Intel and AMD Form x86 Ecosystem Advisory Group

[londiste]

Needed evolutionay steps, to better compete with ARM / RISCV / OTHERS:
For first step, let they standardize chipsets and sockets, so user can replace Intel CPU with AMD CPU, without changing mobo. Then second step, let create x86R (reset), with reduced instructions, well designed memory models, vector processing, usability for phones / tablets, and other aspects that are currently messed and convoluted. Final third step, license it for any other company that want use any core existing (Intel or AMD) or make their custom core and need only ISA.

- Are you seriously talking about standardizing chipsets and sockets in terms of competing with ARM/RISCV? The ecosystem for pretty much everything not x86/64 is highly custom and proprietary and manufacturer support for running anything not intended - say, some Linux distro - is anywhere between lacking and nonexistent.
- The instruction set part is wishful thinking. All x86 implementations today do micro-ops - complexity and legacy are a burden mainly on decoder. Now read your list and compare that to what for example ARM (or Apple) is doing lately, that should be a fun comparison about what direction the trend is.
- Usability for phones/tablets is not a technical question but a market one. Intel did about a generation worth of actual attempt to make phone and tablet SoCs. They were not half bad but serious efforts were discontinued primarily because the profit margin was not good enough vs doing more desktop and data center stuff. If they want to move into that segment, this might not be as hard of a transition as you seem to think. Same applies for AMD - the new handhelds are already halfway there, they are getting a lot of good intel about power consumption challenges.
- I do agree on x86 licensing but unfortunately that is a real mess. The main reason licensing has not been done is fear of real competition - and what better example to illustrate than Nvidia not being able to secure a license back when they tried to get one (and were not yet the behemoth of a company they are today). Lets see if the new Advisory Group changes anything in that regard.

 

[Vayra86]

Hopefully this is prelude to Intel exiting the chip design market such that the industry relies on AMD products manufactured by Intel fabs.

Euh, sounds great, a bit like one step forward and five steps back, that.

Lots of armchair CEO’s ITT.

If it were up to TPU we would have ended world hunger, war, and general discomfort decades ago. Duh.

 

[Vya Domus]

and we saw the recent Windows laptops that are ARM based CPUs

Yeah and they kind of suck, AMD APUs are way better, Apple can use ARM effectively in their products because they do have the aforementioned money and IP to pull that off but they still only make up a tiny share of the total market.

 

[TheLostSwede]

- currently both AMD and Intel chipsets are pretty independent from CPU, and communicate with it using PCIE under diffrent names. Currently at hardware level there is nothing to prevent use Ryzens with Intel chipsets and vice versa. Making PCH as CPU-vendor-independent, could be best way to push truly innovative features to be more competitive., For now they just use 'proprietary refresh'.

You'd think, but it's not quite that simple. The main bus, sure, it's all PCIe, but all the other interfaces such as SPI (or DeSPI as Intel is using) and all the little custom bits that makes the chipset power up alongside the CPU and so on. You have to remember that the chipset block diagrams are vastly simplified and there are more things than just PCIe going on. Just look at the Thunderbolt and USB4 host controllers, they require that extra cable which is home to at least a couple of different interfaces (including USB 2.0, which isn't relevant in this case).

- long use AM4 socket do not prevented adding new features. At Intel you have insanity with LGA 1156, 1155, 1150, 1151 and 1200. Do few pins difference really justify any innovation barriers ? On most mobos you can find third party chipsets for sound / wifi / lan etc

Socket, sure, but it was still a limiting factor and if we look at the AM4 chipsets, there was a huge difference between the X370 and X570 chipset, we went from PCIe 2.0 to PCIe 4.0, something that couldn't have happened without changing the chipset, something you're suggesting shouldn't happen, at least from how you wrote your initial post.
Yes, Intel is changing more than should be required, but AMD only allowed some of the backwards compatibility after there was a community uproar over AMD saying they wouldn't do it. Has that ever happened to Intel?

The reason for third party peripheral chips is because both AMD and Intel lacks the expertise and/or don't want to compete in low margin businesses. Those are also far less complex than a motherboard chipset. As an example, when USB 3.0 was all the rage, AMD made an awful integrated USB 3.0 host controller in their A68 chipset I believe and all their so called FCH chipsets that was slower than any of the third party options in the market. At least they didn't make this mistake with USB4.

- x86S is just too small change. They should make new ISA with 20% instructions, clear programming model and without legacy burden. Then OS / apps will begin to use new ISA, while old soft continue to use current legacy ISA. At some point legacy ISA could by dunped from hardware and be accesible only via emulation.

Well, that's your opinion and there's no way you could drop 80% of the instruction set, so I'm not even going to discuss this with you.

If it were up to TPU we would have ended world hunger, war, and general discomfort decades ago. Duh.

Indeed, as we would've eradicated humankind...

 

[Zareek]

Dare I say, grant Nvidia a license


Horrifically bad idea to allow AMD to become Intel mk II. We need more players not less. Look at what has happened due to Nvidia's dominance.

