AMD Radeon PRO V620 vs NVIDIA CMP 90HX Comparison

The Verdict

The AMD Radeon PRO V620 is the clear overall winner in this comparison. In the only recorded head-to-head benchmark, the Geekbench OpenCL test, the AMD Radeon PRO V620 scores 128,580 against the NVIDIA CMP 90HX's 69,000. That is an 86.3% advantage, a decisive margin. The database places the AMD part at the 96th percentile of all GPUs, while the NVIDIA part sits at the 90th percentile. The Radeon PRO V620 also has a much higher average benchmark score of 136,472 compared to 69,000 for the CMP 90HX.

These two cards target different purposes. The Radeon PRO V620 is a workstation-oriented GPU with 32 GB of memory, while the CMP 90HX is a mining-focused product with no display outputs. For any compute or rendering workload measured by the database, the AMD card is the choice. The NVIDIA card only makes sense if a specific workflow requires its unique architecture traits, but the recorded data does not show any benchmark where it wins.

Architecture Differences

The two GPUs come from different manufacturers and use fundamentally different architectures. The AMD Radeon PRO V620 is built on the Navi 21 chip using RDNA 2.0 architecture, manufactured on a 7 nm process at TSMC. The NVIDIA CMP 90HX uses the GA102 chip with Ampere architecture, manufactured on an 8 nm process at Samsung. The process node difference is significant: 7 nm versus 8 nm, which affects transistor density. The AMD chip packs 26,800 million transistors into a 520 mm² die, giving a density of 51.5M per mm². The NVIDIA chip has 28,300 million transistors on a larger 628 mm² die, yielding 45.1M per mm².

The RDNA 2.0 architecture on the AMD card includes 72 ray tracing cores and 4,608 shading units. The Ampere architecture on the NVIDIA card includes 50 ray tracing cores, 200 tensor cores, and 6,400 shading units. The NVIDIA card has more shading units and adds tensor cores, which the AMD card lacks entirely. However, the AMD card has more texture mapping units at 288 versus 200, and more raster operations processors at 128 versus 80.

The memory subsystems also differ at the architectural level. The AMD card uses GDDR6 memory with a 256-bit bus. The NVIDIA card uses GDDR6X memory with a wider 320-bit bus. Despite the wider bus, the AMD card has substantially more memory capacity. The AMD card supports PCIe 4.0 x16, while the NVIDIA card is limited to PCIe 1.0 x4, a major interface bottleneck for the CMP 90HX.

Both cards support the same API levels: DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. Both have no display outputs, so neither can drive a monitor directly.

Head-to-Head Benchmarks

The only recorded head-to-head benchmark is Geekbench OpenCL. The AMD Radeon PRO V620 scores 128,580, while the NVIDIA CMP 90HX scores 69,000. The AMD card wins by 86.3%, a massive difference that reflects both the clock speed and memory advantages. The AMD card has a boost clock of 2200 MHz, compared to 1710 MHz for the NVIDIA card. The AMD card's FP32 throughput is 20.28 TFLOPS, while the NVIDIA card achieves 21.89 TFLOPS. Interestingly, the NVIDIA card has higher raw FP32 compute, yet it loses decisively in the OpenCL benchmark. This suggests that memory bandwidth and capacity play a larger role in this workload, or that the PCIe 1.0 x4 interface severely hampers the NVIDIA card.

The AMD card also has a much higher pixel rate at 281.6 GPixel/s versus 136.8 GPixel/s for NVIDIA, and a higher texture rate at 633.6 GTexel/s versus 342.0 GTexel/s. These rates align with the AMD card's higher clock speeds and larger number of TMUs and ROPs.

In terms of average benchmark score across all tests in the database, the AMD card scores 136,472, which is nearly double the NVIDIA card's 69,000. The AMD card sits at the 96th percentile of all GPUs, while the NVIDIA card sits at the 90th percentile. The AMD card's nearest rivals are all within 0.9% of its score: the AMD Radeon Pro W6800X Duo at 135,774 (0.5% behind), the AMD Radeon PRO W6800 at 135,396 (0.8% behind), the NVIDIA A10M at 135,230 (0.9% behind), and the NVIDIA RTX 4000 Ada Generation at 135,218 (0.9% behind). The NVIDIA card's nearest rivals include the Intel Arc A770 at 68,809 (0.3% ahead of the CMP 90HX), the AMD Radeon Instinct MI25 at 68,562 (0.6% behind), the AMD Radeon Pro WX 8200 at 69,870 (1.2% ahead), and the NVIDIA Quadro P6000 at 69,986 (1.4% ahead). These rival clusters show that the AMD card competes with top-tier workstation cards, while the NVIDIA card is in a much lower performance tier.

Specification Differences

The two cards differ on nearly every specification. The AMD Radeon PRO V620 uses a 7 nm process from TSMC; the NVIDIA CMP 90HX uses an 8 nm process from Samsung. The AMD card has 26,800 million transistors on a 520 mm² die; the NVIDIA card has 28,300 million on a 628 mm² die. Transistor density is 51.5M per mm² for AMD versus 45.1M per mm² for NVIDIA.

