GPU Comparison

NVIDIA CMP 90HX and AMD Radeon Pro W6600M are both end-of-life products aimed at very different corners of the GPU market, and the benchmark data reflects that split. The NVIDIA CMP 90HX is a mining-focused card with no display outputs, while the AMD Radeon Pro W6600M is a mobile workstation GPU. In the only shared benchmark test, the NVIDIA card holds a clear lead, but the AMD part counters with a second benchmark result and a more modern feature set.

Head-to-Head Benchmarks

The head-to-head comparison contains a single Geekbench OpenCL test, and the NVIDIA CMP 90HX wins it decisively. The NVIDIA card scores 69,000 points, while the AMD Radeon Pro W6600M scores 56,140 points. That is a delta of 22.9% in favor of the NVIDIA part. This is a substantial margin in raw compute throughput, and it aligns with the respective FP32 specifications: the CMP 90HX delivers 21.89 TFLOPS, while the Radeon Pro W6600M delivers 7.290 TFLOPS. The NVIDIA card is nearly three times faster in pure FP32 math, which explains the OpenCL gap.

However, the AMD card has a second benchmark result that the NVIDIA card lacks: a Geekbench Vulkan score of 67,652. That Vulkan score is much closer to the NVIDIA OpenCL result, trailing it by only about 2%. This suggests that when using a modern graphics API, the AMD part is far more competitive than the OpenCL numbers alone would indicate. The Vulkan result also shows that the Radeon Pro W6600M can scale well with API efficiency, even if its raw compute throughput is lower.

Looking at the broader benchmark context, the NVIDIA CMP 90HX sits at the 90th percentile among all GPUs, with an average benchmark score of 69,000. Its nearest rivals include the Intel Arc A770 at 68,809 (0.3% behind) and the AMD Radeon Instinct MI25 at 68,562 (0.6% behind). The NVIDIA card also edges out the AMD Radeon Pro WX 8200 (69,870, which is 1.2% higher) and the NVIDIA Quadro P6000 (69,986, 1.4% higher). So the CMP 90HX is not just ahead of the W6600M, it is also competitively placed against a range of desktop and workstation cards.

The AMD Radeon Pro W6600M, with an average benchmark score of 61,896, sits at the 89th percentile. Its nearest rivals are the AMD Radeon 8050S at 62,108 (0.3% ahead), the NVIDIA GeForce RTX 4090 at 60,347 (2.6% behind), and the Intel Arc Pro A60 at 60,326 (2.6% behind). Notably, the W6600M outperforms the RTX 4090 in this average score, a surprising result that suggests the W6600M's OpenCL and Vulkan scores are strong for a mobile part. The AMD card also leads the Radeon Pro Vega 48 (60,140) by 2.9%.

In summary, the NVIDIA CMP 90HX wins the only direct head-to-head test, and it does so by a wide margin. But the AMD part's Vulkan score shows that it is not a pushover in real-world graphics workloads, and its average score places it just one percentile below the NVIDIA card.

FAQ

Q: Which GPU has the higher Geekbench OpenCL score?

A: The NVIDIA CMP 90HX scores 69,000, which is 22.9% higher than the AMD Radeon Pro W6600M's 56,140.

Q: Does the AMD Radeon Pro W6600M have any benchmark where it performs better?

A: The AMD card has a Geekbench Vulkan score of 67,652, which is much closer to the NVIDIA OpenCL score of 69,000. There is no direct Vulkan comparison available, but the AMD part's Vulkan result is competitive.

Q: How do the two GPUs compare in terms of overall percentile ranking?

A: The NVIDIA CMP 90HX is at the 90th percentile, while the AMD Radeon Pro W6600M is at the 89th percentile. The NVIDIA card has an average score of 69,000 versus 61,896 for the AMD part.

Q: What is a notable rival for the NVIDIA CMP 90HX?

