3 ms·

>Are PC laptops going to catch up? x86 has to scale from embedded boards to laptops to servers. AMD made different trade offs than Apple. Apple chose to achie

>Are PC laptops going to catch up? x86 has to scale from embedded boards to laptops to servers. AMD made different trade offs than Apple. Apple chose to achieve a level of performance by keeping the frequency low and having a high IPC. High IPC means that cores consume a larger size of the die. That helped with lower power consumption. x86 makers like AMD, did the other way around. They dedicated a smaller portion of the die to a core. That makes it cheaper. To make up for that, they used a higher frequency, which also means a higher power consumption. Apple also has a node advantage, which also makes their CPUs more expensive. AMD could totally do what Apple did: have faster IPC at lower power consumption if they would have release more expensive CPUs targeted at the mobile market. But AMD and Apple have different goals. If we stick to the mobile world, Qualcomm is doing with their Orion cores just what Apple did with their M cores. In the first half of this year we will see laptops with Orion cores that will have both strong IPC and low power consumption.

aurareturn 3y ago
>AMD could totally do what Apple did: have faster IPC at lower power consumption if they would have release more expensive CPUs targeted at the mobile market. Doubtful. I'm not sure they have the talent/knowledge to pull it off while Apple Silicon is a moving target with a new generation introduced in each new iPhone.
- DeathArrow 3y ago
  AMD and Apple are in a different business. What makes sense for Apple doesn't make sense for AMD. And AMD employed some of the most talented people in the area like Jim Keller. They totally can do it. Intel, too. Qualcomm is doing it with Orion and Qualcomm has less expertise and resources.
  - BlackFingolfin 3y ago
    Well, Qualcomm at least claims that. Whether they can deliver is to be seen. Wouldn't be the first time they fail to do so...
  - caycep 3y ago
    Granted, Keller was at Apple for a while too...
  - aurareturn 3y ago
    Qualcomm literally bought the team if Apple engineers who led the M1 design in Nuvia. That’s Oryon. For the last 10 years, Qualcomm couldn't come close to Apple designed SoCs. Now they might come close after buying a team of ex-M1 designers. It sounds ridiculous to say that AMD, Intel, and Qualcomm can do what Apple is doing. If they could, they'd be able to compete with Apple Silicon now. It's been 3 years. It's like saying AMD, Intel, Qualcomm can do what Nvidia is doing if they want to.
- ActorNightly 3y ago
  Processor design isn't black voodoo. Its all configuration, most of which is now designed with CAD, and targeting particular use cases. Its kinda like the evolution of bitcoin mining-everyone was doing it with GPUs, AMDs were better because of some hardware optimization where they could hash something in less cycles, then people started making ASIC which are purpose built and the fastest possible way to run the necessary compute. There were no technical advancements here, just economic forces driving the design and manufacture of components that made making ASICs profitable. Apple did the same thing except they designed the chip and the computer as one, and they succeeded because it allowed them to optimize stuff further, and you see this in results where they are the leader but not by much, and often behind systems that are overclocked in single core performance. Power consumption on macs, also varies greatly depending on what you are running, which is again a result of optimization for particular use cases.
  - aurareturn 3y ago
    The iPhone is the most profitable product in the world. It gets massive resources to design a new SoC every single year. That resource spills over to the M series which is just based on whatever the A series does that year. AMD does not have that.
    - ActorNightly 3y ago
      And yet, despite the advantages if all the Apple silicon, its not being used to run the backbone of the web, while AMD and Intel are. Making something work well in general purpose is just as "advanced" as making something super optimized.
      - aurareturn 3y ago
        Apple Silicon isn't designed to be server chips. The M1 team thought it was a good idea but management did not want to do it. So some of the team broke off and started Nuvia to design ARM chips for servers. Qualcomm acquired Nuvia and that became the Qualcomm Oryon SoCs. Anyways, ARM chips are being used for the backbone of the internet. They're called Amazon Graviton chips. Ampere Computing also has ARM server chips. So do many Chinese hyperscalers.
lonjil 3y ago
> Apple also has a node advantage, which also makes their CPUs more expensive. M2 is on TSMC N5. AMD's Zen 4 mobile processors are made with TSMC N4. From die shots, I believe the P-cores in the M2 are about the same size as Zen 4 cores. And yet Apple's M2 laptops are a good deal more power efficient than Zen 4 laptops.
- incrudible 3y ago
  > a good deal more power efficient than Zen 4 laptops Zen4 is competitive when it comes to actual computing. When doing (almost) nothing, the Apple system comes out on top, but that also has a lot to do with its software.
  - chippiewill 3y ago
    Yeah, Apple have been massively optimizing for heterogenous cores and having efficient power states when not doing any work.
    - jwells89 3y ago
      I think macOS is also just better at actually being idle when the user isn’t doing anything. Windows seemingly always has something shuffling around in the background.
      - bee_rider 3y ago
        This is definitely part of it, as anyone who’s run Linux and Windows on the same machine can probably attest Windows is constantly doing… something? It is quite odd as it doesn’t seem to offer much in exchange for all these wasted cycles.
  - aurareturn 3y ago
    > Zen4 is competitive when it comes to actual computing. Source for this? Geekbench and SPEC heavily favors Apple Silicon in both raw performance and perf/watt. The perf per watt is 2-3x. The only CPU benchmark Zen4 wins in is Cinebench R23, which uses Intel Embree engine inside and heavily optimizes for x86 AVX instructions.
pohl 3y ago
> x86 has to scale from embedded boards to laptops to servers Isn’t this also true of ARM?
- geerlingguy 3y ago
  Yes, though Arm is probably a bit more robust at the lower end (think microcontrollers and very low-power industrial cores), and a little weaker at the high end (per-core performance on Neoverse still doesn't seem to match AMD/Intel).
rldjbpin 3y ago
Spot on assessment. In fact, AMD has gone the opposite direction by shrinking die size through their Zen 4c architecture (https://news.ycombinator.com/item?id=36458678 https://news.ycombinator.com/item?id=36458678). To me the more interesting aspect is how memory architecture will be approached going ahead by both parties. Apple can afford to go more bespoke and integrated, but how can x86 maintain a good mix between performance and design flexibility.

aurareturn 3y ago

>AMD could totally do what Apple did: have faster IPC at lower power consumption if they would have release more expensive CPUs targeted at the mobile market. Doubtful. I'm not sure they have the talent/knowledge to pull it off while Apple Silicon is a moving target with a new generation introduced in each new iPhone.

