AMD Radeon 660M vs NVIDIA GeForce GTX 950 Comparison

The AMD Radeon 660M and NVIDIA GeForce GTX 950 represent two very different approaches to graphics processing, separated by nearly seven years of architectural evolution. The data shows a clear overall winner in raw compute benchmarks, but the comparison is far more nuanced when considering the platform context, feature sets, and the specific workloads where each GPU excels. The benchmark results position the GTX 950 ahead in both shared tests, yet the 660M's modern feature set and integrated design tell a compelling story for specific use cases.

Head-to-Head Benchmarks

The head-to-head data is unambiguous in its outcome: the NVIDIA GeForce GTX 950 wins both benchmark tests included in the comparison. In the Geekbench OpenCL test, the GTX 950 scores 15,662 against the Radeon 660M's 12,876, representing a 17.8% advantage for the NVIDIA part. This is a substantial margin in a compute-heavy workload, indicating that the Maxwell 2.0 architecture's larger shading unit count — 768 versus 384 — provides a significant throughput advantage in raw parallel compute tasks.

The Vulkan test tells a similar story, though with a narrower margin. The GTX 950 scores 16,734 while the Radeon 660M achieves 14,748, a difference of 11.9% in favor of the NVIDIA GPU. The smaller gap in Vulkan compared to OpenCL suggests that the Radeon 660M's RDNA 2.0 architecture scales relatively better in modern, lower-overhead graphics APIs. This is consistent with the architectural generational leap between the two parts, where the newer RDNA 2.0 design benefits from more efficient command processing and draw call handling.

Looking at the broader benchmark landscape, the GTX 950's average benchmark score of 13,189 places it in the 53rd percentile of all GPUs, while the Radeon 660M's average of 13,812 places it in the 55th percentile. This is a curious inversion: the GTX 950 wins both individual head-to-head tests, yet the Radeon 660M has a higher average score across the full benchmark suite and a better percentile ranking. This discrepancy suggests that the Radeon 660M performs relatively better in other benchmark categories not captured in the head-to-head comparison, likely benefiting from its newer architecture in specific compute patterns.

The nearest rivals for each GPU provide additional context. The Radeon 660M sits within 0.5% of the AMD Radeon RX 7900 XT in average score, a remarkable result given the massive difference in market positioning. It also trails the NVIDIA RTX A2000 Mobile by just 0.1% and the AMD Radeon RX 570X by 0.4%. The GTX 950, meanwhile, falls 0.8% behind the NVIDIA GeForce GTX 480 and sits 0.9% ahead of the NVIDIA Tesla M2090, placing it in a similar performance tier to decade-old high-end parts.

Where Each One Wins

The GTX 950's victory in both head-to-head benchmarks establishes its dominance in raw compute performance, particularly in OpenCL workloads where its 17.8% lead is most pronounced. This advantage stems from its substantially higher shading unit count (768 versus 384) and texture mapping units (48 versus 24), which directly translate to higher fill rates. The GTX 950 delivers 57.02 GTexel/s in texture throughput and 38.02 GPixel/s in pixel rate, compared to 45.60 GTexel/s and 30.40 GPixel/s respectively for the Radeon 660M.

However, the Radeon 660M demonstrates its strengths in areas beyond the measured benchmarks. Its RDNA 2.0 architecture supports DirectX 12 Ultimate (12_2), while the GTX 950 is limited to DirectX 12 (12_1). This means the 660M has access to hardware-accelerated ray tracing through its 6 dedicated ray tracing cores, a feature entirely absent from the Maxwell 2.0-based GTX 950. The 660M also offers superior FP16 performance at 2.918 TFLOPS with a 2:1 ratio, whereas the GTX 950 has no listed FP16 capability, making the AMD part significantly better suited for workloads that leverage half-precision arithmetic.

The platform context further differentiates the two. The Radeon 660M is an integrated graphics processor with a 40W TDP, requiring no power connectors and fitting into any system with a compatible AMD Rembrandt mobile processor. The GTX 950 is a dual-slot discrete card with a 90W TDP requiring a 1x 6-pin power connector and a suggested 250W power supply. For lightweight or portable systems where power efficiency and physical space are paramount, the Radeon 660M's integrated design is a decisive advantage.

The Verdict

The data supports a clear performance winner: the NVIDIA GeForce GTX 950 outperforms the AMD Radeon 660M in both measured benchmarks, with an 11.9% lead in Vulkan and a 17.8% lead in OpenCL. For users whose primary concern is raw compute throughput in these workloads, the GTX 950 is the superior choice based strictly on the benchmark scores.

However, the verdict becomes more nuanced when considering the full picture. The Radeon 660M holds a higher average benchmark score (13,812 versus 13,189) and a better percentile ranking (55th versus 53rd), indicating superior performance across the broader benchmark suite beyond the two head-to-head tests. This suggests the 660M may be the better all-around performer in mixed workloads, even if it loses the specific tests measured head-to-head.

