AMD Radeon VII vs NVIDIA CMP 90HX Comparison

# NVIDIA CMP 90HX vs AMD Radeon VII

The data is unambiguous: the AMD Radeon VII is the faster card in the only direct head-to-head benchmark available, and it wins decisively. In Geekbench OpenCL, the Radeon VII scores 91,947 against the CMP 90HX’s 69,000, a 25% gap in favor of AMD. However, the CMP 90HX is a mining-specific product with no display outputs, while the Radeon VII is a fully featured consumer GPU with video outputs, so the choice depends entirely on workload. For compute performance in OpenCL, the Radeon VII is the clear pick; for a dedicated mining card with no display capability, the CMP 90HX is the only option that fits that niche.

The Verdict

Pick the AMD Radeon VII if you need raw compute performance, particularly in OpenCL workloads. It beats the CMP 90HX by 25% in the Geekbench OpenCL test, and it also offers higher memory capacity (16 GB vs 10 GB) and significantly more memory bandwidth (1.02 TB/s vs 760.3 GB/s). The Radeon VII also has display outputs (1x HDMI 2.0b and 3x DisplayPort 1.4a), making it a versatile card for any task beyond pure computation. Its average benchmark score is 66,004, placing it in the 90th percentile of all GPUs, and it outperforms its nearest rivals: it is 1.1% ahead of the NVIDIA Tesla T4, 1.4% ahead of the Tesla P40, 2.8% ahead of the AMD Radeon Pro WX 9100, and 3.4% ahead of the NVIDIA CMP 30HX.

Pick the NVIDIA CMP 90HX only if you require a card with no display outputs and a mining-specific design. It has a higher FP32 throughput (21.89 TFLOPS vs 13.44 TFLOPS), more shading units (6400 vs 3840), and it draws 320 W, which is 25 W more than the Radeon VII’s 295 W. Its average benchmark score is 69,000, also in the 90th percentile, and it is virtually tied with its nearest rivals: 0.3% ahead of the Intel Arc A770, 0.6% ahead of the AMD Radeon Instinct MI25, 1.2% behind the AMD Radeon Pro WX 8200, and 1.4% behind the NVIDIA Quadro P6000. However, the CMP 90HX loses the only direct comparison, and its lack of display outputs makes it useless for general-purpose computing.

Architecture Differences

The two cards come from fundamentally different architectures. The NVIDIA CMP 90HX uses the GA102 chip built on Ampere architecture, manufactured on Samsung’s 8 nm process. It packs 28,300 million transistors on a 628 mm² die, yielding a transistor density of 45.1 million per mm². The AMD Radeon VII uses the Vega 20 chip with GCN 5.1 architecture, built on TSMC’s 7 nm process. It has 13,230 million transistors on a 331 mm² die, giving a density of 40.0 million per mm². The CMP 90HX has more than double the transistors and nearly double the die size, but the Radeon VII achieves a slightly lower density on a smaller, more modern node.

Core configurations differ sharply. The CMP 90HX has 6400 shading units, 200 texture mapping units, 80 render output units, 50 ray tracing cores, and 200 tensor cores. The Radeon VII has 3840 shading units, 240 texture mapping units, and 64 render output units, with no dedicated ray tracing or tensor cores. This means the CMP 90HX excels at raw FP32 compute (21.89 TFLOPS) compared to the Radeon VII’s 13.44 TFLOPS, but the Radeon VII offers higher FP16 performance at 26.88 TFLOPS thanks to a 2:1 ratio, versus the CMP 90HX’s 1:1 ratio at 21.89 TFLOPS. Pixel rate favors NVIDIA (136.8 GPixel/s vs 112.0 GPixel/s), while texture rate favors AMD (420.0 GTexel/s vs 342.0 GTexel/s).

Memory subsystems are also distinct. The CMP 90HX has 10 GB of GDDR6X on a 320-bit bus with 760.3 GB/s bandwidth. The Radeon VII has 16 GB of HBM2 on a 4096-bit bus with 1.02 TB/s bandwidth. The Radeon VII’s memory clock is 1000 MHz (2 Gbps effective), while the CMP 90HX runs at 1188 MHz (19 Gbps effective). The Radeon VII’s wider bus and higher capacity give it a clear bandwidth advantage, which likely contributes to its OpenCL win. DirectX support also differs: the CMP 90HX supports DirectX 12 Ultimate (12_2) and Vulkan 1.4, while the Radeon VII only supports DirectX 12 (12_1) and Vulkan 1.3. Both support OpenGL 4.6.

FAQ

Q: Which card is faster in OpenCL?

A: The AMD Radeon VII is 25% faster, scoring 91,947 versus the NVIDIA CMP 90HX’s 69,000 in the Geekbench OpenCL test.

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

A: No. It has no display outputs at all, making it strictly a compute or mining card. The Radeon VII, by contrast, has 1x HDMI 2.0b and 3x DisplayPort 1.4a.

Q: Which card has more memory and bandwidth?

A: The Radeon VII has 16 GB of HBM2 with 1.02 TB/s bandwidth, versus the CMP 90HX’s 10 GB of GDDR6X with 760.3 GB/s bandwidth.

Q: What is the power consumption difference?

A: The CMP 90HX has a TDP of 320 W, while the Radeon VII has a TDP of 295 W. The CMP 90HX also recommends a 700 W PSU, versus 600 W for the Radeon VII.

Q: Do these cards support ray tracing?

A: The CMP 90HX has 50 ray tracing cores, but the Radeon VII has no ray tracing cores at all.

