Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple M1 CPU & GPU speed is very disappointing
- Thread starter Appletoni
- Start date
- Sort by reaction score
- Status
For business and self-employed people = main group of Apple MacBook buyers, time is also money so faster results requiring less time wins.
Intel was stupid to stay with 6 cores and AMD started to increase cores every time -> 8 -> 16 -> 32 -> 64 -> and now with 96 followed by 128 cores.
Obviously this works for AMD because that is exactly what the most people want to have/buy. = Multicore Performance.
The most chess players which I know and I know a lot, use 16 cores AMD notebooks to run Stockfish and other things.
But go and play chess and use your AMD machine (I myself also have a threadripper desktop and am very happily using it).
Leave us alone, in our small and delusional world using our nasty and underpowered M1 Macs.
Godspeed!
Leave us alone, in our small and delusional world using our nasty and underpowered M1 Macs.
Godspeed!
The ridiculous juvenile thing here is the logic used when comparing a low-end 28W laptop CPU with a 280W workstation CPU in an attempt to bash the "weaker" CPU.
I could start a discussion too about "AMD Threadripper 3990X speed is very disappointing" and use Fujitsu A64FX ARM CPU with 104 cores costing 10 times more than 3990X as comparison and say "For some industries time is money so faster results requiring less time wins. That's also why some travelers fly rather than drive."
If people want Threadripper perfomance out of M1 because their time is money then they should go and buy one or research more before they buy M1 and get disappointed and want to "trade their MBA M1 16GB/256GB for gaming laptop with 6800m dGPU."
For chess players time is certainly not money. If your time is money why do you even play chess and Stockfish to begin with? Pointless argument!
Last edited:
That doesn't make sense to me, computers are computers and comparing them is fair game, but would you rather have everyone just not pay attention to the M1 when talking about performance?
I can assure you that performance is a much higher requirement on the list than power draw for anything I buy. (for myself or for my users at work)
That doesn’t make sense to me, automobiles are automobiles and comparing them is fair game, but would you rather have everyone not pay attention to the f150 when talking about performance of semi trucks?
I can assure you that performance is a much higher requirement on the list than fuel economy for everything I buy (for myself or my users at my trucking company)
A semi truck isn't the same thing and don't do the same job, so no I wouldn't compare them. It wouldn't be the same as a computer vs a computer.
That's your job to do, I wouldn't argue against your requirements, I know nothing about them.
Like I said that Threadripper performance in Cinebench is very disappointing...
Yes, both are vehicles and do the same job of transporting you between to cities, but if the job is to transport you and tons of loads the semi truck is better. It doesn't mean F-150 is "disappointing" because it can't transport tons of loads like the semi truck, as little as M1 is disappointing because it has low scores in chess/Stockfish or it's 10 times slower than Threadripper in Cinebench. As for the M1 GPU which the OP also mentions in the thread's title I think I showed that it's the other way around. It's the AMD iGPU Vega 8 that is disappointing in comparison to M1, but maybe that doesn't make sense to you either and you also prefer to compare it to RTX 3090 and find M1 iGPU disappointing?
Last edited:
It might be, I wouldn't know nor care, never shopped for one. A threadripper is a server class CPU, which is a little different, but yeah, it's a similar comparison in the computer vs computer department and valid for comparison. I bet it wouldn't be disappointing in multi thread tests!
False, one's for carrying large amounts of heavy cargo, the other is not.
That comparison doesn't make any difference to me, not in my wheelhouse. (GPU performance) So you, or anyone else can say whatever they like and I probably wont even read it.
Time again to dispel FUD with facts.
M1 isn't 10W. Measured with built-in powermetrics under full CPU load without GPU load running Stockfish measures 19.6W. Now I'm curious what the power draw is for full CPU + GPU loads but that's for another day.
ANE Power: 0 mW
DRAM Power: 276 mW
CPU Power: 19647 mW
Package Power: 20196 mW
Passive cooling on the MBA M1 isn't sufficient to prevent M1 thermal throttling under sustained load and will drop ~24% performance.
