NVIDIA H20 vs NVIDIA N1 20SM Comparison

The database compares two NVIDIA parts with no recorded head-to-head benchmark entries. The H20 is a Hopper-generation server accelerator built around the GH100 chip, while the N1 20SM is a Blackwell 2.0 IGP built around the GB20B chip. Both products have an average benchmark score of 0 and a 50th percentile placement, so the comparison rests on specification-level data.

Where Each One Wins

The recorded data assigns zero wins to both products, so there is no benchmark-based use-case split. The separation comes from the specification fields. The H20 wins on raw throughput: 9984 shading units, 312 tensor cores, 312 TMUs, 39.54 TFLOPS FP32, 79.07 TFLOPS FP16, and 4.03 TB/s memory bandwidth. That combination points to dense compute and memory-bandwidth-bound workloads.

The N1 20SM wins on capacity and integration: 128 GB LPDDR5X memory, 20 RT cores, a 2346 MHz boost clock, a 56.30 GPixel/s pixel rate, and a 1x HDMI display output. That combination points to integrated graphics and memory-capacity-focused roles. The H20 has no display outputs; the N1 is an IGP. The H20 is a 500 W SXM Module with a 900 W suggested PSU, while the N1 has no power connectors and an unknown TDP.

Architecture Differences

Both parts are built on a 5 nm TSMC process, but the underlying chips differ. The H20 uses the GH100 chip on the Hopper architecture in the Server Hopper (Hxx) generation. The N1 20SM uses the GB20B chip on the Blackwell 2.0 architecture in the Blackwell IGP (N1x) generation. The H20 die is 814 mm² and contains 80,000 million transistors at a density of 98.3M per mm². The N1 die is 382 mm²; its transistor count and density are not recorded.

The compute core counts differ sharply. The H20 has 9984 shading units, 312 TMUs, and 312 tensor cores. The N1 has 2560 shading units, 160 TMUs, and 80 tensor cores. Both parts have 24 ROPs. The H20 has no RT core entry, while the N1 has 20 RT cores.

Clock behavior also differs. The H20 runs at a base clock of 1830 MHz and a boost clock of 1980 MHz. The N1 runs at 741 MHz base and 2346 MHz boost. Memory clocks are recorded as 1313 MHz with 5.3 Gbps effective for the H20, and 1067 MHz with 8.5 Gbps effective for the N1.

The memory systems are very different. The H20 uses 96 GB HBM3 on a 6144-bit bus with 4.03 TB/s bandwidth. The N1 uses 128 GB LPDDR5X on a 256-bit bus with 273.2 GB/s bandwidth. In terms of integration, the H20 is an SXM Module with no display outputs, a 500 W TDP, and a 900 W suggested PSU. The N1 is an IGP with no power connectors, an unknown TDP, and one HDMI output. Both list N/A for DirectX, OpenGL, and Vulkan, and both use PCIe 5.0 x16. The database does not include cache sizes for either part.

Head-to-Head Benchmarks

The head-to-head benchmark list for this pair is empty. The database records zero wins for the H20 and zero wins for the N1 20SM. Both products have an average benchmark score of 0 and are placed at the 50th percentile among all GPUs in the database. Because no measured results are present, no exact benchmark win margins can be reported.

The largest numerical advantages visible in the recorded data are specification-level. The H20's FP32 throughput is 39.54 TFLOPS against the N1's 12.01 TFLOPS. Its FP16 throughput is 79.07 TFLOPS (2:1) against 12.01 TFLOPS (1:1). Its memory bandwidth is 4.03 TB/s against 273.2 GB/s. The N1 counters with 128 GB memory against 96 GB, a boost clock of 2346 MHz against 1980 MHz, a pixel rate of 56.30 GPixel/s against 47.52 GPixel/s, and the only RT core count in the pair at 20. These are not benchmark scores, but they are the only quantitative comparisons the database provides for this pair.

FAQ

Q: Does the database record any head-to-head benchmark wins for these two products?

A: No. The database records zero wins for both products. Both have an average benchmark score of 0 and a 50th percentile placement.

