GeForce GTX 1070
The GeForce GTX 1070 again uses the GP104 GPU, but in a stripped down form. The chip is produced on the 16nm FinFET process technology, the number of transistors is the same as 7.2 billion, even if not all of them are used. The GP104 GPU variant on the GeForce GTX 1070 has three Graphics Processing Clusters active instead of four. Each of the GPCs contains five streaming multiprocessors, and they, in turn, have two SMs with 64 stream processors each.
As a result, we get 1.920 stream processors for the GeForce GTX 1070 (3x5x2x64 = 1.920). Accordingly, the number of texture units has decreased from 160 to 120. The memory interface still consists of eight 32-bit controllers, the width is 256 bits. Therefore, the number of raster operation pipelines has not changed, remaining 64. NVIDIA specifies a graphics card thermal package of 150 W. Unlike the GeForce GTX 970, NVIDIA didn't go tricky with the GeForce GTX 1070 by limiting the cache or register block size. The problem with the truncated memory controller is also a thing of the past.
NVIDIA has changed not only the number of stream processors and other architecture components, but also the frequency compared to the GeForce GTX 1080. The base frequency is 1.506 MHz (GeForce GTX 1080: 1.607 MHz), in GPU Boost mode, the frequency increases to at least 1.683 MHz (GeForce GTX 1080: 1.733 MHz). Instead of GDDR5X, NVIDIA used classic GDDR5 memory for the GeForce GTX 1070. But it is set in a very fast version at 2.000 MHz. Thanks to the 256-bit memory bus, the GeForce GTX 1070 provides a memory bandwidth of 256 GB / s (for the GeForce GTX 1080 - 320 GB / s).
Specifications GeForce GTX 1070
|
||||||
Chip
|
||||||
Frequencies
|
||||||
Memory
|
||||||
Interface and TDP
|
The GPU Boost mechanism is very dependent on the temperature of the GPU - especially with reference NVIDIA graphics cards, which are oriented to work in a certain temperature range. Here, the Power Limit power consumption bar does not affect so much. With the GeForce GTX 1070 Founders Edition, we got GPU temperatures of 82-83°C at full load. That is, the video card rested on the temperature limit. Accordingly, the Boost frequency did not reach the theoretical maximum of 1.886 MHz, remaining in the range of 1.722-1.785 MHz.