DeathArrow 3y ago
AMD and Apple are in a different business. What makes sense for Apple doesn't make sense for AMD. And AMD employed some of the most talented people in the area like Jim Keller. They totally can do it. Intel, too. Qualcomm is doing it with Orion and Qualcomm has less expertise and resources.
- BlackFingolfin 3y ago
  Well, Qualcomm at least claims that. Whether they can deliver is to be seen. Wouldn't be the first time they fail to do so...
- caycep 3y ago
  Granted, Keller was at Apple for a while too...
- aurareturn 3y ago
  Qualcomm literally bought the team if Apple engineers who led the M1 design in Nuvia. That’s Oryon. For the last 10 years, Qualcomm couldn't come close to Apple designed SoCs. Now they might come close after buying a team of ex-M1 designers. It sounds ridiculous to say that AMD, Intel, and Qualcomm can do what Apple is doing. If they could, they'd be able to compete with Apple Silicon now. It's been 3 years. It's like saying AMD, Intel, Qualcomm can do what Nvidia is doing if they want to.
ActorNightly 3y ago
Processor design isn't black voodoo. Its all configuration, most of which is now designed with CAD, and targeting particular use cases. Its kinda like the evolution of bitcoin mining-everyone was doing it with GPUs, AMDs were better because of some hardware optimization where they could hash something in less cycles, then people started making ASIC which are purpose built and the fastest possible way to run the necessary compute. There were no technical advancements here, just economic forces driving the design and manufacture of components that made making ASICs profitable. Apple did the same thing except they designed the chip and the computer as one, and they succeeded because it allowed them to optimize stuff further, and you see this in results where they are the leader but not by much, and often behind systems that are overclocked in single core performance. Power consumption on macs, also varies greatly depending on what you are running, which is again a result of optimization for particular use cases.
- aurareturn 3y ago
  The iPhone is the most profitable product in the world. It gets massive resources to design a new SoC every single year. That resource spills over to the M series which is just based on whatever the A series does that year. AMD does not have that.
  - ActorNightly 3y ago
    And yet, despite the advantages if all the Apple silicon, its not being used to run the backbone of the web, while AMD and Intel are. Making something work well in general purpose is just as "advanced" as making something super optimized.
    - aurareturn 3y ago
      Apple Silicon isn't designed to be server chips. The M1 team thought it was a good idea but management did not want to do it. So some of the team broke off and started Nuvia to design ARM chips for servers. Qualcomm acquired Nuvia and that became the Qualcomm Oryon SoCs. Anyways, ARM chips are being used for the backbone of the internet. They're called Amazon Graviton chips. Ampere Computing also has ARM server chips. So do many Chinese hyperscalers.

lonjil 3y ago

> Apple also has a node advantage, which also makes their CPUs more expensive. M2 is on TSMC N5. AMD's Zen 4 mobile processors are made with TSMC N4. From die shots, I believe the P-cores in the M2 are about the same size as Zen 4 cores. And yet Apple's M2 laptops are a good deal more power efficient than Zen 4 laptops.

incrudible 3y ago
> a good deal more power efficient than Zen 4 laptops Zen4 is competitive when it comes to actual computing. When doing (almost) nothing, the Apple system comes out on top, but that also has a lot to do with its software.
- chippiewill 3y ago
  Yeah, Apple have been massively optimizing for heterogenous cores and having efficient power states when not doing any work.
  - jwells89 3y ago
    I think macOS is also just better at actually being idle when the user isn’t doing anything. Windows seemingly always has something shuffling around in the background.
    - bee_rider 3y ago
      This is definitely part of it, as anyone who’s run Linux and Windows on the same machine can probably attest Windows is constantly doing… something? It is quite odd as it doesn’t seem to offer much in exchange for all these wasted cycles.
- aurareturn 3y ago
  > Zen4 is competitive when it comes to actual computing. Source for this? Geekbench and SPEC heavily favors Apple Silicon in both raw performance and perf/watt. The perf per watt is 2-3x. The only CPU benchmark Zen4 wins in is Cinebench R23, which uses Intel Embree engine inside and heavily optimizes for x86 AVX instructions.

pohl 3y ago

> x86 has to scale from embedded boards to laptops to servers Isn’t this also true of ARM?

geerlingguy 3y ago
Yes, though Arm is probably a bit more robust at the lower end (think microcontrollers and very low-power industrial cores), and a little weaker at the high end (per-core performance on Neoverse still doesn't seem to match AMD/Intel).

rldjbpin 3y ago

Spot on assessment. In fact, AMD has gone the opposite direction by shrinking die size through their Zen 4c architecture (https://news.ycombinator.com/item?id=36458678 https://news.ycombinator.com/item?id=36458678). To me the more interesting aspect is how memory architecture will be approached going ahead by both parties. Apple can afford to go more bespoke and integrated, but how can x86 maintain a good mix between performance and design flexibility.