AMD Radeon 660M vs NVIDIA GeForce GTX 780M Comparison

Head-to-Head Benchmarks

The recorded data shows a clear overall victory for the AMD Radeon 660M, which wins both of the direct comparison tests. In the Geekbench OpenCL benchmark, the AMD part scores 12,876 points against 12,769 for the NVIDIA GeForce GTX 780M, a narrow margin of 0.8%. This is essentially a statistical tie in raw compute throughput, with the AMD part holding a very slight edge. The Geekbench Vulkan test tells a different story, however: the Radeon 660M scores 14,748 points versus 12,696 for the GTX 780M, a decisive 16.2% advantage. This large delta in Vulkan performance reflects the architectural generational gap between the two, as the AMD part is built on a much newer design with modern API support.

Looking at the broader average benchmark scores, the AMD Radeon 660M sits at 13,812 points, while the GTX 780M averages 11,261 points. That places the AMD part roughly 22.7% higher on average across all recorded tests. The percentile rankings reinforce this: the Radeon 660M lands in the 55th percentile of all GPUs in the database, while the GTX 780M sits at the 50th percentile. The AMD part is not just faster in the head-to-head tests; it also outperforms the NVIDIA chip in overall database standing.

The nearest rivals for the Radeon 660M include the NVIDIA RTX A2000 Mobile at 13,821 points (just 0.1% higher), the AMD Radeon RX 570X at 13,871 points (0.4% higher), and the AMD Radeon RX 7900 XT at 13,745 points (0.5% lower). This shows the 660M sits in a tightly contested performance band around the 13,800-point mark, remarkably close to a desktop Radeon RX 570X and even the flagship RX 7900 XT in this specific synthetic metric. For the GTX 780M, its nearest rivals are the AMD Radeon Pro WX 3200 at 11,228 points (0.3% higher) and the AMD FirePro W4300 at 11,225 points (0.3% higher), placing it in a cluster around 11,200 to 11,300 points. Interestingly, the NVIDIA RTX PRO 6000 Blackwell Max-Q also appears as a rival at 11,088 points, 1.6% lower, which is an odd comparison given the massive difference in their respective market positions.

The Vulkan result is particularly telling. The Radeon 660M's 16.2% lead in that test likely stems from its modern RDNA 2.0 architecture with native Vulkan 1.4 support, compared to the GTX 780M's older Kepler design which tops out at Vulkan 1.2.175. In OpenCL, the two are nearly identical, suggesting that raw compute throughput is comparable despite the huge difference in shading unit counts, which we will examine later.

FAQ

Q: Which GPU wins in Geekbench OpenCL, and by how much?

A: The AMD Radeon 660M wins with a score of 12,876 against 12,769 for the NVIDIA GeForce GTX 780M, a margin of 0.8%.

Q: How large is the Vulkan performance gap between the two?

A: The AMD Radeon 660M scores 14,748 in Geekbench Vulkan, while the GTX 780M scores 12,696. That is a 16.2% advantage for the AMD part.

Q: What are the average benchmark scores for each GPU?

A: The AMD Radeon 660M has an average benchmark score of 13,812, while the NVIDIA GeForce GTX 780M averages 11,261.

Q: Where does each GPU rank among all GPUs in the database?

A: The AMD Radeon 660M is at the 55th percentile, while the NVIDIA GeForce GTX 780M is at the 50th percentile.

Q: Which GPU has more shading units?

A: The NVIDIA GeForce GTX 780M has 1,536 shading units, far more than the 384 shading units of the AMD Radeon 660M. Despite this, the AMD part still wins both benchmark tests.

Q: What is the memory configuration difference?

A: The GTX 780M has 4 GB of GDDR5 memory on a 256-bit bus with 160.0 GB/s bandwidth. The Radeon 660M uses system shared memory with system dependent bandwidth.

The Verdict

The data points to the AMD Radeon 660M as the superior choice for anyone comparing these two on raw benchmark performance. It wins both head-to-head tests, holds a 22.7% advantage in average score, and sits five percentile points higher in the overall GPU distribution. The Vulkan result alone, a 16.2% lead, is substantial enough to recommend the AMD part for modern workloads that leverage this API. The OpenCL test shows near parity, with the AMD part only 0.8% ahead, so for compute tasks that rely purely on OpenCL, the two are effectively interchangeable.

The GTX 780M, despite having four times the shading units and a much higher texture rate, cannot overcome the architectural disadvantages of its 2013-era Kepler design. Its 50th percentile ranking places it in the middle of the pack, while the Radeon 660M sits slightly above average at the 55th percentile. For a user prioritizing modern API support, higher average performance, and a larger performance headroom in Vulkan, the AMD Radeon 660M is the clear pick.

However, there is a caveat in the memory department. The GTX 780M offers 4 GB of dedicated GDDR5 memory with 160.0 GB/s of bandwidth, which can be advantageous in scenarios where dedicated memory bandwidth matters more than raw compute. The Radeon 660M relies on system shared memory, making its bandwidth system dependent. In real-world gaming with large textures, the dedicated memory of the GTX 780M could mitigate some of its compute deficit, but the benchmark data does not reflect any such advantage. On the measured tests, the AMD part wins.

