My, thank you!
I do mind it being stickied (minding it in a nice, grateful way)!
"Now I can't comment", meaning you're bound to NDA terms, or you just don't want to?
BTW, I think that if NV could finally release a full GF106 (GTS 450) with 192-bit memory and 24 ROP's (like the slower-clocked GTX 460M for notebooks) while maintaining the same 783MHz clock, it just might trade blows with a 5770. Clock it at least 800MHz, and it's very likely to come out ahead. Why is Nvidia waiting so long to do this? Perhaps NV wants to keep 1.5GB for high-end Fermi cards only for now, and afraid that 768MB would scare buyers away from choosing it over a 5750 1GB for example. Or maybe NV is afraid that it would come too close to a GTX 460 768MB (with same 24ROP's, but lower pixel output due to much lower clocks, plus it's already priced low enough, so NV has nowhere to position the full GF106 without pricing the well-selling 128-bit version down the toilet), especially if it does indeed have 48 TMU's instead of 32. Maybe NV just wants to wait and see what AMD has to do with the 5770. It's driving me nuts. Is NV just collecting all of the good chips and using it for 460M low-powered notebooks, or saving it to counter AMD's 5770 replacement, with adjusted clocks and voltages?
See a review on GTX 460M that just came out yesterday: http://www.tomshardware.com/reviews/gef ... 799-2.html
It says that it has the same 32 TMU's, not 48 as I speculated... oh well. ( EDIT-- I keep on forgetting that, the TMU's are usually linear with the shaders, not the ROP's/memory bus. )
Still, this is what GTS 250 should have been, instead of just a renamed 9800GTX+ with 1GB of memory. It would have served NV very, very well during their long depression between the discontinued 285/275 and delayed Fermi. The exact specs including DX11 support were certainly do-able on 55nm, just requiring high power consumption shy of 4890/275 territory. That way, the 5770 would've had less appeal, plus the 5830 would've had no chance but to carry 32ROP's and a true 256-bit bus. IMHO, the GF104 architecture is probably what the canned G90 project (replaced by G92 and GT200) would have looked a lot like (with 384-bit GDDR3 and DX10 instead). Oh, I can feel Nvidia's intermittent "phantom limb syndrome" from having cut off the 9900 Ultra along with its dream G90 project that was under development for 2-3+ years, heh!
GTS 550 (full GF106) speculated specs compared to GTS 450 and G80:
~5% higher clocks than GTS 450 (around 825 MHz) -- fast 192sp's having 10-15% more shader operations/sec than GTX 260-216, at least 20% more texturing power than 8800 Ultra
50-60% more bandwidth than GTS 450 due to 192-bit bus with ~5% faster GDDR5 (~5% more bandwidth than 8800GTX's 384-bit GDDR3)
50% more ROP's (same 24 ROP's as 8800GTX, but at least 40% higher clock and pixel output capacity)
Now, only if it could have at least 48-64 TMU's... then it would've been much, much closer to what the souped-up G90 would've looked like on 55nm (although it would hardly have any more bandwidth than 8800 Ultra, and come in either 768MB or 1.5GB flavors).