AMD Radeon PRO V710 vs NVIDIA CMP 90HX Comparison

The NVIDIA CMP 90HX and AMD Radeon PRO V710 are both professional-grade GPUs designed for compute workloads, but they target different ends of the efficiency and performance spectrum. The CMP 90HX is a mining-specific Ampere card with no display outputs, while the PRO V710 is a RDNA 3.0 workstation card that also lacks display outputs. Benchmark data shows the PRO V710 delivers a massive 68.8% higher OpenCL score, but the CMP 90HX holds a higher overall percentile ranking due to averaging across different test suites. The verdict is straightforward: the PRO V710 is the clear winner for raw compute throughput, but the CMP 90HX holds its own in specific legacy workloads where its architecture excels.

The Verdict

For anyone prioritizing raw compute performance in OpenCL-based applications, the AMD Radeon PRO V710 is the definitive choice. Its Geekbench OpenCL score of 116,460 absolutely dwarfs the NVIDIA CMP 90HX’s 69,000, representing a 68.8% advantage in that specific test. This is not a marginal difference; it is a generational leap in compute capability that makes the PRO V710 the obvious pick for rendering, simulation, or any workload that scales with FP32 throughput. The data shows the PRO V710 delivers 27.65 TFLOPS of FP32 performance compared to the CMP 90HX’s 21.89 TFLOPS, a 26.3% raw compute advantage.

However, the CMP 90HX is not without merit. Its 90th percentile ranking among all GPUs, based on its single benchmark score, is higher than the PRO V710’s 88th percentile. This is because the CMP 90HX’s average benchmark score of 69,000 is its sole data point, while the PRO V710’s average of 58,657 is dragged down by its 3DMark Steel Nomad DX12 score of 853. If your workload is strictly OpenCL compute, the PRO V710 is superior. But if you need a card that performs adequately across a broader mix of tests, the CMP 90HX’s consistency is notable.

Choose the CMP 90HX only if you have a specific legacy workload that favors NVIDIA’s Ampere architecture and you can tolerate its 320W TDP. Choose the PRO V710 for almost everything else, especially if power efficiency matters, as it sips just 158W compared to the CMP 90HX’s 320W. The PRO V710 also offers 28GB of memory versus 10GB, making it the only sensible option for large datasets.

Where Each One Wins

The AMD Radeon PRO V710 wins decisively in OpenCL compute tasks. Its Geekbench OpenCL score is 68.8% higher than the CMP 90HX, and it also wins in the 3DMark Steel Nomad DX12 test, where it scores 853 while the CMP 90HX has no corresponding entry. The PRO V710’s advantage comes from its higher FP32 throughput (27.65 TFLOPS vs 21.89) and its 28GB of GDDR6 memory, which allows it to handle far larger working sets than the CMP 90HX’s 10GB GDDR6X.

The NVIDIA CMP 90HX wins in raw memory bandwidth. Its 760.3 GB/s bandwidth is 50.9% higher than the PRO V710’s 504.0 GB/s, which can be critical for memory-bound algorithms. It also has a higher transistor count (28,300 million vs 28,100 million) and a larger die (628 mm² vs 346 mm²), though these do not translate into compute wins. In terms of architectural features, the CMP 90HX has 200 tensor cores, which the PRO V710 lacks entirely, making it potentially useful for AI inference workloads that rely on tensor operations.

The CMP 90HX also wins on shading unit count with 6,400 units versus 3,456, but this does not help it in the benchmark results. The PRO V710 counters with higher pixel rate (192.0 GPixel/s vs 136.8) and texture rate (432.0 GTexel/s vs 342.0). In practical terms, the PRO V710 wins every benchmark it appears in, while the CMP 90HX has no benchmark wins to its name in this comparison.

Architecture Differences

The two cards represent fundamentally different design philosophies. The NVIDIA CMP 90HX uses the GA102 chip on Samsung’s 8nm process, featuring 28,300 million transistors on a massive 628 mm² die. This yields a transistor density of 45.1 million per square millimeter. The AMD Radeon PRO V710 uses the Navi 32 chip on TSMC’s 5nm process, packing 28,100 million transistors into just 346 mm², for a density of 81.2 million per square millimeter. The 5nm node gives AMD a 2.7x density advantage, enabling a smaller, more power-efficient chip.

The CMP 90HX is built on the Ampere architecture with a boost clock of 1710 MHz, while the PRO V710 uses RDNA 3.0 with a boost clock of 2000 MHz. The PRO V710’s higher clock speed helps it achieve superior FP32 performance despite having nearly half the shading units. Memory configurations differ drastically: the CMP 90HX uses 10GB of GDDR6X on a 320-bit bus, while the PRO V710 uses 28GB of GDDR6 on a 224-bit bus. The CMP 90HX compensates with higher bandwidth (760.3 GB/s vs 504.0), but the PRO V710’s capacity advantage is enormous for large compute jobs.

Feature-wise, the CMP 90HX includes 50 RT cores and 200 tensor cores, while the PRO V710 has 54 RT cores and no tensor cores. Both support DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. Power delivery differs significantly: the CMP 90HX requires a 320W TDP with dual 8-pin connectors and a 700W PSU, while the PRO V710 needs only 158W, a single 8-pin, and a 450W PSU. The CMP 90HX is dual-slot and 285mm long, whereas the PRO V710 is single-slot. Both have no display outputs, emphasizing their compute-only purpose.

FAQ

Q: Which card is faster in OpenCL benchmarks?

