NVIDIA CMP 50HX vs NVIDIA GeForce RTX 5090 Mobile Comparison

The NVIDIA CMP 50HX and NVIDIA GeForce RTX 5090 Mobile represent two radically different design philosophies from the same manufacturer, separated by a generation of GPU architecture and targeting entirely distinct use cases. The data shows a clear performance hierarchy between the two, but the nature of that gap is shaped by their respective roles as a mining-specific product and a mobile flagship.

Head-to-Head Benchmarks

The only two benchmarks shared by both GPUs in the database are Geekbench OpenCL and Geekbench Vulkan, and the results are decisively lopsided. In Geekbench OpenCL, the RTX 5090 Mobile scores 201,834, which is 145,699 points higher than the CMP 50HX’s 56,135. This translates to a delta of -72.2% from the perspective of the CMP 50HX, meaning the mobile part is roughly 3.6 times faster in this compute-heavy test. The Geekbench Vulkan result is even more one-sided: the RTX 5090 Mobile posts 198,405 against the CMP 50HX’s 47,445, a delta of -76.1%. The RTX 5090 Mobile is over four times faster in raw Vulkan compute throughput.

These are not marginal victories; the RTX 5090 Mobile’s scores are in a different league entirely. The CMP 50HX’s average benchmark score of 51,790 places it in the 86th percentile of all GPUs, while the RTX 5090 Mobile’s average of 45,152 sits in the 84th percentile. This apparent contradiction — where the slower card has a higher percentile — is explained by the differing benchmark suites each card was subjected to. The CMP 50HX was only tested on Geekbench, while the RTX 5090 Mobile includes PassMark tests, which drag its average down despite its superior raw performance in the shared tests.

Within its own rival set, the CMP 50HX leads the AMD Radeon RX 6900 XT by 1.6%, the AMD Radeon RX Vega 64 by 3.6%, the NVIDIA GeForce RTX 5070 Ti by 3.7%, and the Intel Arc A550M by 4.1% in average benchmark score. The RTX 5090 Mobile, by contrast, edges out the NVIDIA GeForce RTX 4070 Ti by 0.8%, while trailing the AMD Radeon Pro 5500 XT by 0.5%, the Intel Arc A730M by 1%, and the NVIDIA RTX 5880 Ada Generation by 1.8%. These rival comparisons show that the CMP 50HX is a strong performer relative to its peers, while the RTX 5090 Mobile sits in a tighter, more contested performance band among its own competitors.

Architecture Differences

The architectural gulf between these two parts is substantial. The CMP 50HX is built on the Turing architecture using the TU102 chip, fabricated on TSMC’s 12 nm process. It packs 18,600 million transistors onto a die size of 754 mm², yielding a transistor density of 24.7 million per square millimeter. The RTX 5090 Mobile, in contrast, uses the Blackwell 2.0 architecture with the GB203 chip, built on a 5 nm process. It contains 45,600 million transistors on a die of just 378 mm², achieving a transistor density of 120.6 million per square millimeter — nearly five times denser than the Turing part.

The memory subsystems are equally divergent. The CMP 50HX has 10 GB of GDDR6 memory on a 320-bit bus, delivering 560.0 GB/s of bandwidth. The RTX 5090 Mobile offers 24 GB of GDDR7 memory on a 256-bit bus, with bandwidth reaching 896.0 GB/s. Despite the narrower bus, the newer memory type provides substantially higher bandwidth. The CMP 50HX’s memory clock is 1750 MHz with 14 Gbps effective speed, while the RTX 5090 Mobile also runs 1750 MHz but achieves 28 Gbps effective due to GDDR7’s signaling improvements.

Compute resources tell a similar story of generational leap. The CMP 50HX has 3,584 shading units, 192 texture mapping units, and 80 raster operation units. It also includes 56 ray tracing cores and 448 tensor cores. The RTX 5090 Mobile more than triples the shading units to 10,496, with 328 TMUs and 112 ROPs. It has 82 ray tracing cores and 328 tensor cores — fewer tensor cores in raw count, but the Blackwell architecture’s design makes them far more efficient. The FP32 throughput is 11.07 TFLOPS for the CMP 50HX versus 31.80 TFLOPS for the RTX 5090 Mobile. The FP16 figures are 22.15 TFLOPS (2:1 rate) for the CMP 50HX and 31.80 TFLOPS (1:1 rate) for the RTX 5090 Mobile.

Clock speeds are counterintuitive: the CMP 50HX has a base clock of 1350 MHz and boost of 1545 MHz, while the RTX 5090 Mobile runs a lower 990 MHz base and 1515 MHz boost. The mobile part’s efficiency comes from its architecture rather than raw clocks. The CMP 50HX is rated at 250 W TDP with a dual-slot cooler and two 8-pin power connectors, requiring a 600 W power supply. The RTX 5090 Mobile is an integrated graphics processor (IGP) with a 95 W TDP, no power connectors, and no dedicated slot width — it is soldered to a motherboard.

The CMP 50HX has no display outputs, reflecting its mining-focused design, while the RTX 5090 Mobile’s display outputs are listed as “Portable Device Dependent.” Both support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The CMP 50HX uses a PCIe 1.0 x4 interface, while the RTX 5090 Mobile uses PCIe 5.0 x16 — a significant difference in host connectivity.

FAQ

Q: Which GPU has higher raw FP32 performance?

A: The RTX 5090 Mobile has 31.80 TFLOPS FP32, nearly triple the CMP 50HX’s 11.07 TFLOPS.

Q: How do their memory configurations compare?

A: The CMP 50HX uses 10 GB of GDDR6 on a 320-bit bus with 560.0 GB/s bandwidth. The RTX 5090 Mobile has 24 GB of GDDR7 on a 256-bit bus with 896.0 GB/s bandwidth.

