GT300 reportedly taped out

[josiahsuarez]

GT300 reportedly taped out
Written by Peter Scott
Thursday, 30 July 2009 09:42


Should launch late this year, if all goes well

According to Charlie D's squad of angel moles, the chip everybody likes to call GT300, although that's not its real name, has finally taped out.

This means Nvidia and TSMC have their work cut out for them if they plan to launch the chip before New Year's and counter ATI's Evergreen series parts. Charlie believes that they won't make it, as the timeframe is rather tight, but some other sources indicate the chip will materialize in late November.

TSMC's 40nm process is still far from mature, and there is always a chance something will go wrong, especially with complex, high end parts. However, both ATI and Nvidia seem confident that the process has improved and become viable. It's now simply a matter of yields and volumes, hence there's a chance we will see a paper launch in November, followed by limited availability till Q1.

More here.
(http://www.semiaccurate.com/2009/07/29/miracles-happen-gt300-tapes-out/)
Also read

GT300 to launch in Q4 2009 (http://www.fudzilla.com/content/view/14810/34/)1

[josiahsuarez]

this is not really GT300 related but I thought it was really interesting! worth reading

http://www.eetimes.com/news/latest/showArticle.jhtml;jsessionid=CYJXX10IT5NFSQSNDLRSK HSCJUNN2JVN?articleID=218900011
Performance scaling of single-thread processors stopped in 2002, Dally said, following a period when the industry derived a performance increase of 52 percent per year for more than 20 years. But throughput-optimized processors like graphics processing units (GPUs) are still improving by greater than 70 percent per year, he said.

"Now that we are no longer scaling single-thread processors, you have no alternative," Dally said.

Throughput processors have hundreds of cores today and will have thousands of cores by 2015, Dally said. By then, Nvidia will have GPUs implemented on 11 nm process technology that feature roughly 5,000 cores and 20 teraflops of performance, Dally predicted.

[Maverick123w]

this is not really GT300 related but I thought it was really interesting! worth reading

http://www.eetimes.com/news/latest/showArticle.jhtml;jsessionid=CYJXX10IT5NFSQSNDLRSK HSCJUNN2JVN?articleID=218900011

Unless they find more efficient ways of making these things I find that highly unlikely.

[Heinz68]

Good news for NVIDIA.
Provided it doesn't need any re-spin (which Charlie says would be a small miracle) we might see the GT300 before Christmas and that would be great news.

My first priority right now is upgrading to Windows 7 and SSD so I might wait and see to compare the card to the ATI Graphic Cards most likely introduced sooner.

[LydianKnight]

Unless they find more efficient ways of making these things I find that highly unlikely.

If you're thinking about current process technologies, forget about that, 11nm is far below what current technology can provide at the shrink level, 11nm is forcibly nanoscale.

This probably means if the prediction is (more or less) correct, we could be seeing what I told about future technology, at least... if they're smart enough, but if they play the 'hey, I'm the cowboy here and you're just an idiot' role, in that case could be a very vapor-like prediction...

Vertical interconnects, core layering, nanoscale fabrication... maybe or maybe not...

We'll see sooner or later :P

[josiahsuarez]

it's a very optimistic prediction. if you assume nvidia doubles the number of cores with each new generation (which they are doing with the GT200->GT300 transition), and you also assume they release a new generation every 17 months, they would be in the 5000 core ballpark by 2015. I think he's just speaking from some internal roadmaps that are highly unattainable. call it an aspirational goal :)

[john19055]

I think sooner then later nvidia is going to have to go with a multiple chip set-up,they are just getting to small for everthing needed on them plus more performance.I still don't see why they can't use two or four chips on a single board and let it use all the memory as it does with one chip.

[LydianKnight]

Too small for everything needed on them? like what?

About going multi-chip, the problem remains the same as with current SLI 2/3-way systems... as long as they don't find a near-to-perfection load balancing algorithm, what's the real purpose of having a multi-chip setup if you're getting something in the order of 1-4 frames delay?

And finally, you need good software to catch up with good hardware... you can have the most powerful graphics card ever made in history, but... put Crysis on it, and you're pretty screwed, because you can get astonishing visuals, but this visuals have to be backep up by good coding practices.

What I mean with this is simple... the better code to run on your GPU, the better results you can get with it, it all depends on how you use it, so why do you say it gets small for everything?