GF106 die shot and 512sp GF100

[570091D]

its the same part and pcb as a 480, why would it cost more to make it? They can slap $50 on it and sell it as the GTX+ model.

the pcb shouldn't be the same, the extra cluster should include a wider mem bus and more mem.... i just don't think the 5% performance gain justifies adding another card to the top of thier line-up, unless it's going to beat the 5970.

[spursindonesia]

the pcb shouldn't be the same, the extra cluster should include a wider mem bus and more mem.... i just don't think the 5% performance gain justifies adding another card to the top of thier line-up, unless it's going to beat the 5970.

Incorrect bro, because GTX 480 is already equipped with all of GF 100 available ROP & buswidth (48 ROP & 384 bit respectively). If any, it's the power phase sector of the PCB that would need to be strengthened, quite considerably so if the last leak info from expreview can be trusted (considering nVidia's sh1tty GF 100 yield, if they want to maximize the amount of perfect GF 100 available, we could expect 325-350 w TDP of a 512 SP GF 100 part, and that might be conservative ! ).

[570091D]

Incorrect bro, because GTX 480 is already equipped with all of GF 100 available ROP & buswidth (48 ROP & 384 bit respectively). If any, it's the power phase sector of the PCB that would need to be strengthened, quite considerably so if the last leak info from expreview can be trusted (considering nVidia's sh1tty GF 100 yield, if they want to maximize the amount of perfect GF 100 available, we could expect 325-350 w TDP of a 512 SP GF 100 part, and that might be conservative ! ).

well, i stand corected. however, assuming tdp figures from this article would reflect a consumer part is silly. this is an engineering sample. not meant for customers to play crysis, but meant for engineers to debug silicon. i'm not even sure we should take the performance data seriously...

[spursindonesia]

well, i stand corected. however, assuming tdp figures from this article would reflect a consumer part is silly. this is an engineering sample. not meant for customers to play crysis, but meant for engineers to debug silicon. i'm not even sure we should take the performance data seriously...

Yeah i see your point, but one thing has to be considered, that nVidia failed to launch a full fledged GF 100 for a reason, and fact that the samples of GTX 480 have a very wide variance of power consumption already (only GTX 470 seems more docile & general in average power consumtion number), i can only imagine the designated TDP for GF 100 512 SP if nVidia wants to MAXIMIZE the amount of fully specced GF 100 chip available, the variance IMHO could go through the roof. That's why i suggested that the TDP has to be propped up (while the number for GF 100 family is already very liberal in the first place).