NVIDIA GeForce RTX 5090 Mobile vs NVIDIA H20 Comparison

Head-to-Head Benchmarks

The database contains no direct head-to-head benchmark results between the NVIDIA GeForce RTX 5090 Mobile and the NVIDIA H20. The H20 has no recorded benchmark scores in the database, with an average benchmark score of 0 and an empty benchmark list. Consequently, the data cannot support a direct numerical comparison across any shared workload.

For the RTX 5090 Mobile, the recorded benchmark suite provides a clear performance profile. Its 3DMark Steel Nomad DX12 score is 5871, which places it in the 84th percentile of all GPUs. Its average benchmark score of 45152 is nearly identical to the AMD Radeon Pro 5500 XT at 45384, a difference of -0.5%. It sits 0.8% ahead of the NVIDIA GeForce RTX 4070 Ti, which scores 44795. The Intel Arc A730M scores 45592, putting it 1% ahead of the RTX 5090 Mobile, and the NVIDIA RTX 5880 Ada Generation scores 45972, which is 1.8% ahead. These deltas show that the RTX 5090 Mobile clusters tightly with those four rivals in average performance, neither dominating nor trailing by a wide margin.

Specific API-level tests on the RTX 5090 Mobile show a broad range. Geekbench OpenCL returns 201834, while Geekbench Vulkan returns 198405. Passmark G3D scores 30034, and Passmark GPU Compute scores 13401. The Passmark legacy and specialized tests show lower absolute numbers: DirectX 10 at 183, DirectX 11 at 269, DirectX 12 at 138, DirectX 9 at 324, and G2D at 1057. Because the H20 has no comparable figures, no percentage advantage can be calculated between the two cards in any of these tests.

The absence of H20 benchmarks means the database offers no evidence of wins for either product in a direct confrontation. The RTX 5090 Mobile has 10 recorded tests; the H20 has none. The H20 percentile rank of 50 is listed, but with no underlying score data, that percentile functions only as a placeholder in the absence of measurements.

Architecture Differences

The two GPUs come from different NVIDIA architectures and target separate market segments. The RTX 5090 Mobile uses the GB203 chip under the Blackwell 2.0 architecture, part of the GeForce 50 Mobile generation. The H20 uses the GH100 chip under the Hopper architecture, part of the Server Hopper (Hxx) generation. Both are fabricated on a 5 nm process at TSMC, but the die sizes and transistor counts differ substantially. The RTX 5090 Mobile contains 45,600 million transistors on a 378 mm² die, yielding a transistor density of 120.6M per mm². The H20 contains 80,000 million transistors on a 814 mm² die, with a density of 98.3M per mm². The H20 die is more than twice the area and nearly double the transistor count.

Clock speeds differ as well. The RTX 5090 Mobile has a base clock of 990 MHz and a boost clock of 1515 MHz. The H20 runs at a base clock of 1830 MHz and a boost clock of 1980 MHz, meaning the server chip operates at a higher frequency despite the larger die. Memory technology also separates the two. The RTX 5090 Mobile uses 24 GB of GDDR7 on a 256-bit bus, with a memory clock of 1750 MHz and 28 Gbps effective, producing 896.0 GB/s of bandwidth. The H20 uses 96 GB of HBM3 on a 6144-bit bus, with a memory clock of 1313 MHz and 5.3 Gbps effective, producing 4.03 TB/s of bandwidth. The H20 has four times the memory capacity and over four times the memory bandwidth.

Compute resources show a mixed comparison. The RTX 5090 Mobile has 10496 shading units, 328 TMUs, and 112 ROPs. The H20 has 9984 shading units, 312 TMUs, and only 24 ROPs. The RTX 5090 Mobile includes 82 RT cores, while the H20 lists no RT cores at all. Both have tensor cores: 328 on the RTX 5090 Mobile and 312 on the H20. Pixel rate heavily favors the mobile card at 169.7 GPixel/s versus 47.52 GPixel/s for the H20, due to the H20's low ROP count. Texture rate slightly favors the H20 at 617.8 GTexel/s versus 496.9 GTexel/s for the RTX 5090 Mobile. FP32 performance favors the H20 at 39.54 TFLOPS versus 31.80 TFLOPS. FP16 shows a larger split: the H20 delivers 79.07 TFLOPS with a 2:1 ratio, while the RTX 5090 Mobile delivers 31.80 TFLOPS at a 1:1 ratio.

