Stockfish patch to improve performance on M1 by up to 80%.

[Romain_H]

Was the link to hard for you to click?

Your link provides nothing of value. Plus your assumption that most of the power draw comes from the CPU is based on what exactly? I‘d rather guess most of the power draw when fully maxed out comes from the GPU.
Now, considering your Ryzen example, what would the power draw of both CPU and GPU be in comparison?

 

[Leifi]

Plus your assumption that most of the power draw comes from the CPU is based on what exactly?

The assumption is based on the fact that at M1 macs does pretty most everything using the M1 chip. the only other major power draws are the screen and SSD. The draw of those would be highly constant on CPU-centric benchmarks.

And it is a fair point that using a BFGPU and a significantly faster CPU will warrant more power draw. But as long as M1 chips can't come close to the level of GPU+CPU performance the Watt issue is sort of moot. You also have a lot more freedom to up-clock, downclock and select on-chip GPU or external GPU, etc., etc. on Hw that is not so limited as Apples, more appliance-like devices. it is like comparing a mobile 4nm equipped phone with a Macbook Pro... The Macbook will draw more power in that comparison, but the top-end speed will also be greater. which is the more important metric.

 

[Leifi]

No, it's not bad for Apple. I realize it's hard for people like you to understand, but most people who buy PCs don't ever overclock them.

Charging more money for less functionally and flexibilty, is a winning concept no doubt.

 

[Romain_H]

As usual, you are wrong. For one, M1 beat close to every notebook on the market in terms of CPU speed. Plus: You need to differentiate what part of the package is utilized - there are CPU portion and GPU portion. Not every workload uses the same components; what you tend to do is comparing apples to oranges

 

[thenewperson]

Damn, this is still a troll thread I guess.

 

[leman]

So you compare the full chipset-draw of an AMD with an 8-core compute-core only draw of the M1?

I have no idea what you are trying to say here. You asked for the M1 consumption numbers while running stockfish, that's what I gave you.

It seems the argument has moved from Software is not optimized to power consumption at 5nm at expense of top-performance compared to bigger die chipsets is "king".. (desperate to get to post a "win" for the M1, right?) ??

Again, I have no idea what you are trying to say. The argument for M1 has always been performance per watt, performance in-class and general purpose performance. As I said on multiple occasions, I don't find it at all surprising that an unrestricted Zen3 will outperform M1 in a carefully optimised sustained SIMD workload. They both have the same SIMD width, but Zen3 has higher clocks. If you manage to run your mobile Zen3 at 4Ghz, that should give you around 30% advantage over M1 Pro/Max in these kind of workloads. If that makes you happy, be my guest.

I also am a bit puzzled by your picture posted showing a "package power" of 40.3W...

Package power is the power consumed by the entire chip, which includes GPU, memory and everything else.

The nps shown on your picture seems much much lower of what has been posted for the M1 Max earlier in this thread (11M vs 16M (or 20M w/o nnue)). Can you explain this low nps, in that picture, please?

There is nothing to explain. My screenshot does not show the benchmark summary. Whatever diagnostics information is displayed while the benchmark is running is completely inconsequential. Using the same benchmarks settings as here: http://ipmanchess.yolasite.com/amd--intel-chess-bench-stockfish.php my machine gives the following result:

===========================

Total time (ms) : 87563
Nodes searched : 1410508750
Nodes/second : 16108501

 

[Krevnik]

So you compare the full chipset-draw of an AMD with an 8-core compute-core only draw of the M1?

<…>

I also am a bit puzzled by your picture posted showing a "package power" of 40.3W...

Even if we take your link at face value, you’re trying to compare chipset draw to “at the wall” draw. How is that better?

Package power is the draw of the M1 package as a whole. That’s CPU + GPU + RAM. It’s still covering more than the chipset draw of an AMD chip without an iGPU. And Intel’s iGPUs aren’t quite in the same category as the ones in the M1 Pro/Max. But the numbers should still be mostly comparable when looking at CPU workloads.

