Tesla V100
A data center GPU released by NVIDIA in 2017, based on the Volta architecture. It is available in 16 GB and 32 GB HBM2 memory configurations, supports FP16 Tensor Core acceleration with approximately 31.4 TFLOPS FP16 throughput, and targets AI training and inference workloads.
Timeline (last 90 days)
The NVIDIA Tesla V100 has a 32GB VRAM version that uses HBM2 high-bandwidth memory.
The Tesla V100 lacks full support for modern attention optimization libraries such as FlashAttention 2.
Data center cards (such as the Tesla V100) typically have no video output interfaces and rely on server airflow for cooling, requiring additional cooling solutions for home deployment.
The SXM2 version of the Tesla V100 supports NVLink high-speed interconnect with a bandwidth of up to 300 GB/s and a TDP of 300W.
The PCIe version of the Tesla V100 has a TDP of 250W and is bandwidth-limited by the PCIe bus.
The Volta architecture Tensor Cores of the Tesla V100 only support FP16 and lack native support for BF16 and FP8.
The Tesla V100 has a compute capability of 7.0.
The NVIDIA Tesla V100 is a data-center-grade GPU released in 2017.
As data-center cards, the Tesla series have no active cooling fans and rely on the forced airflow of server chassis.
Mainstream inference frameworks vLLM, llama.cpp, and Ollama still support the V100 with compute capability 7.0.
4 more timeline events
All Facts (14)
Data center cards (such as the Tesla V100) typically have no video output interfaces and rely on server airflow for cooling, requiring additional cooling solutions for home deployment.
50%UnverifiedThe NVIDIA Tesla V100 has a 32GB VRAM version that uses HBM2 high-bandwidth memory.
50%UnverifiedThe Tesla V100 lacks full support for modern attention optimization libraries such as FlashAttention 2.
50%UnverifiedThe SXM2 version of the Tesla V100 supports NVLink high-speed interconnect with a bandwidth of up to 300 GB/s and a TDP of 300W.
50%UnverifiedThe PCIe version of the Tesla V100 has a TDP of 250W and is bandwidth-limited by the PCIe bus.
50%UnverifiedThe Volta architecture Tensor Cores of the Tesla V100 only support FP16 and lack native support for BF16 and FP8.
50%UnverifiedThe Tesla V100 has a compute capability of 7.0.
50%UnverifiedAs data-center cards, the Tesla series have no active cooling fans and rely on the forced airflow of server chassis.
50%UnverifiedMainstream inference frameworks vLLM, llama.cpp, and Ollama still support the V100 with compute capability 7.0.
50%UnverifiedFor single-card scenarios running Qwen3, the performance difference between the SXM2 and PCIe versions is not significant, and the PCIe version has a lower deployment barrier.
50%UnverifiedThe NVIDIA Tesla V100 is a data-center-grade GPU released in 2017.
50%UnverifiedThe Tesla V100 is based on the Volta architecture, manufactured on a 12nm process, and features 5,120 CUDA cores and 640 first-generation Tensor Cores.
50%UnverifiedThe Tesla V100's FP16 compute performance is approximately 31.4 TFLOPS, and up to 125 TFLOPS with Tensor Cores enabled.
50%UnverifiedThe Tesla V100's inference speed for running 8B-class models at FP16 precision is approximately in the range of 20–40 tokens/s.
50%