Power and form factor differ by design intent. The RTX 5090 Mobile has a TDP of 95 W, an IGP slot width, and no power connectors. The H20 has a TDP of 500 W, an SXM Module slot width, and a suggested PSU of 900 W. The H20 has no display outputs, while the RTX 5090 Mobile has portable device dependent outputs. API support also diverges: the RTX 5090 Mobile supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, while the H20 lists N/A for DirectX, OpenGL, and Vulkan. The H20 is a compute-oriented server part without a graphics API stack.

Where Each One Wins

The data points to distinct strengths for each GPU. The RTX 5090 Mobile wins in pixel throughput, with 169.7 GPixel/s compared to 47.52 GPixel/s for the H20. That 3.6x advantage in pixel rate comes from 112 ROPs versus 24, and it translates directly to rasterization-heavy workloads. The RTX 5090 Mobile also has more shading units (10496 versus 9984), more TMUs (328 versus 312), and a higher transistor density (120.6M per mm² versus 98.3M per mm²), which reflects a more compact and efficient design for graphics processing.

The RTX 5090 Mobile carries RT cores, 82 of them, while the H20 has none. Any ray-traced workload falls entirely to the RTX 5090 Mobile. The mobile card also supports the full graphics API set: DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. The H20 has no graphics APIs listed, so it cannot be used for conventional rendering. The RTX 5090 Mobile also has a much lower TDP at 95 W versus 500 W, and its IGP form factor means it fits into portable devices, whereas the H20 requires an SXM module slot.

The H20 wins in memory capacity and bandwidth. Its 96 GB of HBM3 on a 6144-bit bus provides 4.03 TB/s, compared to 24 GB of GDDR7 on a 256-bit bus at 896.0 GB/s. That is a 4x capacity advantage and roughly a 4.5x bandwidth advantage. The H20 also has a higher boost clock at 1980 MHz versus 1515 MHz, and a higher base clock at 1830 MHz versus 990 MHz. Its FP32 output of 39.54 TFLOPS exceeds the RTX 5090 Mobile's 31.80 TFLOPS. In FP16, the H20 reaches 79.07 TFLOPS with a 2:1 ratio, more than double the RTX 5090 Mobile's 31.80 TFLOPS at 1:1. Texture rate also favors the H20 at 617.8 GTexel/s versus 496.9 GTexel/s. The H20's higher transistor count of 80,000 million and larger die of 814 mm² support its server-class compute role.

For use cases, the RTX 5090 Mobile suits graphics-heavy applications: real-time rendering, ray tracing, and DirectX or Vulkan workloads. The H20 suits memory-bound compute tasks: large model inference, high-bandwidth data processing, and FP16 tensor operations. The H20 has no display outputs, so it cannot drive a monitor. The RTX 5090 Mobile can, through portable device dependent outputs.

The Verdict

The database shows two GPUs optimized for opposite ends of the computing spectrum. The NVIDIA GeForce RTX 5090 Mobile is a graphics-first mobile processor. Its 82 RT cores, 112 ROPs, DirectX 12 Ultimate support, and 169.7 GPixel/s pixel rate make it the only choice for rendering tasks. Its 84th percentile ranking among all GPUs and an average benchmark score of 45152, within 2% of four close rivals, confirm a strong competitive position in its class. The H20, with no benchmarks and no graphics APIs, cannot be evaluated for rendering performance and does not participate in the same application space.

