Caporegime
- Joined
- 8 Jul 2003
- Posts
- 30,063
- Location
- In a house
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Global Foundries 28nm wafer spotted
- Thread starter LoadsaMoney
- Start date
Unlikely, although I think they will indeed be fabbed in Global Foundries (assuming September-November launch, usual time).
also the dual-core-only 32nm let-down from Intel.
???? It was on their roadmap and performs as well or better than expected? Were you expecting mroe of it?
In short yes. They seem ready to go and Gate First HKMG design looks a great technology to use. What we might see is 5800's on 32nm and 6800's on 28nm. Or possibly, 40nm TSMC 5800's and Global Foundries 32nm 6800's
A 32nm 5890, in May anyone ?
Last edited:
Associate
- Joined
- 3 Sep 2009
- Posts
- 419
- Location
- Nottingham
Looking forward to 1nm!!!! xD
Believe you cannot go below 11nm, don't know the ins and outs of it though.
Associate
- Joined
- 9 Mar 2008
- Posts
- 1,580
- Location
- Glasgow
I think the 6800 is still expected to be on 40nm
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
I think the main thing AMD are working on right now, is contingency plans. 32nm is likely to be skipped altogther for graphics cards, TSMC are 95% certainly skipping past it and Global look set to skip it for the bulk process also. THE SOI + HKMG is going to be stupidly expensive comparitively and is very silly cost wise to make gpu's on it.
The other issue is TSMC's 40nm process really isn't maturing at all, normally you'd start off with half decent yields and build on them fairly quickly, the 40nm is fairly old now and is still terrible in terms of yields, theres no signs they will be able to vastly improve it, its simply fundamentally lacking things it needs to be less leaky, like HKMG which it will not add to a ongoing process midway through the year.
Its VERY unlikely we'd see a usual double transistor (roughly) next gen part on 40nm, next to impossible we'll see a 32nm part at all(as it seems no one will do a 32nm process), which leaves to real possibilities. A inbetween move at 40nm, no where near double the transistors, its got very little thermal overhead now as it is, minor yield/power improvements won't allow for a massive increase. A tweaked version with an extra few clusters, possibly.
The other likelyhood is a slightly longer gap, hopefully only until Q1 2012, for the next gen and have the usual double transistors(maybe a touch more with a jump all the way to 28nm), the ability most likely to actually increase clocks quite a bit giving their ROP's a very nice boost and likely a return to 48XX type pricing with very very small cores with some pretty awesome low power numbers. They wouldn't increase it too much as 28nm will be around quite a long time so they need room for another generation on the same process most likely.
But again this is where contingency plans come in, they can't for 100% certain know the 28nm process will go off without a hitch, their success's of late have been in planning for the worst case scenario and not getting beaten by it like Nvidia have with Fermi. Its likely they'll have a design ready to go at 40nm TSMC, 28nm tsmc and Global 28nm. With different processes, not always but often cores need tweaking as people make their cores differently. I think other than a minor tweak, a la 4890 >5890, with better overclocking, some tweaks to drop power a little and have higher stock clocks, we likely won't see anything till 28nm and next year, and hopefully at Global Foundries over TSMC.
If their design is compatible with TSMC and Global, and both have a working 28nm process it could also be a win for them on supply having two different sources.
Unlike previous generations with 32nm bulk all but dropped, they won't get a 6 month headstart testing with a mid end chip on the new process, they'll likely be forced to go all in on a new process with their top end chips.
The other issue is TSMC's 40nm process really isn't maturing at all, normally you'd start off with half decent yields and build on them fairly quickly, the 40nm is fairly old now and is still terrible in terms of yields, theres no signs they will be able to vastly improve it, its simply fundamentally lacking things it needs to be less leaky, like HKMG which it will not add to a ongoing process midway through the year.
Its VERY unlikely we'd see a usual double transistor (roughly) next gen part on 40nm, next to impossible we'll see a 32nm part at all(as it seems no one will do a 32nm process), which leaves to real possibilities. A inbetween move at 40nm, no where near double the transistors, its got very little thermal overhead now as it is, minor yield/power improvements won't allow for a massive increase. A tweaked version with an extra few clusters, possibly.