Q: Which card has a higher FP32 throughput?

A: The CMP 90HX achieves 21.89 TFLOPS FP32, significantly higher than the Radeon VII’s 13.44 TFLOPS.

Specification Differences

| Specification | NVIDIA CMP 90HX | AMD Radeon VII |

|---|---|---|

| Chip | GA102 | Vega 20 |

| Architecture | Ampere | GCN 5.1 |

| Process Node | 8 nm (Samsung) | 7 nm (TSMC) |

| Transistors | 28,300 million | 13,230 million |

| Die Size | 628 mm² | 331 mm² |

| Transistor Density | 45.1M / mm² | 40.0M / mm² |

| Base Clock | 1500 MHz | 1400 MHz |

| Boost Clock | 1710 MHz | 1750 MHz |

| Memory Clock | 1188 MHz (19 Gbps effective) | 1000 MHz (2 Gbps effective) |

| Memory Size | 10 GB | 16 GB |

| Memory Type | GDDR6X | HBM2 |

| Memory Bus Width | 320 bit | 4096 bit |

| Memory Bandwidth | 760.3 GB/s | 1.02 TB/s |

| Shading Units | 6400 | 3840 |

| TMUs | 200 | 240 |

| ROPs | 80 | 64 |

| RT Cores | 50 | None |

| Tensor Cores | 200 | None |

| Pixel Rate | 136.8 GPixel/s | 112.0 GPixel/s |

| Texture Rate | 342.0 GTexel/s | 420.0 GTexel/s |

| FP32 | 21.89 TFLOPS | 13.44 TFLOPS |

| FP16 | 21.89 TFLOPS (1:1) | 26.88 TFLOPS (2:1) |

| TDP | 320 W | 295 W |

| Suggested PSU | 700 W | 600 W |

| Bus Interface | PCIe 1.0 x4 | PCIe 3.0 x16 |

| Display Outputs | None | 1x HDMI 2.0b, 3x DisplayPort 1.4a |

| DirectX | 12 Ultimate (12_2) | 12 (12_1) |

| Vulkan | 1.4 | 1.3 |

| Dimensions (L×H×W) | 285 mm × 112 mm | 280 mm × 125 mm × 40 mm |

| Release Date | 2021-07-27 | 2019-02-06 |

| Launch MSRP | None | 699 USD |

Head-to-Head Benchmarks

The only direct comparison available is the Geekbench OpenCL test, and the result is lopsided. The AMD Radeon VII scores 91,947, while the NVIDIA CMP 90HX scores 69,000. This is a 25% advantage for AMD, a significant margin that indicates the Radeon VII’s memory subsystem (16 GB HBM2 with 1.02 TB/s bandwidth) and higher FP16 throughput (26.88 TFLOPS vs 21.89 TFLOPS) provide a tangible compute advantage in this workload. The CMP 90HX, despite its higher FP32 raw throughput (21.89 TFLOPS vs 13.44 TFLOPS), falls well behind in OpenCL, suggesting that the Radeon VII’s architecture is better optimized for this type of parallel compute task.

The CMP 90HX does not win any head-to-head benchmark in the data, though it does have theoretical advantages in FP32, pixel rate, and shading units. Its average benchmark score of 69,000 places it in the 90th percentile, but that score is identical to its OpenCL result, meaning no other benchmark data exists for this card. The Radeon VII, with an average score of 66,004, has additional benchmark results: 2,304 in 3DMark Steel Nomad DX12, 77,975 in Geekbench Metal, and 91,788 in Geekbench Vulkan. These scores show the Radeon VII is strong across multiple APIs, while the CMP 90HX has only a single OpenCL data point.

Where Each One Wins

The AMD Radeon VII wins outright in the only head-to-head test, making it the superior choice for OpenCL compute workloads. Its 25% lead in that benchmark is backed by a larger memory pool (16 GB vs 10 GB) and higher bandwidth (1.02 TB/s vs 760.3 GB/s), which are critical for data-intensive tasks. The Radeon VII also offers display outputs, enabling use in workstations for visualization or general computing, and it supports multiple APIs (Metal, Vulkan, OpenCL) with strong scores across all of them. Its lower TDP (295 W vs 320 W) and lower PSU requirement (600 W vs 700 W) make it easier to integrate into existing systems.

The NVIDIA CMP 90HX wins in raw FP32 throughput, delivering 21.89 TFLOPS versus the Radeon VII’s 13.44 TFLOPS, and it has more shading units (6400 vs 3840) and 50 ray tracing cores, which the Radeon VII lacks entirely. It also supports DirectX 12 Ultimate and Vulkan 1.4, newer API versions than the Radeon VII’s DirectX 12 (12_1) and Vulkan 1.3. However, these theoretical advantages do not translate into a benchmark win, and the card’s lack of display outputs restricts it to mining or compute-only roles. Its PCIe 1.0 x4 interface is also a severe bottleneck compared to the Radeon VII’s PCIe 3.0 x16, potentially limiting data transfer speeds. The CMP 90HX is end-of-life, as is the Radeon VII, but the Radeon VII has a longer track record and a launch MSRP of 699 USD, while the CMP 90HX never had a listed MSRP.

For users prioritizing compute performance, the Radeon VII is the clear winner. For users who need a mining-specific card with no display outputs and higher FP32 throughput, the CMP 90HX is the only one that fits that purpose, but it loses the only available head-to-head benchmark. The verdict is straightforward: the Radeon VII is the better overall GPU, while the CMP 90HX is a niche product with a narrower appeal.