← Back to GPUs
NVIDIA · Data Center
NVIDIA H100 80GB
$22000$30000 MSRP
The NVIDIA H100 is the GPU that trained most of the world's leading AI models. With 80GB of HBM3 at 3.35 TB/s bandwidth, it's in a completely different class from consumer hardware. FP8 Transformer Engine provides massive inference speedups. Used H100s are becoming available as companies refresh to Blackwell, but still cost $20k+. Requires server infrastructure — no display output, no consumer drivers.
Best ForThe fastest AI GPU in existence — for those with the budget and infrastructure
VerdictIf you're asking whether you need an H100, you probably don't. But nothing else comes close.
AI
10/10
Gaming
1/10
Specifications
VRAM80GB HBM3
Memory Bandwidth3350 GB/s
CUDA Cores14,592
Boost Clock1845 MHz
TDP350W
Power Connector1x 8-pin
Length267mm
Form FactorDual Slot
Release Year2023
AI Capabilities
Unrivaled80GB VRAM
Run 70B+ models, no compromises. The AI power user's dream.
Can run (Q4 quantized)
Llama 3.1 70BLlama 3.1 8BQwen 2.5 72BQwen 2.5 32BQwen 2.5 14BMistral 7BDeepSeek R1 70BFLUX.1 DevStable Diffusion XLStable Diffusion 3.5 LargeHunyuanVideoCogVideoX-5BMochi 1LTX VideoStable Video DiffusionWan Video 14BCodestral 22BQwen 2.5 Coder 32BLLaVA 1.6 34BAlphaFold 2ESMFold (ESM-2 15B)ESM-2 3BscGPTRFdiffusionFine-tune Llama 8BFine-tune Llama 70BTrain SDXL LoRATrain FLUX LoRA
Recommended system RAM for AI: 160GB+ (2x GPU VRAM for model overflow)
Performance Estimates
Estimated tokens/sec for LLM inference based on 3350 GB/s memory bandwidth — not hardware benchmarks. Methodology · What is Q4/Q8?
Llama 3.1 70B70B
Q8~32-39 tok/sFastLlama 3.1 8B8B
FP16~131-162 tok/sExcellentQwen 2.5 72B72B
Q8~31-38 tok/sFastQwen 2.5 32B32B
FP16~33-41 tok/sFastQwen 2.5 14B14B
FP16~75-93 tok/sExcellentMistral 7B7B
FP16~150-185 tok/sExcellentDeepSeek R1 70B70B
Q8~32-39 tok/sFastCodestral 22B22B
FP16~48-59 tok/sFastQwen 2.5 Coder 32B32B
FP16~33-41 tok/sFastPros
- +80GB HBM3 at 3.35 TB/s — fastest memory available
- +FP8 Transformer Engine
- +NVLink for multi-GPU scaling
- +Runs any model unquantized
Cons
- -$22,000+ used
- -Requires server chassis
- -No display output
- -No gaming whatsoever
- -Power and cooling requirements
aiworkstation