To folks who still cling to the belief that NV puts gamers first, dream on.
GTX 670 is the gaming card--using a cut down GK104.
Most uncut GK104 chips go into Tesla K10 dual-GPU professional boards that sell for multiple times what a GTX 690 would sell for.
http://www.amazon.com/PNY-DisplayPor.../dp/B0044XUD1U (That's a Quadro board but still gives you an idea of how much a Fermi-based Tesla board might cost.) According to vr-zone, Tesla K10 and K20 have more than 150,000 pre-orders, meaning 300,000 GPUs that won't be going into GeForce cards.
http://vr-zone.com/articles/how-the...-prime-example-of-nvidia-reshaped-/15786.html
The leftover GK104 chips go into GeForce GTX 680. This number will go up as Tesla demand is satisfied, freeing up GK104s for use in GeForce cards.
Similarly, the first batches of GK110 are already claimed by Tesla pre-orders:
http://techreport.com/articles.x/22989/3
http://vr-zone.com/articles/nvidia-...-k20-2013-geforce-and-quadro-cards/15884.html
Nvidia is prioritizing profits over gross revenue, and this is absolutely the right thing to do when capacity-constrained. I always thought it was so stupid of NV to release Fermi to gamers first, but in hindsight maybe we helped debug things for them or whatever, giving them more time to get drivers together for Quadro and Tesla. In any case, I would rather that gamers continue to kvetch about not getting GK110 in GeForce cards this year, and instead having those GPUs go into HPC cards that solve real-world problems like drilling/seismic/financial/medical imaging problems.
If it's any consolation, Quadro is in the same boat we are: they aren't getting any Kepler-based Quadros until Kepler-based Tesla cards ship first, either.
Expect to see cut-down GK110 chips for GeForce GTX gaming-grade cards sometime next year, after some of the initial Tesla demand is met.