NVIDIA GeForce RTX 5090 D vs NVIDIA H20 Comparison

Where Each One Wins

The NVIDIA GeForce RTX 5090 D and NVIDIA H20 occupy entirely different segments of the GPU landscape, and the recorded data reflects this divergence clearly. The RTX 5090 D is a consumer-oriented Blackwell 2.0 part with a full suite of graphics benchmarks, while the H20 is a Hopper-generation server accelerator with no graphics benchmarks recorded and no display outputs.

For the RTX 5090 D, the benchmark data shows wins across every recorded test category. In 3DMark Steel Nomad DX12, it scores 14,326 points. In Geekbench OpenCL, it reaches 310,674, and in Geekbench Vulkan, 376,915. Passmark results show 44,065 in G3D, 28,396 in GPU Compute, and individual DirectX scores of 434 (DX9), 371 (DX11), 231 (DX10), and 219 (DX12). The 2D score is 1,487.

The H20, by contrast, has no benchmark entries in the database. Its percentile rank among all GPUs is 50, and its average benchmark score is recorded as zero. This is not a performance verdict but a data availability gap: the H20 is a server part with no gaming or workstation graphics test results in the database. The wins tally reflects this, with the RTX 5090 D taking all recorded wins and the H20 taking none.

The use-case split is stark. The RTX 5090 D is clearly aimed at high-end desktop graphics, with dual-slot cooling, a 16-pin power connector, and three DisplayPort 2.1b outputs plus one HDMI 2.1b. The H20 is an SXM module with no display outputs, suggesting a compute-only role in data center environments. Its memory configuration reinforces this: 96 GB of HBM3 on a 6,144-bit bus, yielding 4.03 TB/s of bandwidth, compared to the RTX 5090 D's 32 GB of GDDR7 on a 512-bit bus at 1.79 TB/s.

The recorded data shows no overlap in their intended workloads. The RTX 5090 D delivers graphics performance across DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. The H20 lists N/A for DirectX, OpenGL, and Vulkan, indicating it is not designed for traditional graphics APIs at all.

The Verdict

The data supports a clear separation of roles. The RTX 5090 D is the choice for any workload that requires rasterization, ray tracing, or general-purpose graphics through standard APIs. Its 104.8 TFLOPS FP32 throughput, 170 RT cores, and 680 tensor cores provide the computational foundation for those tasks. The H20, with 39.54 TFLOPS FP32 and 312 tensor cores, is configured for server-side compute, likely inference or training workloads, though the database does not contain benchmark results to quantify that.

The RTX 5090 D's percentile rank of 92 among all GPUs places it in the upper tier of recorded hardware. Its average benchmark score of 77,712 is nearly identical to the nearest rival, the AMD Radeon RX 6850M XT at 78,940, a delta of -1.6 percent. The RTX 5090 D also sits just 1.1 percent above the AMD Radeon RX 6650M XT (76,904) and 2.1 percent above the NVIDIA Tesla P100 PCIe 12 GB (79,396). These are narrow margins, indicating that in aggregate benchmark terms, the RTX 5090 D is competitive with those parts, though the specific workload mix matters.

The H20 has no nearest rivals listed and no average score, so no comparative verdict can be drawn from the database. The only recorded metric is its 50th percentile rank, which places it at the median of all GPUs in the database, but without benchmark scores, this is not actionable for performance predictions.

For a buyer or system integrator, the choice is dictated by the workload, not by head-to-head performance. The RTX 5090 D is the only option of the two with any graphics benchmark data. The H20's value cannot be assessed from the database, and its lack of display outputs and graphics API support makes it unsuitable for any desktop or workstation graphics role.

Head-to-Head Benchmarks

The database contains no head-to-head benchmark results between the RTX 5090 D and the H20. The wins counts are zero for both in that specific category. This is expected given that one part has a full benchmark suite and the other has none.

Looking at the RTX 5090 D's individual results provides context for its standing. In 3DMark Steel Nomad DX12, the 14,326 score is a synthetic measure of DirectX 12 performance. The Geekbench OpenCL score of 310,674 and Vulkan score of 376,915 indicate strong compute and graphics API performance respectively. The Vulkan score being higher than OpenCL by roughly 21 percent suggests the architecture favors Vulkan's lower overhead model.

Passmark results show a different picture. The G3D score of 44,065 is the most representative overall graphics metric. The GPU Compute score of 28,396 is lower, indicating that compute tasks do not scale as well as graphics tasks on this part. The DirectX scores are unusual: DX9 at 434 is the highest, DX11 at 371, DX10 at 231, and DX12 at 219. This inversion, where older APIs score higher, may reflect driver optimization priorities or the specific Passmark test methodology.

The nearest rivals for the RTX 5090 D provide a reference frame. The AMD Radeon RX 6850M XT is 1.6 percent ahead in average score, and the Tesla P100 variants are 2.1 and 2.4 percent ahead. These are small margins, all within 3 percent, suggesting that the RTX 5090 D's average score of 77,712 is tightly clustered with these parts. The delta of 1.1 percent above the RX 6650M XT is even smaller, indicating that the RTX 5090 D and that mobile GPU are nearly equivalent in aggregate benchmark terms.

