NVIDIA H200 NVL vs Lisuan Tech LX ULTRA Comparison

Where Each One Wins

The NVIDIA H200 NVL and Lisuan Tech LX ULTRA occupy entirely different segments of the GPU market, and the data reflects that split clearly. The H200 NVL is a server-class accelerator built for massive compute workloads, while the LX ULTRA is a consumer-oriented graphics card with display outputs and full API support.

The H200 NVL wins decisively in raw compute throughput. Its FP32 performance reaches 60.32 TFLOPS, which is roughly 2.45 times the LX ULTRA's 24.58 TFLOPS. In FP16, the H200 NVL delivers 120.6 TFLOPS compared to the LX ULTRA's 49.15 TFLOPS, a 2.45x advantage that matters for AI inference and training workloads. The H200 NVL also holds a massive memory advantage with 141 GB of HBM3e versus 24 GB of GDDR6, and its 4.89 TB/s bandwidth dwarfs the LX ULTRA's 432.0 GB/s by more than 11 times.

The LX ULTRA wins in areas that directly affect gaming and desktop use. It has 4x DisplayPort 1.4a outputs, whereas the H200 NVL has no display outputs at all. The LX ULTRA also supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.3, while the H200 NVL reports N/A for all three APIs. The LX ULTRA's pixel rate of 192.0 GPixel/s is more than four times the H200 NVL's 42.84 GPixel/s, which suggests the LX ULTRA is better suited for rasterization-heavy rendering tasks despite its lower overall compute throughput.

In texture work, the H200 NVL leads with 942.5 GTexel/s versus 384.0 GTexel/s, a 2.45x margin. The LX ULTRA has a higher ROP count at 96 versus 24, which reinforces its suitability for final pixel output in graphics pipelines.

Architecture Differences

The two cards come from different design philosophies. The H200 NVL uses the GH100 chip on NVIDIA's Hopper architecture, built on a 5 nm process at TSMC with 80,000 million transistors on an 814 mm² die. The transistor density works out to 98.3M per mm². The LX ULTRA uses the 7G105 chip on Lisuan Tech's TrueGPU architecture, built on a 6 nm process at TSMC, with transistor count and die size listed as unknown.

The H200 NVL has 16,896 shading units, 528 TMUs, and 24 ROPs. It also packs 528 tensor cores, which are essential for the AI and deep learning workloads this card targets. The LX ULTRA has 6,144 shading units, 192 TMUs, and 96 ROPs, with no tensor cores listed. That absence of tensor cores is a clear differentiator: the H200 NVL is designed for matrix math acceleration, while the LX ULTRA is not.

Memory architecture could not be more different. The H200 NVL uses 141 GB of HBM3e across a 6144-bit bus, achieving 4.89 TB/s. The LX ULTRA uses 24 GB of GDDR6 across a 192-bit bus, achieving 432.0 GB/s. Memory clock rates show the GDDR6 running at 2250 MHz (18 Gbps effective) on the LX ULTRA, while the H200 NVL's memory runs at 1593 MHz (6.4 Gbps effective), but the HBM3e's wider bus and higher total bandwidth make the comparison moot.

Power and interface differences also stand out. The H200 NVL draws up to 600 W with an 8-pin EPS connector and a suggested 1000 W PSU, while the LX ULTRA draws 225 W with a single 16-pin connector and a suggested 550 W PSU. The H200 NVL uses PCIe 5.0 x16, the LX ULTRA uses PCIe 4.0 x16. Both are dual-slot cards, with the H200 NVL at 267 mm (10.5 inches) long and the LX ULTRA at 268 mm (10.6 inches), nearly identical in physical footprint.

The H200 NVL's generation is listed as Server Hopper (Hxx), released on 2024-11-17, with predecessor Server Ada and successor Server Blackwell. The LX ULTRA's generation is 7G100, released on 2026-03-16, with no predecessor or successor listed. The H200 NVL's production status is Active, as is the LX ULTRA's.

Head-to-Head Benchmarks

The database records a single benchmark for the H200 NVL: Geekbench OpenCL score of 334,891. This places it at the 100th percentile among all GPUs, meaning it outperforms every other recorded GPU in that test. The LX ULTRA has no recorded benchmarks, with an average benchmark score of 0 and a percentile of 50.

The H200 NVL's nearest rivals provide context for its standing. The NVIDIA B200 scores 345,482, which is 3.1% higher than the H200 NVL. 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 than the H200 NVL. The NVIDIA L40S scores 295,763, which is 13.2% lower.

The head-to-head benchmark comparison between the H200 NVL and LX ULTRA is empty in the database, and the win counts are zero for both. This means no direct comparative measurements exist, so any analysis must rely on the architectural specifications and the H200 NVL's absolute benchmark score.

Looking at compute metrics directly, the H200 NVL's FP32 output of 60.32 TFLOPS is 2.45 times the LX ULTRA's 24.58 TFLOPS. The FP16 figures follow the same ratio: 120.6 TFLOPS versus 49.15 TFLOPS. Texture rate also runs at 2.45x in favor of the H200 NVL, with 942.5 GTexel/s versus 384.0 GTexel/s.

