Nvidia has just released info on their 1080 card. It looks impressive. 2 times the "VR" performance and 3X the efficiency as a Titan X. shows around 20% faster for regular gaming performance at stock clocks than Titan X.
But I need to stop the praise there. why use the fastest ram available (Gddr5x) and not give it the full potential it can achieve on at least a 384 bit bus?! this card is limited to 320gbs of memory bandwidth while the titan x has 336gbs bandwidth. lowering the bus to 256 significantly limits the amount of gains that can be found from a memory overclock (if the early samples of Gddr5X even allow memory overclocks) to a memory performance that cannot surpass the previous generation of graphics card.
I understand they have done work again with compression to reduce total amount of data that needs to be sent to memory, but that's what they did with Maxwell and look how that turned out for them...
amd's fury x has 2gb less vram and manages to beat the Titan x at 4K simply due to increased memory bandwidth. having more bandwidth is better than having better compression clearly.
Nvidia knows this! look at 16xAA settings compared to comparable amd cards... that bandwidth kills them.
I look at this new 1080 and I just scratch my head as to why. this card could be 35% faster than Titan X and NVidia just wont make the memory bus a little wider to push an easy half terabyte/s bandwidth.
I know what your going to say. "its faster so I don't question it" but you have to at least ask yourself, WHY NOT? it could be even better than it is! is this because the GP100 is expected to disappoint due to wasted die space for DP compute, and NVidia wants this card to be faster than 1080? I'm just at a loss as to why they would make the memory buss so small for this card.
Please help with any possible explanation to why you guys think they are intentionally doing this.
FYI: AMD's Polaris 10 die is expected to be 234mm^2. the 1080 die is 301mm^2. it will be faster don't bring this up in the comments for reasoning.
But I need to stop the praise there. why use the fastest ram available (Gddr5x) and not give it the full potential it can achieve on at least a 384 bit bus?! this card is limited to 320gbs of memory bandwidth while the titan x has 336gbs bandwidth. lowering the bus to 256 significantly limits the amount of gains that can be found from a memory overclock (if the early samples of Gddr5X even allow memory overclocks) to a memory performance that cannot surpass the previous generation of graphics card.
I understand they have done work again with compression to reduce total amount of data that needs to be sent to memory, but that's what they did with Maxwell and look how that turned out for them...
amd's fury x has 2gb less vram and manages to beat the Titan x at 4K simply due to increased memory bandwidth. having more bandwidth is better than having better compression clearly.
Nvidia knows this! look at 16xAA settings compared to comparable amd cards... that bandwidth kills them.
I look at this new 1080 and I just scratch my head as to why. this card could be 35% faster than Titan X and NVidia just wont make the memory bus a little wider to push an easy half terabyte/s bandwidth.
I know what your going to say. "its faster so I don't question it" but you have to at least ask yourself, WHY NOT? it could be even better than it is! is this because the GP100 is expected to disappoint due to wasted die space for DP compute, and NVidia wants this card to be faster than 1080? I'm just at a loss as to why they would make the memory buss so small for this card.
Please help with any possible explanation to why you guys think they are intentionally doing this.
FYI: AMD's Polaris 10 die is expected to be 234mm^2. the 1080 die is 301mm^2. it will be faster don't bring this up in the comments for reasoning.