AMD Radeon PRO W7900 vs NVIDIA CMP 90HX Comparison

FAQ

Q: What are the average benchmark scores for the AMD Radeon PRO W7900 and the NVIDIA CMP 90HX?

A: The AMD Radeon PRO W7900 has an average benchmark score of 110,725, while the NVIDIA CMP 90HX has an average benchmark score of 69,000.

Q: How do the two cards compare in the Geekbench OpenCL test?

A: The AMD Radeon PRO W7900 scores 84,379, which is 22.3% higher than the NVIDIA CMP 90HX's score of 69,000.

Q: Which card has a higher memory capacity and what are the types?

A: The AMD Radeon PRO W7900 has 48 GB of GDDR6 memory, while the NVIDIA CMP 90HX has 10 GB of GDDR6X memory.

Q: What are the process nodes for each GPU?

A: The AMD Radeon PRO W7900 uses a 5 nm process from TSMC, whereas the NVIDIA CMP 90HX uses an 8 nm process from Samsung.

Q: Are both cards still in production?

A: No. The AMD Radeon PRO W7900 is listed as active, while the NVIDIA CMP 90HX is end-of-life.

Q: What is the difference in their percentile rankings among all GPUs?

A: The AMD Radeon PRO W7900 sits in the 94th percentile, while the NVIDIA CMP 90HX is in the 90th percentile.

Architecture Differences

The AMD Radeon PRO W7900 is built on the RDNA 3.0 architecture with the Navi 31 chip, codenamed Plum Bonito, and belongs to the Radeon Pro Navi series. It uses a 5 nm process at TSMC, packing 57,700 million transistors onto a 529 mm² die, yielding a transistor density of 109.1 million per mm². The NVIDIA CMP 90HX, in contrast, employs the Ampere architecture with the GA102 chip and is part of the Mining GPUs generation. Fabricated on Samsung's 8 nm process, it contains 28,300 million transistors across a larger 628 mm² die, resulting in a much lower density of 45.1 million per mm².

The compute resources differ substantially. The AMD card has 6,144 shading units, 384 texture mapping units, 192 ROPs, and 96 ray tracing cores. The NVIDIA card offers 6,400 shading units, but only 200 TMUs, 80 ROPs, and 50 RT cores. Notably, the NVIDIA CMP 90HX includes 200 tensor cores, a feature completely absent from the AMD specification list. The AMD card's FP32 and FP16 throughput is 61.32 TFLOPS each, while the NVIDIA card reaches 21.89 TFLOPS in both, highlighting a 1:1 ratio in each case.

Memory architectures are also distinct. The AMD Radeon PRO W7900 uses 48 GB of GDDR6 on a 384-bit bus, delivering 864.0 GB/s of bandwidth. The NVIDIA CMP 90HX has 10 GB of GDDR6X on a 320-bit bus, providing 760.3 GB/s. The AMD card's memory clock is 2250 MHz (18 Gbps effective), while the NVIDIA card runs at 1188 MHz (19 Gbps effective). The physical design differs as well: the AMD card is a triple-slot unit with 3x DisplayPort 2.1 and 1x mini-DisplayPort 2.1 outputs, whereas the NVIDIA card is a dual-slot unit with no display outputs whatsoever, reflecting its dedicated mining purpose.

Power and interface specifications reinforce their different roles. The AMD card has a 295 W TDP, requires a 600 W suggested PSU, and connects via PCIe 4.0 x16. The NVIDIA card has a 320 W TDP, suggests a 700 W PSU, and uses only PCIe 1.0 x4, which is sufficient for mining workloads. Both cards share the same DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 API support, but their production statuses diverge: AMD is active, NVIDIA is end-of-life.

Head-to-Head Benchmarks

The only recorded head-to-head benchmark is the Geekbench OpenCL test. The AMD Radeon PRO W7900 scores 84,379, while the NVIDIA CMP 90HX scores 69,000. This gives the AMD card a decisive 22.3% advantage, the sole win in the comparison (1 win for AMD, 0 for NVIDIA). Looking at the broader benchmark averages, the AMD card's 110,725 average score far exceeds the NVIDIA card's 69,000, a gap of roughly 60%. To put these numbers into perspective, the AMD card's nearest rivals include the AMD Radeon Pro Vega II (average score 109,617, 1% lower), the NVIDIA RTX A5500 Mobile (113,944, 2.8% higher), the AMD Radeon Pro W6600X (107,342, 3.2% lower), and the NVIDIA Tesla V100 SXM2 16 GB (114,395, 3.2% higher). The AMD card sits comfortably in the 94th percentile of all GPUs, indicating strong overall performance. The NVIDIA CMP 90HX, by contrast, finds its nearest rivals at much lower performance levels: the Intel Arc A770 (68,809, 0.3% lower), the AMD Radeon Instinct MI25 (68,562, 0.6% lower), the AMD Radeon Pro WX 8200 (69,870, 1.2% higher), and the NVIDIA Quadro P6000 (69,986, 1.4% higher). Its 90th percentile ranking shows it is still above most GPUs, but the raw score gap to the AMD card is substantial. The data indicates that in OpenCL compute workloads, the AMD Radeon PRO W7900 is categorically faster, and this single benchmark result aligns with the architectural advantages in FP32 throughput, memory bandwidth, and pixel/texture rates.

