GPU Comparison

The data presents a dramatic generational clash: the NVIDIA GeForce GTX 660, a Kepler-era card from 2012, facing the Intel Arc A380, a modern Xe-HPG design from 2022. The benchmark results show a clear shift in compute capability, but the underlying architectural differences tell a story that goes beyond simple performance numbers. This analysis examines where each card excels and what that means for a potential user.

Head-to-Head Benchmarks

The available head-to-head results are decisive and one-sided. In the Geekbench OpenCL test, the Intel Arc A380 scores 38,224, which is a massive 70.3% higher than the GTX 660’s 11,347. This is not a marginal improvement; it represents a fundamental leap in raw compute throughput. The gap is nearly identical in the Vulkan test, where the Arc A380 scores 36,736 against the GTX 660’s 11,415, a 68.9% advantage. The Intel card dominates both API workloads, leaving the older NVIDIA card far behind.

These results are consistent with the average benchmark scores for each card. The GTX 660 has an average benchmark score of 9,022, placing it in the 45th percentile of all GPUs. The Arc A380’s average score of 8,558 puts it in the 44th percentile. The head-to-head results are so lopsided because they are compute-heavy tests, which heavily favor the Arc A380’s modern architecture and higher FP32 throughput. The GTX 660’s nearest rivals, like the NVIDIA TITAN V CEO Edition with a score of 9,037, show it is clustered with other older or lower-tier cards; the Arc A380’s closest competitor, the AMD FirePro W5170M, scores 8,595, illustrating that the Intel card’s average is held back by its weaker driver optimization in legacy APIs, not by a lack of hardware capability.

Architecture Differences

The architectural chasm between these two cards is enormous, explaining the benchmark results. The GTX 660 is built on a 28 nm process at TSMC, packing 2,540 million transistors into a 221 mm² die. The Intel Arc A380, in contrast, uses a modern 6 nm TSMC process, fitting 7,200 million transistors into a smaller 157 mm² die. The transistor density tells the story: the Arc A380 has 45.9 million transistors per mm², compared to the GTX 660’s 11.5 million per mm². This allows Intel to deliver vastly more compute hardware in a smaller physical footprint.

The compute resources differ significantly. The GTX 660 has 960 shading units, 80 texture mapping units (TMUs), and 24 ROPs. The Arc A380 has 1,024 shading units, only 64 TMUs, but a higher count of 32 ROPs. The Intel card also includes 8 dedicated ray tracing cores, a feature entirely absent from the Kepler-based GTX 660. The FP32 performance is a clear indicator of the generational leap: the Arc A380 offers 4.198 TFLOPS, more than double the GTX 660’s 1.981 TFLOPS. The Intel card also supports FP16 compute at 8.397 TFLOPS, a capability the GTX 660 lacks entirely.

Memory architecture also diverges. The GTX 660 uses 2 GB of GDDR5 on a 192-bit bus, providing 144.2 GB/s of bandwidth. The Arc A380 comes with 6 GB of GDDR6 on a narrower 96-bit bus, yet still achieves higher bandwidth at 186.0 GB/s due to faster memory speeds (15.5 Gbps effective vs. 6 Gbps effective). The Intel card also supports modern display outputs, including 3x DisplayPort 2.0 and HDMI 2.1, whereas the GTX 660 is limited to 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2. The API support is a further divider: the Arc A380 supports DirectX 12 Ultimate (12_2) and Vulkan 1.4, while the GTX 660 is capped at DirectX 12 (11_0) and Vulkan 1.2.175.

The Verdict

The data makes a straightforward case for the Intel Arc A380. It wins the only two head-to-head benchmarks available, and it does so by a commanding margin (70.3% in OpenCL and 68.9% in Vulkan). Its architectural advantages, a modern 6 nm node, double the FP32 throughput, ray tracing support, and triple the memory capacity, are overwhelming. The GTX 660 is a product of a different era, and the benchmarks show it cannot compete in modern compute workloads. The only area where the GTX 660 has a numerical advantage is in its power draw, listed at 140 W, though the Arc A380’s 75 W TDP is actually lower, suggesting better efficiency. The GTX 660’s higher power requirement with vastly inferior performance makes the Arc A380 the clear choice for any application requiring current-generation compute or rendering features.

