AMD Radeon RX 6650M vs NVIDIA GeForce RTX 5090 D Comparison

The NVIDIA GeForce RTX 5090 D and AMD Radeon RX 6650M occupy opposite ends of the GPU spectrum, yet both sit within the 91st–92nd percentile of all GPUs benchmarked in this database. The RTX 5090 D, a desktop flagship built on Blackwell 2.0, delivers an average benchmark score of 77,712, while the RX 6650M, a mobile RDNA 2.0 part, posts 71,768. These scores place them only 8.3% apart on average, but the underlying benchmark data reveals a far wider gulf in raw compute, memory bandwidth, and architectural capability. The head-to-head results show the RTX 5090 D leading by 372.1% in Geekbench OpenCL and 384.9% in Geekbench Vulkan, making the comparison less about competition and more about contrasting design philosophies.

Where Each One Wins

The RTX 5090 D wins decisively in every benchmark where both GPUs have recorded scores. In Geekbench OpenCL, the NVIDIA part scores 310,674 against the AMD’s 65,800 — a 372.1% advantage. In Geekbench Vulkan, the margin widens to 384.9%, with scores of 376,915 versus 77,735. These are the only two tests shared by both products, so the data shows a complete sweep for the RTX 5090 D. The RX 6650M has no benchmark wins in any recorded test.

The RTX 5090 D’s win profile extends beyond the head-to-head list. Its standalone benchmarks include 3DMark Steel Nomad DX12 at 14,326, Passmark G3D at 44,065, and Passmark GPU Compute at 28,396. The RX 6650M has no entries for these tests, so no direct comparison exists, but the scale of the NVIDIA scores — particularly the Passmark G3D figure — indicates a GPU designed for high-throughput desktop workloads. The RX 6650M’s percentile of 91 versus the RTX 5090 D’s 92 shows that despite the score gap, the AMD part still outperforms roughly nine out of ten GPUs in the database, making it a strong mobile option rather than a weak one.

The use-case split is clear from the data: the RTX 5090 D is for compute-heavy, high-resolution, and multi-tasking scenarios where its 32 GB of GDDR7 memory and 1.79 TB/s bandwidth can be fully utilized. The RX 6650M targets portable systems where its 120 W TDP and IGP form factor fit into laptops, trading raw performance for efficiency and physical integration.

Architecture Differences

The architectural gap between these two GPUs is enormous. The RTX 5090 D uses the GB202 chip on TSMC’s 5 nm process, packing 92,200 million transistors into a 750 mm² die — a transistor density of 122.9 million per square millimeter. The RX 6650M uses the Navi 23 chip on TSMC’s 7 nm process, with 11,060 million transistors on a 237 mm² die, yielding 46.7 million per square millimeter. The NVIDIA chip has over eight times the transistor count and more than three times the die area.

Memory subsystems differ fundamentally. The RTX 5090 D features 32 GB of GDDR7 on a 512-bit bus, delivering 1.79 TB/s of bandwidth. The RX 6650M has 8 GB of GDDR6 on a 128-bit bus, providing 224.0 GB/s — roughly one-eighth the bandwidth. Clock speeds are similar in boost terms (2407 MHz for NVIDIA, 2416 MHz for AMD), but the NVIDIA part’s base clock is 2017 MHz versus the AMD’s 2068 MHz, and the AMD has a game clock of 2222 MHz that has no NVIDIA equivalent.

Compute resources are starkly different. The RTX 5090 D 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 listed. This translates to 104.8 TFLOPS FP32 for NVIDIA versus 8.659 TFLOPS FP32 for AMD — a 12.1x difference. The NVIDIA part also achieves 104.8 TFLOPS FP16 (1:1), while the AMD part reaches 17.32 TFLOPS FP16 (2:1), showing that the NVIDIA architecture dedicates equal throughput to FP16 without the rate split.

Pixel and texture rates follow the same pattern: the RTX 5090 D outputs 423.6 GPixel/s and 1,636.8 GTexel/s, while the RX 6650M manages 154.6 GPixel/s and 270.6 GTexel/s. Power and physical design diverge completely — the RTX 5090 D is a dual-slot desktop card with a 575 W TDP, a 16-pin connector, a 950 W suggested PSU, and dimensions of 304 mm by 137 mm by 48 mm. The RX 6650M is an IGP (integrated graphics processor) with a 120 W TDP, no power connectors, no length/height/width data, and portable-device-dependent display outputs.

The Verdict

The data dictates a clear verdict: the RTX 5090 D is the superior GPU by every measurable metric in this FACT PACK. Its 372.1% lead in Geekbench OpenCL and 384.9% lead in Geekbench Vulkan are not incremental improvements but categorical differences. The NVIDIA part also holds a 104.8 TFLOPS FP32 rating versus 8.659 TFLOPS, a 1.79 TB/s memory bandwidth versus 224.0 GB/s, and 32 GB versus 8 GB of VRAM. For any workload that leverages these resources — high-end gaming, rendering, AI inference, or scientific compute — the RTX 5090 D is the only rational choice from this data.

The RX 6650M is not without merit, but its merits are contextual. It sits at the 91st percentile of all GPUs, meaning it outperforms most hardware in the database despite its modest specifications. Its 120 W TDP and IGP form factor make it suitable for laptops where the RTX 5090 D’s 575 W TDP and dual-slot design are physically impossible. The AMD part’s 8 GB of GDDR6 memory and 224.0 GB/s bandwidth are adequate for 1080p gaming and light compute. However, the RX 6650M’s production status is end-of-life, while the RTX 5090 D is active, and the NVIDIA card’s launch MSRP is 2,299 USD.

The practical verdict is that the RTX 5090 D targets desktop users who need maximum throughput and are willing to accommodate a 950 W PSU. The RX 6650M targets mobile users who need a capable GPU that fits into a laptop chassis. Neither GPU competes in the other’s domain. If forced to choose based purely on benchmark performance, the RTX 5090 D wins outright; if forced to choose based on physical integration, the RX 6650M is the only option.

FAQ

Q: How much faster is the RTX 5090 D in Geekbench OpenCL compared to the RX 6650M?

A: The RTX 5090 D scores 310,674 while the RX 6650M scores 65,800, resulting in a 372.1% advantage for the NVIDIA GPU.

Q: What is the memory bandwidth difference between the two GPUs?

A: The RTX 5090 D has 1.79 TB/s of bandwidth from 32 GB of GDDR7 on a 512-bit bus, while the RX 6650M has 224.0 GB/s from 8 GB of GDDR6 on a 128-bit bus — an 8x difference.

Q: Do both GPUs support the same API versions?

A: Yes, both support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, according to the FACT PACK.

Q: Which GPU has more shading units?

A: The RTX 5090 D has 21,760 shading units, while the RX 6650M has 1,792 — a 12.1x difference that aligns with the FP32 TFLOPS gap.

Q: What is the power consumption difference?

A: The RTX 5090 D has a TDP of 575 W and requires a 950 W suggested PSU, whereas the RX 6650M has a TDP of 120 W and no suggested PSU listed.

Q: Is the RX 6650M still in production?

A: No, the RX 6650M has end-of-life production status, while the RTX 5090 D is listed as active.

Head-to-Head Benchmarks

The head-to-head data contains only two shared tests, and both are decisive NVIDIA wins. In Geekbench OpenCL, the RTX 5090 D scores 310,674 against the RX 6650M’s 65,800. This 372.1% delta is consistent with the FP32 compute difference — 104.8 TFLOPS versus 8.659 TFLOPS — and the memory bandwidth gap of 1.79 TB/s versus 224.0 GB/s. The OpenCL workload exercises both compute and memory subsystems, so the NVIDIA part’s advantages in both areas compound into the observed score.

In Geekbench Vulkan, the RTX 5090 D scores 376,915 versus 77,735 for the RX 6650M, a 384.9% delta. Vulkan is a lower-level API that can expose hardware differences more directly than OpenCL. The NVIDIA GPU’s 170 RT cores and 680 tensor cores, which have no direct equivalents in the AMD part (28 RT cores, no tensor cores), likely contribute to this larger margin. The NVIDIA GPU also has a higher texture rate (1,636.8 GTexel/s versus 270.6 GTexel/s) and pixel rate (423.6 GPixel/s versus 154.6 GPixel/s), which matter in graphics-heavy Vulkan workloads.

The RTX 5090 D’s standalone benchmarks provide additional context even though no RX 6650M scores exist for comparison. The 3DMark Steel Nomad DX12 score of 14,326, Passmark G3D of 44,065, and Passmark GPU Compute of 28,396 all indicate a GPU that handles both rasterization and compute tasks at high throughput. The RX 6650M’s absence from these tests means its performance in those specific workloads cannot be quantified here, but given its 8.3% lower average score, it would likely trail significantly.

The wins are not close. The smallest delta in the shared tests is 372.1%, which is not a marginal edge but a generational and class-based gap. The RTX 5090 D’s nearest rivals in the database — AMD Radeon RX 6850M XT at -1.6% and NVIDIA Tesla P100 PCIe 16 GB at -2.4% — are all within 3% of its average score, indicating that its real competition sits elsewhere. The RX 6650M’s nearest rivals — NVIDIA TITAN X Pascal at -0.5% and AMD Radeon Pro Vega 64 at -0.8% — are similarly close, showing that both GPUs perform near the top of their respective peer groups. But those peer groups are separated by roughly 6,000 points of average benchmark score, and the head-to-head deltas confirm that no benchmark in the shared set favors the AMD part.