AMD Radeon Pro Vega 64 vs NVIDIA CMP 90HX Comparison

AMD Radeon Pro Vega 64 and NVIDIA CMP 90HX are two end-of-life graphics cards that serve fundamentally different purposes, yet their benchmark data reveals a closer competition than their disparate architectures suggest. The AMD card, a 2017-era professional mobile solution, edges out the NVIDIA mining-specific part in the single shared benchmark, though the CMP 90HX counters with significantly higher raw compute specifications. This analysis breaks down the numbers to determine which card holds up better in a purely performance-based comparison.

Head-to-Head Benchmarks

The only directly comparable benchmark between these two cards is Geekbench OpenCL, where the AMD Radeon Pro Vega 64 scores 71,094 against the NVIDIA CMP 90HX's 69,000. That gives AMD a 3% lead, a slim but definitive margin. In practical terms, this means the Vega 64 delivers slightly better general-purpose compute performance in OpenCL workloads, despite being four years older and built on a much older architecture. The delta is small enough that real-world differences would be negligible in most tasks, but it is a measurable victory for AMD.

Looking at the broader benchmark landscape, the AMD card has three benchmark scores recorded — Geekbench Metal at 71,868, Geekbench OpenCL at 71,094, and Geekbench Vulkan at 74,174 — giving it an average benchmark score of 72,379. The NVIDIA card has only one recorded score, its OpenCL result of 69,000, which also serves as its average. This means the CMP 90HX's performance profile is less thoroughly documented, but its single score places it in the 90th percentile versus the AMD card's 91st percentile. That one-percentile gap aligns with the 3% benchmark delta.

The nearest rival data reinforces the parity between these cards. The AMD Radeon Pro Vega 64 sits just 0.4% above the NVIDIA TITAN X Pascal (average score 72,098) and 0.9% above the AMD Radeon RX 6650M (71,768). The NVIDIA CMP 90HX, meanwhile, is 0.3% above the Intel Arc A770 (68,809) and 0.6% above the AMD Radeon Instinct MI25 (68,562). Both cards are clustered tightly with their contemporaries, suggesting neither has a dramatic performance advantage over the other in raw compute terms.

Where Each One Wins

The AMD Radeon Pro Vega 64 wins in raw OpenCL compute performance, as demonstrated by its 3% higher score. It also has a more complete benchmark profile, with strong Metal and Vulkan results that the NVIDIA card lacks entirely. The Vulkan score of 74,174 is notably higher than its OpenCL score, suggesting the Vega architecture handles Vulkan workloads particularly well. For users running OpenCL-based applications or Vulkan titles, the AMD card is the clear pick based on available data.

The NVIDIA CMP 90HX wins in raw specification headroom. Its FP32 compute of 21.89 TFLOPS is nearly double the AMD card's 11.06 TFLOPS, and its memory bandwidth of 760.3 GB/s exceeds the Vega 64's 402.4 GB/s by a wide margin. However, these advantages do not translate into a benchmark victory in the one test where they can be compared. The CMP 90HX also offers modern features like 50 ray tracing cores and 200 tensor cores, which the AMD card lacks entirely. For workloads that leverage these specialized units — machine learning inference, ray-traced rendering — the NVIDIA card has capabilities the AMD card simply cannot match, even if the OpenCL score does not reflect it.

Architecture Differences

The architectural divide between these two cards is stark. The AMD Radeon Pro Vega 64 uses the Vega 10 chip on GCN 5.0 architecture, built on a 14 nm process at GlobalFoundries. It packs 12,500 million transistors into a 495 mm² die, yielding a transistor density of 25.3 million per square millimeter. The NVIDIA CMP 90HX uses the GA102 chip on Ampere architecture, fabricated on Samsung's 8 nm process. It contains 28,300 million transistors across a 628 mm² die, for a density of 45.1 million per square millimeter — a 78% higher transistor density that reflects the newer manufacturing process.

Memory configurations differ fundamentally. The AMD card uses 16 GB of HBM2 on a 2048-bit bus, achieving 402.4 GB/s of bandwidth. The NVIDIA card uses 10 GB of GDDR6X on a 320-bit bus, achieving 760.3 GB/s — nearly double the bandwidth despite less capacity and a narrower bus. Clock speeds also favor NVIDIA: the CMP 90HX runs at 1500 MHz base and 1710 MHz boost, versus the Vega 64's 1250 MHz base and 1350 MHz boost. The memory clocks tell a similar story, with the NVIDIA card's 1188 MHz (19 Gbps effective) far exceeding the AMD card's 786 MHz (1572 Mbps effective).

Compute unit configurations differ as well. The NVIDIA card has 6400 shading units, 200 TMUs, and 80 ROPs, along with 50 RT cores and 200 tensor cores. The AMD card has 4096 shading units, 256 TMUs, and 64 ROPs, with no dedicated ray tracing or tensor hardware. The pixel and texture rates reflect these differences: NVIDIA leads in pixel rate at 136.8 GPixel/s versus 86.40 GPixel/s, while the texture rates are nearly identical at 342.0 GTexel/s versus 345.6 GTexel/s. FP16 performance is another differentiator — the AMD card achieves 22.12 TFLOPS via a 2:1 ratio, while the NVIDIA card delivers 21.89 TFLOPS at a 1:1 ratio, meaning its FP16 throughput matches FP32 rather than doubling it.

