GPU Comparison

The AMD Radeon RX 6650M and NVIDIA CMP 90HX occupy opposite ends of the hardware spectrum, yet their benchmark results place them within striking distance of each other in the aggregate. The data shows the NVIDIA part edges ahead in the only direct comparison available, but the architectural and physical disparities between the two are vast. This analysis breaks down the head-to-head results, contextualizes each GPU’s standing among its peers, and clarifies which use cases each part actually serves based strictly on the recorded specifications and benchmark data.

Head-to-Head Benchmarks

The sole direct benchmark comparison in the data is the Geekbench OpenCL test. Here, the NVIDIA CMP 90HX scores 69,000, while the AMD Radeon RX 6650M scores 65,800. That represents a delta of -4.6% for the AMD part, meaning the CMP 90HX is approximately 4.6% faster in this particular compute workload. While a 4.6% lead is not overwhelming, it is consistent across the single data point, and it gives the NVIDIA card the only win in the head-to-head tally (1 win for the CMP 90HX, 0 for the RX 6650M).

It is worth noting that the RX 6650M has a second benchmark result in its record, a Geekbench Vulkan score of 77,735, which is not directly compared against the CMP 90HX, as the NVIDIA part lacks a Vulkan entry in the fact pack. The RX 6650M’s average benchmark score across both tests is 71,768, which is higher than the CMP 90HX’s single-test average of 69,000. This is an important nuance: when averaging all available data, the AMD part actually posts a higher aggregate number, but the direct OpenCL comparison still favors NVIDIA.

Looking at the nearest rivals for each card provides additional context. The CMP 90HX’s average score of 69,000 places it just 0.3% ahead of the Intel Arc A770 (68,809) and 0.6% ahead of the AMD Radeon Instinct MI25 (68,562). On the other side, it trails the AMD Radeon Pro WX 8200 by -1.2% (69,870) and the NVIDIA Quadro P6000 by -1.4% (69,986). This suggests the CMP 90HX is tightly clustered with a group of high-end workstation and enthusiast cards, with no dominant lead or deficit greater than 1.4% among its nearest peers.

The RX 6650M’s nearest rivals tell a similar story of tight competition. Its average score of 71,768 is 1.3% ahead of the AMD Radeon RX 6600 LE (70,829), but it trails the NVIDIA TITAN X Pascal by -0.5% (72,098), the AMD Radeon Pro Vega 64 by -0.8% (72,379), and the AMD Radeon Vega Frontier Edition by -2.2% (73,370). Despite the lower direct head-to-head score, the RX 6650M’s average benchmark percentile of 91 (versus 90 for the CMP 90HX) confirms that both parts sit in the same performance tier overall, with the AMD card holding a marginal edge in aggregate compute output when both its OpenCL and Vulkan results are considered.

The Verdict

The data does not support a clear overall winner. In the direct OpenCL test, the NVIDIA CMP 90HX is 4.6% faster, which is a measurable lead. However, the AMD Radeon RX 6650M posts a higher average benchmark score (71,768 vs. 69,000) and a slightly better percentile ranking (91st vs. 90th percentile of all GPUs). If the selection criterion is raw compute throughput in a single OpenCL workload, the CMP 90HX is the pick. If the criterion is broader compute capability across multiple API workloads, the RX 6650M has the edge based on its Vulkan result.

Beyond raw scores, the physical and functional differences are decisive for real-world selection. The CMP 90HX has no display outputs, making it unusable for any graphics output task. The RX 6650M is described as "Portable Device Dependent" for display outputs, meaning it is designed to drive a screen in a laptop or integrated form factor. The CMP 90HX is a dual-slot, 285 mm long card requiring two 8-pin power connectors and a 700 W suggested power supply, while the RX 6650M is an integrated graphics processor (IGP) with no power connectors and a 120 W TDP. These are not interchangeable products.

For a user needing a GPU to render frames to a display, the RX 6650M is the only viable option. For a user building a dedicated compute or mining rig where display output is irrelevant and power delivery is not a constraint, the CMP 90HX offers a 4.6% performance advantage in the tested OpenCL workload, along with substantially higher raw specs (21.89 TFLOPS FP32 vs. 8.659 TFLOPS FP32). But that performance comes at a 320 W TDP versus 120 W, and the CMP 90HX’s PCIe 1.0 x4 interface is a severe bottleneck for general system integration, whereas the RX 6650M uses PCIe 4.0 x8.

Where Each One Wins

The NVIDIA CMP 90HX wins in scenarios where raw compute throughput is the sole metric and power consumption is secondary. Its 21.89 TFLOPS of FP32 performance is more than double the RX 6650M’s 8.659 TFLOPS, and its 10 GB of GDDR6X memory on a 320-bit bus delivers 760.3 GB/s of bandwidth versus the RX 6650M’s 8 GB GDDR6 on a 128-bit bus at 224.0 GB/s. The CMP 90HX also has 200 tensor cores and 50 ray-tracing cores, which the RX 6650M lacks in tensor cores (listed as null) and has only 28 ray-tracing cores. For compute workloads that leverage these features, such as machine learning inference or certain rendering tasks, the CMP 90HX is clearly superior on paper.

