AMD Radeon Pro WX 8200 vs NVIDIA CMP 90HX Comparison

AMD Radeon Pro WX 8200 vs NVIDIA CMP 90HX: two end-of-life cards aimed at opposite ends of the compute spectrum. The WX 8200 is a professional workstation GPU with display outputs and a 90th-percentile standing, while the CMP 90HX is a mining-focused Ampere part with no video outputs and a PCIe 1.0 x4 interface. The benchmark data shows a razor-thin edge for AMD in the only shared test, but the architectural gap tells a much larger story.

Where Each One Wins

The AMD Radeon Pro WX 8200 wins the only head-to-head benchmark available, the Geekbench OpenCL test, with a score of 69,774 against the NVIDIA CMP 90HX's 69,000. That is a 1.1% delta in favor of AMD. The WX 8200 also has two additional benchmark results in the Fact Pack: a Geekbench Metal score of 70,759 and a Geekbench Vulkan score of 69,076, giving it an average benchmark score of 69,870. The CMP 90HX has only the single OpenCL result, so its average is exactly 69,000.

Neither card dominates the other by a wide margin in raw compute. The WX 8200's advantage is narrow but consistent across its three tests, with all scores clustering between 69,076 and 70,759. The CMP 90HX's single score sits 774 points below the WX 8200's OpenCL result. In terms of percentile ranking, both cards sit at the 90th percentile versus all GPUs, meaning they occupy the same tier in the overall performance distribution.

The real separation is in capabilities, not scores. The WX 8200 offers 4x mini-DisplayPort 1.4a outputs, making it viable for visualization workloads. The CMP 90HX has no display outputs at all, which disqualifies it from any task requiring a monitor. The CMP 90HX also runs a PCIe 1.0 x4 bus interface, a severe bandwidth bottleneck compared to the WX 8200's PCIe 3.0 x16 connection. For compute tasks that stream data over the bus, the WX 8200's interface is categorically superior.

Architecture Differences

The two GPUs come from different foundries, different nodes, and different design philosophies. The WX 8200 uses the Vega 10 chip built on GlobalFoundries' 14 nm process, packing 12,500 million transistors into a 495 mm² die. The CMP 90HX uses the GA102 chip on Samsung's 8 nm node, with 28,300 million transistors on a 628 mm² die. Transistor density tells the story: the CMP 90HX packs 45.1 million transistors per mm² versus 25.3 million for the WX 8200, reflecting the newer, denser process.

The architecture names reflect different eras. The WX 8200 runs GCN 5.0, AMD's fifth-generation compute architecture that powers the Radeon Pro Polaris generation. The CMP 90HX runs Ampere, NVIDIA's second-generation ray tracing architecture. The CMP 90HX includes 50 RT cores and 200 tensor cores, features completely absent from the WX 8200. It also supports DirectX 12 Ultimate (12_2) and Vulkan 1.4, while the WX 8200 tops out at DirectX 12 (12_1) and Vulkan 1.3. Both support OpenGL 4.6.

Memory subsystems diverge sharply. The WX 8200 uses 8 GB of HBM2 on a 2048-bit bus, delivering 512.0 GB/s of bandwidth. The CMP 90HX uses 10 GB of GDDR6X on a 320-bit bus, delivering 760.3 GB/s. That is a 48.5% bandwidth advantage for the NVIDIA card, a significant margin for memory-bound workloads. The CMP 90HX also has more shading units (6,400 vs 3,584), more ROPs (80 vs 64), and slightly fewer TMUs (200 vs 224). The WX 8200 counters with higher texture rate per TMU, but the raw numbers favor NVIDIA.

Clock speeds also favor the CMP 90HX. It runs a 1500 MHz base and 1710 MHz boost, against 1200 MHz base and 1500 MHz boost for the WX 8200. The memory clock is similarly lopsided: 1188 MHz (19 Gbps effective) for the CMP 90HX versus 1000 MHz (2 Gbps effective) for the WX 8200. The FP32 throughput scales accordingly: 21.89 TFLOPS for the CMP 90HX versus 10.75 TFLOPS for the WX 8200, a 103.6% advantage. FP16 is a different story: the WX 8200 hits 21.50 TFLOPS via a 2:1 ratio, nearly matching the CMP 90HX's 21.89 TFLOPS at 1:1.

