[Rumour] Nvidia GT300 to be faster than HD 5870

Now that we have a good idea on HD 5870’s performance, Fudzilla reportsthat its sources have great confidence that GT300 will end up fasterthan the HD 5870. Judging by the “absolute confidence”, it might end upquite a bit faster too.

The GT300 will also have a dual GPUversion, which will outperform Hemlock (dual 5870).

Fudzillaalso suggests a “late November” release date, which is quite earlyrelative to other rumours.

So, there you have it. If, asFudzilla’s sources claim, Nvidia can manage to make GT300 with decentyields and a die size not much bigger than Cypress, releasing lateNovember, suddenly, we have the tables turned – and Nvidia in thedriver’s seat.

We can also expect more information and samplesby the end of this month, as previously rumoured.

Meanwhile, on the other spectrum of rumours, it’s all doom and gloom for Nvidia. Charlie of SemiAccurate reports Nvidia is suffering pathetic yields of 2% on the GT300. While outspoken with his criticism for Nvidia, Charlie is usually right on with the facts, but we will take this figure with a big bag of salt. However, similar news of <10% yields has been confirmed by several sources, including Kyle, editor-in-chief of HardOCP.

It is quite possible that both these news are true – GT300 may indeed be a tremendous chip. At the same time, it may be impractical to fabricate. It might be one of those things “too good to be real”. After all, Fuad didn’t mention yield, and Charlie didn’t mentionperformance.

One of the rules often followed by GPU makers has been – “shrink first, architecture second”. (Reflected by Intel’s tick-tock approach as well) As we know, Nvidia are champions of shrinking in the last few years. This time, however, Nvidia has jumped straight onto the 40nm bandwagon, with a brand new architecture. ATI had the good mind to try out 40nm with RV740 (HD 4770), which did indeed suffer significant yield issues, despite being a relatively straightforward shrink. No doubt this data helped them fine tune Evergreen chips. Till date, Nvidia’s GT200b 40nm shrinks have been largely unavailable, seen only in few OEM systems. It may be true that Nvidia is suffering from yield issues across all 40nm products. Needless to say, moving straight to a troublesome process with a revolutionary architecture and a big die is a major gamble to say the least.

We will keep a close eye on how the GT300 develops.

Reference: Fudzilla, SemiAccurate