Q: Which product has more shading units and tensor cores?

A: The H20 has 9984 shading units and 312 tensor cores. The N1 20SM has 2560 shading units and 80 tensor cores. The H20 also has 312 TMUs versus 160 TMUs, while both parts have 24 ROPs.

Q: How do the memory configurations compare?

A: The H20 uses 96 GB HBM3 on a 6144-bit bus with 4.03 TB/s bandwidth. The N1 20SM uses 128 GB LPDDR5X on a 256-bit bus with 273.2 GB/s bandwidth. The N1 provides more capacity; the H20 provides more bandwidth.

Q: What are the FP32 and FP16 throughput figures?

A: The H20 reaches 39.54 TFLOPS FP32 and 79.07 TFLOPS FP16 (2:1). The N1 20SM reaches 12.01 TFLOPS FP32 and 12.01 TFLOPS FP16 (1:1).

Q: Do either product support display output?

A: The H20 has no display outputs. The N1 20SM has 1x HDMI.

Q: What are the clock speeds?

A: The H20 has a base clock of 1830 MHz and a boost clock of 1980 MHz. The N1 20SM has a base clock of 741 MHz and a boost clock of 2346 MHz. The N1 has the higher boost clock; the H20 has the higher base clock.

Specification Differences

| Field | NVIDIA H20 | NVIDIA N1 20SM |

|---|---|---|

| Architecture | Hopper | Blackwell 2.0 |

| Chip | GH100 | GB20B |

| Generation | Server Hopper (Hxx) | Blackwell IGP (N1x) |

| Transistors | 80,000 million | unknown |

| Die size | 814 mm² | 382 mm² |

| Transistor density | 98.3M / mm² | No entry |

| Base clock | 1830 MHz | 741 MHz |

| Boost clock | 1980 MHz | 2346 MHz |

| Memory clock | 1313 MHz 5.3 Gbps effective | 1067 MHz 8.5 Gbps effective |

| Memory size | 96 GB | 128 GB |

| Memory type | HBM3 | LPDDR5X |

| Memory bus width | 6144 bit | 256 bit |

| Memory bandwidth | 4.03 TB/s | 273.2 GB/s |

| Shading units | 9984 | 2560 |

| TMUs | 312 | 160 |

| RT cores | No entry | 20 |

| Tensor cores | 312 | 80 |

| Pixel rate | 47.52 GPixel/s | 56.30 GPixel/s |

| Texture rate | 617.8 GTexel/s | 375.4 GTexel/s |

| FP32 | 39.54 TFLOPS | 12.01 TFLOPS |

| FP16 | 79.07 TFLOPS (2:1) | 12.01 TFLOPS (1:1) |

| TDP | 500 W | unknown |

| Slot width | SXM Module | IGP |

| Power connectors | No entry | None |

| Suggested PSU | 900 W | No entry |

| Display outputs | No outputs | 1x HDMI |

| Release date | 2024-01-31 | 2026-05-31 |

| Predecessor | Server Ada | No entry |

| Successor | Server Blackwell | No entry |

The Verdict

The database gives both products the same percentile placement and the same average benchmark score, so the verdict must be drawn from specifications. The H20 is the compute-dense server part: an SXM Module with no display outputs, a 500 W TDP, a 900 W suggested PSU, 96 GB HBM3, 4.03 TB/s bandwidth, 9984 shading units, 312 tensor cores, and 39.54 TFLOPS FP32. The N1 20SM is the integrated part: an IGP with no power connectors, a 1x HDMI output, 128 GB LPDDR5X, 273.2 GB/s bandwidth, 2560 shading units, 80 tensor cores, 20 RT cores, and 12.01 TFLOPS FP32.

The H20 leads in FP32, FP16, shading units, tensor cores, TMUs, texture rate, and memory bandwidth. The N1 leads in memory capacity, boost clock, pixel rate, and RT core presence. For workloads that need the recorded HBM3 bandwidth and the higher FP32 and FP16 throughput, the H20 is the only option in this pair. For a system that needs an IGP form factor, a display output, and the larger 128 GB memory pool, the N1 is the only option in this pair.