1st Stockfish run
Total time (ms): 51341
Nodes searched: 614882999
Nodes/second: 11976451
After multiple back to back runs
Total time (ms): 72443
Nodes searched: 658001226
Nodes/second: 9083020
Example for comparison, Threadripper 3990x at 280W scores 168992746 nps on Stockfish
M1 at 19.6W scores 9083020 nps on Stockfish
280W / 19.6W = 14.3x higher power draw for Threadripper 3990x
Without consideration for M1 thermal throttling
168992746 nps / 11976451 nps = 14.1x higher performance for Threadripper 3990x
With consideration for M1 thermal throttling
168992746nps / 9083020 nps = 18.6x higher performance for Threadripper 3990x
So, for workloads similar to Stockfish, Threadripper 3990x has a performance per watt of between 130% and 98% of M1. Add ~10% for faster 280W Threadripper Pro 3995wx.
M1 isn't 10W. Measured with built-in powermetrics under full CPU load without GPU load running Stockfish measures 19.6W. Now I'm curious what the power draw is for full CPU + GPU loads but that's for another day.
ANE Power: 0 mW
DRAM Power: 276 mW
CPU Power: 19647 mW
Package Power: 20196 mW
Passive cooling on the MBA M1 isn't sufficient to prevent M1 thermal throttling under sustained load and will drop ~24% performance.
1st Stockfish run
Total time (ms): 51341
Nodes searched: 614882999
Nodes/second: 11976451
After multiple back to back runs
Total time (ms): 72443
Nodes searched: 658001226
Nodes/second: 9083020
Example for comparison, Threadripper 3990x at 280W scores 168992746 nps on Stockfish
M1 at 19.6W scores 9083020 nps on Stockfish
280W / 19.6W = 14.3x higher power draw for Threadripper 3990x
Without consideration for M1 thermal throttling
168992746 nps / 11976451 nps = 14.1x higher performance for Threadripper 3990x
With consideration for M1 thermal throttling
168992746nps / 9083020 nps = 18.6x higher performance for Threadripper 3990x
So, for workloads similar to Stockfish, Threadripper 3990x has a performance per watt of between 130% and 98% of M1. Add ~10% for faster 280W Threadripper Pro 3995wx.
Last edited:
In my tests the M1 only used around 15W running stockfish (measured with powermetrics). I'd like to see some more measurements that track the power consumption. Unfortunately I don't have my M1 machine so I can't run the tests myself. Not to mention that a MacBook Air is physically incapable of sustaining 20W for a prolonged period of time.
At 15W power usage, the situation looks a bit different, with M1 delivering 30% more perf/watt than the Threadripper. Which is more in line with industry-standard benchmarks.
When the MacBook Air is thermally throttled it uses about 7w. My Mac mini can sustain up to 20w in certain benchmarks ( I never tested stockfish). when the MacBook Air is using 7w, it is only approx 20% slower. If Apple chose to run at 2.6ghz instead of 3.2ghz the performance/watt is amazing.
my fairly unscientific findings for the power usage of the high performance core in the m1 vs frequency.
2.6 ghz - 1.8w
2.8 ghz - 2.5w
3 ghz - 3.4w
3.2 ghz - 5w
note: this is not the absolute max. Certain code can push these values higher. But overall, backing the frequency even slightly back you increase per/watts significantly
Last edited:
Which is a plausible result for a CPU with a TDP of ~10-15W (I don't think Apple have actually published the TDP) when running a CPU-heavy task.
You can't legitimately compare the published default TDP of a Threadripper (which is what that 280W figure is) with the actual power consumption of the M1 measured while running the actual benchmark. Esp. since TDP is a notoriously shonky statistic which manufacturers will calculate... optimistically, and which can often bet configured to a different value by the system builder. That Threadripper may be drawing 150% of TDP during the test.
You're also assuming that performance-per-watt is linear - might be close to that if your test is very highly optimised for multi-threading and isn't limited by RAM or I/O), otherwise it is more likely to be "diminishing returns". Also, the Threadripper has hyperthreading (or whatever AMD call it) and the effectiveness of that will depend hugely on the specific workload.