The AMD Radeon RX 6650M wins in every scenario involving actual graphics output. It is the only one of the two with any display capability, it consumes 200 W less power, and it fits in an IGP form factor with no power connectors. Its Vulkan score of 77,735 is notably higher than its own OpenCL score, suggesting the architecture is well-optimized for modern graphics APIs. The RX 6650M also has a higher pixel rate (154.6 GPixel/s vs. 136.8 GPixel/s) despite lower texture rate (270.6 GTexel/s vs. 342.0 GTexel/s), indicating it may be better suited for resolution-bound tasks rather than texture-heavy workloads. For mobile or compact systems, the RX 6650M’s 7 nm process node and 11,060 million transistors in a 237 mm² die versus the CMP 90HX’s 8 nm process and 28,300 million transistors in a 628 mm² die make it the only practical choice.

FAQ

Q: Which GPU has a higher average benchmark score?

A: The AMD Radeon RX 6650M has an average benchmark score of 71,768, while the NVIDIA CMP 90HX has an average of 69,000.

Q: What is the difference in the direct OpenCL benchmark?

A: The NVIDIA CMP 90HX scores 69,000 versus the RX 6650M’s 65,800, giving NVIDIA a 4.6% lead in that test.

Q: Can the NVIDIA CMP 90HX output video to a display?

A: No. The CMP 90HX lists "No outputs" for display outputs, making it unsuitable for any graphics display task.

Q: What memory configurations do the two cards use?

A: The RX 6650M uses 8 GB of GDDR6 on a 128-bit bus with 224.0 GB/s bandwidth. The CMP 90HX uses 10 GB of GDDR6X on a 320-bit bus with 760.3 GB/s bandwidth.

Q: How do their power requirements differ?

A: The RX 6650M has a 120 W TDP and no power connectors. The CMP 90HX has a 320 W TDP, requires two 8-pin power connectors, and has a suggested power supply of 700 W.

Q: Which card has better API support for modern graphics?

A: Both cards support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. However, only the RX 6650M has a recorded Vulkan benchmark score (77,735), while the CMP 90HX has no Vulkan test data.

Architecture Differences

The two GPUs are built on fundamentally different architectures and process nodes. The AMD Radeon RX 6650M uses the Navi 23 chip based on RDNA 2.0 architecture, fabricated on a 7 nm process at TSMC. It packs 11,060 million transistors into a 237 mm² die, yielding a transistor density of 46.7 million per mm². The NVIDIA CMP 90HX uses the GA102 chip based on Ampere architecture, fabricated on an 8 nm process at Samsung. It contains 28,300 million transistors across a 628 mm² die, for a density of 45.1 million per mm². The AMD chip is smaller, denser, and more power-efficient per transistor count, while the NVIDIA chip is larger and consumes far more power.

Compute resources differ dramatically. The RX 6650M has 1,792 shading units, 112 texture mapping units, and 64 render output units. The CMP 90HX has 6,400 shading units, 200 TMUs, and 80 ROPs. The NVIDIA card also features 200 tensor cores and 50 ray-tracing cores, while the AMD card has 28 ray-tracing cores and no tensor cores listed. This translates to a massive difference in raw FP32 throughput: 21.89 TFLOPS for the CMP 90HX versus 8.659 TFLOPS for the RX 6650M. The FP16 performance also diverges, with the CMP 90HX achieving 21.89 TFLOPS (1:1 ratio) and the RX 6650M reaching 17.32 TFLOPS (2:1 ratio).

Clock speeds and memory architecture further separate the two. The RX 6650M runs at a base clock of 2068 MHz with a boost of 2416 MHz and a game clock of 2222 MHz. The CMP 90HX runs at a lower 1500 MHz base and 1710 MHz boost. Memory clocks are also different: the RX 6650M uses 1750 MHz (14 Gbps effective) GDDR6, while the CMP 90HX uses 1188 MHz (19 Gbps effective) GDDR6X. The bus widths are 128-bit for AMD and 320-bit for NVIDIA, resulting in the aforementioned bandwidth gap.

The bus interfaces and physical form factors are starkly different. The RX 6650M uses PCIe 4.0 x8 and is an IGP with no power connectors, designed for portable devices. The CMP 90HX uses PCIe 1.0 x4, a severely outdated interface that limits data transfer rates, and is a dual-slot card measuring 285 mm in length and 112 mm in height, requiring two 8-pin power connectors. The CMP 90HX is also a mining-specific product with no display outputs, whereas the RX 6650M’s display outputs are "Portable Device Dependent," meaning they exist but vary by the host laptop or system. Both are end-of-life products, with the RX 6650M released in January 2022 and the CMP 90HX in July 2021.