I will repeat my last two posts to keep the thread on track then.
Anyone know Japanese:
http://www.4gamer.net/games/120/G012093/20120306077/
I wonder if Nvidia is using something similar to HT with Kepler??
It might explain why driver optimisations are supposed to be important.
Here is the relevant part translated(badly) by Google Translate:
"Our users, from conventional products by improving performance "only" 1.5 times as much would not be satisfied. Details of Kepler is not clear, and performance per unit of power consumption is up significantly, it should also significantly improve the absolute performance compared with conventional products of the same price range. "
NVIDIA has been seen, from the end of 2011, has begun the publication of element technology adopted by the GPU in the future, some of which are to be adopted and in fact even Kepler.
When you place an example one, NVIDIA will have to introduce the approach in the technical conference of semiconductor-related was held at the end of 2011, in order to improve the efficiency of use of CUDA Core, to be able to handle more threads on each core was. It was spoken there, increase the number of threads (batch = a bundle of thread) "Warp" which is controlled by 32 thread in the current generation of products, by raising the utilization efficiency of the GPU combined with improvements in scheduling, such is the content, trying to balance the optimization and higher performance with low power consumption.
I think the details are want to tell once again the opportunity, as can be seen if Moraere Think of a Hyper-Threading "" of Intel "approach is NVIDIA, to be able to use threading to other free resources CUDA Core It is, and shall raise the processing performance per CUDA Core, be reduced so that the power consumption per performance "that (official NVIDIA).
In addition, officials close to NVIDIA, "in Kepler, and to improve the processing performance of the DirectCompute PhysX" but also said, If we take the approach NVIDIA officials say certainly, than the arithmetic processing graphics processing "light" is not assigned to the free resources of CUDA Core, a, would not be surprised as to improve the performance be measured.
In any case, future-based products than actually running at the venue of CeBIT 2012 thus, not so far, including the model for the desktop PC, details of Kepler is supposed to be clear. I would not have it be nice to see and finally went into the position for takeoff Kepler."