Nvidia has started their 16nm GPU generation with a big GP100 GPU intended for the high performance computing. While first Pascal based gaming cards used a considerably smaller GP104 GPU, everybody has expected a new Titan card to come eventually. It has arrived early this time around.
Nvidia’s latest flagship is named Nvidia Titan X, confusingly similar to the older Maxwell based GeForce GTX Titan X. The change in branding may reflect the fact that Titan does not fit cleanly into either the gaming cards or workstation cards. The new Titan X is based on GP102 GPU with 3584 CUDA cores, running at 1417MHz with 1531MHz boost, leading to 11TFLOPS in FP32. Interestingly, it is the same number of cores as in GP100, but there are reasons to suspect that GP102 sacrifices FP64 performance, but has slightly higher clocks and FP32 performance. Another difference from the Tesla P100 is the use of GDDR5X memory, rather than HBM2. Titan X has 12GB of 10Gbps GDDRX5 over 384-bit bus, providing impressive 480 GB/s bandwidth. This is quite close to the bandwidth of Fury X using HBM1 (512GB/s). Compared to GTX 1080, Titan X has 24% advantage in computational performance and 50% edge in VRAM amount and bandwidth. The increase in performance corresponds with increased power consumption and leads to card’s 250W TDP.
The launch date of Nvidia Titan X is August 2nd and it will only be available from Nvidia themselves for $1199. The benchmarks and more specific details about GP102 should be available then as well. In terms of pricing, it is not surprising that Nvidia is keeping prices higher than usual, while there is nothing from AMD yet that can compete above GTX 1070 performance. As soon as AMD responds with proper competition, we should expect gaming focused “GTX 1080 Ti” for a more reasonable price.