A: The AMD Radeon PRO V710 scores 116,460 in Geekbench OpenCL, which is 68.8% higher than the NVIDIA CMP 90HX’s 69,000. This is the only head-to-head benchmark available, and the PRO V710 wins it decisively.

Q: Does the CMP 90HX have any advantage in memory bandwidth?

A: Yes, the CMP 90HX provides 760.3 GB/s of bandwidth versus the PRO V710’s 504.0 GB/s, a 50.9% advantage. This could benefit memory-intensive workloads, though it does not overcome the PRO V710’s compute lead in the benchmark data.

Q: Can either card be used for display output?

A: No. Both the NVIDIA CMP 90HX and AMD Radeon PRO V710 have no display outputs, making them unsuitable for any task requiring a monitor connection. They are strictly compute or mining accelerators.

Q: What is the difference in power consumption?

A: The CMP 90HX has a TDP of 320W and requires a 700W PSU with dual 8-pin connectors. The PRO V710 has a TDP of 158W, needs only a 450W PSU, and uses a single 8-pin connector. The PRO V710 is more than twice as power-efficient.

Q: Which card has more memory for large datasets?

A: The PRO V710 offers 28GB of GDDR6 memory, which is 2.8x the CMP 90HX’s 10GB of GDDR6X. For workloads that exceed 10GB, the PRO V710 is the only viable option.

Q: How do their overall performance percentiles compare?

A: The CMP 90HX ranks in the 90th percentile of all GPUs, while the PRO V710 ranks in the 88th. However, this is based on different benchmark sets; the PRO V710’s average score is lower because it includes a 3DMark result, whereas the CMP 90HX only has the OpenCL score.

Head-to-Head Benchmarks

The single head-to-head benchmark is Geekbench OpenCL, and it is not close. The AMD Radeon PRO V710 scores 116,460, while the NVIDIA CMP 90HX scores 69,000. This yields a delta of -40.8% for the CMP 90HX, meaning the PRO V710 outperforms it by 68.8%. In absolute terms, the PRO V710 delivers 47,460 more points. This result aligns with the theoretical FP32 numbers: the PRO V710’s 27.65 TFLOPS versus the CMP 90HX’s 21.89 TFLOPS gives the AMD card a 26.3% raw compute advantage, but the benchmark gap is even larger due to architectural efficiency.

The PRO V710 also has an additional benchmark result in 3DMark Steel Nomad DX12, scoring 853, which the CMP 90HX cannot match because it has no entry for that test. While this is not a direct comparison, it shows the PRO V710’s versatility across DX12 workloads. In terms of wins, the PRO V710 wins 1 out of 1 head-to-head tests, while the CMP 90HX wins 0. The PRO V710’s nearest rivals include the NVIDIA P102-100 (0.2% slower), AMD RX 6950 XT (0.5% slower), and Intel Arc A570M (0.7% slower), showing it is well-positioned among its peers. The CMP 90HX’s nearest rivals are the Intel Arc A770 (0.3% slower), AMD Instinct MI25 (0.6% slower), and Radeon Pro WX 8200 (1.2% faster), indicating it sits in a competitive mid-range cluster.

Specification Differences

The table below highlights only the fields where the two GPUs differ, omitting identical specs like DirectX 12 Ultimate support, OpenGL 4.6, Vulkan 1.4, and having no display outputs.

| Specification | NVIDIA CMP 90HX | AMD Radeon PRO V710 |

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

| Chip | GA102 | Navi 32 |

| Architecture | Ampere | RDNA 3.0 |

| Process Node | 8 nm (Samsung) | 5 nm (TSMC) |

| Transistors | 28,300 million | 28,100 million |

| Die Size | 628 mm² | 346 mm² |

| Transistor Density | 45.1M / mm² | 81.2M / mm² |

| Base Clock | 1500 MHz | 1900 MHz |

| Boost Clock | 1710 MHz | 2000 MHz |

| Memory Clock | 19 Gbps effective | 18 Gbps effective |

| Memory Size | 10 GB | 28 GB |

| Memory Type | GDDR6X | GDDR6 |

| Memory Bus | 320 bit | 224 bit |

| Memory Bandwidth | 760.3 GB/s | 504.0 GB/s |

| Shading Units | 6400 | 3456 |

| TMUs | 200 | 216 |

| ROPs | 80 | 96 |

| RT Cores | 50 | 54 |

| Tensor Cores | 200 | None |

| Pixel Rate | 136.8 GPixel/s | 192.0 GPixel/s |

| Texture Rate | 342.0 GTexel/s | 432.0 GTexel/s |

| FP32 | 21.89 TFLOPS | 27.65 TFLOPS |

| TDP | 320 W | 158 W |

| Slot Width | Dual-slot | Single-slot |

| Power Connectors | 2x 8-pin | 1x 8-pin |

| Suggested PSU | 700 W | 450 W |

| Bus Interface | PCIe 1.0 x4 | PCIe 4.0 x16 |

| Dimensions | 285 mm length, 112 mm height | Not specified |

| Release Date | 2021-07-27 | 2024-10-02 |

| Production Status | End-of-life | Not specified |

| Predecessor | None | Radeon Pro Vega |

| Geekbench OpenCL | 69,000 | 116,460 |

| 3DMark Steel Nomad | None | 853 |

| Percentile | 90th | 88th |

| Avg Benchmark Score | 69,000 | 58,657 |

The PRO V710’s PCIe 4.0 x16 interface is vastly superior to the CMP 90HX’s PCIe 1.0 x4, which is a legacy bottleneck. The CMP 90HX is also end-of-life, while the PRO V710’s status is unspecified, and the PRO V710 is 2 years newer.