For users who need dedicated memory and a proven mobile MXM module form factor, the GTX 780M retains some appeal. But based strictly on the recorded benchmarks, the AMD Radeon 660M is the faster, more modern part with a higher overall standing. The verdict is straightforward: the AMD Radeon 660M is the superior GPU in this comparison.

Specification Differences

The two GPUs diverge sharply in nearly every core specification. The AMD Radeon 660M is built on a 6 nm process at TSMC, while the NVIDIA GeForce GTX 780M uses a 28 nm process, also at TSMC. The transistor counts reflect this: the AMD chip packs 13,100 million transistors on a 208 mm² die, giving a transistor density of 63.0 million per square millimeter. The NVIDIA chip has only 3,540 million transistors on a larger 294 mm² die, for a density of 12.0 million per square millimeter. The AMD part is not only denser but also smaller in physical size.

Clock speeds differ substantially. The Radeon 660M runs at a base clock of 1500 MHz and boosts to 1900 MHz. The GTX 780M has a base clock of 771 MHz and a boost of 797 MHz. The AMD part operates at nearly double the frequency. Memory configurations are also distinct: the GTX 780M has 4 GB of GDDR5 on a 256-bit bus with 160.0 GB/s bandwidth, while the Radeon 660M uses system shared memory with system dependent bandwidth. The NVIDIA memory clock is listed as 1250 MHz with 5 Gbps effective.

The compute units differ in composition. The GTX 780M has 1,536 shading units, 128 texture mapping units, and 32 raster output pipelines. The Radeon 660M has 384 shading units, 24 TMUs, and 16 ROPs. Despite the far lower counts, the AMD part achieves a higher pixel rate of 30.40 GPixel/s versus 25.50 GPixel/s for the NVIDIA chip. However, the texture rate is reversed: the GTX 780M reaches 102.0 GTexel/s against 45.60 GTexel/s for the Radeon 660M. The FP32 compute is 2.448 TFLOPS for the NVIDIA part versus 1,459.2 GFLOPS for the AMD part. The Radeon 660M also features 6 ray tracing cores, which the GTX 780M lacks entirely.

Power consumption differs significantly. The Radeon 660M has a TDP of 40 W, while the GTX 780M draws 122 W. The AMD part is an IGP with no power connectors and a PCIe 4.0 x8 interface, while the NVIDIA part is an MXM module using MXM-B (3.0). Both have portable device dependent display outputs. The AMD part supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The NVIDIA part supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. Both are end-of-life products, with the Radeon 660M released on 2022-01-03 and the GTX 780M on 2013-05-10.

Architecture Differences

The architectural gap between these two GPUs is vast, spanning nearly a decade of development. The AMD Radeon 660M uses the RDNA 2.0 architecture on the Rembrandt chip, part of the Navi II IGP generation for mobile platforms. The NVIDIA GeForce GTX 780M uses the Kepler architecture on the GK104 chip, belonging to the GeForce 700M series. Kepler was designed for an era before ray tracing, variable rate shading, or mesh shaders, while RDNA 2.0 natively supports these modern features.

The Radeon 660M includes 6 ray tracing cores, a hardware feature entirely absent from the GTX 780M. This alone makes the AMD part capable of hardware-accelerated ray tracing, which the NVIDIA chip cannot do at all. The RDNA 2.0 architecture also supports DirectX 12 Ultimate with feature level 12_2, whereas the GTX 780M only reaches DirectX 12 (11_0). Vulkan support also differs: the AMD part supports Vulkan 1.4, while the NVIDIA part supports only 1.2.175. These API differences translate directly into the 16.2% Vulkan benchmark gap.

The manufacturing process is another fundamental difference. The 6 nm process for the AMD part allows for 63.0 million transistors per square millimeter, more than five times the density of the 28 nm process used for the GTX 780M. Even though the GTX 780M has a larger die at 294 mm², it contains only 3,540 million transistors, while the smaller 208 mm² AMD die packs 13,100 million transistors. This density advantage enables the AMD part to achieve higher clock speeds, lower power draw, and include modern features like ray tracing in a 40 W package versus 122 W for the NVIDIA chip.

The memory architecture also differs fundamentally. The GTX 780M uses dedicated GDDR5 memory with a 256-bit bus, a traditional discrete GPU design. The Radeon 660M uses system shared memory, meaning it relies on the host system's RAM with bandwidth that is system dependent. This is typical for integrated graphics, but it means the AMD part's memory performance is not fixed in the specification. The NVIDIA part's dedicated 160.0 GB/s bandwidth is a fixed quantity, which can be beneficial in memory-bound scenarios.

The shading unit count difference is striking: 1,536 for the GTX 780M versus 384 for the Radeon 660M. Yet the AMD part wins in pixel rate and overall benchmark scores. This is because RDNA 2.0 is a much more efficient architecture per shading unit, with higher clocks (1900 MHz boost versus 797 MHz) and better instruction-level efficiency. The FP32 throughput of the GTX 780M is higher at 2.448 TFLOPS versus 1,459.2 GFLOPS, but this does not translate to better real-world benchmark results, likely due to driver maturity, API overhead, and architectural inefficiencies in Kepler. The GTX 780M's texture rate is also higher at 102.0 GTexel/s versus 45.60 GTexel/s, yet this advantage does not show up in the recorded scores. The Kepler architecture, while powerful on paper with its many cores, is clearly limited by its older design and lack of modern feature support.