Clock speeds diverge sharply. The AMD card runs at a base clock of 1825 MHz and a boost clock of 2200 MHz. The NVIDIA card runs at 1500 MHz base and 1710 MHz boost. Memory clocks also differ: the AMD card uses 2000 MHz (16 Gbps effective), while the NVIDIA card uses 1188 MHz (19 Gbps effective).

Memory configuration is a major differentiator. The AMD card has 32 GB of GDDR6 memory on a 256-bit bus, delivering 512.0 GB/s of bandwidth. The NVIDIA card has 10 GB of GDDR6X memory on a 320-bit bus, delivering 760.3 GB/s of bandwidth. The NVIDIA card has higher bandwidth, but the AMD card has more than three times the capacity.

Compute resources differ in count and type. The AMD card has 4,608 shading units, 288 TMUs, 128 ROPs, and 72 ray tracing cores. The NVIDIA card has 6,400 shading units, 200 TMUs, 80 ROPs, 50 ray tracing cores, and 200 tensor cores. The AMD card has no tensor cores. FP32 performance is 20.28 TFLOPS for AMD and 21.89 TFLOPS for NVIDIA. FP16 performance is 40.55 TFLOPS (2:1) for AMD and 21.89 TFLOPS (1:1) for NVIDIA.

Power and physical specifications also differ. The AMD card has a TDP of 300 W, while the NVIDIA card has a TDP of 320 W. Both use dual-slot cooling and require 2x 8-pin power connectors, with a suggested 700 W PSU. The AMD card measures 267 mm in length, 120 mm in height, and 50 mm in width. The NVIDIA card measures 285 mm in length and 112 mm in height, with no width recorded. The bus interface is a stark contrast: PCIe 4.0 x16 for AMD versus PCIe 1.0 x4 for NVIDIA.

The release dates differ by several months. The NVIDIA CMP 90HX launched on July 27, 2021, while the AMD Radeon PRO V620 launched on November 3, 2021. Both are end-of-life products. The AMD card's predecessor is the Radeon Pro Vega, while the NVIDIA card has no recorded predecessor or successor.

FAQ

Q: Which GPU has a higher average benchmark score?

A: The AMD Radeon PRO V620 has an average benchmark score of 136,472, while the NVIDIA CMP 90HX has an average score of 69,000.

Q: How much memory does each card have?

A: The AMD Radeon PRO V620 has 32 GB of GDDR6 memory, while the NVIDIA CMP 90HX has 10 GB of GDDR6X memory.

Q: What is the bus interface difference?

A: The AMD Radeon PRO V620 uses PCIe 4.0 x16, while the NVIDIA CMP 90HX uses PCIe 1.0 x4.

Q: Does either card support display outputs?

A: Neither card has display outputs; both are listed as having "No outputs."

Q: Which card has a higher boost clock?

A: The AMD Radeon PRO V620 has a boost clock of 2200 MHz, higher than the NVIDIA CMP 90HX's boost clock of 1710 MHz.

Q: What is the FP16 performance difference?

A: The AMD Radeon PRO V620 achieves 40.55 TFLOPS (2:1) for FP16, while the NVIDIA CMP 90HX achieves 21.89 TFLOPS (1:1).

Where Each One Wins

The AMD Radeon PRO V620 wins the only recorded benchmark, the Geekbench OpenCL test, by 86.3%. It also has a higher average benchmark score, a better percentile ranking (96th versus 90th), and more memory capacity. The AMD card's higher clock speeds, larger number of TMUs and ROPs, and faster pixel and texture rates all support its dominant position in the measured workload. For any task that relies on memory capacity, such as large data sets or high-resolution textures, the 32 GB of GDDR6 memory is a clear advantage over the 10 GB on the NVIDIA card.

The NVIDIA CMP 90HX does have specific strengths in the specification sheet, even if they do not translate into benchmark wins. It has higher raw FP32 compute at 21.89 TFLOPS, more shading units at 6,400, and higher memory bandwidth at 760.3 GB/s. It also has 200 tensor cores, which the AMD card lacks entirely. For workloads that specifically leverage tensor cores, such as AI inference or certain deep learning operations, the NVIDIA card could be relevant. However, the database shows no benchmark where the NVIDIA card wins, and its PCIe 1.0 x4 interface is a severe limitation that likely throttles data transfer in real-world use.

The AMD card also has a higher FP16 throughput at 40.55 TFLOPS versus 21.89 TFLOPS, which matters for mixed-precision compute tasks. The AMD card's smaller die size and higher transistor density indicate a more efficient design. Both cards are end-of-life products, so availability and driver support are equal concerns.

In summary, the AMD Radeon PRO V620 is the superior card for any measured compute workload. The NVIDIA CMP 90HX only has a theoretical edge in raw FP32 and tensor core compute, but the recorded data does not show any scenario where it outperforms the AMD card. The choice is clear: pick the AMD Radeon PRO V620 unless a specific software requirement demands NVIDIA's tensor cores, and even then, the PCIe 1.0 x4 interface may make that impractical.