In early April, NVIDIA company introduced the graphics processor GP100, which is also known as big Pascal, and consists of 15.3 billion transistors and includes 3840 CUDA cores. Also, graphics accelerator Tesla P100 was introduced, based on GP100, which is active only 3584 CUDA cores. Then, later this month, has been presented the GeForce GTX 1080 graphics processor based on the GP104 with 7.2 billion transistors and 2560 CUDA cores. It was assumed that it GP100 will form the base of this new line of flagship - GeForce GTX 1080 Ti, because it was the case with the previous generation of graphics cards when the first came GeForce GTX 980 chip GM204, and then the GeForce GTX 980 Ti with the GM200. It is reported that at the core of GeForce GTX 1080 Ti, as well as the new GeForce GTX the Titan, will be based on GPU GP102, instead of GP100. This chip is rumored to also receive a 3840 CUDA cores, but will differ from GP100 HBM2 by lack of video memory, and a significantly lower performance in double-precision operations (FP64). If the FP64 performance in GP100 chip is 1/2 performance in operations with single precision (FP32), then the GP102, the ratio is equal to 1/32 that scary in the case of consumer graphics cards for gaming. It is reported that the new GeForce GTX Titan, the exact name of which is still unknown, will be based on the full GPU GP102 with all 3840 CUDA cores, which will operate at a frequency of 1507/1621 MHz. The card will receive a 24 GB GDDR5X video memory with a frequency of 10-12 GHz, which will be connected by a 384-bit bus. The TDP level of new items will be equal to 250 watts. In turn, the "youngest" GeForce GTX Ti 1080 will be based on a shortened GP102 with 3456 active CUDA and receive only 12 GB of video memory. All other features are identical . Related Products :
|