When NVIDIA first announced both the GeForce GTX 1080 and GTX 1070 they were pretty hush about the GTX 1070. Many were wondering if the only difference would be that it makes use of GDDR5 instead of GDDR5X, but now we know there is quite a lot that separates the GTX 1070 from the GTX 1080. NVIDIA will be cutting down the CUDA core count, a proportionate number of TMU count, and lower the GPU clock speeds.
The GeForce GTX 1070 is based on the 16nm GP104 silicon, but it is the “GP104-200-A1” ASIC. It will have 1920 CUDA cores, 120 TMUs, and 64 ROPs. So 15 out of the 20 streaming multiprocessors are enabled on the GP104 silicon. It will feature a 256-bit memory bus with an 8GB (GDDR5) standard memory amount. The max GPU boost frequency is 1600 MHz. It will have 6.75 TFLOP/s single-precision floating point performance. The TDP is set at 150W and the card will be powered by a single 8-pin PCIe power connector. Outputs include 3x DisplayPort 1.4 and 1x HDMI 2.0b. The card supports 2-way SLI with the new SLI bridge.
The GTX 1070 will launch June 10th with a MSRP of $379 and the “Founders Edition” will be $449.