Head-to-Head Benchmarks

The only direct comparison in the data is the Geekbench OpenCL test. The WX 8200 scores 69,774, and the CMP 90HX scores 69,000. The 774-point gap translates to a 1.1% delta in AMD's favor. This is a narrow win, especially given the CMP 90HX's theoretical FP32 advantage of more than 2x. The OpenCL benchmark appears to favor memory bandwidth and compute throughput in a balanced way, and the WX 8200's HBM2 implementation evidently compensates for its lower raw FLOPs.

Looking at the nearest rivals provides context. The WX 8200's average score of 69,870 puts it 0.2% behind the NVIDIA Quadro P6000 (69,986), 0.4% behind the NVIDIA RTX A3000 Mobile (70,140), 1.3% ahead of the CMP 90HX (69,000), and 1.4% behind the AMD Radeon RX 6600 LE (70,829). The CMP 90HX's average of 69,000 puts it 0.3% ahead of the Intel Arc A770 (68,809), 0.6% ahead of the AMD Radeon Instinct MI25 (68,562), 1.2% behind the WX 8200, and 1.4% behind the Quadro P6000.

The data shows that the CMP 90HX's massive architectural advantages in FP32, memory bandwidth, and transistor count do not translate into a benchmark win. The WX 8200 edges it out despite having 44% fewer shading units and half the FP32 throughput. This suggests the Geekbench OpenCL workload is not purely FLOP-bound, or that the WX 8200's HBM2 memory subsystem and PCIe 3.0 x16 interface provide offsetting advantages. The CMP 90HX's PCIe 1.0 x4 bus may be a limiting factor in certain OpenCL test components.

The Verdict

The data supports a clear split decision. For any workload requiring a display output, the AMD Radeon Pro WX 8200 is the only choice, as the NVIDIA CMP 90HX has no outputs whatsoever. The WX 8200 also wins the single head-to-head benchmark, making it the better performer in the measured compute test. Its 4x mini-DisplayPort 1.4a outputs and PCIe 3.0 x16 interface make it suitable for workstation tasks, and its 90th-percentile standing confirms it is a capable card.

For pure compute throughput, the NVIDIA CMP 90HX offers overwhelming theoretical advantages: 21.89 TFLOPS FP32 versus 10.75 TFLOPS, 760.3 GB/s memory bandwidth versus 512.0 GB/s, and a newer 8 nm process with 50 RT cores and 200 tensor cores. The data shows it does not win the OpenCL benchmark, but the hardware specifications suggest it could excel in workloads that leverage its tensor cores or require high memory bandwidth. Its lack of display outputs and PCIe 1.0 x4 interface make it impractical for general use, but for a mining or headless compute rig, those are non-issues.

The WX 8200's launch MSRP was 999 USD. The CMP 90HX has no listed launch MSRP. Both are end-of-life products, and the CMP 90HX has no predecessor or successor listed, while the WX 8200 succeeds the Radeon Pro GCN and is succeeded by the Radeon Pro Vega. The WX 8200 also offers lower power requirements: 230 W TDP with a 550 W suggested PSU, versus 320 W TDP with a 700 W suggested PSU for the CMP 90HX.

FAQ

Q: Which card wins the only direct benchmark comparison?

A: The AMD Radeon Pro WX 8200 wins the Geekbench OpenCL test with a score of 69,774 against the NVIDIA CMP 90HX's 69,000, a 1.1% delta.

Q: Can the NVIDIA CMP 90HX be used with a monitor?

A: No. The CMP 90HX has no display outputs listed, while the WX 8200 offers 4x mini-DisplayPort 1.4a outputs.

Q: How do the two cards compare in memory bandwidth?

A: The CMP 90HX provides 760.3 GB/s from 10 GB of GDDR6X on a 320-bit bus, while the WX 8200 provides 512.0 GB/s from 8 GB of HBM2 on a 2048-bit bus.

Q: What is the FP32 performance difference?

A: The CMP 90HX delivers 21.89 TFLOPS FP32, which is 103.6% higher than the WX 8200's 10.75 TFLOPS.

Q: Which card has the higher transistor density?

A: The CMP 90HX, built on Samsung's 8 nm process, has 45.1 million transistors per mm², versus 25.3 million per mm² for the WX 8200 on GlobalFoundries' 14 nm node.

Q: Does either card support ray tracing or tensor cores?

A: Only the CMP 90HX, which includes 50 RT cores and 200 tensor cores. The WX 8200 has no such hardware, and its DirectX support is limited to 12 (12_1) versus the CMP 90HX's 12 Ultimate (12_2).