NVIDIA prepares to launch AMD HD 6970 rival — GeForce GTX 580

A recent news block about upcoming AMD HD 6000 cards triggered new rumors. This time it concerns new mysterious NVIDIA graphic card based on the 40nm GF110 GPU — GTX 580. The reason for such action is that new Cayman based cards that are expected to out perform current high-end Fermi solution on GF100 GPU. Therefore, in order not to yield the palm in single-GPU cards, NVIDIA is acting quickly and is ready to present this its new flagship in December 2010 which is one month later than planned release date of AMD Cayman cards.

Rumors even include some info on possible specification of the novelty: 512 CUDA cores, 128 TMUs, and a 512-bit wide memory interface. Moreover, the cards will hold, 2 GB of GDDR5. GPU and memory clock are yet unknown. The TDP of GTX 580 is likely to be greater but, still close to what GTX 480 has, since both drain addition power through one 6-pin and one 8-pin PCI-Express connectors, which gives us a figure of around 300W.

Direct comparison of GF110 and GF100 (480 CUDA cores, 64 TMUs and 384-bit memory interface) shows that both card have almost the same computational power, however, there is a significant gain in Texturing (+100%) and memory bandwidth (33%). The real question is, will such enhancements, together with possibly higher clocks, result in noticeable overall performance increase? Probably additional 20%-30%, compared to GeForce GTX 480, which should be sufficient to beat new AMD flagship — Cayman XT (HD 6970) which is planned to be released somewhere in November.

So, will we see an improved version of well known and not very successful GF100 GPU or not? Are this the specs GTX 480 should have had in first place?

Anyway, this can do a good competitive job, since AMD will try attract more customers by cutting prices for its product, and it can also effect new ones (Barts, Cayman).
Sound nice.

In the end, remember, these are just rumors and speculations. Nothing of this is officially confirmed. Lets wait for big words from NVIDIA itself.

In addition

Another piece of information leaked: