NVIDIA GeForce RTX 5090 SE vs NVIDIA H200 NVL Comparison

Where Each One Wins

The recorded data separates these two NVIDIA accelerators into entirely different use cases. The GeForce RTX 5090 SE is a client-oriented graphics card built around the Blackwell 2.0 architecture, while the H200 NVL is a server compute accelerator based on Hopper. Neither device has a direct head-to-head benchmark entry in the database, so the comparison relies on their architectural specifications, memory subsystems, and available benchmark results.

The RTX 5090 SE wins in every category tied to rasterization and real-time graphics. It delivers 66.94 TFLOPS of FP32 compute, 380.3 GPixel/s pixel throughput, and 1,045.9 GTexel/s texture fill. It carries 160 ROPs, 440 TMUs, and 110 RT cores, making it the only one of the two with any ray tracing hardware. It also supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, which means it can run modern games and graphics workloads. The H200 NVL has no display outputs, no RT cores, and no graphics API support, so it cannot render frames or run client software at all.

The H200 NVL wins in memory capacity and bandwidth. It packs 141 GB of HBM3e on a 6144-bit bus, delivering 4.89 TB/s of memory bandwidth. The RTX 5090 SE has 24 GB of GDDR7 on a 384-bit bus, yielding 1.34 TB/s. That is a 3.55 TB/s gap in raw bandwidth, and a 117 GB gap in capacity. For large model inference, training datasets, or in-memory databases, the H200 NVL holds a decisive advantage.

The H200 NVL also leads in FP16 throughput. It delivers 120.6 TFLOPS with a 2:1 ratio relative to FP32, while the RTX 5090 SE provides 66.94 TFLOPS at a 1:1 ratio. That makes the H200 NVL nearly twice as fast in half-precision compute, which is the dominant precision for many AI inference and training workloads.

Architecture Differences

The two chips come from different architectural families. The RTX 5090 SE uses the GB202 die, built on Blackwell 2.0, fabricated by TSMC on a 5 nm process. The die measures 750 mm² and integrates 92,200 million transistors, resulting in a transistor density of 122.9 million per mm². The H200 NVL uses the GH100 die, built on Hopper, also on TSMC 5 nm. Its die is larger at 814 mm² but packs fewer transistors at 80,000 million, giving a lower density of 98.3 million per mm². The higher density of the GB202 indicates a more compact logic layout, while the GH100 uses its larger area for a different balance of compute and memory.

The memory architecture is fundamentally different. The RTX 5090 SE uses GDDR7, a conventional discrete memory type mounted on a 384-bit bus. The H200 NVL uses HBM3e stacked memory on a 6144-bit bus. That 16x difference in bus width explains the massive bandwidth gap. The H200 NVL also has far more memory capacity because HBM3e stacks allow much larger total capacities per package.

The compute configurations differ substantially. The RTX 5090 SE has 14,080 shading units, 440 TMUs, 160 ROPs, 110 RT cores, and 440 tensor cores. The H200 NVL has 16,896 shading units, 528 TMUs, only 24 ROPs, no RT cores, and 528 tensor cores. The H200 has 2,816 more shading units and 88 more tensor cores, but its rasterization hardware is minimal, which reflects its server role. The RTX 5090 SE has 136 more ROPs, which is critical for pixel output.

Clock speeds also diverge. The RTX 5090 SE runs at a base clock of 1740 MHz and boosts to 2377 MHz. The H200 NVL runs at 1365 MHz base and 1785 MHz boost. That 592 MHz boost clock difference explains why the RTX 5090 SE achieves higher FP32 throughput despite having fewer shading units. The H200 NVL compensates with a 2:1 FP16 ratio, effectively doubling its half-precision output.

Power delivery and physical design reflect their different installation contexts. The RTX 5090 SE has a 500 W TDP and uses a single 16-pin power connector, with a suggested PSU of 900 W. The H200 NVL has a 600 W TDP and uses an 8-pin EPS connector, with a suggested PSU of 1000 W. Both are dual-slot cards, and both share the same length of 267 mm and height of 111 mm. The RTX 5090 SE has a width of 40 mm; the H200 NVL width is not recorded.

Head-to-Head Benchmarks

The database contains no direct head-to-head benchmark entries between these two products. However, the H200 NVL has one recorded benchmark result: 334,891 in Geekbench OpenCL. That score places it in the 100th percentile of all GPUs in the database. Its nearest rivals provide context. The NVIDIA B200 scores 345,482, which is 3.1% higher. The NVIDIA B300 SXM6 AC scores 369,831, which is 9.4% higher. The AMD Instinct MI300X scores 317,994, which is 5.3% lower. The NVIDIA L40S scores 295,763, which is 13.2% lower. So the H200 NVL sits between the MI300X and the B200, closer to the B200 than to the L40S.

The RTX 5090 SE has no benchmark scores recorded in the database. Its percentile is listed at 50, and its average benchmark score is zero. That means there is no measured performance data to compare directly against the H200 NVL's OpenCL result. The RTX 5090 SE's performance profile must be inferred from its raw specifications.

What the data does show is a clear division of compute strengths. In FP32, the RTX 5090 SE produces 66.94 TFLOPS versus 60.32 TFLOPS for the H200 NVL, a 10.97 TFLOPS advantage for the GeForce card. In FP16, the H200 NVL produces 120.6 TFLOPS versus 66.94 TFLOPS for the RTX 5090 SE, a 53.66 TFLOPS advantage for the Hopper card. Pixel rate favors the RTX 5090 SE massively: 380.3 GPixel/s versus 42.84 GPixel/s, a 337.46 GPixel/s gap. Texture rate favors the RTX 5090 SE by 103.4 GTexel/s, with 1,045.9 versus 942.5.

Memory bandwidth favors the H200 NVL by 3.55 TB/s, with 4.89 TB/s versus 1.34 TB/s. The H200 NVL also has a 117 GB capacity advantage. Those are the dominant factors for AI workloads. The RTX 5090 SE's advantages in pixel rate, texture rate, and FP32 are the dominant factors for graphics workloads.

FAQ

Q: Which card has more memory bandwidth?

A: The H200 NVL has 4.89 TB/s from HBM3e on a 6144-bit bus. The RTX 5090 SE has 1.34 TB/s from GDDR7 on a 384-bit bus. The H200 NVL leads by 3.55 TB/s.

Q: Which card supports ray tracing?

A: Only the RTX 5090 SE. It has 110 RT cores and supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. The H200 NVL has no RT cores and no graphics API support.

Q: How do their FP16 compute rates compare?

A: The H200 NVL delivers 120.6 TFLOPS at a 2:1 FP16 ratio. The RTX 5090 SE delivers 66.94 TFLOPS at a 1:1 ratio. The H200 NVL is 53.66 TFLOPS faster in half-precision.

Q: What is the H200 NVL's nearest rival in the database?

A: The NVIDIA B200 scores 345,482, which is 3.1% higher than the H200 NVL's 334,891. The AMD Instinct MI300X scores 317,994, which is 5.3% lower.

Q: Do both cards use the same process node?

A: Yes, both are fabricated by TSMC on a 5 nm process. The RTX 5090 SE uses the GB202 die at 750 mm², and the H200 NVL uses the GH100 die at 814 mm².

Q: Which card has a higher boost clock?

A: The RTX 5090 SE boosts to 2377 MHz, while the H200 NVL boosts to 1785 MHz. The GeForce card has a 592 MHz higher boost clock.

The Verdict

The data points to a clear split based on workload type. The RTX 5090 SE is the choice for real-time graphics, rasterization, and client-side rendering. It has 160 ROPs, 110 RT cores, 66.94 TFLOPS of FP32, and full graphics API support. It also has a 500 W TDP and a 16-pin connector, which fits a desktop build. The H200 NVL cannot do any of that, as it has no display outputs and no graphics APIs.

The H200 NVL is the choice for AI inference and training that depends on memory capacity and bandwidth. Its 141 GB HBM3e and 4.89 TB/s bandwidth are far beyond what the RTX 5090 SE can offer. Its 120.6 TFLOPS FP16 throughput is the highest compute figure in this comparison. Its 600 W TDP and 8-pin EPS connector suggest a server power delivery system.

For a builder assembling a workstation that needs both graphics and compute, the RTX 5090 SE is the only viable option because the H200 NVL cannot output video. For a data center operator running large models, the H200 NVL is the only option with sufficient memory. The GeForce card wins on pixel throughput, texture throughput, FP32, and clock speed. The Hopper card wins on memory capacity, memory bandwidth, and FP16.

Neither card is a substitute for the other. The RTX 5090 SE occupies a 50th percentile position in the database with no recorded benchmarks, while the H200 NVL sits at the 100th percentile with a 334,891 OpenCL score. That percentile gap reflects the database's composition, which includes server accelerators with measured results and client GPUs without comparable entries. The practical takeaway is that the RTX 5090 SE belongs in a gaming or content-creation rig, and the H200 NVL belongs in a server rack.

Specification Differences

The following fields differ between the two products. The RTX 5090 SE uses the GB202 chip, Blackwell 2.0 architecture, and belongs to the GeForce 50 generation. The H200 NVL uses the GH100 chip, Hopper architecture, and belongs to the Server Hopper generation. The RTX 5090 SE has 92,200 million transistors on a 750 mm² die with a density of 122.9 million per mm². The H200 NVL has 80,000 million transistors on an 814 mm² die with a density of 98.3 million per mm².

Clock speeds differ: the RTX 5090 SE runs at 1740 MHz base and 2377 MHz boost, with memory at 1750 MHz and 28 Gbps effective. The H200 NVL runs at 1365 MHz base and 1785 MHz boost, with memory at 1593 MHz and 6.4 Gbps effective. Memory capacity is 24 GB GDDR7 for the RTX 5090 SE versus 141 GB HBM3e for the H200 NVL. Bus width is 384 bit versus 6144 bit. Bandwidth is 1.34 TB/s versus 4.89 TB/s.

Shading units are 14,080 versus 16,896. TMUs are 440 versus 528. ROPs are 160 versus 24. RT cores are 110 versus none. Tensor cores are 440 versus 528. Pixel rate is 380.3 GPixel/s versus 42.84 GPixel/s. Texture rate is 1,045.9 GTexel/s versus 942.5 GTexel/s. FP32 is 66.94 TFLOPS versus 60.32 TFLOPS. FP16 is 66.94 TFLOPS at 1:1 versus 120.6 TFLOPS at 2:1.

TDP is 500 W versus 600 W. Power connectors are 1x 16-pin versus 8-pin EPS. Suggested PSU is 900 W versus 1000 W. Display outputs are 1x HDMI 2.1b and 3x DisplayPort 2.1b versus no outputs. API support is DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 versus N/A for all three. The RTX 5090 SE has a width of 40 mm; the H200 NVL width is not recorded. Release dates are 2025-12-31 for the RTX 5090 SE and 2024-11-17 for the H200 NVL. The RTX 5090 SE has a launch MSRP of 1,499 USD; the H200 NVL has no recorded launch MSRP.