What’s the fate of K12/ARM at AMD?

[kschendel]

You could make the fastest ARM CPU ever, that ate every x86 design for breakfast. And you'd still go out of business. Too much of the world runs on x86, even if it isn't the best. So Lisa Su did make the right call. And it shows. A fire breathing ARM CPU would be great to see in the wild, but there is too much software out there that would never be rewritten or recompiled to use it.

And yet if you visit some of the Mac-related forums, any number of people are convinced that the ARM-based iMac is a done deal!

I'm inclined to agree that ARM is not going to displace x86 any time soon, not even in a relatively closed ecosystem like Apple's.

 

[scannall]

And yet if you visit some of the Mac-related forums, any number of people are convinced that the ARM-based iMac is a done deal!

I'm inclined to agree that ARM is not going to displace x86 any time soon, not even in a relatively closed ecosystem like Apple's.

Apple is the only one that could pull it off for their products. And they may at some point. But for the rest of the world? It'll take a while.

 

[RetroZombie]

Who is the guy next to Jim Keller, does anyone know?

Do we have to thank tsmc again for providing the 20nm total failure process, we could even have seen the dismiss of amd for that crappy idea of mixing arm with x86.

And in all the interview nothing was clear what they were working on, i get all these ideas but:
- same core with two instruction sets (armv8/x64)
- two different cpus and platforms
- same io for both but two different cpus

I see mark papermaster very enthusiastic about it, the ceo it's the ceo so ok, lisa not very comfortable, jim with the face of i will do it all just don't know how and why, and the other guy what ever.

EDIT: Is there some extended version of the video?

 

[soresu]

Do we have to thank tsmc again for providing the 20nm total failure process, we could even have seen the dismiss of amd for that crappy idea of mixing arm with x86.

20nm was a node too far for HKMG processes without finFET - it simply wasn't providing adequate improvement on 28nm for large, high power chips.

It's certainly a question of whether TSMC are indeed doing the same at 3nm with finFET, as they are rumored to be doing while Samsung go with the new MBCFET/Nanosheet device based node for their own 3nm process.

 

[naukkis]

Do we have to thank tsmc again for providing the 20nm total failure process, we could even have seen the dismiss of amd for that crappy idea of mixing arm with x86.

Those chips was to be build with 20nm from Globalfoundries, TSMC has nothing to do with it.

 

[RetroZombie]

Those chips was to be build with 20nm from Globalfoundries, TSMC has nothing to do with it.

Didn't know that, thank you for the information, then thanks GF

 

[DrMrLordX]

GF's 20nm process was a complete disaster. It's the reason why GF had to license 14nm from Samsung in the first place.

 

[soresu]

GF's 20nm process was a complete disaster. It's the reason why GF had to license 14nm from Samsung in the first place.

TSMC's 20nm was little better, there's a good reason we had so many 28nm GPU generations.

 

[Richie Rich]

I don't agree with what you are saying about Zen borrowing anything significant from BD. There's a video out there where David Kanter is asked about that and he seemed to struggle to come up with much other than the branch predictor (?), not sure. I'll try to find the video. If anything I'd think Zen borrowed some from the Cat Cores and maybe a bit from K10, but I think it was just mostly fresh stuff.

Around 39:00 he is asked about what came from BD, and can't come up with anything. Says ALU's then walks that back and says microcode. Then goes on about planar to FinFet and saying they basically had to redo it all anyway. I take that as basically saying nothing much if anything was taking from BD/XV.

Zen1 has few significant things from BD, like dual cluster FPU concept (two 64-bit FPUs with ability to handle 128-bit AVX as well, in Zen uplifted to 128-bit to be able handle 256-bit AVX2), also SMU and separated INT and FPU. Dozer was bad due to very weak designed back-end (2xALUs) but concept itself was pretty innovative. As Keller said, they took the good things from both DB and Jaguar uarchs. Don't forget that speaking about BD legacy in Zen publicly was not good idea at that time. They needed to attract investors to great product, not to scare them with BD resurrection. IMHO that's why he spoke very vague in this topic.

I'm still having trouble trying to follow your K12 timeline. As @moinmoin pointed out it was on their roadmap alongside Zen (Though not in that image as it's ARM only).