The LX ULTRA counters with a pixel rate of 192.0 GPixel/s, which is 4.48 times the H200 NVL's 42.84 GPixel/s. This is a clear indication that the LX ULTRA was designed with different priorities, favoring final pixel output over raw texture and compute throughput.

Memory bandwidth is the most lopsided specification. The H200 NVL's 4.89 TB/s is 11.3 times the LX ULTRA's 432.0 GB/s. Memory capacity shows a similar gap, with 141 GB versus 24 GB, a 5.9x difference.

The Verdict

The data indicates two purpose-built products with almost no overlap in intended use. The H200 NVL is a compute accelerator for servers, evidenced by its lack of display outputs, N/A API support, tensor cores, 600 W power draw, and PCIe 5.0 interface. Its 100th percentile Geekbench OpenCL score and its position among rivals like the B200 and B300 confirm it sits at the top of the compute hierarchy.

The LX ULTRA is a desktop graphics card, with 4x DisplayPort 1.4a outputs, DirectX 12 Ultimate support, a 225 W power draw, and a 550 W suggested PSU. Its higher pixel rate and ROP count, combined with lower power consumption, point toward traditional rendering workloads rather than AI or HPC tasks.

For buyers who need massive memory capacity, tensor core acceleration, and top-tier compute throughput, the H200 NVL is the clear choice based on the recorded data. Its 141 GB of HBM3e and 4.89 TB/s bandwidth are unmatched in this comparison, and its benchmark score places it at the top of the database.

For buyers who need a graphics card with display outputs, modern API support, and lower power requirements, the LX ULTRA is the only viable option between these two, since the H200 NVL cannot output video at all. The LX ULTRA's 192.0 GPixel/s pixel rate and 96 ROPs make it better suited to rasterization-heavy workloads.

The absence of tensor cores on the LX ULTRA is a decisive factor for AI workloads. The H200 NVL's 528 tensor cores are built for matrix operations, and its FP16 throughput of 120.6 TFLOPS supports that role. The LX ULTRA has no such hardware.

FAQ

Q: Which GPU has higher FP32 compute performance?

A: The NVIDIA H200 NVL delivers 60.32 TFLOPS FP32, which is 2.45 times the Lisuan Tech LX ULTRA's 24.58 TFLOPS.

Q: Does the LX ULTRA support display outputs?

A: Yes, the LX ULTRA has 4x DisplayPort 1.4a outputs. The H200 NVL has no display outputs.

Q: How do their memory bandwidths compare?

A: The H200 NVL has 4.89 TB/s from 141 GB of HBM3e on a 6144-bit bus. The LX ULTRA has 432.0 GB/s from 24 GB of GDDR6 on a 192-bit bus. The H200 NVL's bandwidth is 11.3 times higher.

Q: Which GPU supports DirectX 12 Ultimate?

A: The LX ULTRA supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.3. The H200 NVL lists N/A for DirectX, OpenGL, and Vulkan.

Q: What is the power draw difference?

A: The H200 NVL has a TDP of 600 W with a suggested 1000 W PSU. The LX ULTRA has a TDP of 225 W with a suggested 550 W PSU.

Q: Does the H200 NVL have tensor cores?

A: Yes, the H200 NVL has 528 tensor cores. The LX ULTRA has no tensor cores listed.

Specification Differences

| Specification | NVIDIA H200 NVL | Lisuan Tech LX ULTRA |

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

| Architecture | Hopper | TrueGPU |

| Process Node | 5 nm | 6 nm |

| Transistors | 80,000 million | Unknown |

| Die Size | 814 mm² | Unknown |

| Base Clock | 1365 MHz | None listed |

| Boost Clock | 1785 MHz | None listed |

| Memory Size | 141 GB | 24 GB |

| Memory Type | HBM3e | GDDR6 |

| Memory Bus | 6144 bit | 192 bit |

| Memory Bandwidth | 4.89 TB/s | 432.0 GB/s |

| Shading Units | 16896 | 6144 |

| TMUs | 528 | 192 |

| ROPs | 24 | 96 |

| Tensor Cores | 528 | None |

| FP32 | 60.32 TFLOPS | 24.58 TFLOPS |

| FP16 | 120.6 TFLOPS | 49.15 TFLOPS |

| Pixel Rate | 42.84 GPixel/s | 192.0 GPixel/s |

| Texture Rate | 942.5 GTexel/s | 384.0 GTexel/s |

| TDP | 600 W | 225 W |

| Power Connectors | 8-pin EPS | 1x 16-pin |

| Suggested PSU | 1000 W | 550 W |

| Bus Interface | PCIe 5.0 x16 | PCIe 4.0 x16 |

| Display Outputs | No outputs | 4x DisplayPort 1.4a |

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

| OpenGL | N/A | 4.6 |

| Vulkan | N/A | 1.3 |

| Length | 267 mm | 268 mm |

| Release Date | 2024-11-17 | 2026-03-16 |