The other likelyhood is a slightly longer gap, hopefully only until Q1 2012, for the next gen and have the usual double transistors(maybe a touch more with a jump all the way to 28nm), the ability most likely to actually increase clocks quite a bit giving their ROP's a very nice boost and likely a return to 48XX type pricing with very very small cores with some pretty awesome low power numbers. They wouldn't increase it too much as 28nm will be around quite a long time so they need room for another generation on the same process most likely.
But again this is where contingency plans come in, they can't for 100% certain know the 28nm process will go off without a hitch, their success's of late have been in planning for the worst case scenario and not getting beaten by it like Nvidia have with Fermi. Its likely they'll have a design ready to go at 40nm TSMC, 28nm tsmc and Global 28nm. With different processes, not always but often cores need tweaking as people make their cores differently. I think other than a minor tweak, a la 4890 >5890, with better overclocking, some tweaks to drop power a little and have higher stock clocks, we likely won't see anything till 28nm and next year, and hopefully at Global Foundries over TSMC.
If their design is compatible with TSMC and Global, and both have a working 28nm process it could also be a win for them on supply having two different sources.
Unlike previous generations with 32nm bulk all but dropped, they won't get a 6 month headstart testing with a mid end chip on the new process, they'll likely be forced to go all in on a new process with their top end chips.
Last edited:
Associate
- Joined
- 10 Nov 2009
- Posts
- 1,099
- Location
- London
I think I understood 50% of that 
If TSMC 40nm process is flawed in some way and yeilds are not going to improve much then what else can be done ?
Its stick with TSMC's 40nm and throw money away, jump back to TSMC's 45nm or 55nm and hope for a dramatic improvement in yeilds, or shrink to 32/28nm.
It may be expensive but if the yeilds are better it will probably be worth it. The other thing to remember is 40% from every wafer is going to AMD anyway.
Its stick with TSMC's 40nm and throw money away, jump back to TSMC's 45nm or 55nm and hope for a dramatic improvement in yeilds, or shrink to 32/28nm.
It may be expensive but if the yeilds are better it will probably be worth it. The other thing to remember is 40% from every wafer is going to AMD anyway.
32nm won't be skipped. Qualcomm, a much bigger company than AMD, have just signed a deal with GF for 32nm chips. I think it's also likely that we will see an 5000 series refresh on 32nm sometime in the Q3.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Not sure I know what you mean by that, anyway at the moment theres really no choice, the same core back on 55nm with better wafer yields would end up being not far off the GTX280 in size, and we know Nvidia couldn't produce them cheaply enough to make a decent profit selling them significantly under £200, so it wouldn't really make much of a difference.
They are all basically stuck with TSMC until they get the next process out and I would think Nvidia and AMD are both hoping its a decent process with far less issues, which looks set to be 28nm and next year, hopefully earlier rather than later.
I think we might be able to skip 32nm because, largely, TSMC have been late by 6-12 months on the last 3 processes really, so basically they've slipped way behind the process curve they should be on. 32nm really could/should have been done a while ago, I think also they likely don't have HKMG ready for 32nm because they hoped to introduce it at 28nm, but leakage has become such a problem in the meantime that 32nm without it will be even more useless than their 40nm.
AMD also are a little behind on process's, but now being run by the almost limitlessly funded Global Foundries for production, can afford to rekit their fabs as quickly as possible, which is also likely to let them just bypass the 32nm process for bulk stuff.
I've been saying for some time now Global is the best thing that can happen for Nvidia and AMD for gpu's. TSMC has for FAR too long been completely unopposed in the market, they could screw up hugely(as they did with 2 months last production and 6 weeks of worthless silicon) and both companies still have no choice but to stay with them. Having a viable alternative means some competition in price for production aswell as TSMC having to invest far more heavily in new process's. Being late will cost them millions, maybe hundreds of millions in lost contracts. Before Global being late just meant being late, spending half in R&D and having a bad process 6 months late didn't cost them any business, because there just wasn't an alternative at all.
Even if Nvidia don't go to Global, they'll benefit from TSMC actually trying to compete. TSMC almost doubled their R&D spending for last year after it became clear GLobal weren't far away from taking orders for parts. Unfortunately R&D spending really only improves the next process, or even the one after it(or even the one after that), that spending won't improve teh 40nm process but 28/22nm in the future should be far less problematic as TSMC try a lot harder.
Associate
- Joined
- 3 Sep 2009
- Posts
- 419
- Location
- Nottingham
Yeah it was just a joke, I think even if it was possible, it would be a couple of decades down the line haha.