So if you are talking about power draw of a CPU process like Stockfish, you need to compare it to the CPU draw of other chipsets. The video the link refers to does go over what their stress test was:

- Cinebench to stress the CPU
- Geekbench to stress the GPU
- No mention of display brightness
- Measured from the wall

This is a lot more being measured than the consumption of an AMD or Intel CPU package when we’re looking to measure power draw of a CPU processes.

If you want to get good comparisons of a GPU+CPU workload, such as gaming, you do want to measure the full M1 package power, along with the CPU+GPU consumption of the machine you are comparing it against. So no ignoring that mobile GPU in the AMD/Intel system.

I’ve said it before on threads here that the 16” M1 Max MBP design used up all the thermal headroom they gained from the M1 CPU design to scale up the GPU. It does mean that peak power consumption of the CPU+GPU as a whole is in a similar ballpark as the previous 16” MBP. But in practice, the efficiencies gained means that it’s not operating at that power draw for any real amount of time in many scenarios. It’s pretty clear when the same scenario on my Intel 16” vs the M1 Max 16” will have entirely different needs for cooling.

 

[Leifi]

Even if we take your link at face value, you’re trying to compare chipset draw to “at the wall” draw. How is that better?

I am not "trying to compare". I am just replying to some ridiculous comparisons made here. Personally, I don't care diddley if it draws 35W or 70W.. as long as it performs well.

 

[Leifi]

I have no idea what you are trying to say here. You asked for the M1 consumption numbers while running stockfish, that's what I gave you.

The thing is that you are making a lot of bold claims about the 5900HX power-draw without backing those up apart from some useless links to notebookcheck.net testing complete package power-draw test on the ROG laptop (not comparable tests).. and then compare that to measuring only the cores of an M1, and critizizing links testing complete laptop-draws of Macbook. This look genuinely dishonest IMHO. I just thought you where trying to do a fair comparison here?

 

[Romain_H]

What? Its you who compares the Ryzen power draw only to all of the M1 power draw. Its YOU who makes unfair comaprisons.
Compare CPU vs CPU or compare the CPU+GPU vs CPU+GPU, not CPU Ryzen vs CPU+GPU on the M1.

 

[Leifi]

If you want to get good comparisons of a GPU+CPU workload, such as gaming, you do want to measure the full M1 package power, along with the CPU+GPU consumption of the machine you are comparing it against. So no ignoring that mobile GPU in the AMD/Intel system.

From a power-draw standpoint it would be more relevannt to compare with the 5900H built in GPU rather than external GPU - an options that the Apple completely lack.

 

[Leifi]

What? Its you who compares the Ryzen power draw only to all of the M1 power draw. Its YOU who makes unfair comaprisons.
Compare CPU vs CPU or compare the CPU+GPU vs CPU+GPU, not CPU Ryzen vs CPU+GPU on the M1.

I am re-iterating here. I am not doing any wattage "comparisons" at all.. these are all done by the Apple shills here...

 

[Krevnik]

I am not "trying to compare". I am just replying to some ridiculous comparisons made here. Personally, I don't care diddley if it draws 35W or 70W.. as long as it performs well.

Let’s assume the comparisons you talk about are indeed as ridiculous as you claim. You did nothing to improve the quality of discussion about them by throwing in numbers that aren’t measuring the same thing as anything else discussed up to that point. The point of throwing those numbers in were to prompt a different kind of comparison that made your point of view look more favorable, so I’d call that a comparison, yes.

From a power-draw standpoint it would be more relevannt to compare with the 5900H built in GPU rather than external GPU options that the Apple completely lacks.

If you are looking at a GPU workload and just want to compare how the iGPU in the Ryzen handles it, sure. But then you throw out the benefit of a dGPU if provided in the Intel/AMD laptop, so there’s skews in the results there. For a GPU-bound workflow, there’s clearly some nuance that should be taken into account when presenting results of that kind of comparison.

But keep in mind you are in a thread discussing a CPU-bound workflow. The focus will be on CPU cores and their performance and power consumption. We aren’t going to talk about the GPU much, as during these sort of tests it should be very lightly loaded.