A: The Intel Arc A770 is the closest rival, with an average score of 68,809, just 0.3% behind the NVIDIA card.

Q: What is a notable rival for the AMD Radeon Pro W6600M?

A: The AMD Radeon 8050S is the closest rival, with an average score of 62,108, which is 0.3% ahead of the W6600M.

Q: Which GPU has a higher FP32 compute throughput?

A: The NVIDIA CMP 90HX delivers 21.89 TFLOPS, while the AMD Radeon Pro W6600M delivers 7.290 TFLOPS. The NVIDIA part is roughly three times higher.

Architecture Differences

The two GPUs are built on fundamentally different architectures and process nodes. The NVIDIA CMP 90HX uses the GA102 chip, which is based on the Ampere architecture and manufactured on Samsung's 8 nm process. The die is 628 mm² and contains 28,300 million transistors, yielding a transistor density of 45.1M per mm². The AMD Radeon Pro W6600M, by contrast, uses the Navi 23 chip, based on RDNA 2.0, built on TSMC's 7 nm process. That die is much smaller at 237 mm², with 11,060 million transistors, giving a slightly higher density of 46.7M per mm².

The core configurations differ dramatically. The NVIDIA CMP 90HX has 6,400 shading units, 200 texture mapping units, and 80 ROPs. It also includes 50 ray tracing cores and 200 tensor cores. The AMD Radeon Pro W6600M has 1,792 shading units, 112 TMUs, and 64 ROPs, with 28 ray tracing cores but no tensor cores. This explains the massive FP32 gap: the NVIDIA card has over 3.5 times more shading units.

Memory architecture is another major divergence. The NVIDIA card uses 10 GB of GDDR6X on a 320-bit bus, achieving 760.3 GB/s of bandwidth. The AMD card uses 8 GB of GDDR6 on a 128-bit bus, with 224.0 GB/s of bandwidth. The NVIDIA card has nearly 3.4 times the memory bandwidth, which is critical for compute-heavy workloads.

Clock speeds also differ. The NVIDIA CMP 90HX has a base clock of 1500 MHz and a boost of 1710 MHz. The AMD Radeon Pro W6600M has a lower base clock of 1224 MHz but a much higher boost of 2034 MHz. This higher boost clock helps the AMD part close some of the gap in certain workloads, but it cannot overcome the raw hardware advantage of the NVIDIA card.

The NVIDIA card also has a much higher power envelope: 320 W TDP versus 90 W for the AMD part. This is partly because the NVIDIA card is a dual-slot desktop card requiring 2x 8-pin power connectors and a 700 W suggested PSU, while the AMD card is an IGP (integrated GPU) for mobile devices with no power connectors and no suggested PSU listed.