The H20's absence from head-to-head comparisons is the key finding. Without benchmark data, there is no basis for a quantitative comparison. The database records its architecture, memory, and compute specifications, but no performance measurements.

FAQ

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

A: No. The H20 has an empty benchmarks array, an average benchmark score of zero, and no nearest rivals listed. Its percentile rank is 50, but no individual test scores are recorded.

Q: What is the RTX 5090 D's average benchmark score and percentile rank?

A: The RTX 5090 D has an average benchmark score of 77,712 and ranks in the 92nd percentile among all GPUs in the database.

Q: How does the RTX 5090 D compare to its nearest rival, the AMD Radeon RX 6850M XT?

A: The RX 6850M XT has an average score of 78,940, which is 1.6 percent higher than the RTX 5090 D's 77,712. This is the closest rival in the recorded data.

Q: What memory configurations do the two GPUs use?

A: The RTX 5090 D uses 32 GB of GDDR7 on a 512-bit bus with 1.79 TB/s bandwidth. The H20 uses 96 GB of HBM3 on a 6,144-bit bus with 4.03 TB/s bandwidth.

Q: Which GPU supports graphics APIs like DirectX and Vulkan?

A: The RTX 5090 D supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The H20 lists N/A for all three graphics APIs.

Q: What are the FP32 compute figures for each GPU?

A: The RTX 5090 D delivers 104.8 TFLOPS FP32. The H20 delivers 39.54 TFLOPS FP32, which is approximately 38 percent of the RTX 5090 D's FP32 throughput.

Architecture Differences

The RTX 5090 D and H20 are built on different architectures from different generations. The RTX 5090 D uses the GB202 chip with Blackwell 2.0 architecture, part of the GeForce 50-series generation. The H20 uses the GH100 chip with Hopper architecture, part of the Server Hopper (Hxx) generation.

Both are fabricated on a 5 nm process at TSMC, but the die characteristics differ. The RTX 5090 D has 92,200 million transistors on a 750 mm² die, giving a transistor density of 122.9 million per square millimeter. The H20 has 80,000 million transistors on a larger 814 mm² die, resulting in a lower density of 98.3 million per square millimeter. The H20's die is 8.5 percent larger but contains 13 percent fewer transistors.

Clock speeds also differ. The RTX 5090 D has a base clock of 2017 MHz and a boost clock of 2407 MHz. The H20 has a base clock of 1830 MHz and a boost clock of 1980 MHz. The RTX 5090 D's boost clock is 21.6 percent higher. Memory clocks show a larger gap: the RTX 5090 D runs at 1750 MHz with 28 Gbps effective, while the H20 runs at 1313 MHz with 5.3 Gbps effective. The H20's much wider 6,144-bit bus compensates, yielding 4.03 TB/s bandwidth versus 1.79 TB/s.

The compute unit counts are substantially different. The RTX 5090 D has 21,760 shading units, 680 TMUs, and 176 ROPs. The H20 has 9,984 shading units, 312 TMUs, and only 24 ROPs. The RTX 5090 D has 170 RT cores, while the H20 has none recorded. Tensor core counts are 680 for the RTX 5090 D and 312 for the H20. These counts drive the pixel rate (423.6 GPixel/s for the RTX 5090 D vs 47.52 GPixel/s for the H20) and texture rate (1,636.8 GTexel/s vs 617.8 GTexel/s).

FP32 throughput is 104.8 TFLOPS for the RTX 5090 D and 39.54 TFLOPS for the H20. FP16 performance differs in ratio: the RTX 5090 D achieves 104.8 TFLOPS at a 1:1 ratio, while the H20 achieves 79.07 TFLOPS at a 2:1 ratio. This indicates the H20 is optimized for reduced-precision compute, while the RTX 5090 D handles FP16 and FP32 at the same rate.

Power specifications also differ. The RTX 5090 D has a TDP of 575 W with a suggested PSU of 950 W. The H20 has a TDP of 500 W with a suggested PSU of 900 W. The RTX 5090 D is a dual-slot card with one 16-pin connector, while the H20 is an SXM module with no power connector listed.

Physical and interface differences are significant. The RTX 5090 D measures 304 mm in length, 137 mm in height, and 48 mm in width, using PCIe 5.0 x16. The H20 has no dimensions recorded and also uses PCIe 5.0 x16. The RTX 5090 D has display outputs (one HDMI 2.1b and three DisplayPort 2.1b), while the H20 has no outputs.

Release dates show the H20 launched on 2024-01-31, nearly a year before the RTX 5090 D on 2025-01-29. The RTX 5090 D has a launch MSRP of 2,299 USD, while the H20 has no MSRP recorded. Production status is Active for both. The RTX 5090 D's predecessor is the GeForce 40 series, and its successor is GeForce 60. The H20's predecessor is Server Ada, and its successor is Server Blackwell.