 

[leman]

The thing is that you are making a lot of bold claims about the 5900HX power-draw without backing those up apart from some useless links to notebookcheck.net testing complete package power-draw test on the ROG laptop (not comparable tests).. and then compare that to measuring only the cores of an M1, and critizizing links testing complete laptop-draws of Macbook. This look genuinely dishonest IMHO. I just thought you where trying to do a fair comparison here?

The notebook check review I posted has detailed breakdown of per-core frequencies and power draws… with the CPU cluster drawing approximately 70W in highest performance setting.

As to “not comparable terms”, sure. I’ve been asking folks who own Ryzen laptops to post detailed power consumption numbers when running Stockfisch but all I got back was “my ryzen is 2x,3x,10x faster at sane power draw” gibberish. So I posted the tests I could find. Since power draw generally depends on frequency these should be fairly accurate anyway. Abs of course, that ROG laptop is not “normal Ryzen” - exactly what I have been trying to show. It runs with much higher power level than what’s expected from these chip. An average 5900HS will be slower.

Anyway, you got a Ryzen? Then post your results and power consumption figures.

 

[Leifi]

The watt comparison is useless if you just want to get the best performance from the mobile-chess analysis. It is as useless as comparing power-draw between the M1 Max laptop and the Mediatek Dimensity 9000 phones ...

If you are plugged in the difference really is how much you save on your electricity bill. If you are on the road other factors like time used for different use-cases and battery size, and a possibility to use battery packs are more important than only the power draw of the cores...

 

[tomO2013]

Hi Leifi,

I get a sense that performance in stock fish is incredibly important to you and a core driver in any hardware purchase decisions.

Would you mind sharing your stock fish power metrics from your AMD 5900H laptop here please?

Thanks

 

[robco74]

If we were comparing desktops, then power draw might not be a fair criterion to use. But for laptops it's rather important and shouldn't be dismissed. Having lots of power can mean little if you can't get decent performance for any length of time on battery. If you have to be plugged in all the time, you don't have a laptop, you have a portable.

Not only do AMD/Intel laptops tend to have poorer battery life, they throttle significantly while on battery to last as long as they do. Apple Silicon OTOH, gives consistent performance and strong battery life.

So I suppose it depends on how long you run Stockfish for, and if you want performance when you're away from an outlet.

 

[JimmyjamesEU]

Not today Satan! Nice try.

 

[Romain_H]

the Apple shills here...

Name calling won't help your cause, really. You are way too focused on one metric and one metric only, ignoring anything else.
Actually the "shills" are mostly correct in what they are saying, whereas you clearly lack understanding on the issue, as evidenced by your claim that there are no jumps in performance due to software optimisation to be expected. Of course, within just a couple of days your claim has been debunked.

 

[cmaier]

Name calling won't help your cause, really. You are so focused on one metric and one metric only, ignoring anything else.
Actually the "shills" are mostly correct in what they are saying, whereas you clearly lack understanding on the issue, as evidenced by your claim that there are no jumps in performance due to software optimisation to be expected. Of course, within just a couple of days your claim has been debunked.

The latter point is key. A bunch of folks who don’t understand CPU design and trade offs jump on the only data point they could find that made Apple look bad, and declared it was proof of inherent flaws in M1 and could never be fixed. 2 weeks later a software update makes Apple look great, and most of them scurry away.

 

[dgdosen]

The watt comparison is useless if you just want to get the best performance from the mobile-chess analysis. It is as useless as comparing power-draw between the M1 Max laptop and the Mediatek Dimensity 9000 phones ...

If you are plugged in the difference really is how much you save on your electricity bill. If you are on the road other factors like time used for different use-cases and battery size, and a possibility to use battery packs are more important than only the power draw of the cores...

How do we know you're not hooking up some kind of oscillation overthruster while performing your test runs? I think it's important to note the power used...

 

[leman]

The watt comparison is useless if you just want to get the best performance from the mobile-chess analysis. It is as useless as comparing power-draw between the M1 Max laptop and the Mediatek Dimensity 9000 phones ...

If you are plugged in the difference really is how much you save on your electricity bill. If you are on the road other factors like time used for different use-cases and battery size, and a possibility to use battery packs are more important than only the power draw of the cores...

