Apple doesn't need to manufacture their stuff at TSMCs latest protest, neither do AMD or Qualcomm or NVidia.
Taking the offer with the most money attached is not a good idea in the long term. They're currently providing unprecedented leverage to Apple.
Apple Silicon needs the performance crown as the comment you're replying to mentioned. Having a good market share in the premium sphere is not good enough - if instead of trading blows but still ending up ahead of AMD, AMD processors also were on 5nm with DDR5 memory they'd be plainly slower. Then Apple would be in the position where their laptops are slower and have worse battery life than the competition and also Docker doesn't work right for developers and some of your VSTs don't work at all, and few people would take the pain of changing architecture for worse performance.
I find quite interesting how you frame Apple designing good SoCs as a nefarious plot to take over the world and somehow kill companies they’re not competing with. That’s some impressive mental gymnastics.
The architecture itself is worse than it's competitor. Even with a node advantage and 3x faster RAM at the same power consumption, it cannot beat a 5980HS in single thread by a large margin (https://www.anandtech.com/show/17024/apple-m1-max-performanc...) and it can only beat it in multicore by using almost twice as much power. In the phone space, their chips are also
only ahead by 10-20% despite a huge process advantage (35% higher density).
Their SoCs are pretty good. But they're only good because Apple has almost exclusive access to TSMC 5nm.
I'd love to hear about how the M1 isn't competing against AMD and NVidia.
> The chips here aren’t only able to outclass any competitor laptop design, but also competes against the best desktop systems out there, you’d have to bring out server-class hardware to get ahead of the M1 Max – it’s just generally absurd. (Anandtech as you linked to)
You’re making claims about hypothetical AMD / Nvidia performance at different nodes which are impossible to verify.
And Apple is clearly taking a huge risk by investing massively in the latest node. At the minimum they deserve credit for that.
They're pretty easy to verify. Find the average difference in performance at a given TDP and substract the efficiency improvement, this breaks down in single-core somewhat but not in multi-core.
There is very little speed differential between desktop and mobile single-core performance nowadays. The M1 Max is nowhere even close to the best desktop systems which would be the 3990X or the upcoming 5990WX. Despite having much faster RAM - which is the biggest reason why it can perform so well in FP workloads - in integer workloads it cannot compete. I'm citing Anandtech on their data, not on their opinion.
How do you go from “we found one CPU that has similar performance” to “their core is worse”? The source you quoted says that the M1 has a 20% performance lead on average, so you’ll have to define “significant margin”. The M1 package also almost never reaches 35W without stressing the GPU.
In any case, if someone said that the M1 was unquestionably faster than anything else on Earth it was not me. It does not change the point that asserting that better is somehow worse is still stupid.
> I'd love to hear about how the M1 isn't competing against AMD and NVidia.
The M1 Max exceeds 35W for all but one test and even averages 62W for one test that does not use the GPU.
The point is not that it is not faster. A 20% performance advantage on average given 3x faster memory and an up to 50% more efficient process means that the architecture is slower on average but that it makes it up in (third-party) memory and process.
>To whom are they selling their GPUs?
A lot of professionals that would otherwise buy high-end AMD or NVidia GPUs for now, and eventually they will probably be used for virtualization.
Please stop throwing around that 62W wall power measurement when discussing package power software estimates. You aren't taking enough caution to interpret or present disparate measurements in a valid context.
Some of the conclusions you're presenting in this thread are correct. But you're doing a terrible job of justifying them despite the available data, and making unnecessary exaggerations. This discussion deserves a bit more rigor.
It's not wall power consumption. It's wall minus idle. Package power consumption metrics are often unreliable. If you have another explanation for where that power went I'm all ears, but my experience tells me that the on-board power consumption meter is simply off.
Actually looking closer at the data, the reported package power is often even higher than wall minus idle, which means it's almost certainly inaccurate.
So you do understand at least some of the limitations in the different measurement methodologies, but you still choose to compare with the less similar of the two available numbers?
If you want to respond to someone who specifically referred to package power reported on Apple's chip, or if you want to make comparisons against package power and TDP reported on an AMD chip, why do you choose to respond with the wall power measurement? Subtracting out idle power doesn't remove all the potential sources of error from measuring at the wall, and in particular it cannot remove the error introduced by inconsistently including all the inefficiencies of converting from wall power to the low voltage DC the chip actually runs on. You seem to be disingenuously cherry-picking by going with the 62W wall power rather than the corresponding 44W package power measurement that was published a few pixels above it.
AMD CPUs typically have a very accurate SoC power measurements that's within 95% of reality. Obviously that is not the case for the M1 Max, which is not abnormal. This is probably because of the offsetting Apple does to avoid counting RAM power consumption, storage power consumption, etc..., that on the M1 Mac Mini amounts to something like 5W.
On the other hand, using wall minus idle also has some advantages - you don't take into account the baseline power consumption of the memory controller, the baseline power consumption of the GPU, or the PCI/storage controllers, or the baseline power consumption of the USB controllers, and so on and so forth - that's easily 2-4W even on an M1, on a Ryzen chip this is included in package power, whereas it's not for the M1 Max.
Because of that I judged the two numbers to be the closest to each other, and the minimal power consumption losses are more than offset by excluding a fair amount of power consumption that is reflected in the package for an AMD chip (and indeed any chip).
I didn't explain and justify all of this because I didn't have the time to do so and I don't think it was necessary to clutter up the text this much. I'm not cherry-picking either. There is a reason why wall-power is included for Intel and M1 reviews but very rarely for AMD, including in AnandTech's reporting, and it's obvious from reading the Apple data on idle power usage for the M1 Mac Mini, from everything that the M1 chip does on-package and from the 7.2W idle, from which only the screen, fans and NAND chips (not controller) are outside the package, that this is not a package power reading, but an estimate of what on the PC World would be called "Core+SOC", which is typically significantly lower than actual package power. If there was a Core+SOC number provided for the 5980HS I'd compare that but I don't have one and I didn't find any. In either case in the real world the two measurements are going to be at most 1-2W apart and I'm not sure if it advantages the M1 Max or the 5980HS.
> There is a reason why wall-power is included for Intel and M1 reviews but very rarely for AMD, including in AnandTech's reporting,
So now you're casting aspersions, too? The AnandTech article you've referred to numerous times explains exactly why wall power measurements were not included for the AMD chip, and the reason is not what you're trying to insinuate.
As for your point about charger efficiency: there's more than one step of voltage conversion to get from AC wall power to the CPU die. The losses at each step add up.
Taking the offer with the most money attached is not a good idea in the long term. They're currently providing unprecedented leverage to Apple.
Apple Silicon needs the performance crown as the comment you're replying to mentioned. Having a good market share in the premium sphere is not good enough - if instead of trading blows but still ending up ahead of AMD, AMD processors also were on 5nm with DDR5 memory they'd be plainly slower. Then Apple would be in the position where their laptops are slower and have worse battery life than the competition and also Docker doesn't work right for developers and some of your VSTs don't work at all, and few people would take the pain of changing architecture for worse performance.