Two weeks ago at GDC, Nvidia announced the GTX Titan X but didn’t really offer much information about it. Today, we have all the spec details we’ve been waiting for, as the green GPU maker announced more at the GTC (GPU Technology Conference).
Inside the GM200 GPU that’s aboard the GeForce GTX Titan X, you’ll find a grand total of 3072 CUDA cores, which are spread out over 24 SMMs. Beyond the 128 CUDA cores that each SMM carries, they also carry eight TMUs, making for a total TMU count of 192. It also has 96 ROPs on board. The base frequency of the GPU is 1000 MHz, and it’ll boost up to around 1075 MHz so long as the thermal headroom is available.
Nvidia stated that the Titan X can handle 7 teraflops (SP) and 0.2 teraflops (DP).
Naturally, the GPU is based on the Maxwell architecture, but it is still fabricated on the 28 nm lithographic process. Because it also has eight billion transistors on board, it’s physically huge. In fact, it’s 601 square mm huge, making it 9 percent larger than the GPU found in the original GTX Titan. The card in its entirety has a TDP of 250 W.
On the card itself, you’ll also find a majestic 12 GB of GDDR5 memory that runs at 7.0 GHz, which is addressed over a 384-bit memory interface. This gives it a memory bandwidth of a respectable 336.5 GB/s. There’s no backplate on the card.
Finally, there’s the price. The Titan X will land with a price tag of $999, which is about what one would have expected on the consumer side.
