SemiAccurate have got hold of a few performance figures for the Geforce GTX 480. Earlier rumours indicated at the GTX 480 being upto 60% faster than the HD 5870. However, SemiAccurate’s sources report that the GTX 480, on average, will end up being a mere 5% faster than the HD 5870 at higher resolutions, despite costing Nvidia a lot more to manufacture. The GTX 470 was not tested in these internal evaluations, and will most certainly end up slower than the HD 5870.

The GTX 480 is set to contain 512 shaders, as initially expected, but far lower clocks than the clock target of 750/1500 Mhz. Instead, it will be clocked betwen 600-625 MHz for the core, 1200-1250 MHz for the shaders (or, as Nvidia calls it, CUDA cores).

More next page.


SemiAccurate have got hold of a few performance figures for the Geforce
GTX 480. Earlier rumours indicated at the GTX 480 being upto 60% faster
than the HD 5870. However, SemiAccurate’s sources report that the GTX
480, on average, will end up being a mere 5% faster than the HD 5870 at
higher resolutions, despite costing Nvidia a lot more to manufacture.
The GTX 470 was not tested in these internal evaluations, and will most
certainly end up slower than the HD 5870.

The GTX 480 is set to
contain 512 shaders, as initially expected, but far lower clocks than
the clock target of 750/1500 Mhz. Instead, it will be clocked betwen
600-625 MHz for the core, 1200-1250 MHz for the shaders (or, as Nvidia
calls it, CUDA cores).

Reports from CES which suggested a hot and noisy card are reflected. Idle temperatures are reported to be 70C, despite running at a very noisy 70% fan speed. Of course, the final cooling solution can be expected to be more aggressive, but there is no doubt – this is one hot, power guzzling GPU, as you would expect on a ~550 m2 chip on a troubled fabrication process.

The GF100 architecture does have its advantages, especially in tessellation situations, where all 512 shaders can work on tessellation. In the Unigine Heaven benchmark, the GTX 480 scores twice as much as the HD 5870, blazing past the HD 5970 along the way. However, several glitches were noticed on the same benchmark, and real life gaming applications aren’t expected to reflect GTX 480’s strong showing  in the synthetic benchmark.

Of course, everything mentioned in Charlie’s article must be taken as a rumour and nothing more. For what it is worth, however, Charlie and his sources have been spot on about Fermi right from the beginning, however outlandish their claims might sound initially. Given GF100’s neverending list of problems, it wouldn’t be a surprise if it does barely beat the HD 5870. We would certainly hope that is not the case, but it would be no surprise.

Unfortunately for Nvidia, ATI have got a fair amount of pricing buffer on the HD 5870. Remember, the HD 5870’s first MSRP was $379, and there’s no doubt ATI can drop the price to $349 or even lesser, if they so wish. The Geforce GTX 480 is unlikely to be profitable anywhere near that price. Rumours suggest Nvidia are producing only 5000-8000 GTX 480s in the first run. If true, it is a sign that the product is simply to maintain a PR image, rather than make any real money. In such a case, pulling an Intel (canceling the first version of Larrabee) would have been the wiser decision. 

The other issue worth considering is – will there be a dual-GPU GF100? A certain “Gemini” card has been rumoured in the past, but the feasibility of a dual-GF100 remains to be seen. It will depend on the GTX 480’s final thermal characteristics.

Meanwhile, all ATI have to do to blast past the GTX 480 is release a slightly overclocked refresh. 

Reference: SemiAccurate