FAQ

Q: Which card is faster in compute-heavy workloads?

A: The Intel Arc A380 is significantly faster, winning the Geekbench OpenCL test with a score of 38,224 versus the GTX 660’s 11,347, a 70.3% difference. It also wins the Vulkan test with 36,736 points against 11,415, a 68.9% advantage.

Q: Does the GTX 660 have any ray tracing capability?

A: No. The GTX 660 has no ray tracing cores, while the Intel Arc A380 includes 8 dedicated ray tracing cores.

Q: How do the memory capacities compare?

A: The Intel Arc A380 has 6 GB of GDDR6 memory, which is three times the 2 GB of GDDR5 found on the GTX 660.

Q: What is the difference in their average benchmark scores?

A: The GTX 660 has an average benchmark score of 9,022, placing it in the 45th percentile. The Arc A380 has an average score of 8,558, placing it in the 44th percentile, despite winning the head-to-head tests.

Q: Which card supports newer display interfaces?

A: The Intel Arc A380 supports 3x DisplayPort 2.0 and 1x HDMI 2.1. The NVIDIA GeForce GTX 660 is limited to 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2.

Q: What is the DirectX feature level of each card?

A: The Intel Arc A380 supports DirectX 12 Ultimate (12_2), while the NVIDIA GeForce GTX 660 only supports DirectX 12 (11_0).

Where Each One Wins

The Intel Arc A380 is the winner in all measured compute and rendering scenarios. The data shows it dominates in OpenCL and Vulkan workloads, which are common in modern games and GPU-accelerated applications. Its higher FP32 throughput (4.198 TFLOPS vs 1.981 TFLOPS) and support for ray tracing make it suitable for current-generation gaming features. The 6 GB memory buffer also provides more headroom for high-resolution textures and larger datasets. The GTX 660 has no benchmark wins in this comparison. Its only potential advantage lies in its legacy API support for older DirectX 11 titles, but even then, the Arc A380 has a Passmark DirectX 11 score of 38, a metric that is likely more indicative of driver overhead than raw capability. The GTX 660’s lower FP32 performance and lack of modern features mean it cannot claim victory in any compute-heavy task.

Specification Differences

The specifications where the two cards differ are extensive and define their respective capabilities.

| Specification | NVIDIA GeForce GTX 660 | Intel Arc A380 |

| :--- | :--- | :--- |

| Process Node | 28 nm | 6 nm |

| Transistors | 2,540 million | 7,200 million |

| Die Size | 221 mm² | 157 mm² |

| Transistor Density | 11.5M / mm² | 45.9M / mm² |

| Base Clock | 980 MHz | 2000 MHz |

| Boost Clock | 1032 MHz | 2050 MHz |

| Memory Clock | 1502 MHz (6 Gbps effective) | 1937 MHz (15.5 Gbps effective) |

| Memory Size | 2 GB | 6 GB |

| Memory Type | GDDR5 | GDDR6 |

| Memory Bus Width | 192 bit | 96 bit |

| Memory Bandwidth | 144.2 GB/s | 186.0 GB/s |

| Shading Units | 960 | 1024 |

| TMUs | 80 | 64 |

| ROPs | 24 | 32 |

| RT Cores | 0 | 8 |

| Pixel Rate | 20.64 GPixel/s | 65.60 GPixel/s |

| Texture Rate | 82.56 GTexel/s | 131.2 GTexel/s |

| FP32 Performance | 1.981 TFLOPS | 4.198 TFLOPS |

| FP16 Performance | N/A | 8.397 TFLOPS (2:1) |

| TDP | 140 W | 75 W |

| Power Connectors | 1x 6-pin | 1x 8-pin |

| Suggested PSU | 300 W | 250 W |

| Bus Interface | PCIe 3.0 x16 | PCIe 4.0 x8 |

| Display Outputs | 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2 | 1x HDMI 2.1, 3x DisplayPort 2.0 |

| DirectX Support | 12 (11_0) | 12 Ultimate (12_2) |

| Vulkan Support | 1.2.175 | 1.4 |