I totally agree on the second part, but Nvidia is not the company we want making CPUs. They've proven over and over that they will use their market position and power to push everyone around, including consumers. All corporations are not to be trusted and put profits above all else. Nvidia is however, dare I say it, EVIL! Intel isn't much better, but it's been a while since they had the market power to be consumer abusive.

 

[igormp]

For first step, let they standardize chipsets and sockets, so user can replace Intel CPU with AMD CPU, without changing mobo.

That's irrelevant, can't really standardize chipsets and sockets when those are related to features that need updating in the end. No other CPU features this.
You seem to be only thinking of desktop usage, whereas this partnership is more about the ISA itself, and the main selling points of x86 are server and laptops, desktops are just an afterthought at this point.

Then second step, let create x86R (reset), with reduced instructions, well designed memory models, vector processing

X86S already goes away with some legacy boot stuff, I guess this group will make it easier to coordinate a push to this update.
What other memory model do you want to see in the ISA?
What other vector processing apart from SSE/AVX/AVX512 would you like to see?

usability for phones / tablets, and other aspects that are currently messed and convoluted.

Phone/tablet stuff is not really related to the ISA at all, but rather how AMD and Intel design their chips.
Apple's M lineup could've been x86 as well and would've been great. ISA is just a matter of front-end and decoder for your compiler to target, after all.

Final third step, license it for any other company that want use any core existing (Intel or AMD) or make their custom core and need only ISA.

That I agree with. Nvidia for sure would be interested in licensing such core.

Intel tried phones and lost billions on it, x86 isn't suitable for that form factor, let's not go there.

ISA has nothing to do with that, it's just a matter of the underlying µarch. Intel just shoved a crappy desktop-level CPU in there and tried to run with it.

Licensing cores aren't going to work, neither company has designed their architectures for something like that and it would not be cost effective, as x86 chips are not like Arm or RISC-V that was designed for modularity.

They haven't done so, but nothing would stop them planning for that. Licesing the ISA itself is also an option, ARM makes a profit from both ways.

ARM cannot ever really succeed in displacing x86 in that market because it's not backed by any huge corporation with billions to spare on R&D and the IPs of AMD or Intel, that's what you need to consistently deliver competitive chips year after year.

Introducing Grace

A CPU Purpose-Built for Servers Running Terabytes of Data

www.nvidia.com

Google Cloud Introduces Tau VMs | Google Cloud Blog

Compute Engine’s new Tau VMs based on AMD EPYC processors provide leading price/performance for scale-out workloads on an x86-based architecture.

cloud.google.com

Loading…

www.oracle.com

ARM Processor - AWS Graviton Processor - AWS

AWS Graviton processors deliver the best price performance for your cloud workloads, optimized for a range of general-purpose, compute, memory, and storage-intensive workloads.

aws.amazon.com

Those look pretty big to me. But I don't get the fixation folks have with one ISA replacing another, nothing stops both from just co-existing, it's as simple as that.
ARM CPUs are just another CPU, pretty much like how you can pick between Epyc or Xeon when spinning up instances in your favorite cloud provider.

They should make new ISA with 20% instructions, clear programming model and without legacy burden.

And see no adoption whatsoever because no one is going to recompile everything for this new ISA without backwards compatibility for their current software?
And what benefit would such reduced ISA have? Even ARM keeps piling up new instructions with each new revision.

 

[TheLostSwede]

ISA has nothing to do with that, it's just a matter of the underlying µarch. Intel just shoved a crappy desktop-level CPU in there and tried to run with it.

Well, it clearly has something to do with it, as it simply doesn't seem to be possible to "compact" it down in the same way Arm has with their ISA.

They haven't done so, but nothing would stop them planning for that. Licesing the ISA itself is also an option, ARM makes a profit from both ways.

Well, what's stopping them is that Intel and AMD owns different part of the ISA and if one or the other company says no, it's not happening. x86/amd64 is not unified, nor controlled by a single entity, which is why it would be a lot harder to license. Obviously we used to have Centaur/VIA doing their stuff, but that was always a bit janky and now that's in the hands of Zhaoxin, but it's not exactly progressing at any sensible pace, so it's way behind and simply not competitive.

 

[igormp]

Well, it clearly has something to do with it, as it simply doesn't seem to be possible to "compact" it down in the same way Arm has with their ISA.

No, that's a matter of µarch. Intel did a shitty µarch back then, that was it.
ARM has many different µarches going around, from big server ones (such as the neoverse cores), to smaller ones, all under the same ISA.
You won't be getting an AmpereOne core compact enough to a phone, as an example.

Well, what's stopping them is that Intel and AMD owns different part of the ISA and if one or the other company says no, it's not happening. x86/amd64 is not unified, nor controlled by a single entity, which is why it would be a lot harder to license.

Yeah, hence why this advisory group and them getting into agreement. If that will ever happen, or if they'll consider this a good idea, who knows.

 

[Nhonho]

ARM is rising and in 10 years there could be nothing but ARM machines.

Yes, and that's the choice of a new generation...