AMD Radeon 660M

The AMD Radeon 660M is an integrated graphics processor that delivers desktop-class performance in a 40-watt package, with benchmark data showing it sits in the 54th percentile of all GPUs. Its average benchmark score of 13,812 places it in a dead heat with the NVIDIA GeForce GTX 680, a discrete card from a previous generation, indicating that modern integrated graphics have closed a significant gap. The 660M is an end-of-life product from the Rembrandt generation, built on TSMC's 6 nm process with RDNA 2.0 architecture, and it offers a compelling combination of features and performance for portable devices.

Benchmark Performance

The data shows the Radeon 660M achieving an average benchmark score of 13,812, which places it in the 54th percentile of all GPUs. This is a strong result for an integrated processor, as it outperforms roughly half of all discrete graphics cards ever benchmarked. In the individual tests, the 660M scores 12,876 in Geekbench OpenCL and 14,748 in Geekbench Vulkan, showing a noticeable advantage in the Vulkan API which suggests the architecture scales well with modern, lower-overhead graphics interfaces.

When compared directly to its nearest rivals, the 660M's performance is remarkably balanced. It shows a 0% delta against the NVIDIA GeForce GTX 680, meaning the two are exactly matched in average score. Against the NVIDIA RTX A2000 Mobile, the 660M trails by only 0.1%, a margin that is effectively imperceptible in real-world usage. The AMD Radeon RX 570X is 0.4% faster, again a negligible difference. The only rival where the 660M holds a clear lead is the NVIDIA P106-090, which the 660M beats by 0.7%. These sub-1% deltas indicate that the 660M performs in a very tight performance band, and any of these GPUs would deliver nearly identical frame rates in most scenarios.

The FP32 compute throughput of 1,459.2 GFLOPS, combined with a texture rate of 45.60 GTexel/s and a pixel rate of 30.40 GPixel/s, supports these benchmark results. The 384 shading units and 24 texture mapping units provide enough raw processing power to handle modern game engines at reduced settings, while the 16 raster operation units keep pixel throughput competitive with entry-level discrete solutions.

Ray Tracing and Feature Set

The Radeon 660M includes 6 dedicated ray tracing cores, making it one of the few integrated GPUs capable of hardware-accelerated ray tracing. This is a significant feature for an IGP, as it allows portable devices to run games with ray-traced effects enabled, albeit at lower resolutions and with performance compromises. The ray tracing cores work in conjunction with the RDNA 2.0 architecture to accelerate the BVH traversal and ray-intersection calculations required for real-time ray tracing.

The feature set is further bolstered by support for DirectX 12 Ultimate (12_2), which is the most recent DirectX feature level. This enables the 660M to support advanced rendering techniques such as variable rate shading, mesh shaders, and sampler feedback, all of which are part of the DirectX 12 Ultimate specification. The GPU also supports Vulkan 1.4 and OpenGL 4.6, ensuring broad compatibility with modern and legacy applications alike. Notably, the 660M lacks tensor cores, meaning it has no dedicated hardware for AI acceleration or deep learning super sampling; any such features would rely on the general-purpose shading units.

The memory interface is system shared, meaning the GPU accesses main system memory rather than having its own dedicated VRAM. This design choice is typical for integrated processors and keeps the overall system cost and power consumption lower, but it does mean that memory performance is system dependent. The bus interface is PCIe 4.0 x8, which provides sufficient bandwidth for most integrated graphics workloads.

Power and Cooling

The Radeon 660M has a thermal design power of 40 watts, which is remarkably low for the level of performance it delivers. This low TDP makes it suitable for thin and light laptops where cooling is limited and battery life is a priority. The GPU requires no power connectors, drawing all its power through the motherboard, and the slot width is listed as IGP, confirming its integrated nature.

Because of this low power draw, there is no suggested PSU recommendation in the data, as the 660M is not intended for desktop use where a separate power supply would be required. The cooling solution is likewise portable device dependent, meaning the thermal solution is designed by the laptop manufacturer to fit their specific chassis. The 6 nm process node from TSMC contributes to the efficiency, allowing 13,100 million transistors on a 208 mm² die, resulting in a transistor density of 63.0 million transistors per square millimeter. This high density enables the 660M to pack a substantial amount of compute hardware into a small, power-efficient package.