If you look to prior performance on stockfish over every other aspect of a laptop, then a desktop replacement portable Zen3 machine that dispenses with the CPU power limit certainly seems like a better choice for you. Not every product caters to niche use cases like yours and that’s perfectly fine.

Some others - me for example - prefer to use a machine thst offers the performance of a desktop workstation without tethering us to a desk, without sacrificing battery life, portability or thermals. Different tools, different people. Thst said, if my primary computing interest were sustained SIMD workloads I would not pick Apple Silicon - I’d go with Threadripper or something. Also, new Intel chips are going to be great for these kind of workloads - that’s exactly what multiple core architectures excel at.

 

[Leifi]

If you look to prior performance on stockfish over every other aspect of a laptop, then a desktop replacement portable Zen3 machine that dispenses with the CPU power limit certainly seems like a better choice for you. Not every product caters to niche use cases like yours and that’s perfectly fine.

Agreed and I fully appreciate that you may value different things than me... Some just care about the color... Who am I or you to say they are wrong... But the original threads of disucssion was about Apple M1 (not even the Max and Pro) not living up to the performance-hype when running chess related benchmarks (both CPU and GPU bound benches).

Some others - me for example - prefer to use a machine thst offers the performance of a desktop workstation without tethering us to a desk, without sacrificing battery life, portability or thermals. Different tools, different people. Thst said, if my primary computing interest were sustained SIMD workloads I would not pick Apple Silicon - I’d go with Threadripper or something. Also, new Intel chips are going to be great for these kind of workloads - that’s exactly what multiple core architectures excel at.

My major point would be that if you buy a faster AMD+BFGPU laptop you would also not be tethered to a desk. For use-cases not involving gaming, chess, etc. You still have the option to run many hours in power-saving modes on the bigger batteries of these alternatives. I would argue that thermals+cooling usually are better designed and thought through in most gaming laptops as well.

Thst said, if my primary computing interest were sustained SIMD workloads I would not pick Apple Silicon - I’d go with Threadripper or something. Also, new Intel chips are going to be great for these kind of workloads - that’s exactly what multiple core architectures excel at.

I could not agree more with you on these final points made. I am very keen on trying out a new high-end alderlake laptop, when I can get one.

 

[mr_roboto]

So if you are talking about power draw of a CPU process like Stockfish, you need to compare it to the CPU draw of other chipsets. The video the link refers to does go over what their stress test was:

- Cinebench to stress the CPU
- Geekbench to stress the GPU
- No mention of display brightness
- Measured from the wall

This is a lot more being measured than the consumption of an AMD or Intel CPU package when we’re looking to measure power draw of a CPU processes.

Was it clear from the video whether they took pains to ensure the battery wasn't charging during these benchmarks?

 

[leman]

But the original threads of disucssion was about Apple M1 (not even the Max and Pro) not living up to the performance-hype when running chess related benchmarks (both CPU and GPU bound benches).

Except is absolutely does. It's a low power CPU that reaches ~10mnps in a passively cooled chassis. That's as good as it gets for that class of laptops.

My major point would be that if you buy a faster AMD+BFGPU laptop you would also not be tethered to a desk. For use-cases not involving gaming, chess, etc. You still have the option to run many hours in power-saving modes on the bigger batteries of these alternatives.

But I don't want to run "many hours in power-saving modes". I want to run in full performance mode. I don't want my laptop to slow down to a crawl just because its not plugged in. And I very much doubt that any AMD or Intel laptop is going to be faster than my M1 Max for the work I do. You are measuring performance by looking at one extremely niche streamlined workload which favours x86 CPUs with their higher clock and SMT. In the meantime my M1 Max runs my data manipulation scripts around 3x faster than my i9-9980HK laptop, while using between 2x-5x less power and thus giving me twice as longer useful battery life.

I would argue that thermals+cooling usually are better designed and thought through in most gaming laptops as well.

If your laptop needs over 100W to run the work it is designed for, sure, it better have a good designed cooling system. I prefer to have a laptop that can do the same at much lower power consumption though. I don't need scalding air getting blown all over my legs.