- Sunday, 8 November 2009
- INQ Mobile
- RSS
Historically, America has never invaded a country that has McDonalds - it's a fact - US Marine quoted on BBC 4
TSMC's 40nm problem revealed
WE'VE BEEN wondering about the seemingly endless delays to TSMC's 40ish nm process, and finally found someone to spill the beans as to why. The culprit is leakage.
People close to the foundry tell us that the TSMC 40nm process is something akin to a screen door on a submarine when it comes to electrons.
The Taiwanese elves have been badgering away at the problem for a long time, and while they do have a cunning plan, we are told it doesn't really work.
This means that the power savings from the 55 -> 40 shrink are going to be minimal to negative, best case results are minor gains.
If there is an optical shrink or a minor tweak to the parts, you will basically end up in the same place speed and capability wise. If you are looking for better GPUs from ATI or Nvidia this spring and summer, don't hold your breath.
It will result in cost savings for the manufacturers though, so cheaper may be in order, but faster is extremely unlikely. µ
Share this:
ShareBusiness Intelligence Software
Customer Relationship Management (CRM)
Enterprise Accounting Software
Enterprise Resource Planning (ERP)
Enterprise Systems Management
Well done Charlie! An article that doesn't bash nVidia! (well, at least not directly!) I thought I'd never see the day! Little-by-little, hopefully, you'll get back to interesting reporting again! (Maybe?!)
Phaps Charlie is bored with bashing nvidia (it gettin way to easy the past 12 months), time to pick on a diff company...
Mebbie for 2010, the inq readers can vote for charlies next target ;)
Yall do realize that NVDA is trying to go into production on TSMC 40nm, right? Or in your hardon to poke at charlie did you miss that fact?
I for one love Charlie's articles because they generate so much controversy and heated debate. As an engineer, if there are problems with manufacturers, I'd like to hear about it in detail, and Charlie's done a great job in reporting them. As for Charlie's opinions about Nvidia, it's called FREE SPEECH! How often have we expressed our own opinions to other people? Have we ever thought if others would like to hear about it? NO! Keep up the good work Charlie!
I figured as much, if the process was stellar, we would see a successor to the 4850/70 line coming on 40nm pretty quick... However, instead we know it'll be a simplified core for use in a 4830 ish variant. Cheap and the new base mainstream. Leakage won't both it too much, as its entire Raison d'etre is to drop card costs / raise margins for a performance level that's taking expensive (relatively) silicon away from 4850 and 4870 production.
So it may not be a stellar overclocking 4830 replacement... but it will be cheap to make.
I've already got a chip out in the 40nm. It's not that bad. Yes, it leaks like a sieve worst case but that's what happens, so you design around the worst case. You optimize your paths using the usual tricks and you can easily pull your leakage down to where 40 will approach 50 and your performance can be much better. If you're one of the push-the-button Synopsis types, well, yes, this process will suck. You have to be smarter and do the right optimizations in the right order to do well down here. (If TSMC can deliver what they're promising on 28nm they'll be doing well and things will be better.)
Like Darkskypoet said, it would be a nice idea, like they did with the X800XL with their smaller 110nm process, a full 16 pixel pipeline which consumed less power and was cheaper to manufacture and gave a run to it's money against the 6800GT. They could do the same with the HD 4830 having only 640 stream processors and that's it, instead of 640 of 800 which still occupying space in the die, I bet that the HD 4800 series yields are higher than before and probably they're neutering working HD 4850 to HD 4830 cards.
How about Seagate over the ongoing firmware fiasco that toasted gazillions of their 7200.11 drives, and Seagate says "a small number". Check their forums for screaming (former) customers.
Hint: Seagate seems to have stopped looking for causes when they found the first one. How else does one explain their serial "checker" telling people with affected drives that their drives are "unaffected" (mine included).
If cheaper 4870's are coming our way in time for Crimbo then that's bonzer!
Charlie, could you do an investigation into the paragraph bug at The Inq? Cheers.
nvidia sucks!! The fresh new bash of nvidia those who craved it.