, but works flawlessly. I've removed ALL asm code from crafty and am now using built-in functions for msb() and such. About the only quirk is that using a built in spin lock requires you to include a builtin barrier() as the ARM / M1 architecture is beginning to play fast and loose with memory write ordering to improve speed. It can compile the entire crafty source (one big file) in about 2.5 seconds, for example.)

highly recommended. And it appears that since the load/store architecture is the future, it's worth consideration.

Good to have you back on the forum but you seems to have been living under the rock for a past few years.
20 core Intel Nehlam is an absolute crap that is slower than an 8 core 5800H Zen3. M1 Pro is ok but still not on par with much cheaper laptops with top of the line Zen 3 or Intel Gen 12 mobile CPUs.
Performance per W is useless for computer chess. For light workloads it's great with battery life though.

however, I spent an hour two days ago tracking down a bug relating to high-performance vs high efficiency cores. Multiple compiles with profiling, etc. When I finished, battery was only at 98% of full charge.

But in any case, my interest was in a much more rational architecture. At last count intel had right at 1,000 opcodes. I am sure the compiler guys are studying all of that to pick exactly the most efficient instructions for a specific piece of code it is compiling. Even Intel has admitted that their external architecture is crap and converted to RISC/load-store internally.

I was impressed with the performance, but that was not my main interest...

Arm is pretty cute for programming.

[Ras]

About the only quirk is that using a built in spin lock requires you to include a builtin barrier() as the ARM / M1 architecture is beginning to play fast and loose with memory write ordering to improve speed.

ARM has generally had weak memory ordering for a long time, not only in the M1. However, the M1 has a special feature that it can be used also with strong memory ordering. That is crucial for Rosetta2 to work as translator, not as software emulator.

[Sopel]

Back in 2020, I bought a new apple pro with the 2.3ghz intel i7 processor, and used it a lot. But earlier this year, I decided that I wanted to play around with the ARM architecture, which is a cousin of the Sun Sparc, MIPS, etc (the so-called RISC computers that I prefer to call "load store" architectures.)

So, last week I went and traded in my 2020 MacBook pro for a new 2022 MacBook pro with the new M1 Pro chip (8 performance cores, 2 efficiency cores, 16 GPU cores.). I was not really sure what to expect but I wanted to fiddle with ARM.

So, first thing I noticed was it is _really_ fast. IE running Crafty with 8 cores produces similar numbers compared to the Intel 20 core Nehlam machine I used right before I retired. Crafty is reaching beyond 80M NPS which I didn't expect.

So, in summary, it is really a good development platform that is faster than any laptop ought to be. Bit pricey, but works flawlessly. I've removed ALL asm code from crafty and am now using built-in functions for msb() and such. About the only quirk is that using a built in spin lock requires you to include a builtin barrier() as the ARM / M1 architecture is beginning to play fast and loose with memory write ordering to improve speed. It can compile the entire crafty source (one big file) in about 2.5 seconds, for example.)

highly recommended. And it appears that since the load/store architecture is the future, it's worth consideration.

Good to have you back on the forum but you seems to have been living under the rock for a past few years.
20 core Intel Nehlam is an absolute crap that is slower than an 8 core 5800H Zen3. M1 Pro is ok but still not on par with much cheaper laptops with top of the line Zen 3 or Intel Gen 12 mobile CPUs.
Performance per W is useless for computer chess. For light workloads it's great with battery life though.

But in any case, my interest was in a much more rational architecture. At last count intel had right at 1,000 opcodes. I am sure the compiler guys are studying all of that to pick exactly the most efficient instructions for a specific piece of code it is compiling. Even Intel has admitted that their external architecture is crap and converted to RISC/load-store internally.

Arm is pretty cute for programming.

Sounds like you're indoctrinated. Nothing of what you said here matters.

[bob]

not sure what any of that means, nor what it has to do with my original post.

[Sopel]

I'll explain what that means. It means that you're providing meaningless facts as arguments in favor of and trying to justify your purchase. You're disregarding any other modern hardware and only compare it to either an older apple device with known issues or a 15 year old CPU. You run an ancient engine and praise M1 for a single result that's just outside your expectations. This is on topic because your post is nothing but an advertisement for apple.

[kiroje]

Back in 2020, I bought a new apple pro with the 2.3ghz intel i7 processor, and used it a lot. But earlier this year, I decided that I wanted to play around with the ARM architecture, which is a cousin of the Sun Sparc, MIPS, etc (the so-called RISC computers that I prefer to call "load store" architectures.)

So, last week I went and traded in my 2020 MacBook pro for a new 2022 MacBook pro with the new M1 Pro chip (8 performance cores, 2 efficiency cores, 16 GPU cores.). I was not really sure what to expect but I wanted to fiddle with ARM.

So, first thing I noticed was it is _really_ fast. IE running Crafty with 8 cores produces similar numbers compared to the Intel 20 core Nehlam machine I used right before I retired. Crafty is reaching beyond 80M NPS which I didn't expect.

So, in summary, it is really a good development platform that is faster than any laptop ought to be. Bit pricey, but works flawlessly. I've removed ALL asm code from crafty and am now using built-in functions for msb() and such. About the only quirk is that using a built in spin lock requires you to include a builtin barrier() as the ARM / M1 architecture is beginning to play fast and loose with memory write ordering to improve speed. It can compile the entire crafty source (one big file) in about 2.5 seconds, for example.)

highly recommended. And it appears that since the load/store architecture is the future, it's worth consideration.

Is that with the bench test ?

I get around 9.1M NPS for 25.3 and 1 thread here on my AMD 5800HS here - that is with an windows compile from here https://github.com/MichaelB7/Crafty/releases