H100 NVL 94GB

The H100 NVL 94GB is a Hopper-generation NVIDIA GPU built on the GH100 architecture, manufactured on a TSMC 4N process node with 80 billion transistors. Released in 2023, it delivers 1979 TFLOPS of FP16 throughput and 1979 TFLOPS BF16 — the two precision formats most commonly used for transformer model training and inference. FP8 precision is supported at 3958 TFLOPS — roughly 2.0× the FP16 rate — enabling near-doubled throughput for inference workloads that can tolerate reduced numerical precision with calibration. The GH100 architecture represents NVIDIA's approach to balancing compute throughput, memory bandwidth, and power efficiency for data center AI workloads. At 400W TDP, the H100 NVL 94GB sits in the mid-range data center tier (400W), fitting standard GPU server form factors.

Memory capacity is 94GB of HBM3 with 3938 GB/s bandwidth. This determines which models can run without quantization: approximately 47B parameters at FP16 (2 bytes/param), 94B at INT8 (1 byte/param), or up to 188B 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 — the ratio of compute (1979 TFLOPS) to memory bandwidth (3938 GB/s) — is approximately 503 FLOP/byte. Transformer inference is typically memory-bound below this threshold, meaning the 3938 GB/s bandwidth figure is the primary determinant of tokens-per-second for most LLM serving workloads.

Multi-GPU configurations are a first-class use case for the H100 NVL 94GB. NVLink 4.0 / PCIe 5.0 provides 900 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 188GB of pooled VRAM — sufficient for 94B parameter models at FP16 — while a four-card setup reaches 376GB. 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 H100 NVL 94GB are Large context LLM inference, Memory-bound workloads, Multi-GPU NVLink setups. NVLink variant of H100 with 94GB HBM3 — 18% more VRAM than standard H100. Designed for large context windows and memory-bound inference. Key limitations to factor into your evaluation: Premium pricing over standard H100 PCIe; Limited availability vs standard H100; Same compute as H100 — only memory capacity improvement. 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 H100 NVL 94GB is a 3-year-old architecture that is still widely deployed in cloud data centers. CUDA compatibility is a significant advantage: the H100 NVL 94GB 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.