-
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.
AMD: 40nm yields have stabilized
- Thread starter gurusan
- Start date
Soldato
- Joined
- 25 Sep 2009
- Posts
- 10,215
- Location
- Billericay, UK
I wonder what yields they are achieving now?
Soldato
- Joined
- 25 Sep 2009
- Posts
- 10,215
- Location
- Billericay, UK
Even if AMD can double the 40nm yields I doubt it will make much difference to the price as in the short term demand for these card is so high AMD can afford to whack a sizeable premium on its 5000 series cards which hasn't dented demand. As consumers we are reliant on Nvidia to release Fermi and assuming it's competitive we may then see prices start to drop.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
People are completely misunderstand the "yield" issue that happened with TSMC recently and its really irritating.
I see no quote from AMD there, and AMD haven't commented on a yield drop per waifer and neither have TSMC.
Around October time TSMC announced a piece of equipment was working incorrectly, they lost 6 weeks of waifers, the yield per waifer didn't go down, the waifers were simply made badly and all were lost, the yield per waifer on other waifers didn't go up or down and they didn't promise they would go up or down by the end of the year. THey promised the replaced equipment would be in place by January(which probably means installed this month and tested, and in production in January).
THis doesn't increase yields per waifer, it increases the FACTORY yield. it would be much easier if they refered to it as simply factory output and not mix factory yield, waifer yield and so forth. They were capable of making 9k waifers a month at 40nm, they dropped to 6k waifers, with the fixed equipment they will be back at 9k waifers a month.
This is all well and good, it means more supply which is great, but each waifer costs the same to make, exactly the same as before, and with the same number of chips per waifer, theres no cost change for each core that comes out, just more cores from more waifers.
Fermi has all kinds of issues, leakage, the type thats a problem with 40nm is a HUGE problem and Nvidia's shader cores run at almost twice the speed and are exponentially more effected by the leakage than a core running half the speed. Add to that a seemingly not great design and the rumours are of course they have at least 2 shader clusters in every core not fully functioning.
The current guestimate on AMD's yields are significantly higher than Nvidia's recent Fermi yields, which were supposedly sub 5% for for 512sp parts on the A2 revision. AMD are supposedly at around 60%, though I wouldn't be surprised if that was only true for the 57XX parts and below, and actually lower for the higher end parts. AFAIK they had something clost to 85-90% yields on the 55nm process which was also around 1/3 cheaper per waifer anyway. a 40nm waifer costs $5000 to go through, a 55nm waifer is only around $3500.
I see no quote from AMD there, and AMD haven't commented on a yield drop per waifer and neither have TSMC.
Around October time TSMC announced a piece of equipment was working incorrectly, they lost 6 weeks of waifers, the yield per waifer didn't go down, the waifers were simply made badly and all were lost, the yield per waifer on other waifers didn't go up or down and they didn't promise they would go up or down by the end of the year. THey promised the replaced equipment would be in place by January(which probably means installed this month and tested, and in production in January).
THis doesn't increase yields per waifer, it increases the FACTORY yield. it would be much easier if they refered to it as simply factory output and not mix factory yield, waifer yield and so forth. They were capable of making 9k waifers a month at 40nm, they dropped to 6k waifers, with the fixed equipment they will be back at 9k waifers a month.
This is all well and good, it means more supply which is great, but each waifer costs the same to make, exactly the same as before, and with the same number of chips per waifer, theres no cost change for each core that comes out, just more cores from more waifers.
Fermi has all kinds of issues, leakage, the type thats a problem with 40nm is a HUGE problem and Nvidia's shader cores run at almost twice the speed and are exponentially more effected by the leakage than a core running half the speed. Add to that a seemingly not great design and the rumours are of course they have at least 2 shader clusters in every core not fully functioning.
The current guestimate on AMD's yields are significantly higher than Nvidia's recent Fermi yields, which were supposedly sub 5% for for 512sp parts on the A2 revision. AMD are supposedly at around 60%, though I wouldn't be surprised if that was only true for the 57XX parts and below, and actually lower for the higher end parts. AFAIK they had something clost to 85-90% yields on the 55nm process which was also around 1/3 cheaper per waifer anyway. a 40nm waifer costs $5000 to go through, a 55nm waifer is only around $3500.
I'd have thought that was just rumour... if they had atleast 2 cluster failures per core they could not make a target of 448sp for tesla parts and only maybe 448 for gaming parts. If it were true they'd be dropping back to 416. They are almost deffinatly having some issues with the shaders tho as can be seen from the fact that they can't get tesla parts to full spec with the more rigurous requirements a workstation card has to meet.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
512 sp's, 16 cluster, 32 shaders per cluster, 512 -(2x32) = 448?
Your misunderstanding me... if they are having atleast 2 cluster failure per core - the number of good 448 parts would be not worth the effort let alone the number of 512 parts... theres no way they could produce tesla parts with 448sp if that was the case... they'd have to drop back to 416sp.