NVIDIA H800 (2023)
Product Overview
NVIDIA H800 is the first export-compliant AI accelerator NVIDIA launched for the China market, based on the Hopper architecture, officially released in early 2023.
Restricted by US export controls (the October 2022 version), H800's interconnect bandwidth was cut to comply:
- FP16: 1,979 TFLOPS (sparse, same as the H100)
- FP8: 3,958 TFLOPS (sparse, same as the H100)
- Memory: 80GB HBM3 (3.35 TB/s bandwidth, same as the H100)
- Interconnect: NVLink cut (H100 900 GB/s → H800 ~400-600 GB/s)
H800's positioning: compute unchanged, interconnect bandwidth cut — staying as competitive as possible within compliance.
Core Specifications
| Parameter | Value |
|---|---|
| Architecture | Hopper GH100 (export-compliance cut-down) |
| Process | TSMC 4N (4nm) |
| FP8 | 3,958 TFLOPS (sparse) |
| FP16 | 1,979 TFLOPS (sparse) |
| TF32 | 989 TFLOPS (sparse) |
| FP32 | 67 TFLOPS |
| INT8 | 3,958 TOPS (sparse) |
| Memory Capacity | 80GB HBM3 |
| Memory Bandwidth | 3.35 TB/s |
| Interconnect | NVLink (downgraded, ~400-600 GB/s) |
| TDP | 350W (SXM5) |
| Release | Early 2023 |
| Discontinued | October 2023 (new US export control rules) |
Export Compliance Background
US Export Control Timeline (H800-related)
| Date | Event |
|---|---|
| 2022-10 | US BIS issues initial export controls restricting A100/H100 exports to China |
| 2023-03 | NVIDIA launches H800 (NVLink bandwidth cut for compliance) |
| 2023-10 | US issues stricter new export control rules |
| 2023-10 | H800 also falls under the controls, shipments to China halted |
H800 vs H100 Compliance-Cut Comparison
| Parameter | H100 | H800 | Cut |
|---|---|---|---|
| FP16 compute | 1,979 TFLOPS | 1,979 TFLOPS | None |
| FP8 compute | 3,958 TFLOPS | 3,958 TFLOPS | None |
| Memory | 80GB HBM3 | 80GB HBM3 | None |
| Memory bandwidth | 3.35 TB/s | 3.35 TB/s | None |
| NVLink bandwidth | 900 GB/s | ~400-600 GB/s | Cut ~33-55% |
| TDP | 700W | 350W | Reduced 50% |
Strategy: NVIDIA chose to keep compute and memory (important for single-card inference) while cutting interconnect bandwidth and power (impacting large-scale training clusters).
H800 vs H20
| Metric | H800 | H20 | Difference |
|---|---|---|---|
| FP16 | 1,979 TFLOPS | 148 TFLOPS | H800 13.4x |
| FP8 | 3,958 TFLOPS | 296 TFLOPS | H800 13.4x |
| Memory | 80GB HBM3 | 96GB HBM3 | H20 +20% |
| Bandwidth | 3.35 TB/s | 4.0 TB/s | H20 +19% |
| NVLink | ~400-600 GB/s (downgraded) | 900 GB/s (downgraded) | Similar |
| TDP | 350W | 400W | H20 +14% |
| Lifecycle | 2023.03 - 2023.10 (7 months only) | 2024.02 - 2025.07 | H20 longer |
Conclusion: H800 compute far exceeds H20, but its lifecycle was extremely short (only 7 months). H20 is NVIDIA's second-generation compliant product after the October 2023 rules.
Comparison with Domestic Chips
H800 vs Hygon DCU K100
| Metric | NVIDIA H800 | Hygon DCU K100 | Difference |
|---|---|---|---|
| FP16 | 1,979 TFLOPS | 192 TFLOPS | H800 10.3x |
| Memory | 80GB HBM3 | 64GB HBM3 | H800 +25% |
| Bandwidth | 3.35 TB/s | 3.2 TB/s | H800 +5% |
| Ecosystem | CUDA (complete) | ROCm (compatible) | H800 advantage |
| Supply | Discontinued October 2023 | Stable supply | K100 advantage |
Note: H800 is discontinued, so this comparison is largely moot. The current battleground for domestic substitution is H20 vs domestic chips.
Use Cases (historical review)
- ✅ Short-term procurement in 2023 (no longer procurable)
- ✅ Single-card inference (compute uncut, 80GB large memory)
- ❌ Large-scale training clusters (NVLink bandwidth cut, multi-card interconnect performance drops)
- ❌ Procurement after 2024 (discontinued, no supply)
Discontinuation Impact (October 2023)
In October 2023, the US issued stricter new export control rules, and H800 also fell under the controls:
Impact on China's AI industry:
- Customers who bought H800: no short-term impact, but no expansion possible
- Customers who did not: forced to move to H20 (heavily cut compute) or domestic chips
- Domestic substitution timeline: brought forward 1-2 years
Vendor Information
| Parameter | Value |
|---|---|
| Company | NVIDIA Corporation |
| Architecture | Hopper (first-generation export-compliance cut-down) |
| Release | Early 2023 |
| Discontinued | October 2023 (new US export control rules) |
| China market positioning | First compliant replacement for the H100 |
| Lifecycle | Only ~7 months (shortest ever) |
Related Cards
- NVIDIA H100 — Full Hopper architecture (banned from sale in China)
- NVIDIA H200 — Upgraded H100 (banned from sale in China)
- NVIDIA H20 — Second-generation compliant version (2024-2025)
- Hygon DCU K100 — Domestic alternative
- Kunlunxin P800 — Domestic alternative