GPU comparison
H100 SXM 80GB vs B200 (HGX, per GPU): specs and rental cost comparison
Direct answer
H100 SXM 80GB vs B200 (HGX, per GPU): H100 SXM 80GB is cheaper on the lowest tracked rental row at $1.80 per GPU-hour, versus $5.89 for B200 (HGX, per GPU). That makes B200 (HGX, per GPU) about 227% more expensive per GPU-hour on the cheapest rows in the ByteCosts GPU catalog. H100 SXM 80GB has 80 GB VRAM and B200 (HGX, per GPU) has 180 GB VRAM. The tables compare source-backed specs and value metrics only where both GPUs have a real spec field and a tracked rental rate. Rental prices are list rates per GPU-hour, before committed-use discounts, capacity limits, or region constraints.
Compare GPU rental across providers - Cheapest per-GPU-hour by accelerator →
H100 SXM 80GB vs B200 (HGX, per GPU): rental value
Cheapest tracked on-demand rental rows and normalized value metrics.
| Metric | H100 SXM 80GB | B200 (HGX, per GPU) |
|---|---|---|
| Cheapest rental / GPU-hour | $1.80 | $5.89 |
| $ / GB VRAM | $0.022 | $0.033 |
| $ / FP16 TFLOP-hour | $0.0018 | $0.0026 |
| $ / FP8 TFLOP-hour | $0.0009 | $0.0013 |
H100 SXM 80GB vs B200 (HGX, per GPU): specs
Hardware specs from the committed GPU spec catalog. Rows with missing source fields on either GPU are omitted.
| Spec | H100 SXM 80GB | B200 (HGX, per GPU) |
|---|---|---|
| Architecture | hopper | blackwell |
| VRAM | 80 GB | 180 GB |
| Memory bandwidth | 3350 GB/s | 7700 GB/s |
| FP16 tensor compute | 989.5 TFLOPS | 2250.0 TFLOPS |
| FP8 tensor compute | 1979.0 TFLOPS | 4500.0 TFLOPS |
| NVLink bandwidth | 900 GB/s | 1800 GB/s |
| TDP | 700 W | 1000 W |
Cheapest tracked rental rows
| GPU | Provider | Instance | Region | GPUs | Instance/hr | Per-GPU/hr |
|---|---|---|---|---|---|---|
| H100 SXM 80GB | Vast.ai | 1x-H100_SXM-32-65536 | Florida, US, NA | 1 | $1.80 | $1.80 |
| B200 (HGX, per GPU) | RunPod | 1x_B200_SECURE | CA | 1 | $5.89 | $5.89 |
Frequently asked questions
Is H100 SXM 80GB or B200 (HGX, per GPU) cheaper to rent?
H100 SXM 80GB is cheaper on the lowest tracked row at $1.80 per GPU-hour, compared with $5.89 for B200 (HGX, per GPU). Region, capacity, spot interruption, and discounts can change a production bill.
Which is better value, H100 SXM 80GB or B200 (HGX, per GPU)?
Use the value table above: $/GB-VRAM is always shown from real VRAM and rental rate, while $/TFLOP rows are shown only when both GPUs have the relevant sourced compute spec. Do not compare a missing field as zero.
Where do these GPU comparison numbers come from?
Rental rates come from the ByteCosts GPU rental catalog normalized from the SkyPilot catalog. Hardware specs come from the committed GPU spec catalog and its per-GPU source URLs.
Cite this page
H100 SXM 80GB vs B200 (HGX, per GPU) GPU comparison. ByteCosts. Updated June 15, 2026. https://bytecosts.com/gpu/compare/h100-sxm-80gb-vs-b200/