V100 16GB

The V100 16GB is a Volta-generation NVIDIA GPU built on the GV100 architecture, manufactured on a TSMC 12nm process node with 21.1 billion transistors. Released in 2017, it delivers 112 TFLOPS of FP16 throughput and 112 TFLOPS BF16 — the two precision formats most commonly used for transformer model training and inference. The GV100 architecture represents NVIDIA's approach to balancing compute throughput, memory bandwidth, and power efficiency for data center AI workloads. At 250W TDP, the V100 16GB sits in the moderate power envelope (250W), suitable for both workstation and data center deployments.

Memory capacity is 16GB of HBM2 with 900 GB/s bandwidth. This determines which models can run without quantization: approximately 8B parameters at FP16 (2 bytes/param), 16B at INT8 (1 byte/param), or up to 32B parameters at INT4/GGUF quantization (0.5 bytes/param). These figures are theoretical maximums — actual capacity is reduced by KV cache, framework overhead, and activation memory, typically by 10–20% for inference and 30–40% for training. The arithmetic intensity ceiling is approximately 124 FLOP/byte (112 TFLOPS ÷ 900 GB/s). Most autoregressive LLM inference falls well below this threshold, making the 900 GB/s memory bandwidth the binding constraint on tokens-per-second throughput rather than raw TFLOPS.

Multi-GPU configurations are a first-class use case for the V100 16GB. NVLink 2.0 / PCIe 3.0 provides 300 GB/s of bidirectional NVLink bandwidth between cards, enabling tensor-parallel inference across multiple GPUs with near-linear VRAM scaling. A two-card configuration provides 32GB of pooled VRAM — sufficient for 16B parameter models at FP16 — while a four-card setup reaches 64GB. NVLink's low-latency, high-bandwidth fabric makes all-reduce operations in data-parallel training significantly faster than PCIe-based alternatives, which top out at ~64 GB/s bidirectional for PCIe 5.0 x16.

The primary workloads for the V100 16GB are Legacy ML training, Budget inference, Experimentation. Original Volta flagship. Cheapest way to access NVLink. Limited by 16GB VRAM for modern models. Key limitations to factor into your evaluation: Only 16GB HBM2 — limits to ~7B models at FP16; No BF16 or FP8 hardware support; Older Volta architecture — significantly slower than A100/H100. When comparing this GPU against alternatives at similar price points, the most important metrics are memory bandwidth (for inference throughput), VRAM capacity (for model size), and FP16/BF16 TFLOPS (for training speed). Raw TFLOPS figures can be misleading for inference — a GPU with lower TFLOPS but higher memory bandwidth will often outperform a higher-TFLOPS card on tokens-per-second for memory-bound autoregressive generation.

In the broader GPU market, the V100 16GB is a 9-year-old design that is increasingly being replaced by newer architectures in cloud deployments, though it remains available at competitive rental rates. CUDA compatibility is a significant advantage: the V100 16GB benefits from the full NVIDIA software ecosystem including cuDNN, TensorRT, FlashAttention, and all major inference frameworks (vLLM, TGI, TensorRT-LLM). CUDA's maturity means optimized kernels are available for virtually every model architecture. For cloud rental, availability varies significantly by provider — some specialize in this GPU tier while others may have limited stock. Compare on-demand and spot pricing across providers using the rental comparison table on this page, and factor in region availability if latency is a concern for your inference workload.