Thermal design power - Wikipedia
AMD Ryzen™ Threadripper™ 3990X Processor
AMD Ryzen™ Threadripper™ 3990X is the world's first 64-Core desktop processor that delivers incredible performance for many demanding workloads. Learn More!
Apple M1 - Benchmark, Test and Specs
Apple M1 - Benchmark, Geekbench 5, Cinebench R20, Cinebench R23, Cinebench R15 and FP32 iGPU (GFLOPS) benchmark results
If my vaguely remembered Physics serves, current goes up in proportion to clock speed (frequency/speed of switching), and power goes up with the square of current - which is why more, slower cores are theoretically better - but only if the software can be optimised to actually use all of those cores - which is hard.
Trouble with this whole argument: the M1 was primarily designed for low-power ultra-portable applications. It had one job to do - thrash the integrated chipsets used in the Intel Macbook Air and low-end MBP - and it seems to do that with a vengeance. It has considerably pushed the envelope in what these ultra-portable systems can do - which is needed when even "regular consumers" want to edit video for YouTube, use no-green-screen-required virtual backdrops in their video conferences . What it isn't is a credible replacement for higher-end personal workstations - unless you're lucky and your workload hits one of the M1s sweet-spots (e.g. certain video codecs). It doesn't have enough CPU cores - and has the wrong balance of efficiency/performance cores - or GPU cores, can't access enough RAM and can't drive more than two displays... None of which is a problem in a MacBook Air, or even a 24" iMac for personal productivity (and occasional Garageband jams) - but it's not really a consideration for people looking at a 64 core Threadripper (...which doesn't have to run off a battery and can be stuck to a cooler the size of Manhatten). We haven't seen Apple's horse for that course yet.
Regarding your physics:Which is a plausible result for a CPU with a TDP of ~10-15W (I don't think Apple have actually published the TDP) when running a CPU-heavy task.
You can't legitimately compare the published default TDP of a Threadripper (which is what that 280W figure is) with the actual power consumption of the M1 measured while running the actual benchmark. Esp. since TDP is a notoriously shonky statistic which manufacturers will calculate... optimistically, and which can often bet configured to a different value by the system builder. That Threadripper may be drawing 150% of TDP during the test.
You're also assuming that performance-per-watt is linear - might be close to that if your test is very highly optimised for multi-threading and isn't limited by RAM or I/O), otherwise it is more likely to be "diminishing returns". Also, the Threadripper has hyperthreading (or whatever AMD call it) and the effectiveness of that will depend hugely on the specific workload.
Thermal design power - Wikipedia
en.wikipedia.org
AMD Ryzen™ Threadripper™ 3990X Processor
AMD Ryzen™ Threadripper™ 3990X is the world's first 64-Core desktop processor that delivers incredible performance for many demanding workloads. Learn More!www.amd.com
Apple M1 - Benchmark, Test and Specs
Apple M1 - Benchmark, Geekbench 5, Cinebench R20, Cinebench R23, Cinebench R15 and FP32 iGPU (GFLOPS) benchmark resultswww.cpu-monkey.com
If my vaguely remembered Physics serves, current goes up in proportion to clock speed (frequency/speed of switching), and power goes up with the square of current - which is why more, slower cores are theoretically better - but only if the software can be optimised to actually use all of those cores - which is hard.
Trouble with this whole argument: the M1 was primarily designed for low-power ultra-portable applications. It had one job to do - thrash the integrated chipsets used in the Intel Macbook Air and low-end MBP - and it seems to do that with a vengeance. It has considerably pushed the envelope in what these ultra-portable systems can do - which is needed when even "regular consumers" want to edit video for YouTube, use no-green-screen-required virtual backdrops in their video conferences . What it isn't is a credible replacement for higher-end personal workstations - unless you're lucky and your workload hits one of the M1s sweet-spots (e.g. certain video codecs). It doesn't have enough CPU cores - and has the wrong balance of efficiency/performance cores - or GPU cores, can't access enough RAM and can't drive more than two displays... None of which is a problem in a MacBook Air, or even a 24" iMac for personal productivity (and occasional Garageband jams) - but it's not really a consideration for people looking at a 64 core Threadripper (...which doesn't have to run off a battery and can be stuck to a cooler the size of Manhatten). We haven't seen Apple's horse for that course yet.
