Why bother at this point? Even the most power-efficient x86 design is going to get smoked in performance per watt (if not raw performance) by Apple Silicon.
1. AMD and apple use different methods for measuring effect. So AMD 35W and apples 18W are not directly comparable.
2. Despite what HN top posts want you to believe newset Zen3 has a performance advantage over M1. Hence at this point it make sense to concentrate on reducing power usage.
3. AMD64 is basically internally a RISC.
4. Why does _any_ post on _anything_ quickly turns into praising apple? You are stealing attention from the real discussion.
The above instruction computes quite a lot of stuff (first the read address, then 8-wide multiply+add), yet on most modern CPUs including AMD Zen 2 it decodes into a single micro-op.
More often the opposite happens, AMD64 instructions are merged into a single micro-op, this is called macro-op fusion.
Ah yes why bother. Why didn't AMD just give up when it was down and Intel processors smoked them. Why didn't Intel just give up when AMD processors where better. It's a matter of time until AMD, Intel or a different company smokes Apple silicon. If everyone would just gave up at the drop of a hat it would mean stagnation.
Because Apple silicon is not for sale? (unless you think buying 3k laptops so you can harvest their CPU is some kind of sane business model)
There are plenty of things you could use such processor in.
Unless Apple decides and starts selling it's own silicon, their performance doesn't matter at all to people who need to source CPU to build whatever they are building.
You can't ship "but in benchmarks it beats competition"
x86 tends to have more I/O than ARM Phone SoCs where a reasonably powerful CPU is simply trapped in the SoC with no meaningful way to take advantage of it either in laptops or for desktops. The Raspberry Pi has 1 PCIe 2.0 lane. A lot of x86 chips have 4 lanes and usually at least PCIe 3.0 if not PCIe 4.0.
The other problem is that "Apple Silicon" is just one slice of the market. When you think about how you didn't even mention ARM it kind of shows that you have very little confidence in ARM yet you still bash x86.
Well, the guy in the video calls it "AMD van go". If you're going to use someone's name, at least learn how to sort of pronounce it: https://youtu.be/z2YU87AqDcU?t=62
Yes, I've been looking for something like that. I think there are two great under-explored use-cases for this:
1. Laptop ergonomics are really poor for taller people who need to hunch over to see a screen. Decoupling the screen from the compute unit would help here, while also enabling users to mix-and-match compute and display technologies, and upgrade them independently.
When compared to something like the Quest 2, this would allow for a much more comfortable headset due to reduced weight and size, while providing a much more performant system.
14 comments
[ 2.4 ms ] story [ 49.6 ms ] thread1. AMD and apple use different methods for measuring effect. So AMD 35W and apples 18W are not directly comparable.
2. Despite what HN top posts want you to believe newset Zen3 has a performance advantage over M1. Hence at this point it make sense to concentrate on reducing power usage.
3. AMD64 is basically internally a RISC.
4. Why does _any_ post on _anything_ quickly turns into praising apple? You are stealing attention from the real discussion.
> 3. AMD64 is basically internally a RISC.
They decode some rarely used instructions into multiple micro-ops, but these micro-ops can be very complex still. Example:
The above instruction computes quite a lot of stuff (first the read address, then 8-wide multiply+add), yet on most modern CPUs including AMD Zen 2 it decodes into a single micro-op.More often the opposite happens, AMD64 instructions are merged into a single micro-op, this is called macro-op fusion.
There are plenty of things you could use such processor in.
Unless Apple decides and starts selling it's own silicon, their performance doesn't matter at all to people who need to source CPU to build whatever they are building.
You can't ship "but in benchmarks it beats competition"
The other problem is that "Apple Silicon" is just one slice of the market. When you think about how you didn't even mention ARM it kind of shows that you have very little confidence in ARM yet you still bash x86.
* will be based on the TSMC 7nm process node and will feature a monolithic design composed of the CPU, GPU, and I/O IPs
* includes the Zen 2 CPU and RDNA 2 GPU architectures
* Van Gogh APUs will feature a total TDP of around 7.5-15W
* designed for the FF3 socket which is a compact low power socket to compete against Tiger Lake-U
I am looking forward to more passively cooled high performance laptops in near future :)
1. Laptop ergonomics are really poor for taller people who need to hunch over to see a screen. Decoupling the screen from the compute unit would help here, while also enabling users to mix-and-match compute and display technologies, and upgrade them independently.
2. That same compute unit could also be paired with a VR headset, creating mobile VR not unlike what MagicLeap was doing with their compute units: https://www.roadtovr.com/wp-content/uploads/2018/07/magic-le...
When compared to something like the Quest 2, this would allow for a much more comfortable headset due to reduced weight and size, while providing a much more performant system.