New Posts  All Forums:

Posts by tpi2007

You're right, but I didn't say that.A 1280 Maxwell Cuda core GPU would be the GTX 860 Ti.Here is what I wrote in another thread:
Exactly, the architecture is very important. For example with the current Maxwell chips Nvidia managed to save power by doing away with one memory channel (when compared to the GTX 650 Ti Boost) or in general (comparing the GTX 750 to the GTX 650 Ti) by reducing the amount of I/O outside the chip, while at the same time improving or retaining the same performance bracket by increasing the L2 cache 8 fold. The GTX 750 Ti is almost 15% faster than the R7 260X. The first...
Hmm, thanks for that info, I wasn't aware of that. Although they don't suggest that the bigger dies won't have more than 2 MB of L2 cache, just that it's not necessarily a predictable added amount compared to the recently released Maxwell parts. Say, I thought that (until you mentioned this) the GTX 880 would have 8 MB of L2, with the midrange having 4 MB. What he is saying is that it's not necessarily like I said. The GTX 880 may end up having 6 MB of L2 and the GTX 860...
With 8 Megabytes of L2 cache (which means less need for GDDR5 memory bandwidth), more than 2800 Maxwell Cuda cores, and a die size of more than 500 mm2, it's very possible they can make something between 19% and 35% faster than the 780 Ti, still on 28nm.If they go the same route as the GTX 680, which at launch was only 19% faster (26% now) than the GTX 580, they can always get away with it. CPUs don't get this kind of yearly improvement and they still sell. Also, these...
But a mobile chip is easier and less expensive to make than a big die GPU (besides, Apple more than likely takes precedence and what little 20nm capacity there is goes to them), especially since Maxwell has a lot more L2 cache than Kepler. The GTX 750 Ti and 750, for example have 2 Megabytes of L2 cache, that is 8 times more than the chip it's replacing, and it's even more than the 1.5 Megabytes GK110 has (GTX 780 Ti, Titan Black, etc), so the midrange and high-end Maxwell...
You're at 40nm, not 45nm.28nm all the way this year, it's feasible as long as AMD has an architecture as efficient as Nvidia's Maxwell.
Thanks!It's time to lock this one then.
Keep it civil, respect others' opinions.
Locked for cleaning. Edit: reopened. Like Twerk says, everybody keep it on-topic. The thread was also moved from PC gaming to Nvidia drivers.
Congrats on your new position!
New Posts  All Forums: