5 ms·

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 n

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.

kergonath 5y ago
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.
- sudosysgen 5y ago
  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-performance-review/4 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.
  - kergonath 5y ago
    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. To whom are they selling their GPUs?
    - sudosysgen 5y ago
      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.
      - wtallis 5y ago
        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.
        sudosysgen 5y ago
        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.
        wtallis 5y ago
        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.
  - klelatti 5y ago
    > 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.
    - sudosysgen 5y ago
      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. I'll give them credit for investing a lot in 5nm.

kergonath 5y ago

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.

sudosysgen 5y ago
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-performance-review/4 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.
- kergonath 5y ago
  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. To whom are they selling their GPUs?
  - sudosysgen 5y ago
    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.
    - wtallis 5y ago
      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.
      - sudosysgen 5y ago
        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.
        wtallis 5y ago
        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.
- klelatti 5y ago
  > 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.
  - sudosysgen 5y ago
    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. I'll give them credit for investing a lot in 5nm.