Power equals capacitance times voltage squared times clock frequency (there’s also a ½ factor depending on how you measure frequency). At least for the dynamic power.
You can't compare a 280W Threadripper to the M1 - they aren't even in the same class of processor. The former is a high-end desktop-class CPU targeted at the server and high-end data processing markets, while the M1 is an entry-level mobile processor. You should be comparing the M1 to Intel's i5 processors used in the notebook market if you want a fair comparison. It's telling that you continue to double down on this meaningless comparison just to bolster your agenda while simultaneously ignoring/rejecting any and all argumentation which proves your basis of comparison and your underlying methodology to be flawed at its very core. But if making that flawed comparison is the only way you can support your agenda, go ahead...
I disagree actually! Comparing processors of different classes can be a very interesting exercise to learn details about certain design choices and predict scalability. The problem with mi7chi’s post is that the power usage estimates are likely incorrect. In my tests the sustained power consumption of M1 was 15W, not 20W, and it’s not very clear whether Threadripper indeed uses 280 watts on this particular text (although I would believe it does as other tests indicate that it indeed peaks out at 280 watts).
My problem with his comparison is that he is comparing the SoC equivalent of a Toyota Camry (great performance and efficiency, not a lot of cargo/towing capacity or top end speed) to the AMD equivalent of a race engine, and trying to use the raw performance numbers to discredit the M1 on its face. It makes just as little sense as the people complaining because the M1 Macs aren't the best gaming machines on the market. They aren't even aimed at the same users or use cases as Threadripper-based systems. My own testing shows that the M1 outclasses my 9th generation Intel i7 gaming rig (6 cores with HT) in every processor-based test. The only area the M1 falls short in is anything GPU-intensive, as the Intel machine also has a GTX 1660Ti. I don't have an 11th gen Intel machine to compare the M1 against, but from what I have read, the performance difference between 9th and 11th generation Intel (mobile) parts is probably not enough to catch up to the M1, let alone surpass Apple's SoC in computational performance. I also used a combination of synthetic benchmarks and real-world applications to compare the two systems, instead of a chess app I'll never install or have any use for.
From a generational perspective, the last big jump in performance on the Intel side was the move from the 7th gen to 8th gen i-series processors, which coincided with a move to a smaller process node). What gains have been achieved in the move to the 9th, 10th, and 11th generation processors were largely due to Intel bumping up clock speeds, as they had hit a wall in terms of their own fabrication of smaller processes. With the M1, we can only go by the example of the A-series processors at this time, and Apple has had a very consistent rate of improvement going back to the A5 SoC. If Apple can maintain that rate of improvement with the M-Series, people like Mi7chy are going to have even less of a platform to stand on going forward, as the A-series has outpaced Intel's rate of improvement for years at this point.
Well, let's have a look at that, shall we?
Ryzen 5950X (high-end Zen 3 desktop CPU with 16 cores): https://www.anandtech.com/show/1621...-review-5950x-5900x-5800x-and-5700x-tested/10
83 SPECint, 64 SPECfloat, 120W power consumption (see page on TDP in the same review)
Apple M1 Mini: https://www.anandtech.com/show/16252/mac-mini-apple-m1-tested/5
28.8 SPECint, 38.7 SPECfloat, let's just say 20W power consumption (although the average is closer to 15W)
For the Zen 3 CPU: 0.7 SPECint/W (5.2 points per core), 0.53 SPECfloat/W (4 points per core)
For the M1: 1.44 SPECint/W (7.2 points per core), 1.9 SPECfloat/W (9.6 points per core)
This is SPEC2017, an industry standard benchmark that combines multiple representative workloads. Note that I count the M1 as a quad-core CPU for the per-core point computation, since the four efficiency cores contribute as much as the Zen's SMT). As you can see, M1 has 2-3 times better energy efficiency and around 30-100% higher performance per core depending on the workload. In the later metric you can see M1's superior energy efficiency really play out — it doesn't need to downclock nearly as much as Zen 3 to keep all cores running. And of course, the fact that M1 has twice as many floating point units as Zen 3 helps as well.
