NVIDIA Tesla K80 24GB GDDR5 graphics card

Dual-GPU Kepler Architecture

The NVIDIA Tesla K80 pairs two GK210 processors on a single PCI Express 3.0 x16 board. Each GPU contributes 12GB of GDDR5 memory for a total of 24GB across a 384-bit interface per chip. The card delivers 4992 CUDA cores aimed at high-performance computing and deep-learning workloads in server environments.

Sustained Compute Throughput

Dual GPUs share the thermal envelope of one double-slot form factor, so clock rates settle at levels that maintain continuous throughput rather than short bursts. The 384-bit GDDR5 buses feed each Kepler engine with steady bandwidth for matrix operations and scientific simulations. Workloads that scale across both processors see linear memory capacity gains without host-side bottlenecks.

Server Installation Notes

The board occupies one PCIe 3.0 x16 slot and draws auxiliary power through standard server connectors. Passive cooling requires directed chassis airflow; insufficient front-to-back velocity is the most common cause of thermal throttling. Driver stacks must enumerate both logical GPUs correctly, and workload schedulers should pin tasks to individual devices to avoid cross-GPU latency.