Nvidia Quadro GP100
Nvidia has brought its biggest fully enabled Pascal GPU out of the custom server-only market and into workstations, the Nvidia Quadro GP100. Dubbed the Quadro GP100, the new graphics card features the same GPU as the Tesla P100, but it comes packed into Nvidia’s trademark blower-style shroud, complete with a standard PCIe interface. The Quadro P6000 was the previous top-end Quadro card, and it was based on the GP102 GPU used in the consumer-focused Titan X Pascal.
The Nvidia Quadro GP100 has a very different architecture and focus compared to Nvidia’s consumer-facing cards, or even other Quadros. Instead of every Streaming Multiprocessor (SM) containing a collection of CUDA cores dedicated to the single precision (FP32) floating point calculations favoured by computer graphics applications like video games, the Nvidia Quadro GP100 has an additional 32 FP64 CUDA cores per SM. This, Nvidia says, results in around 5 teraflops of FP64 performance or half the FP32 rate—a dramatic increase over the 1/32 FP32 rate of the P6000.
Typically, FP64 is favoured in scientific applications where 64-bit double precision math is required. The Nvidia Quadro GP100 also features double-speed (packed) FP16 instructions.
Nvidia has paired the Quadro P100 GPU with 16GB of stacked (through-silicon via) ECC HBM2 (High Bandwidth Memory) for 720GB/s of bandwidth. The reduction in total memory (24GB on the P6000) may cause problems for users with large data sets, though. Nvidia is hoping to offset the difference with the inclusion of a unique NVLink connector.
Where the Tesla P100 communicates entirely over Nvidia’s proprietary NVLink standard—which allows multiple GPUs to connect directly to each other or supporting CPUs at a much higher bandwidth than PCIe 3.0—the Quadro P100 includes a NVLink connector on top of the card in place of an SLI connector. This allows two cards to be linked together for faster memory access and low latency data transfer, while still maintaining PCIe compatibility.
The Nvidia Quadro GP100 is expected to launch in March. Pricing is yet to be announced, but expect it to be extremely expensive. The Quadro P6000, for example, retails for over £5000 (~$6225).
The arrival of more parts based on GP100 hints that Nvidia may have a few more consumer cards up its sleeve to combat the upcoming launch of AMD’s Vega graphics cards, which will feature HBM2 memory along with an all-new architecture. The GTX 1080 Ti has long been rumoured for release, and like the GTX 980 Ti before it is expected to be a slightly cut down, but cheaper version of the Titan X Pascal. Nvidia could also surprise with a new version of the Titan X, based on a fully-enabled GP102 GPU.
Such a move wouldn’t be unprecedented: the original GTX Titan was followed up by the GTX Titan Black, which featured a full-fat GK110 GPU and more memory.
Those with shallower pockets should check out Ars’ budget graphics card round up, which features its favourite for under £200/$200.