Users who prioritize modern features should favor the Radeon 660M. Its DirectX 12 Ultimate support, hardware ray tracing, and FP16 capabilities are absent from the GTX 950. Additionally, the 660M's integrated nature with a 40W TDP makes it the only practical choice for compact, power-constrained systems, while the GTX 950 requires a dedicated slot, a 6-pin power connector, and a 250W power supply. The production status of both is end-of-life, but the Radeon 660M's 2022 release date makes it a more recent design with architectural advantages that may matter for future software.

FAQ

Q: Which GPU performs better in Geekbench OpenCL?

A: The NVIDIA GeForce GTX 950 scores 15,662 compared to the AMD Radeon 660M's 12,876, giving the GTX 950 a 17.8% advantage in this test.

Q: What is the difference in Vulkan benchmark scores?

A: The GTX 950 scores 16,734 while the Radeon 660M scores 14,748, resulting in an 11.9% lead for the NVIDIA part.

Q: Which GPU has a higher average benchmark score?

A: The AMD Radeon 660M has an average benchmark score of 13,812, which is higher than the GTX 950's 13,189, despite losing both head-to-head tests.

Q: Does the Radeon 660M support ray tracing?

A: Yes, the Radeon 660M has 6 dedicated ray tracing cores and supports DirectX 12 Ultimate (12_2), while the GTX 950 does not support ray tracing and is limited to DirectX 12 (12_1).

Q: What are the power requirements for each GPU?

A: The Radeon 660M has a 40W TDP and requires no power connectors, while the GTX 950 has a 90W TDP, requires a 1x 6-pin power connector, and has a suggested power supply of 250W.

Q: How do their memory configurations differ?

A: The GTX 950 has 2GB of GDDR5 memory on a 128-bit bus with 105.8 GB/s bandwidth, while the Radeon 660M uses system shared memory with bus width and bandwidth described as system dependent.

Architecture Differences

The architectural gap between these two GPUs is substantial, reflecting nearly seven years of semiconductor evolution. The Radeon 660M is built on TSMC's 6nm process node, while the GTX 950 uses a 28nm process from the same foundry. This process difference is stark: the 660M packs 13,100 million transistors into a 208 mm² die, achieving a transistor density of 63.0 million per square millimeter. The GTX 950, by contrast, contains only 2,940 million transistors on a larger 228 mm² die, with a density of just 12.9 million per square millimeter.

The Radeon 660M employs the RDNA 2.0 architecture from AMD's Rembrandt chip, placing it in the Navi II IGP generation. The GTX 950 uses NVIDIA's Maxwell 2.0 architecture on the GM206 chip, belonging to the GeForce 900 generation. These are fundamentally different design philosophies: RDNA 2.0 is a modern, efficiency-focused architecture with support for advanced features like hardware ray tracing, while Maxwell 2.0 was designed for the discrete GPU market of 2015 with a focus on raw throughput.

The 660M features 384 shading units, 24 texture mapping units, and 16 render output units, complemented by 6 dedicated ray tracing cores. The GTX 950 doubles the shading units to 768 and TMUs to 48, with 32 ROPs, but has no ray tracing cores or tensor cores. The 660M's FP32 performance is 1,459.2 GFLOPS, while the GTX 950 achieves 1.825 TFLOPS, and the 660M additionally provides 2.918 TFLOPS of FP16 performance through a 2:1 ratio, a capability the GTX 950 lacks entirely.

Specification Differences

The clock speeds reveal the efficiency advantage of the newer process. The Radeon 660M operates at a 1500 MHz base clock and 1900 MHz boost clock, significantly higher than the GTX 950's 1024 MHz base and 1188 MHz boost. This higher clock speed on the 660M partially compensates for its lower core count, though not enough to overcome the GTX 950's raw throughput advantage in the measured benchmarks.

Memory configurations are fundamentally different. The GTX 950 uses 2GB of GDDR5 memory with a 128-bit bus, delivering 105.8 GB/s of bandwidth. The Radeon 660M relies on system shared memory, with its size, type, and bus width all listed as "System Shared" and bandwidth described as "System Dependent." This makes the 660M's memory performance entirely dependent on the host system's RAM configuration, while the GTX 950 has dedicated, predictable memory bandwidth.

The interface and physical specifications also diverge. The Radeon 660M uses a PCIe 4.0 x8 interface and is an integrated GPU with no slot width and no power connectors. The GTX 950 uses PCIe 3.0 x16, is a dual-slot card measuring 202 mm in length, requires a single 6-pin power connector, and has a suggested power supply of 250W. The 660M's display outputs are portable device dependent, while the GTX 950 offers 1x DVI, 1x HDMI 2.0, and 3x DisplayPort 1.2 outputs. Both support DirectX 12, OpenGL 4.6, and Vulkan 1.4, though the 660M's DirectX 12 Ultimate (12_2) support exceeds the GTX 950's DirectX 12 (12_1) level.