Q: What is the power consumption difference?

A: The CMP 50HX is rated at 250 W TDP with two 8-pin connectors and a 600 W suggested PSU. The RTX 5090 Mobile is rated at 95 W TDP with no power connectors.

Q: Why is the CMP 50HX’s percentile higher than the RTX 5090 Mobile’s despite losing all shared benchmarks?

A: The CMP 50HX was tested only on Geekbench OpenCL and Vulkan, while the RTX 5090 Mobile includes PassMark tests in its average. The PassMark scores (G3D of 30,034, GPU Compute of 13,401) lower its average, while the CMP 50HX’s average reflects only its stronger Geekbench results.

Q: Which GPU has more shading units?

A: The RTX 5090 Mobile has 10,496 shading units, compared to 3,584 for the CMP 50HX.

Q: Are both GPUs currently in production?

A: No. The CMP 50HX is marked as End-of-life, while the RTX 5090 Mobile is Active.

The Verdict

The data presents a clear case: the RTX 5090 Mobile is the superior performer in every shared benchmark, and its architectural advantages are overwhelming. The Geekbench OpenCL and Vulkan deltas of -72.2% and -76.1% respectively demonstrate that the mobile part is not merely faster — it is in a different performance class entirely. The RTX 5090 Mobile’s 31.80 TFLOPS FP32 versus the CMP 50HX’s 11.07 TFLOPS, combined with 896.0 GB/s of memory bandwidth versus 560.0 GB/s, leaves no ambiguity about which GPU delivers more compute capability.

However, the CMP 50HX is not without merit. Its 86th percentile placement against all GPUs, its 250 W TDP (which allows for sustained load in a desktop context), and its 10 GB GDDR6 memory on a wider 320-bit bus could make it suitable for specific workloads that favor memory width over capacity. Its end-of-life status and lack of display outputs, though, severely limit its utility outside of compute tasks.

For any user seeking general-purpose graphics or compute performance, the RTX 5090 Mobile is the only rational choice from this comparison. The 84th percentile placement for the mobile part is misleading given its PassMark-inclusive average; the shared Geekbench scores show it is dramatically ahead. The CMP 50HX might appeal to a niche audience needing a mining-specific or headless compute card with no display requirements, but its performance deficit and end-of-life status make it hard to justify against the mobile part’s capabilities.

Specification Differences

| Specification | NVIDIA CMP 50HX | NVIDIA GeForce RTX 5090 Mobile |

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

| Architecture | Turing | Blackwell 2.0 |

| Process Node | 12 nm | 5 nm |

| Transistors | 18,600 million | 45,600 million |

| Die Size | 754 mm² | 378 mm² |

| Transistor Density | 24.7M / mm² | 120.6M / mm² |

| Base Clock | 1350 MHz | 990 MHz |

| Boost Clock | 1545 MHz | 1515 MHz |

| Memory Size | 10 GB | 24 GB |

| Memory Type | GDDR6 | GDDR7 |

| Memory Bus | 320 bit | 256 bit |

| Memory Bandwidth | 560.0 GB/s | 896.0 GB/s |

| Shading Units | 3584 | 10496 |

| TMUs | 192 | 328 |

| ROPs | 80 | 112 |

| RT Cores | 56 | 82 |

| Tensor Cores | 448 | 328 |

| FP32 | 11.07 TFLOPS | 31.80 TFLOPS |

| FP16 | 22.15 TFLOPS (2:1) | 31.80 TFLOPS (1:1) |

| TDP | 250 W | 95 W |

| Slot Width | Dual-slot | IGP |

| Power Connectors | 2x 8-pin | None |

| Suggested PSU | 600 W | None |

| Bus Interface | PCIe 1.0 x4 | PCIe 5.0 x16 |

| Display Outputs | No outputs | Portable Device Dependent |

| Production Status | End-of-life | Active |

| Release Date | 2021-06-23 | 2025-03-26 |

| Predecessor | None | GeForce 40 Mobile |

Where Each One Wins

The RTX 5090 Mobile wins every shared benchmark category. In Geekbench OpenCL, it scores 201,834 versus 56,135 — a 145,699-point advantage that translates to roughly 3.6 times the performance. In Geekbench Vulkan, the gap is 150,960 points (198,405 versus 47,445), representing over four times the performance. The mobile part also wins on memory capacity (24 GB versus 10 GB), memory bandwidth (896.0 GB/s versus 560.0 GB/s), shading unit count (10,496 versus 3,584), and FP32 throughput (31.80 TFLOPS versus 11.07 TFLOPS).

The CMP 50HX wins on none of the shared benchmarks, but it does hold advantages in specific areas: it has a higher base clock (1350 MHz versus 990 MHz) and boost clock (1545 MHz versus 1515 MHz), a wider memory bus (320 bit versus 256 bit), more tensor cores (448 versus 328), and a larger die area (754 mm² versus 378 mm²). It also has a higher transistor density figure when calculated per square millimeter? No — the RTX 5090 Mobile has higher density at 120.6M / mm² versus 24.7M / mm². The CMP 50HX’s advantages are largely irrelevant for real-world compute workloads, as its lower shading unit count and older architecture negate the clock speed benefits.

For use-case selection: the RTX 5090 Mobile is the pick for any workload involving modern gaming, ray tracing, AI inference, or general-purpose compute where its 82 RT cores, 328 tensor cores, and 24 GB GDDR7 memory provide both speed and capacity. The CMP 50HX, with its 10 GB GDDR6, no display outputs, and end-of-life status, is only viable for specific headless compute tasks where its 250 W TDP and 2x 8-pin connectors fit a desktop mining or server setup that predates modern standards. The data overwhelmingly favors the RTX 5090 Mobile for virtually any application.