Ampere Is Designing Their Own Arm Server CPU Cores, Coming In 2022

[pal666]

x86 is RISC under the hood since decades. compatibility layer is relatively small part of the die.

its interface to program is handicapped and you can't get around it. 5 available registers is a joke

[pal666]

will x86 survive soon ?

microsoft wants to make an ARM server.
apple wants to make an ARM server.
nvidia says you don't need x86 cpu because you use our DPUs in the ocp. They also own arm cpus.
intel will make x86 cpu and then who will sell it? what if microsoft says we just do our products for our ARM server and that's it.
apple makes their os for their ARM server.
you also have thunderx3 arm and fujitsu arm.
amazon has it's own arm cpu.
google will probably copy and make their own arm cpu?

will intel drive linux x86 alone ?

well, the answer is clear from your list. microsoft doesn't want to make an arm personal computer, so nothing threatens x86 there

[smitty3268]

its interface to program is handicapped and you can't get around it. 5 available registers is a joke

Register renaming mostly works around that.

[drakonas777]

will x86 survive soon ?

microsoft wants to make an ARM server.
apple wants to make an ARM server.
nvidia says you don't need x86 cpu because you use our DPUs in the ocp. They also own arm cpus.
intel will make x86 cpu and then who will sell it? what if microsoft says we just do our products for our ARM server and that's it.
apple makes their os for their ARM server.
you also have thunderx3 arm and fujitsu arm.
amazon has it's own arm cpu.
google will probably copy and make their own arm cpu?

will intel drive linux x86 alone ?

It's not about what some specific companies want. It's about factual global market share. Until ARM has at least 20% of enterprise I won't consider it a threat to x86. Part of the enterprise won't event switch from Xeons to EPYCs due differences in the same ISA implementations from two different vendors despite the fact that AMD has been offering almost 2x benefits (power/efficiency/density and whatnot) in some cases. Good luck talking such companies into switching to ARM

Future most likely gonna be heterogeneous. Some part of the share will be ARM (mostly in a form of in-house chips at megacorps), other x86.

[edwaleni]

I used X-Gene and eMAG and breathed a sigh of relief when Ampere went to Neoverse, which generally does not suck, for Altra. I sincerely hope the new Ampere cores are not like the old APM cores were.

X-Gene/eMag were inherited from Applied Micro when Ampere bought it from them and are legacy. Neoverse/Altra is their first gen product to get them established in the performance market.

[edwaleni]

Ampere is already a GPU architecture of Nvidia, so incoming lawsuit in 3.. 2.. 1..

Always nice to se unnecessary and stupid lawsuits though, so bring it on, dudes!

The use of "Ampere" by NVidia is simply a code name for a particular project and is not a trademark. Unfortunately the press repeats these project names constantly (like Intel's Alder lake, Ice lake, etc.) so that people begin to think they are actual product names. Carl Sagan sued Apple for calling one of their internal projects "Sagan". (He lost)

NVidia is also an investor in Ampere and if their ARM acquisition goes through will be a the largest investor of the firm ahead of Lenovo and a few others.

[schmidtbag]

Opening the door is one thing, getting your new fancy sofa inside and up the stairs is quite another.

For historical reason general purpose desktop hardware need to support some degree of x86 binary for some fairly low-level stuff (certain printer peripheral drivers for example) . But because ARM and x86 have slightly incompatible memory models, near-native performance requires a degree of hardware support for a translation mode.

Agreed. But I mentioned binfmt_misc mostly as a last resort. I think of it like Wine: you don't want to depend on it, but it has a good chance of helping you out if you need it. Lately, more and more programs are being ported to ARM. Using your printer example (which I've actually been affected by), that too has become less relevant since many printers work over a network and don't need specialized drivers.

[pal666]

Register renaming mostly works around that.

register renaming has zero relevance to that. it allows out of order execution, but if your code says "push register a to stack"(because you need sixth value in a register), some register will be pushed to stack

[coder]

Ampere, if you're listening... partnering with another outfit to make a stripped-down 8-core version on some standard form factor with PCIe (mini-itx?) and UEFI would help drive adoption.

Unlike RISC V, the beauty of ARM is that there's a very high degree of uniformity to the implementations. So, if you want a decent ARM desktop, you could use a CPU from pretty much anyone, and then take your code and run it on a ARM server CPU, with reasonable expectations about compatibility and performance.

That said, there are some optional ISA extensions, but not too many that show up in userspace code. The main exception to that is various SIMD extensions. ARMv9 resets the baseline ISA, though I think Siryn is arriving a little soon to be ARMv9.

Just being able to rack one up as a DEV box or get these into nerds' hands will drive adoption in ways that aren't immediately apparent.

I'm currently getting by with an ODROID-N2+. I'd upgrade it to something more PC-like, if I could, but I'm not anywhere close to buying a Mac Mini,

MediaTek announced it's licensing Nvidia GPU IP for future SoCs and plans to incorporate them in SoCs destined for SFF PCs, in addition to the standard places their chips go.

[coder]

it's unclear how do you imagine competition based on process node. do you think any server customer goes like this "system x uses 6nm process, system y uses 5nm process, obviously 5nm is better, we should select it"? and then "oh, we have two 5nm alternatives, now it's time to compare price and performance"

Um, I took gigi 's post to mean they'll be competing for 5 nm production capacity.