The NVIDIA H20 is a compute and memory capacity specialist. Its 96 GB HBM3 pool and 4.03 TB/s bandwidth, combined with 39.54 TFLOPS FP32 and 79.07 TFLOPS FP16, serve workloads where data volume and throughput outrank pixel output. Its 500 W TDP and SXM module form factor indicate a data center installation, not a client device. The absence of any recorded benchmark scores means the database cannot quantify its real-world compute performance, but the architectural data shows a deliberate design for memory-intensive server tasks.

A user selecting between these two would choose the RTX 5090 Mobile for any graphics application, portable deployment, or API-driven rendering. The H20 would be the selection for large-scale compute and memory expansion, provided the workload does not require display output or graphics APIs. The benchmark results do not place these products in direct competition; they occupy separate segments with separate performance metrics. The recorded data supports the RTX 5090 Mobile as a measured performer and the H20 as an unmeasured but architecturally distinct alternative.

FAQ

Q: Does the NVIDIA H20 have any recorded benchmark scores in the database?

A: No. The H20 has an empty benchmark list and an average benchmark score of 0.

Q: How does the RTX 5090 Mobile compare to the RTX 4070 Ti in average benchmark score?

A: The RTX 5090 Mobile scores 45152, which is 0.8% ahead of the RTX 4070 Ti's 44795.

Q: What is the memory bandwidth difference between the two GPUs?

A: The H20 provides 4.03 TB/s from HBM3, while the RTX 5090 Mobile provides 896.0 GB/s from GDDR7.

Q: Which GPU has more shading units?

A: The RTX 5090 Mobile has 10496 shading units, compared to 9984 on the H20.

Q: Does the H20 support ray tracing?

A: The H20 lists no RT cores, while the RTX 5090 Mobile has 82 RT cores.

Q: What is the FP16 performance for each GPU?

A: The H20 delivers 79.07 TFLOPS with a 2:1 ratio, and the RTX 5090 Mobile delivers 31.80 TFLOPS with a 1:1 ratio.

Specification Differences

| Specification | NVIDIA GeForce RTX 5090 Mobile | NVIDIA H20 |

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

| Chip | GB203 | GH100 |

| Architecture | Blackwell 2.0 | Hopper |

| Generation | GeForce 50 Mobile | Server Hopper (Hxx) |

| Transistors | 45,600 million | 80,000 million |

| Die Size | 378 mm² | 814 mm² |

| Transistor Density | 120.6M / mm² | 98.3M / mm² |

| Base Clock | 990 MHz | 1830 MHz |

| Boost Clock | 1515 MHz | 1980 MHz |

| Memory Size | 24 GB | 96 GB |

| Memory Type | GDDR7 | HBM3 |

| Memory Bus Width | 256 bit | 6144 bit |

| Memory Clock | 1750 MHz 28 Gbps effective | 1313 MHz 5.3 Gbps effective |

| Memory Bandwidth | 896.0 GB/s | 4.03 TB/s |

| Shading Units | 10496 | 9984 |

| TMUs | 328 | 312 |

| ROPs | 112 | 24 |

| RT Cores | 82 | None |

| Tensor Cores | 328 | 312 |

| Pixel Rate | 169.7 GPixel/s | 47.52 GPixel/s |

| Texture Rate | 496.9 GTexel/s | 617.8 GTexel/s |

| FP32 Performance | 31.80 TFLOPS | 39.54 TFLOPS |

| FP16 Performance | 31.80 TFLOPS (1:1) | 79.07 TFLOPS (2:1) |

| TDP | 95 W | 500 W |

| Slot Width | IGP | SXM Module |

| Power Connectors | None | Not listed |

| Suggested PSU | Not listed | 900 W |

| Display Outputs | Portable Device Dependent | No outputs |

| DirectX Support | 12 Ultimate (12_2) | N/A |

| OpenGL Support | 4.6 | N/A |

| Vulkan Support | 1.4 | N/A |

| Release Date | 2025-03-26 | 2024-01-31 |

| Predecessor | GeForce 40 Mobile | Server Ada |

| Successor | None | Server Blackwell |