AMD Radeon PRO W7600 vs NVIDIA CMP 90HX Comparison

The AMD Radeon PRO W7600 and NVIDIA CMP 90HX represent two very different approaches to GPU hardware, despite both being professional or specialized parts. The W7600 is an active, single-slot workstation card with display outputs, while the CMP 90HX is an end-of-life mining GPU with no video outputs at all. Benchmark data shows the W7600 leads in the only common test, but the specifications suggest each card serves a distinct purpose that goes beyond raw compute scores.

FAQ

Q: Which card has a higher average benchmark score?

A: The AMD Radeon PRO W7600 has an average benchmark score of 87,108, placing it in the 93rd percentile of all GPUs. The NVIDIA CMP 90HX scores 69,000, which puts it in the 90th percentile.

Q: How much faster is the W7600 in the Geekbench OpenCL test?

A: The W7600 scores 81,528 versus 69,000 for the CMP 90HX, a delta of 18.2% in favor of the AMD card.

Q: What are the closest rivals to each card?

A: The W7600's nearest rival is the NVIDIA Quadro GP100, which scores 87,445 (0.4% higher). The CMP 90HX's nearest rival is the Intel Arc A770, which scores 68,809 (0.3% higher).

Q: Do both cards support the same API level?

A: Yes, both cards support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Q: Which card has more memory and what type?

A: The CMP 90HX has 10 GB of GDDR6X on a 320-bit bus, while the W7600 has 8 GB of GDDR6 on a 128-bit bus. The CMP 90HX's memory bandwidth is 760.3 GB/s, versus 288.0 GB/s for the W7600.

Q: What is the production status of each card?

A: The W7600 is listed as "Active" and was released on 2023-08-02. The CMP 90HX is "End-of-life" and was released on 2021-07-27.

Architecture Differences

The two GPUs are built on fundamentally different architectures and process nodes. The W7600 uses AMD's RDNA 3.0 architecture, specifically the Navi 33 chip with the codename "Hotpink Bonefish." It is fabricated on a 6 nm process at TSMC, with 13,300 million transistors on a 204 mm² die. That yields a transistor density of 65.2 million transistors per mm². The CMP 90HX, in contrast, uses NVIDIA's Ampere architecture with the GA102 chip. It is built on Samsung's 8 nm process, packing 28,300 million transistors into a much larger 628 mm² die, giving a transistor density of 45.1 million per mm².

The compute resources differ substantially. The W7600 has 2048 shading units, 128 texture mapping units (TMUs), 64 render output units (ROPs), and 32 ray tracing cores. It has no dedicated tensor cores. The CMP 90HX is a larger chip with 6400 shading units, 200 TMUs, 80 ROPs, 50 ray tracing cores, and 200 tensor cores. Despite having more than three times the shading units, the CMP 90HX's FP32 throughput is only 21.89 TFLOPS versus 19.99 TFLOPS for the W7600, a small 9.5% advantage. This is due to clock speeds: the W7600 boosts to 2440 MHz while the CMP 90HX boosts to only 1710 MHz. The W7600 also has a higher base clock at 1720 MHz versus 1500 MHz.

Memory architecture is another clear differentiator. The W7600 uses 8 GB of GDDR6 at 18 Gbps effective, with a 128-bit bus yielding 288.0 GB/s bandwidth. The CMP 90HX uses 10 GB of GDDR6X at 19 Gbps effective, with a 320-bit bus delivering 760.3 GB/s — a 2.6x bandwidth advantage. The power situation is also starkly different. The W7600 has a 130 W TDP and requires a single 6-pin power connector with a 300 W suggested PSU. The CMP 90HX has a 320 W TDP, needs dual 8-pin connectors, and calls for a 700 W PSU. The W7600 is single-slot, while the CMP 90HX is dual-slot.

Head-to-Head Benchmarks

The only direct benchmark comparison available is Geekbench OpenCL, and it is a clear win for the AMD card. The W7600 scores 81,528, while the CMP 90HX scores 69,000. That is an 18.2% advantage for the W7600. This is notable because the CMP 90HX has significantly more memory bandwidth and a much higher FP32 compute rating in its specifications. Despite the theoretical hardware advantages of the CMP 90HX, the W7600 delivers better real-world OpenCL performance.

Looking at the nearest rivals helps contextualize this result. The W7600's average score of 87,108 is slightly below the NVIDIA Quadro GP100 (87,445, a 0.4% gap) and the NVIDIA RTX A4500 (91,671, a 5% gap). It beats the NVIDIA CMP 40HX (85,637, a 1.7% lead). The CMP 90HX's average score of 69,000 sits just above the Intel Arc A770 (68,809, a 0.3% lead) and the AMD Radeon Instinct MI25 (68,562, a 0.6% lead). It trails the AMD Radeon Pro WX 8200 (69,870, a 1.2% deficit) and the NVIDIA Quadro P6000 (69,986, a 1.4% deficit).