As moinmoin mentioned their first ARM CPU AMD A1100 based on Cortex A57 was delayed two years (2014->2016). High priority project Zen was delayed half year. Side project as K12 with all those challenges (bigger engine, new ISA) was probably delayed much more, probably at least one year more than Zen. I remember there was rumor about K12 was taped out in the same time as Zen however it doesn't make sense. It could be tape out of delayed A1100 as we know today. So K12 would reach market probably between 2018-2019.

Here begins the Zen 3 thing. AMD had half finished new K12 uarch with bigger engine in 2015 and Zen 3 development began about the same time. Here starts the questions:
- was AMD rich enough to waste 3 years of engineering work on K12? IMHO no.
- was good things developed for K12 used in Zen 3 design? IMHO yes, same way they used BD stuff for Zen.
- is Zen 3 reworked K12 to x86? IMHO Probably not, but it depends on how much ""bigger" the K12 engine was.
- does it make sense to build Zen 3 on similar 4xALU K12 engine like Zen1/2 just little bigger? Probably not.
- does it make sense to build Zen 3 on bigger 6xALU K12 engine? Probably yes.

EDIT: @Olsan: You're right. BD had 2x pipes of 128-bit FMAC, Excavator 3x pipes and Zen 4x pipes. So in fact Zen FPU is even more similar to BD/XV than I thought.

 

[naukkis]

As moinmoin mentioned their first ARM CPU AMD A1100 based on Cortex A57 was delayed two years (2014->2016). High priority project Zen was delayed half year. Side project as K12 with all those challenges (bigger engine, new ISA) was probably delayed much more, probably at least one year more than Zen. I remember there was rumor about K12 was taped out in the same time as Zen however it doesn't make sense. It could be tape out of delayed A1100 as we know today. So K12 would reach market probably between 2018-2019.

You got it wrong. Zen was K12's side project as their naming suggest, AMD bet everything on ARMv8 first. They changed their priorities later.

 

[Jan Olšan]

Zen1 has few significant things from BD, like dual cluster FPU concept (two 64-bit FPUs with ability to handle 128-bit AVX as well, in Zen uplifted to 128-bit to be able handle 256-bit AVX2), also SMU and separated INT and FPU. Dozer was bad due to very weak designed back-end (2xALUs) but concept itself was pretty innovative.

Actually, BD used the same 128bit unit width as Zen1, doing 256bit AVX ops in two passes (this actually supports your argument more). Same for Jaguar BTW.
The last AMD cores to use 64bit units (doign SSE* ops in two passes) were K8 and Bobcat. K10 already had full-width SSE* units, but its weakness was that they did not support SSSE3 and SSE4.

I agree that the FPU - which was essentially SMTed so it might have been a stepping stone to SMT in Zen - and many other traits (the FPU/ALU+AGU split concept) clearly show that they were made by people previously working on BD-XV and Jaguar, clearly some DNA, knowledge and experiences had to make it in too, even if the actual implementation had to be changed to use lower-power logic in Zen. Power Management is basically an evolution of Excavator too. It is definitely a new architecture because the overhaul was from the grounds up, but it was not made on a green field (or on some external DNA, like K6 was, or theoretically mimicking apple core concepts via Keller).

You got it wrong. Zen was K12's side project as their naming suggest, AMD bet everything on ARMv8 first. They changed their priorities later.

Based on what?

 

[naukkis]

Based on what?

AMD did name their high end cpu projects to Kx. Bulldozer was K11, so K12 is at least named as direct successor.

 

[NostaSeronx]

AMD did name their high end cpu projects to Kx. Bulldozer was K11, so K12 is at least named as direct successor.

Hammer(K8) -> Greyhound/Husky(K9 refresh of K8) -> Bulldozer(K10)

K12 codename is a mountain peak, not an iteration of the above. (Hence, the image of a mountain on the slide: https://forums.anandtech.com/threads/what’s-the-fate-of-k12-arm-at-amd.2566911/page-3#post-40043293 )

AMD no longer used K for architecture but rather used the hexadecimal family number instead.
Greyhound isn't K10, while Bulldozer was K10. Which confused people because: K7 => 07h, K8 => 0Fh, K9 => 10h, K10 => 15h.

