Originally Posted by jcde7ago
Another reason why I opted for the 690 this time around - resell value. As someone who likes to try some flavor of every generation of GPUs, investing in something that will retain a high resell value in order to help move up to the next generation is paramount in this ever-expensive hobby of ours. I've literally got buyers lined up for my 590, at about 90% of the cost for what I originally bought it for.
If GK104 is a pathway to GK110, I don't think i'll have any issues reselling my 690 and moving into GK110 instead of skipping it (which I thought I might do), considering that i'll be able to recoup a significant amount of the money I originally invested.
Having said that, while the prospect of a GK110 flagship performing leagues above GK104 seems promising, we really do need AMD to be able to stand toe-to-toe with Nvidia in the next gen., or Nvidia is going to have a stranglehold on the GPU market if GCN's successor is not up to par with GK110 (and it doesn't look too promising with this generation's cards and given how Nvidia changed their strategy), and I can picture even mid-range GK110's being priced to start in the $500 range. Nvidia knows that they've now set the bar for dual-GPU performance as well with the 690, and that people have no issues lining up for a $1,000 card, so undoubtedly, whatever GK110's flagship dual-GPU card is, it can be assumed to cost AT LEAST $1,000.
I myself am partial to the Green side, primarily because i've had a few too many issues with AMD in the past (I still own 4000/5000 series cards though), and while I do not see that changing any time soon, I am absolutely rooting for AMD to be able to compete with whatever Nvidia comes out with, not only from a consumer standpoint, but from a PC-enthusiast standpoint as well. No such thing as bad competition in this hobby.
The GK104 is savagely compromised as a compute GPU and thats one reason for its perf efficiency in gaming. In fact GK104 in compute falls behind GF110 in a few cases. GK110 will have to devote a lot of die space towards compute features like Double Precision FP, ECC, larger caches for compute performance, wider memory subsystem (384 or 512 bit ) . A lot of these factors will eat into the performance efficiency (perf/watt) wrt GTX 680. And the GTX 780 needs to achieve performance within 250w. It does not have an unlimited power budget.
If you are talking of a 550+ sq mm die chip with 7 billion transistors thats a huge chip. Given that Nvidia is having difficulty making a 294 sq mm die and publicly accepted to 28nm yield problems in its last quarter earnings call while AMD on the contrary said they met their 28nm yield targets I would be concerned about the volume availability of GTX 780. The GTX 780 card needs to be manufacturable in decent quantities given the fact that Nvidia will prioritise GK110 for professional markets (HPC, workstations) with Tesla and Quadro and only then will it be available for consumer market.
As far as AMD is concerned Tahiti is bottlenecked by its front end resources given the fact that scaling from Pitcairn to Tahiti is nowhere near 50%. Tahiti has 60% more shaders ( (2048 vs 1280). If AMD addresses its bottlenecks and increases shader count by 25% they can get 30 - 40% better performance than Tahiti. That should give it enough performance to compete with GK110. I am also a firm believer in competition benefiting the end user. So fingers crossed Edited by raghu78 - 5/8/12 at 8:49am