AMD Radeon PRO W6600 vs NVIDIA CMP 90HX Comparison

# AMD Radeon PRO W6600 vs NVIDIA CMP 90HX

The AMD Radeon PRO W6600 and NVIDIA CMP 90HX are two end-of-life GPUs that took radically different paths to the same market segment. The W6600 is a professional workstation card built around a 7 nm Navi 23 die, while the CMP 90HX is a mining-focused Ampere part with no display outputs whatsoever. The benchmark data shows a single head-to-head result: in Geekbench OpenCL, the W6600 scores 73,514 against the CMP 90HX’s 69,000, giving AMD a 6.5% victory. That is the only direct comparison available, and it sets the tone for a matchup where architectural philosophy matters as much as raw compute.

Head-to-Head Benchmarks

The only shared benchmark between these two cards is Geekbench OpenCL, and the results are closer than the specifications might suggest. The AMD Radeon PRO W6600 posts 73,514 points, while the NVIDIA CMP 90HX trails at 69,000. That is a 6.5% margin in favor of the AMD card — a meaningful but not overwhelming gap. Interestingly, the W6600’s average benchmark score across all tested workloads is 81,995, which is significantly higher than its OpenCL result alone. This suggests the W6600 performs even better in Metal and Vulkan tests, where it scores 94,042 and 78,428 respectively. The CMP 90HX has no such additional data points, so its 69,000 OpenCL figure is also its entire benchmark profile.

Looking at the broader competitive landscape, the W6600 sits at the 92nd percentile of all GPUs, with an average score of 81,995. Its nearest rivals include the AMD Radeon Pro Vega 64X (80,959, just 1.3% behind) and the NVIDIA GeForce RTX 5090 (79,842, 2.7% behind). The fact that a 2021 workstation card edges out a flagship consumer GPU in this aggregated metric is notable. The CMP 90HX, by contrast, lands at the 90th percentile with its 69,000 average score. Its nearest rivals are the Intel Arc A770 (68,809, only 0.3% behind) and the AMD Radeon Instinct MI25 (68,562, 0.6% behind). The CMP 90HX is also beaten by the AMD Radeon Pro WX 8200 (69,870, -1.2% delta) and the NVIDIA Quadro P6000 (69,986, -1.4% delta), meaning it sits at the bottom of its immediate competitive cluster.

The head-to-head delta of 6.5% in OpenCL is consistent with the percentile gap — 92nd versus 90th — but it does not capture the full story. The W6600’s average score is 18.8% higher than the CMP 90HX’s single result, which points to a wider performance envelope when other API workloads are considered. The data implies that the AMD card is not just faster in one test but maintains its advantage across different compute frameworks.

Architecture Differences

The architectural chasm between these two cards is vast. The AMD Radeon PRO W6600 uses the Navi 23 chip on a 7 nm TSMC process, packing 11,060 million transistors into a 237 mm² die. That yields a transistor density of 46.7 million per square millimeter. The NVIDIA CMP 90HX, on the other hand, uses the GA102 chip on Samsung’s 8 nm node, with 28,300 million transistors spread across a massive 628 mm² die. Despite the larger absolute transistor count, the density is actually lower at 45.1 million per square millimeter, reflecting the less advanced process node.

The compute resources tell a similar story of divergence. The CMP 90HX has 6,400 shading units, 200 texture mapping units, and 80 ROPs, dwarfing the W6600’s 1,792 shading units, 112 TMUs, and 64 ROPs. The NVIDIA card also includes 50 RT cores and 200 tensor cores, while the AMD card has 28 RT cores and no tensor cores at all. In raw FP32 throughput, the CMP 90HX delivers 21.89 TFLOPS versus the W6600’s 9.247 TFLOPS. That is a 2.37x advantage in theoretical compute. The FP16 numbers are equally lopsided: the CMP 90HX maintains 21.89 TFLOPS at a 1:1 ratio, while the W6600 achieves 18.49 TFLOPS but only at a 2:1 ratio, meaning half the throughput when actually used.

