AMD Server Roadmap 2010-2013 : 10C/20C 28 nm ???

[Olivon]

http://news.mydrivers.com/1/200/200891.htm

Original PDF

[Hawkeye4077]

Interesting to see that C32 and G34 will have had the shortest socket lifespan of any server socket they have ever made...

Still waiting to see some bulldozer C32 cpus pop up...

[Oliverda]

Half-node for high end CPUs? Interesting...

[Manicdan]

pair them up with fusion and it makes sense
size goes down like 20% and they fill that right back up with a few shaders

[Hans de Vries]

That would presumably be the 28nm HPP (High Performance Plus) process:

http://www.slideshare.net/marketinge...ac-2011-update

The claims are for >3GHz ARM cortex-A9 on this process (slides 4,5,6)

Interestingly, the presentation also makes some claims for 4 GHz (ARM)
on the follow up 20nm HPP process. (slide 15) Well, of course, seeing is
believing. This is early marketing material from GlobalFoundries with
respect to 20nm.


Regards, Hans

[Mechanical Man]

They might try to get some of intels lead on process tech with that half node.

[kuroikenshi]

That would presumably be the 28nm HPP (High Performance Plus) process:

http://www.slideshare.net/marketinge...ac-2011-update

The claims are for >3GHz ARM cortex-A9 on this process (slides 4,5,6)

Interestingly, the presentation also makes some claims for 4 GHz (ARM)
on the follow up 20nm HPP process. (slide 15) Well, of course, seeing is
believing. This is early marketing material from GlobalFoundries with
respect to 20nm.


Regards, Hans

Good one Hans, I really enjoyed those slides

[freeloader]

Does anyone here believe in any "roadmaps" that AMD releases? Add 1 to 2 years to it and you'll have a better time line. It would be nice if they were able to keep their timelines.

[Manicdan]

so a map from a few months ago is off about products coming out in 1-2 years, by 1-2 years. sounds a hair excessive.

[mattkosem]

so a map from a few months ago is off about products coming out in 1-2 years, by 1-2 years. sounds a hair excessive.

Would you rather they not have a roadmap? Complaining that a company is planning for the future is bonkers.

--Matt

[Manicdan]

Would you rather they not have a roadmap? Complaining that a company is planning for the future is bonkers.

--Matt

i was saying i believe the map is more accurate than freeloader believes it is.

[Raqia]

That die shot on page 7 is a Pentium 4...

That would presumably be the 28nm HPP (High Performance Plus) process:

http://www.slideshare.net/marketinge...ac-2011-update

The claims are for >3GHz ARM cortex-A9 on this process (slides 4,5,6)

Interestingly, the presentation also makes some claims for 4 GHz (ARM)
on the follow up 20nm HPP process. (slide 15) Well, of course, seeing is
believing. This is early marketing material from GlobalFoundries with
respect to 20nm.


Regards, Hans

[freeloader]

i was saying i believe the map is more accurate than freeloader believes it is.

I'll stick by my guess. At least one year or more off. AMD has not kept to a roadmap in years now.

[Manicdan]

I'll stick by my guess. At least one year or more off. AMD has not kept to a roadmap in years now.

so no 10/20core BD server option until 2013?
no 28nm till 2014?

going from deneb to thuban wasnt so bad was it?

[Lanek]

I'll stick by my guess. At least one year or more off. AMD has not kept to a roadmap in years now.

Half of this roadmap is allready kept .... 2011-2012 ... only 2013 roadmap is unknown .... but you are surely a lot better of me for tell the future.

What have take time to AMD, was not so much the process, but redesign the way they was think new architectures after A64 ... now, it's done .. Bulldozer is the first new architecture till A64. Now they have plan to make it evolve over time .. process is only the end of the chain for make it evolve.


In this Roadmap, there's something a lot more interessant of the years, or process, .. it's the number of cores, if you watch well you see it will stabilisze to 20 cores after Terramar. What do you think it can mean for thoses architecutres ?

AMD look to have really well plan what will happend to their cpu's thoses next 4 years ... and if you want to be competitive, this is the minimum to have, have a good plan for don't stall behind.

[Chumbucket843]

i dont understand the advantage of going from 32nm to 28nm. i haven't seen performance numbers on GF's 28nm processes but i doubt it's worth the time and effort for the gains.

[RVWinkle]

i dont understand the advantage of going from 32nm to 28nm. i haven't seen performance numbers on GF's 28nm processes but i doubt it's worth the time and effort for the gains.

Seems like a waste to me too. I haven't seen Intel's server roadmap but I assume this will compete with 22nm finfet from Intel. AMD probably needs 28nm and everything else they can throw at Intel just to maintain their current position.

Edit: If it's HP like Hans says then they'll clock pretty high

[freeloader]

so no 10/20core BD server option until 2013?
no 28nm till 2014?

going from deneb to thuban wasnt so bad was it?

I definitely don't think we'll see a 20 core processor in 2012. However, I do believe 28nm will happen in 2012. Deneb to Thuban took about 15 months if memory serves me right.
I hope AMD can stick to it's schedule and perhaps even pull the schedule in by a full quarter. It's great for the consumer when Intel has some legitimate competition.

[uOpt]

Man am I glad I'm not holding a G34 platform right now...

[marten_larsson]

Seeing as how they're to release a 10-core desktop processor in 2012 it isn't that hard to see a 20-core server part...

[mattkosem]

i dont understand the advantage of going from 32nm to 28nm. i haven't seen performance numbers on GF's 28nm processes but i doubt it's worth the time and effort for the gains.

A 25% shrink (32² vs 28²) is a fairly sizeable jump, no? All things held constant, that should cut power consumption and heat down by ~25%. It's not more than half like the jump to 22nm, but it's definitely something.

--Matt

[freeloader]

A 25% shrink (32² vs 28²) is a fairly sizeable jump, no? All things held constant, that should cut power consumption and heat down by ~25%. It's not more than half like the jump to 22nm, but it's definitely something.

--Matt

My math may be rusty, but going from 32nm to 28nm is only a 12.5% reduction. ???

[Manicdan]

(28*28)/(32*32) = .765628

77% the size of 32nm

[bamtan2]

i dont understand the advantage of going from 32nm to 28nm. i haven't seen performance numbers on GF's 28nm processes but i doubt it's worth the time and effort for the gains.

you can guarantee they wouldn't be doing it if it wasn't worth the time and effort for the gains. 100% guarantee. how many geniuses do you think they have running these places anyway? it must be measured in the tens of geniuses. I think they've got it all under control.

[BoredByLife]

i dont understand the advantage of going from 32nm to 28nm. i haven't seen performance numbers on GF's 28nm processes but i doubt it's worth the time and effort for the gains.

28nm is considered a half-node process. It is easy to implement without changing your designs on 32nm. I dont know if you have any backgrond in IC design? But it takes a lot of time and is very complicated. Especially when designing a processor. If you can just use a better process without changing the desing much, you win time, money and performance. I don't see anything bad about it.