AMD CPU speculation... and expert conjecture

[Cazalan]

So the Steamroller Server delays are official now. The bigger problem is they're still showing 28nm in 2014. They must not have confidence in GF/TSMC to hit a 20nm node in time. That's going to hurt when Intel has moved to the 14nm node.

http://www.cpu-world.com/news_2012/2012120301_AMD_server_CPU_roadmap_for_2013_and_2014.html

Maybe there's still a chance for Steamroller APU next year? News on that is still mixed.

The other speculation was AMD might scale up their new Jaguar quad core to higher speeds. Which is basically what the module skeptics have been saying from the beginning. Keep progressing the K10.5 core.

 

[JAYDEEJOHN]

If they dont wait for 20nm, theyll need to make sure its shrinkable, 20nm cant be that far off.
Im afraid of cadence on this as well

 

[jdwii]

Why cant Amd just release steamroller and the next gen server CPU on 32nm who cares if its on 28nm just wait to do that next year they need to stop waiting on fabs and just release what product they have. They really need to stop counting on Global foundries if they don't they wont stop going under. Not only that but a 28nm TSMC tiny APU for tablets/laptops would be nice as well is it really this hard Amd or are you just waiting to go bankrupt our hoping someone buys you out?

 

[esrever]

doesn't work like that. You need more die area if you are adding in transistors for performance. you need to go down a node to do that.

 

[jdwii]

doesn't work like that. You need more die area if you are adding in transistors for performance. you need to go down a node to do that.

its simple stop making designs that need more transistors and use the area they do have more efficiently, seriously they are making the same mistake with steamroller as bulldozer counting on some new design from GF.

Their delays will never stop until they change same as their decrease in sales. If they think i'm waiting until 2014 to upgrade to steamroller and also buying a new board after spending 200$ then they do need to get out of the pure CPU market.

However i will be buying a steamroller laptop.

 

[esrever]

the whole driving force for better performance is moore's law which depends on progressing die shrinks. The gain you get from the same node isn't anywhere as significant as a 50% increase in transistors.

 

[jdwii]

the whole driving force for better performance is moore's law which depends on progressing die shrinks. The gain you get from the same node isn't anywhere as significant as a 50% increase in transistors.

But does 50% more transistors mean 50% more performance? I think everyone here knows Amd can improve performance with even less transistors.

 

[esrever]

may or may not be the case but its not economical.

 

[JAYDEEJOHN]

Imagine it as roots of a plant.
The plant having more roots is capable of drawing water in more ways, or, a cpu having more trannys has more connects/pipes,front end capacity etc etc.
One could physically be the same size as the other, but the one which has more roots/transistors has many more resources, and more capability.
Sorry for the plant analogy, just tired of using cars

 

[amdfangirl]

^ Yay an analogy I understand :3.

 

[mayankleoboy1]

this been posted ?

http://www.xbitlabs.com/news/cpu/display/20121203164423_AMD_to_Release_Next_Generation_High_Performance_Microprocessors_in_the_Second_Half_of_2014.html

 

[earl45]

Imagine it as roots of a plant.
The plant having more roots is capable of drawing water in more ways, or, a cpu having more trannys has more connects/pipes,front end capacity etc etc.
One could physically be the same size as the other, but the one which has more roots/transistors has many more resources, and more capability.
Sorry for the plant analogy, just tired of using cars

Not for nothing I'm still waiting for all the profits AMD is going to make from GF
and all the new customers, While Intel go broke from being fab heavy.

 

[viridiancrystal]

http://www.crazyjoys.com/advanced-micro-devices-nyseamd-server-roadmap-28nm-cpus-and-apus-to-be-released-in-2014,20121984

Well, it's official at this point. Steamy is all-or-nothing. AMD is doing one of two things:
1.) Assuming they are getting income from what they have now, plus the Wii U, and thinking it will be enough to keep them going.
2.) They are putting everything into what comes next. Steamroller may be the effort of everything they have, as it may the only way they have a chance of recovering from this slump.

I'm going with the former, because of the 2014 release. If they wanted to keep going as they are they could easily just settle for Steamroller being another Piledriver, mild gains in performance and power [strike]in[/strike]efficiency. That would be set on a year-over-year release, you would think. Waiting the extra while to put it out may be to make it as best they can.

Just a theory. If it is true or even relevant, I'm skeptical at best. That would be around Broadwell (that's what its called, right?) which could be 20% better than Ivy in performance/watt. Steamy would have to be 30-40% better than Piledriver for what they would put into it. Unfortunately, I don't see that happening.

