AIXX Pedia › Model Comparison › B300 × H100
| Parameters | NVIDIA B300 · Blackwell Ultra | NVIDIA H100 · Hopper Generation |
|---|---|---|
| Architecture | Blackwell Ultra | Hopper |
| Process node | TSMC 4NP | TSMC 4N |
| Per-card memory | 288GB HBM3e (2.3TB across eight cards) | 80GB HBM3 (640GB total across eight GPUs) |
| Memory Bandwidth | Approx. 8TB/s per card (nominal) | 3.35TB/s / card |
| GPU-to-GPU Interconnect | NVLink 5 · 1.8TB/s | NVLink 4 · 900GB/s |
| Board Power (TBP/TGP) | 1100W / card (available in both Air-cooled and Liquid-cooled variants) | Up to 700W (configurable) |
| Form Factor and Cooling | 8-GPU air-cooled (NVL8) · 16-GPU liquid-cooled (NVL16) · GB300 superchip series | HGX H100 8-GPU baseboard (mostly air-cooled towers · liquid-cooled variants available from OEMs) |
| Launch / Delivery | Announced at GTC 2025-03 · shipping from H2 2025 | Announced 2022-03 · shipping since 2022-10 |
| China-market Cadence | Circulating in mainland spot markets from H2 2025 · supply constraints ongoing | Subject to China export controls since 2022-10 · not on the approved list for China |
| Mainland China Street Price Band | Approx. 15.0–17.0M RMB/unit (entry basis 2026-10); market listings as of 10-07: 10.72–15.02M in stock (31 listings) | Approx. 5.02M RMB/unit (hall real-time as of 2026-10-07 · 1 unit · whole-batch asking basis) |
Specs follow NVIDIA's official public materials and the two entry pages; price bands reflect live listings in this Pedia's marketplace and fluctuate with inventory — see the entry pages' 'market band' for the latest. A '—' in both columns means the parameter is not yet cataloged in the entry.
AIXX Index (0–100): weighted estimate from public specs — memory 40% + interconnect 30% + ecosystem 30%, with weights adjusted by scenario. 90+ Flagship · 75+ Excellent · 60+ Usable · below 60 Not applicable. The methodology is transparent and verifiable, but is not a performance guarantee; the scores match the published figures on the two entry pages, and this page does not score separately.
| Scenario Index | B300 | H100 |
|---|---|---|
| LLM Inference | 94 | 85 |
| Model Training / Fine-tuning | 86 | 88 |
| Video generation | 82 | —Not Listed |
| Development / Workstation | 35 | —Not Listed |
| Scientific Computing / HPC | 38 | —Not Listed |
| HPC / FP64 compute | —Not Listed | 92 |
| Cost-Sensitive Inference | —Not Listed | 90 |
| Large-Memory Scenarios (Long Context / Full-Size R1) | —Not Listed | 58 |
| Condition / Reliability / Value Retention (Experience Index) | Pending Bench Testing | Pending Bench Testing |
“—” means this model's entry doesn't list a score for this scenario; this page does not fill it in. Experience Index (condition/reliability/resale value) awaits hands-on testing — batch inspection data is being accumulated and each unit will light up as its Inspection Report is published.
After this comparison, return to the entry pages to verify details and the latest price bands: