AMD Radeon RX 6650M vs NVIDIA GeForce RTX 5090 Comparison

# Head-to-Head Benchmarks

The data shows a decisive, one-sided contest between the NVIDIA GeForce RTX 5090 and the AMD Radeon RX 6650M. Across the two shared benchmark tests, the RTX 5090 wins both, with margins that are nothing short of staggering. In Geekbench OpenCL, the NVIDIA card scores 334,370 against AMD's 65,800, a delta of 408.2%. The Vulkan result is similarly lopsided: 376,728 for the RTX 5090 versus 77,735 for the RX 6650M, a 384.6% advantage. These are not incremental gains; they represent a generational chasm in raw compute capability.

To contextualize these numbers, consider where each card sits relative to its own competitive set. The RTX 5090's average benchmark score is 79,842, which places it just 0.3% ahead of the NVIDIA Tesla P100 PCIe 16 GB and 1.1% ahead of the AMD Radeon RX 6850M XT. Its percentile ranking among all GPUs is 92, meaning it outperforms the vast majority of the field. The RX 6650M, by contrast, posts an average score of 71,768, sitting 0.5% behind the NVIDIA TITAN X Pascal and 2.2% behind the AMD Radeon Vega Frontier Edition. Its percentile is 91, which is surprisingly close to the RTX 5090's 92, but that near-parity in percentile masks the enormous absolute gap in performance — the 5090 averages nearly 8,000 points higher.

Breaking down the individual benchmark suites, the RTX 5090's dominance is consistent. In Passmark G3D, it scores 39,650, while its DirectX 11 result is 341 and DirectX 9 is 395. The RX 6650M has no Passmark scores in the data, so direct comparison is impossible there, but the Geekbench results tell the story clearly enough. The RTX 5090's compute score in Passmark GPU Compute is 26,756, and its 3DMark Steel Nomad DX12 score is 18,355 — both figures that dwarf anything the RX 6650M could plausibly produce given its Geekbench performance.

The verdict is unambiguous: the RTX 5090 is not just faster, it is in a different performance tier entirely. The RX 6650M's 8.659 TFLOPS FP32 peak and 224.0 GB/s memory bandwidth are respectable for a mobile GPU, but they are simply overwhelmed by the RTX 5090's 104.8 TFLOPS and 1.79 TB/s. In every measurable way, the NVIDIA card delivers over five times the compute throughput and over seven times the memory bandwidth of its AMD rival.

Architecture Differences

The architectural divide between these two GPUs is as wide as the performance gap. The RTX 5090 is built on NVIDIA's Blackwell 2.0 architecture, utilizing the GB202 chip, while the RX 6650M uses AMD's RDNA 2.0 architecture on the Navi 23 die. This is a matchup of two very different design philosophies: Blackwell 2.0 is a flagship desktop monster, while RDNA 2.0 in this implementation is a mobile-oriented part.

The process node tells part of the story. The RTX 5090 is fabricated on TSMC's 5 nm process, whereas the RX 6650M uses TSMC's 7 nm node. That two-generation leap in lithography translates directly into transistor density: the GB202 packs 92,200 million transistors into a 750 mm² die, yielding a density of 122.9M transistors per mm². The Navi 23, by contrast, houses 11,060 million transistors on a 237 mm² die, for a density of just 46.7M per mm². The RTX 5090 has over eight times the transistor count on roughly three times the die area.

Core configurations diverge dramatically. The RTX 5090 fields 21,760 shading units, 680 texture mapping units, and 176 ROPs. It also carries 170 ray tracing cores and 680 tensor cores, the latter being a feature entirely absent from the RX 6650M, which has no tensor core equivalent. The RX 6650M offers 1,792 shading units, 112 TMUs, and 64 ROPs, along with 28 ray tracing cores. In raw throughput, the RTX 5090's FP32 compute is 104.8 TFLOPS, while the RX 6650M manages 8.659 TFLOPS — a 12x disparity. The FP16 figures are even more telling: the RTX 5090 delivers 104.8 TFLOPS at a 1:1 ratio with FP32, while the RX 6650M's 17.32 TFLOPS comes at a 2:1 ratio, meaning it sacrifices precision for speed.

Memory subsystems are equally divergent. The RTX 5090 uses 32 GB of GDDR7 on a 512-bit bus, achieving 1.79 TB/s of bandwidth. The RX 6650M uses 8 GB of GDDR6 on a 128-bit bus, with 224.0 GB/s. Clock speeds are surprisingly close — the RTX 5090 boosts to 2407 MHz, the RX 6650M to 2416 MHz — but the memory clocks differ: 1750 MHz (28 Gbps effective) for the NVIDIA card versus 1750 MHz (14 Gbps effective) for AMD.

Power and physical characteristics highlight the different design targets. The RTX 5090 has a 575 W TDP, requires a 950 W power supply, uses a dual-slot cooler with a 16-pin connector, and measures 304 mm in length. The RX 6650M is an integrated graphics processor (IGP) with a 120 W TDP, no power connectors, and dimensions listed as portable device dependent. The RTX 5090 interfaces via PCIe 5.0 x16; the RX 6650M uses PCIe 4.0 x8. Both support DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4.

FAQ

Q: How much faster is the RTX 5090 in Geekbench OpenCL?

