This has given birth to GF104. This is a chip manufactured in the same 40-nm production process, but containing a bit under two billion transistors, unlike the three billion contained by the monstrous GF100. Instead of 512 CUDA cores on GF100, 480 of which were active on GTX 480, then 448 on GTX 470 and 352 on GTX 465, the new GF104 chip has 384 cores, 336 of which are active on GTX 460. According to some sources, GF104 will also be used on GTX 475 with all cores active, as well as the dual-GPU GeForce GTX 495, expected by the end of the summer.

The interesting thing about the new GeForce GTX 460 is that it comes in two variants. The basic difference between these is the width of the memory bus and the quantity of memory present, so we have a version with a 192-bit bus and 768 MB GDDR5 memory, as well as a version with a 256-bit bus and an entire 1 GB video RAM. The ROP unit number is also different - 24 on the weaker and 32 on the stronger model, which could impact performance in antialiasing mode, albeit not as much as on Radeon HD 5830, for example, which had its ROP unit number halved compared to its stronger brethren.

Very significant data in relation to this new chip comes in form of consumption and pricing info. Instead of 200 to 250 W (in more extreme cases, up to 300) with GF100, the new GF104 is declared to 160 W, and the cards themselves should cost about 200 to 250€ depending on the memory quantity and bus width. Even before any tests were run, it was clear to us that this card has a dramatically improved price/performance ratio compared to previous Fermis. Only 768 MB samples are currently available, while the 1 GB ones will appear a bit later, in order to safeguard what little sales GTX 465 has managed to ensure in the limited quantities it has been shipped in. GTX 460’s faster version will definitely reach and surpass the speed of GTX 465, with a price difference of around 100€ in favour of the newer model.

The first model came in from ASUS. Interestingly enough, it was a TOP version, i.e. a factory-shipped overclocked model, which always comes with a very attractive cooling system and “Voltage Tweak” regulation of the graphics chip. Instead of the default 675 MHz for the GPU and 1,350 MHz for CUDA cores, ENGTX460 TOP works at 700 and 1,400 MHz respectively, while memory clocks have been upped from 3600 to 3680 MHz. The cooler that keeps all this under control is dual-slot and rather massive, but not entirely shielded, which improves better ventilation. Two 8mm heatpipes are present and in direct contact with the graphics chip (dubbed “DirectCU”), which gives a cooling performance improvement of up to 20% according to ASUS, compared to Nvidia’s stock cooling solution. Very promising indeed!