H100 80GB vs H200 141GB
Specs, price, and performance compared
Specs
▲ indicates the higher value for that spec. TDP: lower is better for density.
H100 80GB vs H200 141GB: Workload Fit
Both GPUs share the GH100 architecture. The H100 80GB delivers 1979 TFLOPS of FP16/BF16 throughput, compared to 1979 TFLOPS on the H200 141GB. Both support FP8: the H100 80GB at 3958 TFLOPS and the H200 141GB at 3958 TFLOPS.
Memory capacity and bandwidth are the primary constraints for large language model workloads. The H200 141GB provides 141GB of HBM3e versus 80GB of HBM3 on the H100 80GB. Memory bandwidth favors the H200 141GB at 4.8 TB/s — 1.4× the 3.4 TB/s of the H100 80GB. Both support NVLink: the H100 80GB at 900 GB/s and the H200 141GB at 900 GB/s bidirectional.
Power consumption shapes rack density and operating cost. The H100 80GB draws 700W TDP — the same as the H200 141GB. On a TFLOPS-per-watt basis, the H100 80GB is the more efficient option, delivering 1979 TFLOPS at 700W.
The H100 80GB is best suited for LLM training, Large-scale inference, Scientific HPC. The H200 141GB targets 70B+ model inference, Memory-bound LLM serving, Large context windows. For models that exceed the smaller card's VRAM, the H200 141GB's 141GB is the deciding factor — a 141GB card can hold roughly 70B parameters at FP16 or 282B at INT4.
Workload Cost Comparison
Total job cost = observed median $/hr × GPU count × hours. Spot pricing typically 40–70% lower.
The H200 141GB costs 1.4× more than the H100 80GB at current observed median prices. The H200 141GB's 141GB VRAM (vs 80GB on the H100 80GB) allows larger models and batch sizes without offloading, which commands a premium in the cloud market. Memory bandwidth is 1.4× higher on the H200 141GB (4800 GB/s vs 3350 GB/s), which directly improves token throughput for memory-bound LLM inference.
Based on observed on-demand median prices. Actual cost depends on provider, region, and availability.
Strengths & Weaknesses
- Less VRAM than the H200 141GB (80GB vs 141GB)
- Lower memory bandwidth than the H200 141GB (3350 GB/s vs 4800 GB/s)
- 2× the VRAM of the H100 80GB (141GB vs 80GB)
- Higher memory bandwidth (4800 GB/s vs 3350 GB/s on the H100 80GB)
- Newer architecture (GH100, 2024) vs GH100 (2022)
- LLM training
- Large-scale inference
- Scientific HPC
- Highest cost per hour of widely available GPUs
- SXM5 form factor requires specialized server infrastructure
- Overkill for inference-only workloads where A100 suffices
- 70B+ model inference
- Memory-bound LLM serving
- Large context windows
- Premium pricing over H100 — ~40–60% higher hourly cost
- Limited availability vs H100 — fewer providers offer it
- Same compute as H100 — only memory bandwidth improvement