AMD's 40nm GPU release possibly Q1 09

[PhoenixUK]

Actually, with this release we'll get 40nm IGPs hopefully so better battery life in laptops.

But then they'll be adding 4850 sort of cards by defualt to laps and probably add some more extra stuff meaning the battery life will be even shorter than it is .

Look at the mobile-phone market, the batteries are getting weaker and weaker ever since the colour tfts and other stuff were introduced .

 

[drunkenmaster]

Errm, seems very unlikely, the fab ATi use is having problems and have publically stated they won't have 45nm ready until the very end of Q1(was planned for end of this year, then pushed to feb/march and now pushed back another month or two). Its very unliekly to be 40nm, just because its called rv740(if infact it is) being the reason they claim its 40nm, is rv770 70nm? TSCM, or tcsm, tsmc, tcms(i always forget which way round it is) seem to be pushing to 45nm for next year and probably not till april/may by the sounds of it so the rumour is that neither ATi nor Nvidia will be able to get a 45nm core out till then.

 

[Lightnix]

I saw an article about that, but it was just no products based on a 45/40nm manufacturing process will be arriving until Q1 next year (as opposed to Q4 this year as was first anticipated).

 

[pompeyboy]

About time really, 45nm tech is already out there ,i was wondering when ati or nvidia would switch.

 

[Lightnix]

Also, RV770 is 55nm. Not sure what they were trying to get at with that RV740/40nm sentiment.

 

[drunkenmaster]

About time really, 45nm tech is already out there ,i was wondering when ati or nvidia would switch.

Intel have their OWN fabs which produce stuff at 45nm, nvidia and ATi make their stuff at TCSM(or whatever the name is, some combination of those 4 letters i never remember right). THey have flat out said the 45nm, not 40 but 45nm process simply won't be ready until the VERY end of Q1 and quite possibly a month or two after that. Intel has a lot of money and a ton of fabs and switching between them is ridiculously easy for them, AMD with 4 fabs, and TCSM with however many they have taking orders from companies have a monumentally harder time changing a fab over to a new process as they are under such heavy strain all the time with contracts and targets to be met. Intel have contracts with, themselves, and have so many fabs a little downtime in any single one has next to no impact.

AS for 2000 shaders, did it not hint that would be the "dual die" product, you realise that as of right now, a RV770 has 800 shaders. a R700 ALREADY has 1600 shaders, 2000 shaders is a SMALL increase. its 200 per core extra thats it.

Based on the current config of 10 units with 16 sp's in each unit, it would most easily change over to 10 units of 20, with each of the 20 being able to do from 1-5 operations a pop. But if thats the case its again the situation of lots of possibly unused shaders.

THe best way is how Anandtech shows the numbers http://www.anandtech.com/video/showdoc.aspx?i=3341&p=3

in the table, essentially what its saying is the RV770 now in the very worst case scenario can only do 160 operations(when 1 operation in each of the 160 SP's is done) and at best can do 800(when all 5 can be used) and tbh its rare that it hits 5, very rare, but fairly often it can do more than 1. Nvidia can do 240 every time, but never more than 480 so its when you can use 2-3 more operations that the RV770 is better.

But up to 1000 shaders, you're only taking the minimum throughput to 200 operations, with the max at 1000. Obviously its better, and halfs the distance to the 240 Nvidia's at and it might just be enough but that is only assuming Nvidia don't up their numbers again for the next generation.

 

[psychas]

waiting till 8 x 12 nm chips . good more competition is better faster speeds good smaller power consumtion even better

 

[DarkMatter]

all the bashing the hd2900xt i believe my hd2900xt has held the test of time in comparison ot the 8800GTS 320 and 640 and it cost me less...... agreeably its noisy as hell and costs me a fortune to run but still lol

 

[The_Abyss]

How's that AA working out for you?

 

[Scougar]

Maybe he doesn't use it?

The best way is to look at TSMC and see what they will be offering in Q1 09, rather than speculate on products. Any delay with them will lead to a delay for ATI (or Nvidia if they decide to splash the cash on a smaller process!). Hopefully this is ATI implementing the tick tock process like Intel use for their product life cycles.

As mentioned (in case it was missed), it is "only" 1000 shaders per die and not 2000 per die. It's 2000 on the X2 style setup. A smaller process may allow that but if they have already used the space up in the design then this may not be feasible, unless they are using the same pad layout/size [on the pcb].

End of the day, any progress is good and I actually think that ATI is in a VERY good position if they are able to unlink the shaders clock, as it is actually running very slowly (and very cool in comparison to NVidia's high frequency I assume). Would be pretty interesting to see if they could up the clock with the shaders with RV870, but I can't remember if they said they were going to or not with RV870... anyone remember?

As a side note: What is this about Nvidia being in the same position as ATI with the 2900xt? Frankly wrong. The 280 design is still a faster card than anything out at the moment (regardless of price), where the reverse was true as the 8800GTX was kicking the 2900XT's butt. Later driver revisions helped out, but not enough to close the gap.

Matthew