Memory is another point of separation. The CMP 90HX features 10 GB of GDDR6X on a 320-bit bus, delivering 760.3 GB/s of bandwidth. The W6600 has 8 GB of GDDR6 on a 128-bit bus, with 224.0 GB/s. The NVIDIA card has over three times the memory bandwidth, which is critical for the mining workloads it was designed for. The AMD card compensates with higher clocks — 2,331 MHz base and 2,580 MHz boost versus 1,500 MHz base and 1,710 MHz boost — but the sheer memory bandwidth deficit is impossible to overcome.

The API support is identical on paper — both cards support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 — but the underlying hardware capabilities diverge sharply. The CMP 90HX’s tensor cores are absent from the W6600 entirely, which affects any AI or machine learning workloads. Conversely, the W6600’s display outputs (4x DisplayPort 1.4a) are completely absent on the CMP 90HX, which offers no outputs at all. The bus interfaces also differ: the W6600 uses PCIe 4.0 x8, while the CMP 90HX is limited to PCIe 1.0 x4, a stark reminder of its mining-only purpose.

The Verdict

The data paints a clear picture: the AMD Radeon PRO W6600 is the better all-around GPU. It wins the only head-to-head benchmark, holds a higher percentile ranking (92nd versus 90th), and has a substantially higher average benchmark score (81,995 versus 69,000). The W6600 also supports display outputs, making it usable for actual graphics work, while the CMP 90HX is a compute-only device with no video outputs.

However, the verdict is not one-sided in every dimension. The NVIDIA CMP 90HX offers more than double the FP32 throughput (21.89 TFLOPS versus 9.247 TFLOPS), three times the memory bandwidth (760.3 GB/s versus 224.0 GB/s), and 200 tensor cores that the AMD card lacks entirely. If the workload is pure compute — particularly tasks that can leverage tensor cores or massive memory bandwidth — the CMP 90HX has theoretical advantages that the benchmark data does not directly measure.

The practical reality is that the W6600 wins where it matters for a professional user: it is faster in the available OpenCL test, it supports multiple graphics APIs effectively, and it does so at a fraction of the power draw (100 W versus 320 W). The CMP 90HX is a specialized tool for a narrow use case, and the benchmark evidence suggests it is not even particularly good at that use case relative to its specs.

Specification Differences

The two cards differ in nearly every measured specification. The AMD Radeon PRO W6600 uses a 7 nm process, while the NVIDIA CMP 90HX uses 8 nm. The W6600 has 11,060 million transistors on a 237 mm² die, while the CMP 90HX has 28,300 million transistors on a 628 mm² die. Transistor density is close (46.7M/mm² versus 45.1M/mm²), but the NVIDIA chip is nearly three times larger.

Clock speeds favor AMD: the W6600 runs at 2,331 MHz base and 2,580 MHz boost, while the CMP 90HX runs at 1,500 MHz base and 1,710 MHz boost. Memory clocks are also different, with the W6600 at 1,750 MHz (14 Gbps effective) and the CMP 90HX at 1,188 MHz (19 Gbps effective). The memory configuration diverges sharply: 8 GB GDDR6 on a 128-bit bus versus 10 GB GDDR6X on a 320-bit bus, yielding 224.0 GB/s versus 760.3 GB/s of bandwidth.

Compute unit counts are dramatically different: 1,792 shading units, 112 TMUs, and 64 ROPs for the W6600 versus 6,400 shading units, 200 TMUs, and 80 ROPs for the CMP 90HX. The RT core counts are 28 versus 50, and the CMP 90HX adds 200 tensor cores that the W6600 does not have. Pixel rate favors the W6600 at 165.1 GPixel/s versus 136.8 GPixel/s, but texture rate favors the CMP 90HX at 342.0 GTexel/s versus 289.0 GTexel/s. FP32 is 9.247 TFLOPS versus 21.89 TFLOPS, and FP16 is 18.49 TFLOPS (2:1) versus 21.89 TFLOPS (1:1).

