The M1 is an Arm+ in the sense that it supports a stronger memory model option that plain Arm doesn’t, and it’s this feature that is used by Apple’s OS and the Rosetta translation engine that gives the performance boost for x86 translated programs.
Of course the Rosetta isn’t needed by Linux and assuming that all software is recompiled for Arm natively then you probably wouldn’t need it, but I suspect that is what the GP was referring to.
It may be a dud, it may be a stud. If they target end user machines on windows it will probably be a dud, but for AWS this might be a big step up from graviton. Actually has anyone tried to compare graviton vs m1?
AMD has been selling the (64 bit Arm) Opteron A1100 series for a while now. It honestly, needs updated as it's starting to get a little old. However, overall interest in these chips was not as high as hoped.
AMD has experimented with ARM in the past and is selling some ARM 64 chips. such as the Opteron A1100 series chips. However, purchase interest has seemed limited. So AMD definitely has been working on ARM. They may dedicate more resources now that it's a bit of buzz because Apple. Theses chips definitely could use an update. However, like I said in the past interest seemed limited, but these chips definitely could use an update.
Years and years ago they originally tried to release a ARM chip alongside ryzen that was supposed to be pin compatible to their x86 chips. They abandoned that effort so as to focus on ryzen.
AMD please please make a ARM cpu that I can just drop into my desktop mobo.
AMD's strategy has been to focus on a narrow set of initiatives that they can execute well, rather than extend themselves into a bunch of different well-defended markets (as Intel has done to their detriment). In that regard, AMD developing an "M1" competitor, when they already have class-leading x86 processors, doesn't make sense.
Renoir -- AMD's current-generation mobile and desktop processors -- is already competitive with M1 (although integrated GPU performance is lackluster), and their next-generation Cezanne processors are right around the corner.
On the server side (since that's mentioned elsewhere in this thread), AMD previously released ARM-based Opteron processors, but they were slow and unpopular, and further work on that lineup was cancelled in favor of EPYC. Given that EPYC has single-handedly killed the entire ARM server ecosystem practically overnight (with the survivors being relegated to custom processors for supercomputers and hyperscalers), it's highly unlikely that AMD would revive their ARM efforts for this market.
The only situation where an ARM-based AMD processor would make sense, would be as a custom processor where someone else is footing the R&D and manufacturing bill. The only company that I could potentially see interested in such a thing (and would want an ARM-based processor specifically) is Apple themselves -- they need a high-performance ARM processor for the Mac Pro/iMac Pro, and a Threadripper variant with an ARM ISA might make more sense than developing their own in-house high-performance processor for a low-volume line. But even then, it would be an M1 complement, not a competitor.
Overall, I wouldn't put too much stock in this particular rumor.
Supplying the chips for the Xbox and playstation consoles has been pretty successful for amd. I suspect both Microsoft and Sony will be looking long and hard at an arm soc for their next generation consoles and it would be wise for AMD to have an answer to that when it comes.
Why would they if AMD's x86-64bit offering remaining competitive, affordable, will offer easy backwards compatibility and easy PC game porting for developers?
Sony and Microsoft are going nowhere but back to AMD/x86 if the next generation of console remain to same and the whole concept doesn't change drastically.
Is ARM transition worth? Maybe both arch doesn't matter for develop new games, but x86 is better to easily achieve PS4/5 Xbox One/Series compatibility.
ARM's dominance in the mobile and low-power market was largely a historical accident. In particular:
- At the dawn of the smartphone market, existing x86 platforms were designed for high-performance applications that were too power-hungry for a mobile phone.
- Intel wasn't interested in pursing low power mobile phone applications, and sold their XScale line to Marvell.
- AMD couldn't afford to invest in low power mobile phone applications, and sold their entire handset division to Qualcomm.
- VIA wasn't in a position to invest in much of anything, much less mobile phones.
- MIPS was stagnant and didn't really have a clear roadmap or future, and RISC-V didn't exist yet.
- ARM was available for third-parties to license, whereas x86 was not.
As for efficiency, modern multi-core processors spend the overwhelming majority of their transistor budget on logic units and SRAM -- the cost of decoding instructions is trivial. In addition, many of the more esoteric x86 instructions are emulated in microcode. As such, I don't see how AArch64 would be inherently more efficient than x86-64.
20 comments
[ 2.6 ms ] story [ 51.3 ms ] threadOf course the Rosetta isn’t needed by Linux and assuming that all software is recompiled for Arm natively then you probably wouldn’t need it, but I suspect that is what the GP was referring to.
System on a chip is nothing new.
What AMD has going for it is software choices. Maybe microsoft plus linux.
It could mean "any kind of ARM based SoC". Or "an ARM based SoC with x86_64 memory model emulation".
It could however also mean "a tightly coupled package with CPU, GPU and RAM" - even if that's x86 based.
Years and years ago they originally tried to release a ARM chip alongside ryzen that was supposed to be pin compatible to their x86 chips. They abandoned that effort so as to focus on ryzen.
AMD please please make a ARM cpu that I can just drop into my desktop mobo.
Renoir -- AMD's current-generation mobile and desktop processors -- is already competitive with M1 (although integrated GPU performance is lackluster), and their next-generation Cezanne processors are right around the corner.
On the server side (since that's mentioned elsewhere in this thread), AMD previously released ARM-based Opteron processors, but they were slow and unpopular, and further work on that lineup was cancelled in favor of EPYC. Given that EPYC has single-handedly killed the entire ARM server ecosystem practically overnight (with the survivors being relegated to custom processors for supercomputers and hyperscalers), it's highly unlikely that AMD would revive their ARM efforts for this market.
The only situation where an ARM-based AMD processor would make sense, would be as a custom processor where someone else is footing the R&D and manufacturing bill. The only company that I could potentially see interested in such a thing (and would want an ARM-based processor specifically) is Apple themselves -- they need a high-performance ARM processor for the Mac Pro/iMac Pro, and a Threadripper variant with an ARM ISA might make more sense than developing their own in-house high-performance processor for a low-volume line. But even then, it would be an M1 complement, not a competitor.
Overall, I wouldn't put too much stock in this particular rumor.
Sony and Microsoft are going nowhere but back to AMD/x86 if the next generation of console remain to same and the whole concept doesn't change drastically.
Plus on the XBox side, one of the plus is being able to have the titles on Windows store as well, without much effort to the devs.
That is wishful thinking.
The Switch is already using ARM from NVidia.
- At the dawn of the smartphone market, existing x86 platforms were designed for high-performance applications that were too power-hungry for a mobile phone.
- Intel wasn't interested in pursing low power mobile phone applications, and sold their XScale line to Marvell.
- AMD couldn't afford to invest in low power mobile phone applications, and sold their entire handset division to Qualcomm.
- VIA wasn't in a position to invest in much of anything, much less mobile phones.
- MIPS was stagnant and didn't really have a clear roadmap or future, and RISC-V didn't exist yet.
- ARM was available for third-parties to license, whereas x86 was not.
As for efficiency, modern multi-core processors spend the overwhelming majority of their transistor budget on logic units and SRAM -- the cost of decoding instructions is trivial. In addition, many of the more esoteric x86 instructions are emulated in microcode. As such, I don't see how AArch64 would be inherently more efficient than x86-64.