Trumpstyle said:
derpysquirtle64 said:
"It will be AMD who builds the SOC, not Sony/MS. Most likely ARM(Cpu) + RDNA(gpu)."
But what if AMD would not want to build ARM-based SoCs? Yes, as I've mentioned in my post, I don't see Sony being able to create their own ARM SoC. So they would have to rely either on Nvidia with Tegra or AMD (if they'll jump on board with ARM). Another options can be Qualcomm or Samsung I suppose. By the way if I'm not mistaken Samsung is partnering with AMD for some of their newer Exynos SoCs to improve graphics. I think I've heard it somewhere.
"Second point, the main advantage ARM has over X86 is its ipc advantage, I don't know much about this but just what I read from very highly knowledgable people, X86 will lose against ARM because it's basically stuck at 4 decoders, it's very hard for a x86 cpu to go beyond this while ARM can go unlimited. Skylake from intel supposedly have 5 decoders in some press news but this is inaccurate, it has 4 decoders same as zen3. M1 has 8 decoders. It's because of the decoders ARM can have much higher ipc (Perf/ghz).
M1 has around 1.53x higher ipc than Zen3, the new snapdragon that will launch soon around 15%."
I'm not really sure what is meant here by decoders, I guess it is probably refering to GPRs (General Purpose Registers), but not sure. The issue is with IPC comparison. It's not indicative of performance at all. Because x86 is a CISC processor and ARM is a RISC processor. They have completely different instruction sets. Just an example from the internet, in x86 assembly a=b/c operation looks like this:
That's the perfect example of why you can't compare IPC between ARM and x86, as with ARM's reduced instruction set, it is usually required to run more instructions compared to x86 to perform the same operation.
|
If Amd won't build them a ARM CPU I don't think they will go arm, but I don't expect this will be a problem as AMD can just license ARM cpu from ARM. They don't need to build a custom CPU themself.
For decoders stuff, I just have to use google translate from sweden as it's only way I can find quick source:
"What makes Apple's design so potent is that the simplicity of ARM is complemented by a very broad design, where eight instructions can be initiated (issue-width) per clock cycle. In comparison, the x86 camp is much narrower, where until recently both Intel and AMD were limited to being able to decode a maximum of four instructions per clock cycle. In terms of how many instructions can be initiated with the latest generation of Cove architectures, Intel has broadened its design from four to five, while AMD and the Zen 3 architecture stay at four."
https://translate.google.com/translate?hl=en&sl=auto&tl=en&u=https%3A%2F%2Fwww.sweclockers.com%2Fartikel%2F30952-apple-m1-vi-synar-apples-forsta-arm-processor-i-sommarna%2F2
When I compare ipc, I mean real performance 1 cpu core in benchmarks like cinebench, Geekbench 5 and the very higly respected SPEC2006 that pc users love.
https://www.anandtech.com/show/16252/mac-mini-apple-m1-tested/4
M1 is clocked at 3.2 ghz and Zen3 5950x clocked above 4.5ghz. I actually haven't done the math I just made the assumption that A13 has 1.8x higher ipc than Zen2.
|