How It Compares

vs. NVIDIA GeForce GTX 680: The 660M matches the GTX 680 exactly with a 0% delta in average score. This is a remarkable achievement for an integrated GPU, as the GTX 680 was a flagship discrete card in its time. The 660M achieves this parity while consuming a fraction of the power and taking up no expansion slot space, making it a superior choice for portable systems.

vs. NVIDIA RTX A2000 Mobile: The 660M trails the RTX A2000 Mobile by just 0.1%, putting the two within a hair's breadth of each other. The RTX A2000 Mobile is a professional mobile workstation GPU, so this near-parity suggests the 660M can handle professional workloads almost as well, though the 660M lacks the tensor cores that the A2000 Mobile likely uses for AI-accelerated tasks.

vs. AMD Radeon RX 570X: The RX 570X is 0.4% faster than the 660M. This is a negligible difference that would not be noticeable in any real-world application. The RX 570X is a discrete desktop card, so the fact that the 660M can nearly match it speaks to the efficiency of the RDNA 2.0 architecture in integrated form.

vs. NVIDIA P106-090: The 660M holds a 0.7% lead over the P106-090, a mining-oriented GPU with limited display outputs. This is the largest performance delta among the nearest rivals, but it is still under 1%. The 660M's advantage here is more about its full feature set and driver support than raw performance.

Memory Subsystem

The Radeon 660M uses system shared memory, meaning its VRAM size, type, and bus width are all listed as "System Shared." This is a fundamental design characteristic of integrated graphics, where the GPU dynamically allocates a portion of the system's main memory for graphics workloads. The bandwidth is listed as "System Dependent," which means the achievable memory bandwidth depends entirely on the system's memory configuration, including whether dual-channel memory is used and the speed of the RAM.

This memory architecture has significant implications for high-resolution gaming. At 1080p and above, the shared memory interface can become a bottleneck, as the GPU must compete with the CPU for memory bandwidth. However, the PCIe 4.0 x8 bus interface helps mitigate some of these concerns by providing a fast connection between the GPU and the memory controller. For users with fast dual-channel memory, the 660M can achieve adequate bandwidth for 1080p gaming, but at 1440p or higher, the shared memory subsystem will likely limit performance more than the GPU's compute capabilities.

Who Should Consider It

Based on the benchmark data, the Radeon 660M is well-suited for users who primarily play games at 1080p with medium to low settings. Its score of 14,748 in Geekbench Vulkan suggests it handles modern APIs well, and the 54th percentile ranking indicates it will run most titles at playable frame rates. The 6 ray tracing cores enable some ray-traced effects, but users should expect significant performance hits when enabling them, so ray tracing is best reserved for lighter titles or lower resolutions.

The 660M is an excellent choice for users who need a portable device that can handle light gaming, media consumption, and productivity tasks without requiring a discrete GPU. Its 40-watt TDP means it can be passively cooled in some chassis or paired with a quiet fan solution, making it ideal for ultrabooks and thin-and-light laptops. However, users who demand high-refresh-rate gaming at 1440p or above should look elsewhere, as the shared memory subsystem and 1,459.2 GFLOPS of FP32 performance will struggle at those resolutions. For the budget-conscious gamer who prioritizes portability and efficiency over maximum frame rates, the 660M delivers a balanced experience that punches well above its integrated weight class.

FAQ

Q: How does the Radeon 660M compare to the NVIDIA GeForce GTX 680?

A: The 660M matches the GTX 680 exactly in average benchmark score, with a 0% delta. Both GPUs achieve an average score of 13,812, indicating identical overall performance.

Q: Does the Radeon 660M support ray tracing?

A: Yes, the 660M includes 6 dedicated ray tracing cores based on the RDNA 2.0 architecture, which provides hardware acceleration for ray-traced effects in supported games.

Q: What is the power consumption of the Radeon 660M?

A: The 660M has a TDP of 40 watts and requires no power connectors, drawing all power through the motherboard. There is no suggested PSU because it is an integrated GPU.

Q: What API features does the Radeon 660M support?

A: It supports DirectX 12 Ultimate (12_2), Vulkan 1.4, and OpenGL 4.6, enabling modern rendering techniques like variable rate shading and mesh shaders.

Q: How much VRAM does the Radeon 660M have?

A: The 660M has no dedicated VRAM; it uses system shared memory, with the size, type, and bus width all dependent on the host system's configuration.

Q: Is the Radeon 660M suitable for 1440p gaming?

A: The benchmark data indicates it performs in the 54th percentile of all GPUs, but the shared memory subsystem will likely bottleneck at higher resolutions. It is best suited for 1080p gaming at medium to low settings.