AMD Radeon RX 6650M XT vs NVIDIA CMP 90HX Comparison

# AMD Radeon RX 6650M XT vs NVIDIA CMP 90HX

FAQ

Q: Which GPU wins in the available benchmark comparison?

A: The AMD Radeon RX 6650M XT wins the single head-to-head benchmark (Geekbench OpenCL) with a score of 76,904 versus the NVIDIA CMP 90HX’s 69,000, a 11.5% advantage.

Q: What is the performance percentile ranking for each card?

A: The RX 6650M XT sits at the 91st percentile among all GPUs, while the CMP 90HX ranks at the 90th percentile. Despite the close percentile placement, the AMD card holds a clear edge in the measured test.

Q: How does the CMP 90HX compare to its nearest rivals?

A: The CMP 90HX scores 0.3% above the Intel Arc A770 (68,809) and 0.6% above the AMD Radeon Instinct MI25 (68,562), but trails the AMD Radeon Pro WX 8200 by 1.2% and the NVIDIA Quadro P6000 by 1.4%.

Q: What are the memory specifications of each card?

A: The RX 6650M XT has 8 GB of GDDR6 on a 128-bit bus with 256.0 GB/s bandwidth, while the CMP 90HX has 10 GB of GDDR6X on a 320-bit bus with 760.3 GB/s bandwidth.

Q: Which card has higher raw compute throughput?

A: The CMP 90HX delivers 21.89 TFLOPS FP32 and 21.89 TFLOPS FP16, while the RX 6650M XT delivers 9.896 TFLOPS FP32 and 19.79 TFLOPS FP16. The NVIDIA card has a 2.2x FP32 advantage.

Q: What are the power requirements for each GPU?

A: The RX 6650M XT has a 120 W TDP with no power connectors (IGP form factor), while the CMP 90HX has a 320 W TDP, requires two 8-pin power connectors, and recommends a 700 W PSU.

Architecture Differences

The two GPUs come from fundamentally different design philosophies. The AMD Radeon RX 6650M XT is built on the Navi 23 chip using RDNA 2.0 architecture, fabricated on TSMC’s 7 nm process. It integrates 11,060 million transistors on a 237 mm² die, yielding a transistor density of 46.7M per mm². In contrast, the NVIDIA CMP 90HX uses the GA102 chip with Ampere architecture, manufactured on Samsung’s 8 nm process, packing 28,300 million transistors into a massive 628 mm² die at 45.1M per mm² density.

The CMP 90HX is clearly a mining-specific product. It has no display outputs, uses a PCIe 1.0 x4 bus interface, and comes in a dual-slot form factor measuring 285 mm in length and 112 mm in height. The RX 6650M XT is a mobile part with an IGP slot width, no power connectors, and display outputs described as "Portable Device Dependent." Its bus interface is PCIe 4.0 x8.

Compute resources tell a stark story. The CMP 90HX has 6,400 shading units, 200 TMUs, 80 ROPs, 50 RT cores, and 200 tensor cores. The RX 6650M XT has 2,048 shading units, 128 TMUs, 64 ROPs, and 32 RT cores — but no tensor cores. The NVIDIA card’s FP32 throughput of 21.89 TFLOPS is more than double the AMD card’s 9.896 TFLOPS. In FP16, the gap narrows significantly: the CMP 90HX still outputs 21.89 TFLOPS (1:1 ratio), while the RX 6650M XT reaches 19.79 TFLOPS via a 2:1 ratio.

Memory architecture diverges sharply as well. The RX 6650M XT uses 8 GB of GDDR6 on a 128-bit bus at 2000 MHz (16 Gbps effective) for 256.0 GB/s bandwidth. The CMP 90HX uses 10 GB of GDDR6X on a 320-bit bus at 1188 MHz (19 Gbps effective) for 760.3 GB/s — nearly three times the bandwidth. The NVIDIA card’s pixel rate (136.8 GPixel/s) is lower than the AMD card’s (154.6 GPixel/s), but its texture rate (342.0 GTexel/s) is higher than the AMD card’s (309.2 GTexel/s).

Both GPUs support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. Both are end-of-life products, with the CMP 90HX released in July 2021 and the RX 6650M XT in January 2022.

Head-to-Head Benchmarks

The only direct comparison available is the Geekbench OpenCL test, and the result is unambiguous. The AMD Radeon RX 6650M XT scores 76,904 against the NVIDIA CMP 90HX’s 69,000. That is an 11.5% victory for the AMD part, giving it a 1-0 win record in head-to-head tests.

Context from nearest rivals reinforces the gap. The RX 6650M XT’s score places it just 1% behind the NVIDIA GeForce RTX 5090 D (77,712), 2.6% behind the AMD Radeon RX 6850M XT (78,940), and 3.1% to 3.4% behind the Tesla P100 variants (79,396 and 79,605 respectively). The CMP 90HX, meanwhile, is clustered tightly with mid-range competitors: 0.3% above the Intel Arc A770 (68,809), 0.6% above the AMD Radeon Instinct MI25 (68,562), but 1.2% below the AMD Radeon Pro WX 8200 (69,870) and 1.4% below the NVIDIA Quadro P6000 (69,986).

