Graphics Cards

Nvidia volta v100 pcie: 5120 cuda cores, 16gb hbm2, 300w

Table of contents:

Anonim

Nvidia has already released its Tesla V100 graphics card based on the Volta architecture and NVLink interface, now they take a new step by announcing the details of the new V100 GPU based on the same architecture and the more traditional PCI Express interface.

Nvidia V100 looks impressive

The new Nvidia Volta V100 architecture is manufactured by TSMC using its advanced 12nm FinFET process to improve the energy efficiency offered by today's Pascal architecture. Inside it is the impressive figure of 21 billion transistors that translate into 5120 CUDA cores, 80 SMs and 40 TPCs that operate at a speed of 1, 370 MHz in base mode and 1, 455 MHz in turbo mode. In the memory section we have no less than 16 GB HBM2 with a 4, 096-bit interface and a bandwidth of 900 GB / s.

If we compare this new card with the Radeon Instinct MI25 we see that the Nvidia option is somewhat more powerful but without a big difference, anyway let's not forget that Nvidia is much more efficient with the use of TFLOPs so in practice if that there could be an important difference.

  • Radeon Instinct MI25: FP16 @ 24.6 TFLOPs NVIDIA Volta V100 PCIe: FP16 @ 28 TFLOPs

Source: tweaktown

Graphics Cards

Editor's choice

Back to top button