The W7600's percentile ranking of 93 versus 90 for the CMP 90HX reflects this difference, but the margin between those percentiles is modest. What stands out is that the W7600 achieves its higher score with far less power (130 W vs 320 W) and a smaller die. The CMP 90HX's 21.89 TFLOPS FP32 rating and 760.3 GB/s bandwidth do not translate into a winning OpenCL result, likely due to the mining-focused nature of the card and its older architecture.

The Verdict

The data clearly favors the AMD Radeon PRO W7600 for any task that relies on OpenCL compute. It wins the head-to-head benchmark by 18.2%, has a higher average score (87,108 vs 69,000), and sits at a higher percentile (93rd vs 90th). It is also an active product with a launch MSRP of 599 USD, while the CMP 90HX is end-of-life with no MSRP listed.

The CMP 90HX does have advantages in raw specifications: 10 GB of GDDR6X memory, 760.3 GB/s bandwidth, 6400 shading units, and 200 tensor cores. But none of these translate into a benchmark win. The card also has no display outputs, making it unsuitable for any traditional workstation use. The W7600, with its four DisplayPort 2.1 outputs, is clearly the only one of the two that can drive monitors.

For a builder choosing between these two, the decision is straightforward from the data. The W7600 is the faster, more efficient, and more versatile card. The CMP 90HX is a legacy mining part with high power draw and no video output, making it difficult to recommend for any practical purpose beyond niche compute tasks where its memory bandwidth might be useful.

Specification Differences

| Specification | AMD Radeon PRO W7600 | NVIDIA CMP 90HX |

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

| Architecture | RDNA 3.0 | Ampere |

| Process Node | 6 nm (TSMC) | 8 nm (Samsung) |

| Transistors | 13,300 million | 28,300 million |

| Die Size | 204 mm² | 628 mm² |

| Transistor Density | 65.2M / mm² | 45.1M / mm² |

| Base Clock | 1720 MHz | 1500 MHz |

| Boost Clock | 2440 MHz | 1710 MHz |

| Memory Size | 8 GB | 10 GB |

| Memory Type | GDDR6 | GDDR6X |

| Memory Bus | 128 bit | 320 bit |

| Memory Bandwidth | 288.0 GB/s | 760.3 GB/s |

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

| Shading Units | 2048 | 6400 |

| TMUs | 128 | 200 |

| ROPs | 64 | 80 |

| Ray Tracing Cores | 32 | 50 |

| Tensor Cores | None | 200 |

| FP32 | 19.99 TFLOPS | 21.89 TFLOPS |

| FP16 | 39.98 TFLOPS (2:1) | 21.89 TFLOPS (1:1) |

| TDP | 130 W | 320 W |

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

| Power Connectors | 1x 6-pin | 2x 8-pin |

| Suggested PSU | 300 W | 700 W |

| Bus Interface | PCIe 4.0 x8 | PCIe 1.0 x4 |

| Display Outputs | 4x DisplayPort 2.1 | No outputs |

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

| Release Date | 2023-08-02 | 2021-07-27 |

Where Each One Wins

The AMD Radeon PRO W7600 wins in every measurable benchmark category. It has a higher Geekbench OpenCL score (81,528 vs 69,000), a higher average benchmark score (87,108 vs 69,000), and a higher percentile ranking (93rd vs 90th). It also wins on efficiency, drawing 130 W versus 320 W, and on practicality, with display outputs and an active production status. Any workload that uses OpenCL — such as rendering, simulation, or general compute — will favor the W7600 based on this data.

The NVIDIA CMP 90HX wins on raw memory and compute specifications. It offers 10 GB of GDDR6X versus 8 GB of GDDR6, and its 760.3 GB/s bandwidth is more than double the W7600's 288.0 GB/s. It also has more shading units (6400 vs 2048), more ray tracing cores (50 vs 32), and 200 tensor cores where the W7600 has none. Its FP32 rating of 21.89 TFLOPS is slightly higher than the W7600's 19.99 TFLOPS. For workloads that are bottlenecked by memory bandwidth or that can leverage tensor cores, the CMP 90HX could theoretically have an edge — but the benchmark data does not show this in OpenCL.

In practical terms, the W7600 is the only choice for a workstation that needs to output video. The CMP 90HX has no display outputs, so it cannot be used as a typical graphics card. Its PCIe 1.0 x4 interface is also severely limited compared to the W7600's PCIe 4.0 x8, which could bottleneck data transfer in many scenarios. Given that the W7600 is also the faster card in the available benchmark, the CMP 90HX's only real advantage is its larger memory pool and bandwidth, which may matter for specific high-throughput compute tasks that do not rely on OpenCL performance.