The delta between the two cards in this test — 7,904 raw points — is larger than the entire gap between the CMP 90HX and any of its nearest rivals. In practical terms, the RX 6650M XT occupies a performance tier roughly equivalent to a high-end desktop GPU, while the CMP 90HX competes with workstation and mid-range gaming parts. The percentile rankings (91 vs 90) understate the difference because they are coarse buckets; the raw score gap is decisive.

Specification Differences

The differences between these two GPUs are extensive and define their distinct purposes.

Process and die: The RX 6650M XT uses a 7 nm TSMC process with 11,060 million transistors on 237 mm². The CMP 90HX uses an 8 nm Samsung process with 28,300 million transistors on 628 mm². The AMD chip has slightly higher transistor density (46.7M vs 45.1M per mm²).

Clocks: The RX 6650M XT runs at 2068 MHz base and 2416 MHz boost, with a 2162 MHz game clock. The CMP 90HX runs at 1500 MHz base and 1710 MHz boost, with no game clock specified. The AMD card’s 706 MHz boost advantage is substantial.

Memory: 8 GB GDDR6 on 128-bit vs 10 GB GDDR6X on 320-bit. Bandwidth: 256.0 GB/s vs 760.3 GB/s. Memory clock: 2000 MHz (16 Gbps effective) vs 1188 MHz (19 Gbps effective).

Compute units: 2,048 shading units, 128 TMUs, 64 ROPs, 32 RT cores, no tensor cores vs 6,400 shading units, 200 TMUs, 80 ROPs, 50 RT cores, 200 tensor cores.

Output rates: Pixel rate 154.6 vs 136.8 GPixel/s (AMD wins). Texture rate 309.2 vs 342.0 GTexel/s (NVIDIA wins). FP32 9.896 vs 21.89 TFLOPS (NVIDIA wins). FP16 19.79 vs 21.89 TFLOPS (NVIDIA wins, but narrowly).

Power and form factor: 120 W TDP, IGP slot, no power connectors, no PSU recommendation vs 320 W TDP, dual-slot, 2x 8-pin connectors, 700 W PSU recommendation. The CMP 90HX measures 285 mm x 112 mm; the RX 6650M XT has no listed dimensions.

Interface and outputs: PCIe 4.0 x8 with portable-device-dependent display outputs vs PCIe 1.0 x4 with no display outputs.

Release dates: January 2022 for AMD, July 2021 for NVIDIA. Both end-of-life.

Where Each One Wins

AMD Radeon RX 6650M XT wins on:

• Benchmark performance: 11.5% higher Geekbench OpenCL score (76,904 vs 69,000).
• Pixel fill rate: 154.6 vs 136.8 GPixel/s.
• Clock speeds: 2068 MHz base and 2416 MHz boost vs 1500 MHz base and 1710 MHz boost.
• Power efficiency: 120 W TDP vs 320 W TDP, with no external power connectors required.
• Portability: IGP form factor and display outputs make it usable in mobile systems.
• Release date: six months newer.

NVIDIA CMP 90HX wins on:

• Raw FP32 compute: 21.89 vs 9.896 TFLOPS — a 121% advantage.
• FP16 compute: 21.89 vs 19.79 TFLOPS.
• Memory capacity: 10 GB vs 8 GB.
• Memory bandwidth: 760.3 vs 256.0 GB/s — nearly 3x higher.
• Texture fill rate: 342.0 vs 309.2 GTexel/s.
• Tensor cores: 200 vs none.
• RT cores: 50 vs 32.
• Memory type: GDDR6X vs GDDR6.

The CMP 90HX is a compute and bandwidth monster, but it cannot display output and relies on a PCIe 1.0 x4 connection. The RX 6650M XT is a balanced mobile GPU that wins the only benchmark measured despite its smaller memory bus and lower raw specs.

The Verdict

The data shows a clear winner for general-purpose benchmark performance: the AMD Radeon RX 6650M XT. Its Geekbench OpenCL score of 76,904 beats the NVIDIA CMP 90HX by 11.5%, and it does so while consuming 120 W versus 320 W and requiring no external power connectors. For any workload that shows up in OpenCL benchmarks, the AMD card is the better choice.

However, the CMP 90HX is not without merit. Its 21.89 TFLOPS FP32 and FP16 throughput, combined with 760.3 GB/s memory bandwidth and 200 tensor cores, make it a purpose-built compute device. The 10 GB GDDR6X frame buffer and 320-bit bus give it a massive bandwidth advantage that could benefit memory-bound workloads. The absence of display outputs and the PCIe 1.0 x4 interface, though, severely limit its versatility.

For users who need a GPU in a mobile or compact system with display functionality, the RX 6650M XT is the only viable option between the two. For those building a dedicated compute or mining rig where display output is irrelevant and power is available, the CMP 90HX offers 2.2x the FP32 compute and 3x the memory bandwidth — but it will not win OpenCL benchmarks.

The percentile rankings (91st vs 90th) suggest near-parity, but the raw scores tell the real story: the RX 6650M XT outperforms the CMP 90HX in the measured test by a margin larger than the CMP 90HX’s distance to its own nearest rivals. The AMD card’s higher clocks (2416 MHz boost vs 1710 MHz) and superior pixel rate compensate for its smaller memory subsystem in the benchmark that matters most.

Choose the RX 6650M XT for benchmark-driven performance, efficiency, and portability. Choose the CMP 90HX only if raw compute throughput, tensor core availability, or memory bandwidth — not OpenCL scores — are your primary criteria.