GTX 780 Ti
Cores: 2880
Memory Clock: 6GHz or more
Memory Bus: 384-bit
Memory Capacity: 3GB or 4GB
Price: $649
GTX 780
Cores: 2688
Memory Clock: 6GHz or more
Memory Bus: 320-bit
Memory Capacity: 2.5GB or 3GB
Price: $499
GTX 770
Cores: 2496
Memory Clock: 6GHz or more
Memory Bus: 320-bit
Memory Capacity: 2.5GB or 3GB
Price: $399
GTX 760 Ti
Cores: 1536
Memory Clock: 6GHz
Memory Bus: 256-bit
Memory Capacity: 2GB
Price: $299
GTX 760
Cores: 1344
Memory Clock: 6GHz
Memory Bus: 256-bit
Memory Capacity: 2GB
Price: $249
And, perhaps, a dual-GPU card with the performance of 2x 780s (non-Ti)
GTX 790
Cores: 2x 2880
Memory Clock: 6GHz
Memory Bus: 2x 384-bit
Memory Capacity: 2x 3GB or 4GB
Price: $999
Note that those are the highest prices NVIDIA should charge; ideally they would be $499, $349, $299, $249, $219 and $699 for the dual-GPU solution since were talking about the refresh of an architecture on a matured process (TSMCs 28nm in this case).
Whether or not a fully enabled GK110 should/can be available is another matter, if not the 2688 cores configuration would have to be flagship and thus ship with a 384-bit bus (rather than 320-bit), but shouldnt cost more than $499.
With Kepler NVIDIA deviated from tradition by introducing their medium-sized GPU first. Theres absolutely nothing wrong with that, however, they also did something else. They priced the video card based on that GPU at twice the usual price and advertised it as a flagship product. Even if NVIDIA felt it should cost more, it shouldnt have been more than $299. When they introduced their big GPU, a year late and with 1 SMX disabled, the price was again two times more at $999.
In my opinion, the fact that your architecture performs really well, doesnt mean you have to charge twice the norm for it; thats practically ripping off the consumer just because you can. What would happen if Intel suddenly decided they have to charge more for new architectures and, say, ask $460 for a quad-core K-series i5 part By charging so much NVIDIA is not simply making new hardware less accessible to the public, but also effectively impeding microprocessor evolution The Way its Meant to be, ouch. Not to mention software development as well as a game dev you cant afford to target hardware most people cannot afford.
Yes, Intel charges $999 for some processors but thats very different. Those are off-norm, usually six-core configuration (when most consumer software targets two/four cores) which are meant for a different market. The mobile extreme series CPUs are a bit overkill at $999, but thats okay since unlocked mobile parts for overclocking are truly a niche. The flagship consumer CPU costs $329 and most people wouldnt need anything more than an i5 K part for $229. $499 for a GPU is high enough already, granted youre buying a whole video card (PCB, memory, more complicated cooler than Intels stock solution, etc.) but thats all factored in already. Also, we cant escape the fact that OEMs charge even more for premium offerings.
Dont get me wrong, I like NVIDIA very much the processors they design, their developer relations and driver team. The prices theyve recently started to pursue are vulgar and unacceptable though. AMDs GCN architecture is sound and theres room for improvement, their developer relations are also picking up speed thing look much brighter than before. Intel has finally entered the GPU space with a tiny, but very efficient architecture. Theres nowhere with developer relations and driver support, yet they certainly have the resources to change that. Image what would a hypothetical dedicated Intel GPU with 160 EUs and stacked DRAM do. NVIDIA has to get back to normal pricing, obviously it wouldnt really happen with this generation its already too late. Heres hoping that Maxwell brings balance to the force once again.
Cores: 2880
Memory Clock: 6GHz or more
Memory Bus: 384-bit
Memory Capacity: 3GB or 4GB
Price: $649
GTX 780
Cores: 2688
Memory Clock: 6GHz or more
Memory Bus: 320-bit
Memory Capacity: 2.5GB or 3GB
Price: $499
GTX 770
Cores: 2496
Memory Clock: 6GHz or more
Memory Bus: 320-bit
Memory Capacity: 2.5GB or 3GB
Price: $399
GTX 760 Ti
Cores: 1536
Memory Clock: 6GHz
Memory Bus: 256-bit
Memory Capacity: 2GB
Price: $299
GTX 760
Cores: 1344
Memory Clock: 6GHz
Memory Bus: 256-bit
Memory Capacity: 2GB
Price: $249
And, perhaps, a dual-GPU card with the performance of 2x 780s (non-Ti)
GTX 790
Cores: 2x 2880
Memory Clock: 6GHz
Memory Bus: 2x 384-bit
Memory Capacity: 2x 3GB or 4GB
Price: $999
Note that those are the highest prices NVIDIA should charge; ideally they would be $499, $349, $299, $249, $219 and $699 for the dual-GPU solution since were talking about the refresh of an architecture on a matured process (TSMCs 28nm in this case).
Whether or not a fully enabled GK110 should/can be available is another matter, if not the 2688 cores configuration would have to be flagship and thus ship with a 384-bit bus (rather than 320-bit), but shouldnt cost more than $499.
With Kepler NVIDIA deviated from tradition by introducing their medium-sized GPU first. Theres absolutely nothing wrong with that, however, they also did something else. They priced the video card based on that GPU at twice the usual price and advertised it as a flagship product. Even if NVIDIA felt it should cost more, it shouldnt have been more than $299. When they introduced their big GPU, a year late and with 1 SMX disabled, the price was again two times more at $999.
In my opinion, the fact that your architecture performs really well, doesnt mean you have to charge twice the norm for it; thats practically ripping off the consumer just because you can. What would happen if Intel suddenly decided they have to charge more for new architectures and, say, ask $460 for a quad-core K-series i5 part By charging so much NVIDIA is not simply making new hardware less accessible to the public, but also effectively impeding microprocessor evolution The Way its Meant to be, ouch. Not to mention software development as well as a game dev you cant afford to target hardware most people cannot afford.
Yes, Intel charges $999 for some processors but thats very different. Those are off-norm, usually six-core configuration (when most consumer software targets two/four cores) which are meant for a different market. The mobile extreme series CPUs are a bit overkill at $999, but thats okay since unlocked mobile parts for overclocking are truly a niche. The flagship consumer CPU costs $329 and most people wouldnt need anything more than an i5 K part for $229. $499 for a GPU is high enough already, granted youre buying a whole video card (PCB, memory, more complicated cooler than Intels stock solution, etc.) but thats all factored in already. Also, we cant escape the fact that OEMs charge even more for premium offerings.
Dont get me wrong, I like NVIDIA very much the processors they design, their developer relations and driver team. The prices theyve recently started to pursue are vulgar and unacceptable though. AMDs GCN architecture is sound and theres room for improvement, their developer relations are also picking up speed thing look much brighter than before. Intel has finally entered the GPU space with a tiny, but very efficient architecture. Theres nowhere with developer relations and driver support, yet they certainly have the resources to change that. Image what would a hypothetical dedicated Intel GPU with 160 EUs and stacked DRAM do. NVIDIA has to get back to normal pricing, obviously it wouldnt really happen with this generation its already too late. Heres hoping that Maxwell brings balance to the force once again.