Specification Differences

The two cards differ across nearly every major specification category. The process node is 5 nm for AMD versus 8 nm for NVIDIA, and the foundries are TSMC versus Samsung. Transistor counts are 57,700 million versus 28,300 million, and die sizes are 529 mm² versus 628 mm², leading to transistor densities of 109.1M/mm² versus 45.1M/mm². Base clocks are 1760 MHz versus 1500 MHz, and boost clocks are 2495 MHz versus 1710 MHz. Memory size is 48 GB versus 10 GB, with GDDR6 versus GDDR6X types. Bus widths are 384 bit versus 320 bit, and bandwidths are 864.0 GB/s versus 760.3 GB/s. Shading units are 6,144 versus 6,400, TMUs are 384 versus 200, ROPs are 192 versus 80, and RT cores are 96 versus 50. The NVIDIA card has 200 tensor cores, while the AMD card lists none. Pixel rates are 479.0 GPixel/s versus 136.8 GPixel/s, texture rates are 958.1 GTexel/s versus 342.0 GTexel/s, and FP32/FP16 are 61.32 TFLOPS versus 21.89 TFLOPS. TDP is 295 W versus 320 W, slot width is triple-slot versus dual-slot, and suggested PSU is 600 W versus 700 W. Bus interfaces are PCIe 4.0 x16 versus PCIe 1.0 x4. Display outputs are present on the AMD card but absent on the NVIDIA card. Dimensions differ: the AMD card is 280 mm long, 110 mm tall, and 51 mm wide, while the NVIDIA card is 285 mm long and 112 mm tall (width not specified). Production status is active versus end-of-life, release dates are 2023-05-25 versus 2021-07-27, and the AMD card has a launch MSRP of 3,999 USD, while the NVIDIA card has none. The AMD card's predecessor is the Radeon Pro Vega, whereas the NVIDIA card has no predecessor listed.

The Verdict

The data clearly favors the AMD Radeon PRO W7900 in raw performance. Its 22.3% lead in the Geekbench OpenCL test and its far higher average benchmark score (110,725 versus 69,000) establish it as the superior compute card. The AMD card also offers more than four times the memory capacity (48 GB versus 10 GB), higher memory bandwidth, and a larger set of rendering resources (more TMUs, ROPs, and RT cores). The NVIDIA CMP 90HX, while having more shading units and tensor cores, falls behind in every measured benchmark and has no display outputs, making it unsuitable for general graphics work. The AMD card is still in active production, whereas the NVIDIA card is end-of-life. For any workload that relies on OpenCL compute, the AMD Radeon PRO W7900 is the clear choice based on the recorded data. The NVIDIA card's only advantages are its smaller physical footprint (dual-slot versus triple-slot), lower TDP (320 W versus 295 W is actually higher, so this is not an advantage), and tensor cores, but these do not translate into higher benchmark scores. The verdict is straightforward: choose the AMD Radeon PRO W7900 for performance, memory capacity, and longevity. The NVIDIA CMP 90HX, given its mining-specific design and end-of-life status, is not a competitive alternative in standard compute benchmarks.

Where Each One Wins

The AMD Radeon PRO W7900 wins in every recorded benchmark category. In the Geekbench OpenCL test, it scores 84,379 versus 69,000, a 22.3% advantage. Its average benchmark score of 110,725 versus 69,000 further underscores its dominance. The AMD card is also the only one with display outputs, making it suitable for graphics rendering, workstation tasks, and general-purpose compute. Its 48 GB memory capacity and 864.0 GB/s bandwidth provide substantial headroom for large datasets, and its higher pixel and texture rates (479.0 GPixel/s and 958.1 GTexel/s) suggest superior rasterization throughput. The NVIDIA CMP 90HX, on the other hand, has no wins in the head-to-head comparison. Its strengths are limited to specific hardware features: it has 200 tensor cores, which may benefit tensor-based workloads, and its 10 GB of GDDR6X memory offers a higher effective memory clock (19 Gbps versus 18 Gbps), though lower overall bandwidth. Its dual-slot design and shorter length (285 mm versus 280 mm, though the AMD is actually shorter) are minor physical considerations. However, without any benchmark victories and with no display outputs, the NVIDIA card's practical use cases are confined to mining, as its name suggests. In a performance-driven comparison, the AMD Radeon PRO W7900 is the only card with recorded wins, and the data does not support selecting the NVIDIA CMP 90HX for any compute-intensive role. For workloads requiring high memory capacity, high FP32 throughput, or display output, the AMD card is the unambiguous winner. The NVIDIA card's tensor cores might be relevant for AI inference, but no benchmark data in the database validates that advantage. Therefore, based strictly on the recorded measurements, the AMD Radeon PRO W7900 is the recommended choice for all tested scenarios, while the NVIDIA CMP 90HX has no demonstrated winning use case.