 

[gamerk316]

http://www.crazyjoys.com/advanced-micro-devices-nyseamd-server-roadmap-28nm-cpus-and-apus-to-be-released-in-2014,20121984

Well, it's official at this point. Steamy is all-or-nothing. AMD is doing one of two things:
1.) Assuming they are getting income from what they have now, plus the Wii U, and thinking it will be enough to keep them going.
2.) They are putting everything into what comes next. Steamroller may be the effort of everything they have, as it may the only way they have a chance of recovering from this slump.

I'm going with the former, because of the 2014 release. If they wanted to keep going as they are they could easily just settle for Steamroller being another Piledriver, mild gains in performance and power [strike]in[/strike]efficiency. That would be set on a year-over-year release, you would think. Waiting the extra while to put it out may be to make it as best they can.

Just a theory. If it is true or even relevant, I'm skeptical at best. That would be around Broadwell (that's what its called, right?) which could be 20% better than Ivy in performance/watt. Steamy would have to be 30-40% better than Piledriver for what they would put into it. Unfortunately, I don't see that happening.

Here's the problem: AMD isn't making a lot of money on their console wins. The Wii U BOM indicates that the biggest cost of the Wii U is the $140 to create the tablet for crying out loud. AMD is making ZIP profit for processor sold in that space. Nice win, but no profit margin. That accomplishes NOTHING financially for AMD. All it does is tie up resources that would be better put to use where they can actually turn a profit.

 

[fazers_on_stun]

this been posted ?

http://www.xbitlabs.com/news/cpu/display/20121203164423_AMD_to_Release_Next_Generation_High_Performance_Microprocessors_in_the_Second_Half_of_2014.html

http://www.xbitlabs.com/picture/?src=/images/news/2012-12/amd_optero_roadmap.jpg

Sheesh - that's exactly what I predicted not more than 4-5 pages ago ..

I should win a prize or sumpin'..

Maybe "THG Prognosticateur Sans Pareil" ..

 

[fazers_on_stun]

a cpu having more trannys..

Sorry, but I prefer my CPUs straight!

 

[Cazalan]

Here's the problem: AMD isn't making a lot of money on their console wins. The Wii U BOM indicates that the biggest cost of the Wii U is the $140 to create the tablet for crying out loud. AMD is making ZIP profit for processor sold in that space. Nice win, but no profit margin. That accomplishes NOTHING financially for AMD. All it does is tie up resources that would be better put to use where they can actually turn a profit.

They must have got some decent NREs to do the co-development with IBM/Nintendo. In other words it kept them from having to lay off even more employees.

Not huge profits but more of a royalty with low overhead on their part. Some of that depends on how popular Wii U becomes.

 

[twelve25]

the whole driving force for better performance is moore's law which depends on progressing die shrinks. The gain you get from the same node isn't anywhere as significant as a 50% increase in transistors.

What if you just made a bigger processor with same size transistors? Do I really care if my CPU is the size of a saltine vs the size of a postage stamp? Not really. They could go back to slot CPU for all I care.

 

[esrever]

What if you just made a bigger processor with same size transistors? Do I really care if my CPU is the size of a saltine vs the size of a postage stamp? Not really. They could go back to slot CPU for all I care.

if you are willing to pay (change in area)^2 more money then sure.

 

[Cazalan]

What if you just made a bigger processor with same size transistors? Do I really care if my CPU is the size of a saltine vs the size of a postage stamp? Not really. They could go back to slot CPU for all I care.

The Broadwell motherboards will probably end up smaller than the old Pentium 2 cartridge/slot.

The slot design has made a comeback. Especially for servers.

 

[twelve25]

The slot design has made a comeback. Especially for servers.

Which server chips are you referring to?

 

[-Fran-]

if you are willing to pay (change in area)^2 more money then sure.

Not only that, a bigger chip means lower yields for top end parts.

Cheers!

 

[de5_Roy]

amd says they will keep supporting socketed cpus
http://techreport.com/news/24009/amd-were-not-abandoning-socketed-cpus
tbf, intel never said they will abandon socketed cpus entirely. chances are hedt platform will retain lga cpus (thus forcing interested users pay moar moniez).

Mubadala Provides AMD With Financial Room to Breathe
http://www.tomshardware.com/news/amd-mubadala,19550.html

 

[esrever]

amd says they will keep supporting socketed cpus
http://techreport.com/news/24009/amd-were-not-abandoning-socketed-cpus
tbf, intel never said they will abandon socketed cpus entirely. chances are hedt platform will retain lga cpus (thus forcing interested users pay moar moniez).

only the extreme cpus costs $1000 would likely be socketed if that was the case.

 

[twelve25]

I work with large amounts of servers and there is just no chance those are going to be soldered on the motherboard. So I might suspect that in the future, enthusiasts will be buying a lot more Xeon.