 

[Richie Rich]

If it was actually 100x faster(I know you are exaggerating, but hear me out), nevermind x86, everyone would switch to that CPU. Sure, it would take some time, but it will happen.

But you don't do it for 15-20%. Not even for 50%. Because if you want to run your business and you need to recompile everything, see if it works with existing infrastructure and they encounter serious issues, they'd say no.

And emulation has problems beyond just performance. It has compatibility issues. It'll never be 100%, it can't be. It's trying to be something its not.

Why we are talking about so nonsense things like 100x faster CPU or emulation?

• - Emulation for server products doesn't make sense as most Linux/Unix code is available in multiple ISA binaries.
• - They'd not switch to 50% faster CPU? What? Did you noticed that Amazon is working on 64-core Graviton2 (Cortex A76 core has about 86% IPC of SkyLake) so they do switch even when CPU core is 14% slower? Economically it makes sense (price and power).
• And new A77 is +22% faster and Acorn promised 25% IPC jump every year. In two years ARM will make sense also performance wise.
• And Nuvia with ex-Apple engineers will bring their monstrous +83% IPC advantage.

A13@2.6 GHz is beating the fastest (carefully binned silicon) Ryzen 3950X@4.6 GHz. This is just ridiculous.
If they would clocked their A13@3.0 GHz this will be equal to Zen 2 @5.5 GHz (lets forget about scaling as both would suffer same).
And A13@3.5 Ghz == Zen2@6.4 Ghz.
And upcoming A14(assuming +10% IPC rise over A13) clocked for laptops at 3.5 GHz is eqal to Zen2 clocked at 7.06 GHz (+102% IPC).



IMHO canceling K12 ARM core was bad decision in long term when you see Apple's core performance and power consumption. K12 could be no.1 in ARM server world no doubt. But AMD chose to play safe and be always the second in x86 world. Not very ambitious decision.

IPC calculations based on SPECint2006:

• - 9900K .... 54.28/5 GHz = 10.86 pts/GHz
• - 3950X .... 50.02/4.6 GH = 10.87 pts/GHz
• - A76 ........ 26.65/2.84 GHz = 9.38 pts/GHz ...... -14% IPC over 9900K
• - A77 ........ 33.32/2.84 GHz = 11.73 pts/GHz ...... +8% IPC over 9900K
• - A11 ........ 36.80/2.39 GHz = 15.40 pts/GHz .... +42% IPC over 9900K (first 6xALU core)
• - A12 ........ 45.32/2.53 GHz = 17.91 pts/GHz .... +65% IPC over 9900K
• - A13 ........ 52.82/2.65 GHz = 19.93 pts/GHz .... +83% IPC over 9900K

 

[RetroZombie]

Hammer(K8) -> Greyhound/Husky(K9 refresh of K8) -> Bulldozer(K10)

Actually K9 is dual core version of K8.
K10 is Phenom.
K10.5 is Phenom II.
K11 might be the bulldozer family but only searching.

 

[Kuiva maa]

I am surprised that K12's fate hasn't been widely communicated when details have been known for a while now.

Amazon's homegrown 2.3GHz 64-bit Graviton processor was very nearly an AMD Arm CPU

A little bit of history and a little more info on customized SoC

www.theregister.co.uk

It seems the main reason AMD was very keen on talking about K12 around 2014-2015 wasn't their faith in the superiority of ARM or anything like that but just the fact they had a tangible, potential deal with amazon in their hands. This was during their darkest days well before Zen helped them turn things around. My guess is that once the Amazon deal fell through there was no reason for K12 to exist and they focused 100% to x86.

 

[lobz]

Can you PLEASE make this '83% higher IPC sheet' your signature? It'd make life so much easier for a lot of us on this forum. Not for you though, as I guess you permanently have it on your clipboard.

 

[moinmoin]

But AMD chose to play safe and be always the second in x86 world.

If ARM is as strong as you never tire to describe then companies other than AMD and Apple should be able to make it happen. As long as x86 is dominant AMD playing a strong role in that market is a good thing, and unlike with ARM where anybody can join the x86 market is essentially a duopoly right now. Better not to let it become a de facto monopoly again while it dominates over ARM (or any other ISA).

 

[Panino Manino]

This is still on going... all this K12 savior talk is tiresome, no?