FAQ

Q: Which card has the higher average benchmark score?

A: The AMD Radeon Pro Vega 64 has an average benchmark score of 72,379, compared to the NVIDIA CMP 90HX's 69,000. The AMD card also holds a higher percentile ranking at 91 versus 90 for the NVIDIA card.

Q: Does the NVIDIA CMP 90HX support display outputs?

A: No. The CMP 90HX has no display outputs, making it unsuitable for any use case requiring a monitor connection. The AMD Radeon Pro Vega 64's display outputs are listed as portable device dependent, meaning they vary by implementation.

Q: What is the power draw difference between these cards?

A: The NVIDIA CMP 90HX has a TDP of 320 W and requires two 8-pin power connectors with a suggested 700 W power supply. The AMD Radeon Pro Vega 64 has a TDP of 250 W and uses no power connectors, as it is designed as an integrated graphics processor (IGP) solution.

Q: Which card supports the newer DirectX version?

A: The NVIDIA CMP 90HX supports DirectX 12 Ultimate (12_2), while the AMD Radeon Pro Vega 64 supports DirectX 12 (12_1). The NVIDIA card also supports Vulkan 1.4 versus Vulkan 1.3 on the AMD card.

Q: How do the memory bandwidths compare?

A: The NVIDIA CMP 90HX offers 760.3 GB/s of bandwidth from its 10 GB GDDR6X memory on a 320-bit bus. The AMD Radeon Pro Vega 64 offers 402.4 GB/s from its 16 GB HBM2 memory on a 2048-bit bus. NVIDIA's bandwidth is nearly double, despite the smaller capacity.

Q: Which card has more shading units?

A: The NVIDIA CMP 90HX has 6400 shading units, compared to 4096 on the AMD Radeon Pro Vega 64. The NVIDIA card also has more ROPs (80 versus 64), though the AMD card has more TMUs (256 versus 200).

Specification Differences

| Specification | AMD Radeon Pro Vega 64 | NVIDIA CMP 90HX |

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

| Architecture | GCN 5.0 | Ampere |

| Process Node | 14 nm | 8 nm |

| Transistors | 12,500 million | 28,300 million |

| Die Size | 495 mm² | 628 mm² |

| Transistor Density | 25.3M / mm² | 45.1M / mm² |

| Base Clock | 1250 MHz | 1500 MHz |

| Boost Clock | 1350 MHz | 1710 MHz |

| Memory Size | 16 GB HBM2 | 10 GB GDDR6X |

| Memory Bus | 2048 bit | 320 bit |

| Memory Bandwidth | 402.4 GB/s | 760.3 GB/s |

| Shading Units | 4096 | 6400 |

| TMUs | 256 | 200 |

| ROPs | 64 | 80 |

| RT Cores | None | 50 |

| Tensor Cores | None | 200 |

| FP32 Compute | 11.06 TFLOPS | 21.89 TFLOPS |

| FP16 Compute | 22.12 TFLOPS (2:1) | 21.89 TFLOPS (1:1) |

| Pixel Rate | 86.40 GPixel/s | 136.8 GPixel/s |

| Texture Rate | 345.6 GTexel/s | 342.0 GTexel/s |

| TDP | 250 W | 320 W |

| Slot Width | IGP | Dual-slot |

| Power Connectors | None | 2x 8-pin |

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

| Display Outputs | Portable Device Dependent | No outputs |

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

| Vulkan | 1.3 | 1.4 |

| Release Date | 2017-06-26 | 2021-07-27 |

The Verdict

Based strictly on the benchmark data, the AMD Radeon Pro Vega 64 is the better-performing card overall. It wins the only head-to-head test with a 3% margin in OpenCL, holds a higher average benchmark score (72,379 versus 69,000), and ranks one percentile higher. It also offers the flexibility of display outputs and a lower 250 W TDP, making it usable in a wider range of scenarios. The NVIDIA CMP 90HX, with no display outputs and a mining-specific design, is a more specialized tool.

However, the NVIDIA card's architectural advantages cannot be ignored for certain workloads. Its 21.89 TFLOPS FP32 compute, 760.3 GB/s memory bandwidth, and dedicated RT and tensor cores make it a more capable part for compute-heavy tasks that leverage those features — even though the OpenCL benchmark does not reflect that advantage. The CMP 90HX also supports DirectX 12 Ultimate and Vulkan 1.4, offering newer API compatibility.

For users who need a general-purpose card with proven OpenCL and Vulkan performance, the AMD Radeon Pro Vega 64 is the safer choice based on the data. For users who know their workloads can exploit the NVIDIA card's higher raw compute and specialized cores, the CMP 90HX offers more theoretical headroom, but the lack of display outputs and the single benchmark result make it a riskier proposition. The benchmark results indicate that AMD's older architecture still holds its own against NVIDIA's newer silicon in the one test where they compete directly.