Power and physical specs also differ: the W6600 draws 100 W with a single 6-pin connector and a 300 W suggested PSU, while the CMP 90HX draws 320 W with dual 8-pin connectors and a 700 W suggested PSU. The W6600 is single-slot at 241 mm long, while the CMP 90HX is dual-slot at 285 mm long and 112 mm tall. The bus interface is PCIe 4.0 x8 for AMD versus PCIe 1.0 x4 for NVIDIA. Display outputs are 4x DisplayPort 1.4a on the W6600 and none on the CMP 90HX. The W6600 launched on 2021-06-07 with a launch MSRP of 649 USD, while the CMP 90HX launched on 2021-07-27 with no MSRP listed.

FAQ

Q: Which card performs better in the only shared benchmark?

A: The AMD Radeon PRO W6600 wins Geekbench OpenCL with a score of 73,514 against the NVIDIA CMP 90HX’s 69,000, a 6.5% advantage.

Q: Does the NVIDIA CMP 90HX have any performance advantages?

A: Yes, the CMP 90HX has significantly higher theoretical compute: 21.89 TFLOPS FP32 versus 9.247 TFLOPS, and 760.3 GB/s memory bandwidth versus 224.0 GB/s.

Q: Can the NVIDIA CMP 90HX be used for display output?

A: No, the CMP 90HX has no display outputs at all, while the AMD Radeon PRO W6600 offers 4x DisplayPort 1.4a.

Q: How do the two cards compare in terms of power consumption?

A: The AMD Radeon PRO W6600 has a 100 W TDP with a single 6-pin connector and a 300 W suggested PSU, while the NVIDIA CMP 90HX has a 320 W TDP with dual 8-pin connectors and a 700 W suggested PSU.

Q: Which card ranks higher among all GPUs?

A: The AMD Radeon PRO W6600 sits at the 92nd percentile with an average benchmark score of 81,995, while the NVIDIA CMP 90HX sits at the 90th percentile with an average score of 69,000.

Q: Does the CMP 90HX support tensor core workloads?

A: The NVIDIA CMP 90HX includes 200 tensor cores, while the AMD Radeon PRO W6600 has none, so the CMP 90HX is the only one of the two capable of tensor operations.

Where Each One Wins

The AMD Radeon PRO W6600 wins in every measured benchmark and in overall practical usability. Its 6.5% OpenCL lead over the CMP 90HX is complemented by higher scores in Metal (94,042) and Vulkan (78,428), though those tests have no direct CMP 90HX comparison. The W6600 also wins decisively on efficiency, drawing 100 W versus 320 W while still delivering a higher benchmark score. Its 4x DisplayPort 1.4a outputs make it a functional workstation card, and its single-slot 241 mm form factor fits in far more chassis. The 92nd percentile ranking versus the CMP 90HX’s 90th is another clear win.

The NVIDIA CMP 90HX wins on raw specification sheets, even if those advantages do not translate to benchmark victories. Its 21.89 TFLOPS FP32 throughput is more than double the W6600’s 9.247 TFLOPS, and its 760.3 GB/s memory bandwidth is over three times higher. The 200 tensor cores and 50 RT cores provide hardware capabilities the AMD card simply lacks. For workloads that can exploit massive parallelism, high bandwidth, or tensor operations — such as certain AI inference or data processing tasks — the CMP 90HX has the theoretical edge. Its 10 GB of GDDR6X memory also exceeds the W6600’s 8 GB of GDDR6.

The data suggests a clear division of purpose. The W6600 is a general-purpose professional GPU that happens to outperform the CMP 90HX in the available benchmarks. The CMP 90HX is a mining-specific part whose specification sheet promises more than the benchmark results deliver. If the choice is based purely on measured performance, the AMD Radeon PRO W6600 is the pick. If the workload requires tensor cores or extreme memory bandwidth, the CMP 90HX’s hardware features justify consideration despite its benchmark deficit.