Even if it never had even be considered, more ALU will come naturally no matter what. Zen is what it is and there are some ways how it can evolve. Whatever AMD can do AMD will do, give time and opportunity.

Also, after some delay Qualcomm reacted and now the ARM market is more competitive than before, K12 would have a place to exist.

 

[Markfw]

This is still on going... all this K12 savior talk is tiresome, no?

Even if it never had even be considered, more ALU will come naturally no matter what. Zen is what it is and there are some ways how it can evolve. Whatever AMD can do AMD will do, give time and opportunity.

Also, after some delay Qualcomm reacted and now the ARM market is more competitive than before, K12 would have a place to exist.

K12 tiresome ? Yes, ARM at AMD tiresome ? Yes Most of this thread tiresome ? yes.

Can we do anything about it ? Stop posting here and let it die.

 

[Thunder 57]

- was AMD rich enough to waste 3 years of engineering work on K12? IMHO no.
- was good things developed for K12 used in Zen 3 design? IMHO yes, same way they used BD stuff for Zen.
- is Zen 3 reworked K12 to x86? IMHO Probably not, but it depends on how much ""bigger" the K12 engine was.
- does it make sense to build Zen 3 on similar 4xALU K12 engine like Zen1/2 just little bigger? Probably not.
- does it make sense to build Zen 3 on bigger 6xALU K12 engine? Probably yes.

No company wants to throw three years worth of effort away, but there is also such a thing as throwing good money after bad. For example, Intel should have cut their losses rather than keep trying to push x86 phones/tablets. Contra Revenue just didn't work.

Secondly, I still have yet to see any proof that K12 was as wide as you say. Could you please provide a source for that, or is it just speculation?

 

[scannall]

No company wants to throw three years worth of effort away, but there is also such a thing as throwing good money after bad. For example, Intel should have cut their losses rather than keep trying to push x86 phones/tablets. Contra Revenue just didn't work.

Exactly. You need to go to the market. Intel tried to pull the market to them. And failed miserably. AMD, with Zen went to where the market is. Is ARM better? Yes, it's a modern robust ISA. But, that doesn't matter all that much. For good or ill, we are stuck with the choices IBM made 40 years ago. And will be for some time yet.

 

[amd6502]

AMD did name their high end cpu projects to Kx. Bulldozer was K11, so K12 is at least named as direct successor.

I believe Zen was originally K12 and the acorn sister core Zen. Then the marketing team had some input about it and came back with "proof" that Zen would help much more with the branding than a plain ordinary name like 'K12'.

The names were then swapped.

BD had 2x pipes of 128-bit FMAC, Excavator 3x pipes

Is this really true? XV's FPU is 50% more capable than that of PD/SR?

 

[Thunder 57]

I believe Zen was originally K12 and the acorn sister core Zen. Then the marketing team had some input about it and came back with "proof" that Zen would help much more with the branding than a plain ordinary name like 'K12'.

The names were then swapped.

I wouldn't be surprised. Guess it doesn't matter these days, though.

Is this really true? XV's FPU is 50% more capable than that of PD/SR?

I don't think so. But nobody really did a "Deep dive" on XV that I am aware of. Lack of interest I'd guess.

 

[amd6502]

I think I read somewhere that the A1100 was the end result of a commision by or joint venture with Amazon. The partnership didn't really work out and Amazon decided to part ways and do their own totally in-house develpment. The A1100 seemed like it was just quickly wrapped up into a pretty practical (networking oriented) package (likely with Amazon buying some number of these.).

If things had worked out between AMD and Amazon, the K12 might very well have happened. Without a guaranteed large scale customer it would have been a bad move by management to get into a crowded business that was still (very very slowly) building its market and ecosystem. I think it would've been a bit nuts, especially when financially strapped. Focusing on x86 Zen and re-evaluating the acorn server market every few years is a good call.

Skybridge sounds like a good idea though. Why reinvent all the boards when you can just do it once.

Secondly, I still have yet to see any proof that K12 was as wide as you say. Could you please provide a source for that, or is it just speculation?

To me it sounded like a (twin) sister architecture. So it would have been essentially the same design as Zen1 (14nm), except with an ARM front end. So that'd be a 4+2 wide integer core.