At the beginning of the SOLIDWORKS World conference, NVIDIA revealed a handful of new Quadro cards. The cards targeted professional graphics workloads. NVIDIA’s presentation covered affordable choices like the Quadro P4000 and P2000. The company also unveiled the most fascinating high-end product, the NVIDIA Quadro GP100.
NVIDIA brought out its Pascal GPU from the custom-made server-only market into workstations. Dubbed the Quadro GP100, the brand-new graphics card included the same GPU as the Tesla P100. However, the company packed the card with NVIDIA’s trademark blower-style shroud, complete with a standard PCIe interface. PCR Online mentioned that the Quadro P6000 was the previous top-end Quadro card. NVIDIA based the P6000 on the GP102 GPU used in the consumer-focused Titan X Pascal.
According to WCCF Tech, the Quadro GP100 has a different architecture compared to NVIDIA’s consumer-facing cards, including other Quadros. The Quadro GP100 has an additional 32 FP64 CUDA cores per Streaming Multiprocessor (SM). The additional CUDA cores is way better than having every SM contain a collection of CUDA cores committed to a single precision. NVIDIA stated that this results in around five teraflops of FP64 efficiency or half the FP32 rate. It is a remarkable boost over the 1/32 FP32 rate of the P6000.
Generally, scientific applications that required 64-bit double accuracy math favored FP64. The Quadro GP100 likewise featured double-speed (packed) FP16 guidelines. NVIDIA matched the Quadro GP100 GPU with 16GB of stacked ECC HBM2 (High Bandwidth Memory) for 720GB/s of bandwidth. The reduction in overall memory (24GB on the P6000) may trigger problems for users with large data sets. NVIDIA intended to offset the difference with the addition of a special NVLink connector.
The Tesla P100 communicated entirely on NVIDIA’s exclusive NVLink standard. This allowed several GPUs to link directly to each other or support CPUs at a much higher bandwidth than PCIe 3.0. The Quadro GP100 GPU consisted of a NVLink adapter on top of the card in location of an SLI adapter. This allowed two cards to be linked together for faster memory gain access and low latency data transfer, while still keeping PCIe compatibility.
NVIDIA Quadro GP100 Release Date and Price Details
This announcement marked the very first time consumers saw NVLink executed at an add-in card level. The execution left the possibility for the innovation’s usage on multi-GPU setups for customers in 2017. NVIDIA offered the Quadro P6000 for $4500 on Amazon.
Since the GP100 is more advanced that the P6000, consumers can expect the GPU at the $8000-10,000 range. It will be fascinating to see NVIDIA’s method with GP100 and the P100 GPU. The Quadro GP100 is expected to be launched in March.