M1 is just the first product in the series, and a product for entry-level hardware at that. I really don't understand what is the point you are trying to make here. I mean, that would be silly to judge Zen 3 by Ryzen 3 5400U, would it?
For the record the 3995 won't use over 280W of power.Which is a plausible result for a CPU with a TDP of ~10-15W (I don't think Apple have actually published the TDP) when running a CPU-heavy task.
You can't legitimately compare the published default TDP of a Threadripper (which is what that 280W figure is) with the actual power consumption of the M1 measured while running the actual benchmark. Esp. since TDP is a notoriously shonky statistic which manufacturers will calculate... optimistically, and which can often bet configured to a different value by the system builder. That Threadripper may be drawing 150% of TDP during the test.
You're also assuming that performance-per-watt is linear - might be close to that if your test is very highly optimised for multi-threading and isn't limited by RAM or I/O), otherwise it is more likely to be "diminishing returns". Also, the Threadripper has hyperthreading (or whatever AMD call it) and the effectiveness of that will depend hugely on the specific workload.
Thermal design power - Wikipedia
AMD Ryzen™ Threadripper™ 3990X Processor
AMD Ryzen™ Threadripper™ 3990X is the world's first 64-Core desktop processor that delivers incredible performance for many demanding workloads. Learn More!Apple M1 - Benchmark, Test and Specs
Apple M1 - Benchmark, Geekbench 5, Cinebench R20, Cinebench R23, Cinebench R15 and FP32 iGPU (GFLOPS) benchmark results
If my vaguely remembered Physics serves, current goes up in proportion to clock speed (frequency/speed of switching), and power goes up with the square of current - which is why more, slower cores are theoretically better - but only if the software can be optimised to actually use all of those cores - which is hard.
Trouble with this whole argument: the M1 was primarily designed for low-power ultra-portable applications. It had one job to do - thrash the integrated chipsets used in the Intel Macbook Air and low-end MBP - and it seems to do that with a vengeance. It has considerably pushed the envelope in what these ultra-portable systems can do - which is needed when even "regular consumers" want to edit video for YouTube, use no-green-screen-required virtual backdrops in their video conferences . What it isn't is a credible replacement for higher-end personal workstations - unless you're lucky and your workload hits one of the M1s sweet-spots (e.g. certain video codecs). It doesn't have enough CPU cores - and has the wrong balance of efficiency/performance cores - or GPU cores, can't access enough RAM and can't drive more than two displays... None of which is a problem in a MacBook Air, or even a 24" iMac for personal productivity (and occasional Garageband jams) - but it's not really a consideration for people looking at a 64 core Threadripper (...which doesn't have to run off a battery and can be stuck to a cooler the size of Manhatten). We haven't seen Apple's horse for that course yet.
Yes the "consumer" 5000 series chips can pull more than rated TDP. It is probably why AMD doesn't include a cooling solution and recommends watercooling the 5950/5900.
Take a look on that thing:
www.tomshardware.com
Origin PC's NS-15 Workstation Laptop Is Ryzen 9 5950X Ready
16 Zen 3 Cores in a 15.6-inch form factor
Editor's note, Nov. 10, 11:42 a.m. ET: Origin has informed Tom's Hardware that the listing of the Ryzen 5000 series processor was an error involving its website configurator, and that it is not offering any laptops with the 5000 series chips. The original story is below.Take a look on that thing:
Origin PC's NS-15 Workstation Laptop Is Ryzen 9 5950X Ready
16 Zen 3 Cores in a 15.6-inch form factor
- Status