AMD Radeon Pro Vega 64X vs NVIDIA CMP 90HX Comparison

# AMD Radeon Pro Vega 64X vs NVIDIA CMP 90HX

The AMD Radeon Pro Vega 64X and NVIDIA CMP 90HX represent two very different answers to the question of what a high-end GPU should be. The Vega 64X is a 2019 professional Mac-oriented part built on GCN 5.0, while the CMP 90HX is a 2021 Ampere-based mining card with no display outputs. In the single available head-to-head benchmark, the AMD card wins decisively, but the two cards occupy entirely different performance envelopes, architectural philosophies, and feature sets. This analysis breaks down where each card wins, how their underlying designs differ, and what the data actually says about their respective capabilities.

Where Each One Wins

The benchmark data yields one clear head-to-head result: in Geekbench OpenCL, the AMD Radeon Pro Vega 64X scores 78,467 against the NVIDIA CMP 90HX’s 69,000, a 13.7% advantage for AMD. That is the only direct comparison available, and it puts the Vega 64X ahead in raw compute throughput as measured by that workload.

However, the story is more nuanced when looking at each card’s standing among its own rivals. The Vega 64X posts an average benchmark score of 80,959 across its two tests (Geekbench Metal at 83,450 and Geekbench OpenCL at 78,467), placing it in the 92nd percentile of all GPUs. Its nearest rivals include the AMD Radeon PRO W6600 (average score 81,995, which is 1.3% higher), the NVIDIA GeForce RTX 5090 (79,842, 1.4% lower), the NVIDIA Tesla P100 PCIe 16 GB (79,605, 1.7% lower), and the Tesla P100 PCIe 12 GB (79,396, 2.0% lower). This shows the Vega 64X sits in a tight cluster at the top end of compute performance, essentially trading blows with much newer and more expensive hardware.

The CMP 90HX, by contrast, has a single benchmark score of 69,000 in Geekbench OpenCL, placing it in the 90th percentile. Its nearest rivals are the Intel Arc A770 (68,809, 0.3% higher), the AMD Radeon Instinct MI25 (68,562, 0.6% higher), the AMD Radeon Pro WX 8200 (69,870, 1.2% lower), and the NVIDIA Quadro P6000 (69,986, 1.4% lower). The CMP 90HX is thus clustered with mid-to-high-end workstation cards, but it sits roughly 11,959 points (or about 17.3%) behind the Vega 64X in average score.

Where each card wins depends on the workload. The Vega 64X wins the compute benchmark outright and also offers a Metal score, which the CMP 90HX cannot match because it has no display outputs and thus no graphics API support in the traditional sense. The CMP 90HX’s strengths lie elsewhere: it has a higher boost clock (1,710 MHz vs 1,468 MHz), more shading units (6,400 vs 4,096), and substantially higher FP32 throughput (21.89 TFLOPS vs 12.03 TFLOPS). In raw theoretical compute, the CMP 90HX is 81.9% higher, but the Geekbench result shows that architectural efficiency and memory subsystem design matter more than raw FLOP counts in at least one real-world workload.

Architecture Differences

The two cards could hardly be more different under the hood. The AMD Radeon Pro Vega 64X uses the Vega 10 chip built on GCN 5.0 architecture, fabricated 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, built on Samsung’s 8 nm process, with 28,300 million transistors on a 628 mm² die — a density of 45.1 million per square millimeter. That means the CMP 90HX has more than double the transistor count and nearly double the density, reflecting the newer process node and more complex design.

Memory configurations diverge sharply. The Vega 64X features 16 GB of HBM2 on a 2048-bit bus, delivering 512.0 GB/s of bandwidth with memory clocked at 1000 MHz (2 Gbps effective). The CMP 90HX has 10 GB of GDDR6X on a 320-bit bus, but its memory runs at 1188 MHz (19 Gbps effective), yielding 760.3 GB/s of bandwidth — 48.5% more than the Vega 64X. This bandwidth advantage is notable for mining workloads, which are typically memory-bound, and it helps explain why the CMP 90HX was designed for that purpose despite its lower benchmark score.

Compute resources tell a similar story of generational shift. The Vega 64X has 4,096 shading units, 256 TMUs, and 64 ROPs, with pixel and texture rates of 93.95 GPixel/s and 375.8 GTexel/s respectively. Its FP32 throughput is 12.03 TFLOPS, with FP16 at 24.05 TFLOPS via a 2:1 ratio. The CMP 90HX has 6,400 shading units, 200 TMUs, and 80 ROPs, with pixel rate of 136.8 GPixel/s (45.6% higher) and texture rate of 342.0 GTexel/s (9.0% lower). Its FP32 is 21.89 TFLOPS, and FP16 is also 21.89 TFLOPS at a 1:1 ratio — meaning no FP16 acceleration advantage. The CMP 90HX also includes 50 RT cores and 200 tensor cores, features entirely absent from the Vega 64X, which has neither.

Power and form factor differences are stark. The Vega 64X is rated at 250 W TDP, is an IGP (integrated graphics package) with no power connectors, and has display outputs described as "Portable Device Dependent." The CMP 90HX draws 320 W, is a dual-slot card measuring 285 mm (11.2 inches) in length and 112 mm (4.4 inches) in height, requires two 8-pin power connectors, and a 700 W suggested PSU. It has no display outputs at all. The bus interfaces also differ: the Vega 64X uses PCIe 3.0 x16, while the CMP 90HX uses PCIe 1.0 x4 — a deliberate limitation for mining cards to reduce cost and complexity.

