Repeating that canned GK100 crap is not making it more credible. As a matter of fact if NV choses to not release GK110 for desktop and battle the ground there just with performance SKUs as up to now, there's not a single necessity for any GK100 ghosts being cancelled, GK110 supposedly suck in gaming or any other funky theory that anyone could come up with.
In all likeliness it could simply mean that they can actually finance GK110 expenses from professional market sales only and it would be a quite simple business decision.
Apart from that GK104 has 8 SMXs (192SPs/SMX), 4 GPCs, 16 TMUs/SMX, 32 ROPs, 256bit while GK110 has 15 SMXs (192SPs/SMX), 5 GPCs, 16 TMUs/SMX, 48 ROPs, 384bit which as mentioned over and over again above doesn't suggest any 3D weakness, rather the contrary and even more so since both belong to the same architectural family.
Just for the record's sake there are rumors floating around that the GK110 HPC chips shipped are 14 SMX cores clocked in the high 700s. I can't of course vouch that it's true, however it's at least an indication that things might not be as bad as with GF100. Even if they'd just go for a 14 SMX desktop core at say 850MHz it still would be at least 40% faster than a GTX680 on estimate.
On top of that someone should actually ask himself why it's damn hard with initial runs for high complexity chips to not be able to supply with the top bin. Why isn't Intel shipping any Xeon Phis for HPC with all of its 64 cores enabled and why aren't they pushing frequencies even further as they will later on. One would think that Intel with its own foundries and manufacturing process advantage would have a magic wand and get everything right from the start at maximum capacity. As a matter of fact both Intel and NVIDIA are for the very same reason in a hurry to supply their HPC partners with their respective offerings, as none of the two would want the other to have any significant lead time. Sweet irony both offerings will be either at 1 TFLOPs DGEMM or slightly above
In all likeliness it could simply mean that they can actually finance GK110 expenses from professional market sales only and it would be a quite simple business decision.
Apart from that GK104 has 8 SMXs (192SPs/SMX), 4 GPCs, 16 TMUs/SMX, 32 ROPs, 256bit while GK110 has 15 SMXs (192SPs/SMX), 5 GPCs, 16 TMUs/SMX, 48 ROPs, 384bit which as mentioned over and over again above doesn't suggest any 3D weakness, rather the contrary and even more so since both belong to the same architectural family.
Just for the record's sake there are rumors floating around that the GK110 HPC chips shipped are 14 SMX cores clocked in the high 700s. I can't of course vouch that it's true, however it's at least an indication that things might not be as bad as with GF100. Even if they'd just go for a 14 SMX desktop core at say 850MHz it still would be at least 40% faster than a GTX680 on estimate.
On top of that someone should actually ask himself why it's damn hard with initial runs for high complexity chips to not be able to supply with the top bin. Why isn't Intel shipping any Xeon Phis for HPC with all of its 64 cores enabled and why aren't they pushing frequencies even further as they will later on. One would think that Intel with its own foundries and manufacturing process advantage would have a magic wand and get everything right from the start at maximum capacity. As a matter of fact both Intel and NVIDIA are for the very same reason in a hurry to supply their HPC partners with their respective offerings, as none of the two would want the other to have any significant lead time. Sweet irony both offerings will be either at 1 TFLOPs DGEMM or slightly above