A: The RTX 5090 scores 334,370 versus the RX 6650M's 65,800, a delta of 408.2% in NVIDIA's favor.

Q: Does the RX 6650M have any benchmark where it beats the RTX 5090?

A: No. In the two head-to-head tests (Geekbench OpenCL and Vulkan), the RTX 5090 wins both, giving NVIDIA 2 wins and AMD 0.

Q: What is the memory configuration difference?

A: The RTX 5090 has 32 GB of GDDR7 on a 512-bit bus with 1.79 TB/s bandwidth. The RX 6650M has 8 GB of GDDR6 on a 128-bit bus with 224.0 GB/s bandwidth.

Q: Are the architectures fundamentally different?

A: Yes. The RTX 5090 uses NVIDIA's Blackwell 2.0 on a 5 nm process, while the RX 6650M uses AMD's RDNA 2.0 on a 7 nm process. The RTX 5090 also has 680 tensor cores; the RX 6650M has none.

Q: What is the power draw difference?

A: The RTX 5090 has a 575 W TDP and requires a 950 W power supply, while the RX 6650M has a 120 W TDP and no external power connectors.

Q: Which card has a higher boost clock?

A: The RX 6650M boosts slightly higher at 2416 MHz versus the RTX 5090's 2407 MHz, though the RTX 5090 has a much higher base clock of 2017 MHz versus 2068 MHz for the AMD part.

Specification Differences

The specification sheets for these two GPUs diverge on nearly every field. The process node differs: 5 nm for the RTX 5090 versus 7 nm for the RX 6650M. Transistor counts are 92,200 million versus 11,060 million, with die sizes of 750 mm² and 237 mm² respectively. The RTX 5090's transistor density is 122.9M per mm², more than double the RX 6650M's 46.7M per mm².

Core counts show a similar gulf. The RTX 5090 has 21,760 shading units, 680 TMUs, 176 ROPs, 170 RT cores, and 680 tensor cores. The RX 6650M has 1,792 shading units, 112 TMUs, 64 ROPs, and 28 RT cores, with no tensor cores. Pixel rates are 423.6 GPixel/s versus 154.6 GPixel/s, and texture rates are 1,636.8 GTexel/s versus 270.6 GTexel/s. FP32 compute is 104.8 TFLOPS versus 8.659 TFLOPS; FP16 is 104.8 TFLOPS (1:1) versus 17.32 TFLOPS (2:1).

Memory is a major differentiator: 32 GB GDDR7 versus 8 GB GDDR6, 512-bit versus 128-bit bus, 1.79 TB/s versus 224.0 GB/s bandwidth. Power envelopes are vastly different: 575 W TDP versus 120 W. The RTX 5090 is dual-slot with a 16-pin connector and 950 W suggested PSU; the RX 6650M is an IGP with no connectors. Bus interfaces are PCIe 5.0 x16 versus PCIe 4.0 x8. Display outputs are 1x HDMI 2.1b and 3x DisplayPort 2.1b for NVIDIA, versus portable device dependent for AMD. The RTX 5090 has physical dimensions (304 mm length); the RX 6650M has none listed. Release dates differ: 2025-01-29 for the RTX 5090 versus 2022-01-03 for the RX 6650M. The RTX 5090 has a launch MSRP of 1,999 USD; the RX 6650M has none listed.

Fields that are identical include the foundry (TSMC for both), the API support (DirectX 12 Ultimate, OpenGL 4.6, Vulkan 1.4 for both), and the memory clock speed of 1750 MHz (though effective rates differ: 28 Gbps versus 14 Gbps).

Where Each One Wins

The RTX 5090 wins everywhere measurable. In compute workloads, its 104.8 TFLOPS FP32 and 104.8 TFLOPS FP16 give it a 12x and 6x advantage over the RX 6650M respectively. For high-resolution gaming with ray tracing, the 170 RT cores and 680 tensor cores provide hardware acceleration that the RX 6650M's 28 RT cores and absent tensor cores cannot match. The 32 GB GDDR7 frame buffer with 1.79 TB/s bandwidth is ideal for 4K and 8K textures, large datasets, and AI workloads where memory capacity is the bottleneck. The RTX 5090's PCIe 5.0 x16 interface also offers double the bandwidth of the RX 6650M's PCIe 4.0 x8.

The RX 6650M's advantages are confined to practical deployment scenarios rather than performance. Its 120 W TDP makes it suitable for thin-and-light laptops where the RTX 5090's 575 W requirement would be impossible. As an IGP, it needs no external power connectors, simplifying system integration. Its portable device dependent display outputs mean it can adapt to various laptop form factors. Its smaller die size (237 mm²) and lower transistor count also suggest lower manufacturing cost, though pricing data is not available for this part. The RX 6650M's higher boost clock (2416 MHz versus 2407 MHz) is a trivial win that does not translate into real-world performance.

For users who need maximum frames, compute throughput, or ray tracing capability, the RTX 5090 is the only choice. For ultraportable systems where power efficiency and physical footprint are paramount, the RX 6650M offers a workable solution, but it sacrifices enormous performance to achieve that mobility. The benchmark data leaves no ambiguity: this is a matchup between a flagship desktop GPU and a mid-range mobile part, and the desktop GPU wins by every metric that matters.