API support reflects their different target markets. The Vega 64X supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3. The CMP 90HX supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, giving it newer API versions. But the CMP 90HX’s lack of display outputs means those APIs are moot for any visual workload — it cannot drive a monitor, making it useless for gaming or traditional graphics work.

The Verdict

The data points to a clear conclusion: the AMD Radeon Pro Vega 64X is the better card for any compute workload that relies on OpenCL or Metal, which is the only benchmark ground they share. Its 13.7% advantage in Geekbench OpenCL is significant, and its 92nd percentile standing among all GPUs (versus the CMP 90HX’s 90th) reinforces that it is the higher-performing part overall.

For users who need a GPU for general-purpose compute, professional graphics, or any task requiring display output, the Vega 64X is the only choice — the CMP 90HX literally cannot output video. The Vega 64X’s 16 GB of HBM2 memory is also more generous than the CMP 90HX’s 10 GB of GDDR6X, which matters for large datasets.

However, the CMP 90HX has its own rationale. Its 760.3 GB/s memory bandwidth is 48.5% higher than the Vega 64X, and its FP32 throughput of 21.89 TFLOPS is 81.9% higher. For memory-bound or FP32-heavy workloads that do not require display output — and that are not captured by the Geekbench OpenCL test — the CMP 90HX could plausibly outperform the Vega 64X. Its higher transistor count (28,300 million vs 12,500 million) and newer 8 nm process also suggest better raw computational headroom.

The practical verdict depends on use case. For a workstation or development machine where you need to see results, the Vega 64X wins outright. For a headless compute node or mining rig where only throughput matters, the CMP 90HX’s higher theoretical specs and bandwidth make it the more compelling option — though the single benchmark available does not confirm this in practice.

FAQ

Q: Which card has the higher average benchmark score?

A: The AMD Radeon Pro Vega 64X has an average benchmark score of 80,959, versus 69,000 for the NVIDIA CMP 90HX. That is a difference of 11,959 points.

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

A: No. The CMP 90HX has "No outputs" listed for display outputs, making it unsuitable for any task requiring a monitor.

Q: How much memory bandwidth does each card have?

A: The AMD Radeon Pro Vega 64X has 512.0 GB/s from its 16 GB HBM2 memory on a 2048-bit bus. The NVIDIA CMP 90HX has 760.3 GB/s from its 10 GB GDDR6X memory on a 320-bit bus.

Q: What is the transistor count difference between the two cards?

A: The Vega 64X has 12,500 million transistors, while the CMP 90HX has 28,300 million transistors — more than double.

Q: Which card has the higher FP32 throughput?

A: The NVIDIA CMP 90HX has 21.89 TFLOPS of FP32, while the AMD Radeon Pro Vega 64X has 12.03 TFLOPS. The CMP 90HX is 81.9% higher.

Q: What is the percentile ranking for each card among all GPUs?

A: The AMD Radeon Pro Vega 64X is in the 92nd percentile, while the NVIDIA CMP 90HX is in the 90th percentile.

Head-to-Head Benchmarks

The only direct head-to-head benchmark available is Geekbench OpenCL, and it is a decisive win for the AMD Radeon Pro Vega 64X. The AMD card scores 78,467 against the NVIDIA CMP 90HX’s 69,000, a delta of 13.7% in AMD’s favor. This result is notable because it contradicts the raw specification sheet: the CMP 90HX has 81.9% higher FP32 throughput, 48.5% higher memory bandwidth, and 126.4% more transistors, yet it loses by a substantial margin in this compute workload.

The win suggests that the Vega 64X’s GCN architecture is more efficient at translating raw resources into actual compute performance, at least in OpenCL. The Vega 64X’s HBM2 memory, while lower in bandwidth, may have lower latency or better access patterns for the benchmark’s workload. Additionally, the CMP 90HX’s PCIe 1.0 x4 interface could be a bottleneck in certain data-transfer scenarios, though it should not affect a purely compute-bound kernel.

Looking at the rivals for context, the Vega 64X’s score of 78,467 in OpenCL is below its own Metal score of 83,450, indicating that the card performs better in Apple’s API. The CMP 90HX has no Metal score at all. The Vega 64X’s nearest rival, the AMD Radeon PRO W6600, averages 81,995 — 1.3% higher — showing that the Vega 64X is competitive with much newer AMD workstation parts. The CMP 90HX’s nearest rival, the Intel Arc A770, scores 68,809, just 0.3% below, meaning the CMP 90HX is essentially tied with a mainstream consumer card in this metric.

The broader picture from the benchmark data is that the Vega 64X belongs in a higher performance tier. Its 92nd percentile ranking, combined with its 13.7% head-to-head win, places it clearly above the CMP 90HX in the one test they share. The CMP 90HX’s 90th percentile is respectable, but it trails the Vega 64X by roughly 17% in average score. For any buyer choosing between these two based on benchmark performance alone, the AMD card is the winner — provided the workload is compatible with its older GCN architecture and its IGP form factor.