Specification Differences

• Process Node: NVIDIA CMP 90HX is on Samsung 8 nm; AMD Radeon Pro W6600M is on TSMC 7 nm.
• Transistors: NVIDIA has 28,300 million; AMD has 11,060 million.
• Die Size: NVIDIA is 628 mm²; AMD is 237 mm².
• Transistor Density: NVIDIA is 45.1M / mm²; AMD is 46.7M / mm².
• Base Clock: NVIDIA is 1500 MHz; AMD is 1224 MHz.
• Boost Clock: NVIDIA is 1710 MHz; AMD is 2034 MHz.
• Memory Size: NVIDIA is 10 GB; AMD is 8 GB.
• Memory Type: NVIDIA is GDDR6X; AMD is GDDR6.
• Memory Bus: NVIDIA is 320 bit; AMD is 128 bit.
• Memory Bandwidth: NVIDIA is 760.3 GB/s; AMD is 224.0 GB/s.
• Shading Units: NVIDIA is 6400; AMD is 1792.
• TMUs: NVIDIA is 200; AMD is 112.
• ROPs: NVIDIA is 80; AMD is 64.
• RT Cores: NVIDIA is 50; AMD is 28.
• Tensor Cores: NVIDIA is 200; AMD has none.
• Pixel Rate: NVIDIA is 136.8 GPixel/s; AMD is 130.2 GPixel/s.
• Texture Rate: NVIDIA is 342.0 GTexel/s; AMD is 227.8 GTexel/s.
• FP32: NVIDIA is 21.89 TFLOPS; AMD is 7.290 TFLOPS.
• FP16: NVIDIA is 21.89 TFLOPS (1:1); AMD is 14.58 TFLOPS (2:1).
• TDP: NVIDIA is 320 W; AMD is 90 W.
• Slot Width: NVIDIA is Dual-slot; AMD is IGP.
• Power Connectors: NVIDIA is 2x 8-pin; AMD is None.
• Suggested PSU: NVIDIA is 700 W; AMD is not listed.
• Bus Interface: NVIDIA is PCIe 1.0 x4; AMD is PCIe 4.0 x16.
• Display Outputs: NVIDIA is No outputs; AMD is Portable Device Dependent.
• Dimensions: NVIDIA is 285 mm / 11.2 inches long and 112 mm / 4.4 inches high; AMD has no listed dimensions.
• Release Date: NVIDIA is 2021-07-27; AMD is 2021-06-07.
• Predecessor: NVIDIA has none listed; AMD is FirePro Mobile.

The Verdict

The data points to a clear performance hierarchy, but the intended use cases are completely different. The NVIDIA CMP 90HX is a compute monster. Its 22.9% lead in Geekbench OpenCL over the AMD Radeon Pro W6600M is backed by a 21.89 TFLOPS FP32 throughput, 760.3 GB/s of memory bandwidth, and a 90th percentile ranking. If you need raw compute performance and do not need display output, the NVIDIA card is the obvious choice from the benchmark results.

The AMD Radeon Pro W6600M, on the other hand, is a mobile GPU. It draws only 90 W, has no external power connector, and is designed as an IGP for portable devices. Despite its much lower raw compute, its Vulkan score of 67,652 is close to the NVIDIA card's OpenCL number, and it sits at the 89th percentile. For a mobile workstation part, the data shows it is highly competitive in modern API workloads.

There is also a question of interface compatibility. The NVIDIA card uses PCIe 1.0 x4, which is an unusual and limited bus interface, while the AMD card uses PCIe 4.0 x16. This means the AMD part has far better host connectivity, which can matter for data transfer and system integration.

The verdict is straightforward: the NVIDIA CMP 90HX wins on raw performance and is meant for dedicated mining or compute rigs where display output is irrelevant. The AMD Radeon Pro W6600M wins on efficiency, mobility, and modern connectivity, making it the better fit for a laptop or compact workstation. Neither card is a general-purpose desktop GPU replacement.

Where Each One Wins

NVIDIA CMP 90HX wins on:

• Raw compute: 22.9% higher OpenCL score, 21.89 TFLOPS FP32.
• Memory bandwidth: 760.3 GB/s versus 224.0 GB/s.
• Higher percentile: 90th versus 89th.
• Memory capacity: 10 GB versus 8 GB.
• Pixel and texture rates: 136.8 GPixel/s and 342.0 GTexel/s versus 130.2 GPixel/s and 227.8 GTexel/s.
• Tensor core support: 200 tensor cores versus none.

AMD Radeon Pro W6600M wins on:

• Efficiency: 90 W TDP versus 320 W.
• Boost clock: 2034 MHz versus 1710 MHz.
• Bus interface: PCIe 4.0 x16 versus PCIe 1.0 x4.
• Mobility: IGP form factor versus dual-slot desktop card.
• Display support: Portable Device Dependent versus no outputs.
• FP16 throughput: 14.58 TFLOPS (2:1) versus 21.89 TFLOPS (1:1), the AMD part is more efficient at half-precision relative to its FP32.
• Modern process node: 7 nm versus 8 nm.