AMD Radeon 660M vs AMD Radeon RX 570X Comparison

The AMD Radeon RX 570X and the AMD Radeon 660M occupy nearly identical positions in the overall performance hierarchy, yet they achieve that status through radically different designs. The data places both at the 55th percentile among all GPUs, with average benchmark scores of 13,871 and 13,812 respectively—a razor-thin 0.4% margin that puts the RX 570X nominally ahead. This near-parity is remarkable given that the 660M is an integrated graphics processor while the RX 570X is a dedicated add-in board from an earlier generation. The single head-to-head benchmark available, Geekbench OpenCL, tells a completely different story, however, with the RX 570X delivering 38,939 points against the 660M’s 12,876—a 202.4% advantage that seems to contradict the aggregate scores. The explanation lies in the other benchmarks each card reports, suggesting that the OpenCL test heavily favors the discrete card’s dedicated memory and raw throughput, while the 660M’s overall standing is buoyed by its Vulkan performance.

Head-to-Head Benchmarks

The only direct comparison in the data is the Geekbench OpenCL test, and it is a landslide. The RX 570X scores 38,939, which is more than triple the 660M’s 12,876. That 202.4% delta is the largest performance gap recorded in this comparison, and it highlights a fundamental difference in how each GPU handles compute workloads. The RX 570X’s dedicated 8 GB of GDDR5 memory with 224.0 GB/s of bandwidth provides a massive advantage in OpenCL’s memory-intensive operations, whereas the 660M relies on system-shared memory with bandwidth described as “System Dependent.” The RX 570X also benefits from its 2,048 shading units running at a 1,244 MHz boost clock, delivering 5.095 TFLOPS of FP32 throughput. In contrast, the 660M’s 384 shading units at 1,900 MHz produce 1,459.2 GFLOPS FP32—a 3.5x deficit in raw compute that the OpenCL score reflects almost exactly.

Looking at the aggregate benchmarks, the picture becomes more nuanced. The RX 570X reports a Passmark G3D score of 1,923 and a Passmark G2D score of 750, while the 660M counters with a Geekbench Vulkan score of 14,748—a test the RX 570X does not report. This missing data point is crucial: the 660M’s Vulkan result is 14.5% higher than its own OpenCL score, suggesting that its RDNA 2.0 architecture is substantially more efficient in modern graphics APIs. If both cards had been tested across the same suite, the aggregate scores might shift, but as it stands, the 0.4% difference in average score (13,871 vs 13,812) places them in a statistical tie. The nearest rival data reinforces this: the RX 570X is 0.4% ahead of the RTX A2000 Mobile, while the 660M is 0.1% behind that same NVIDIA part, and both are within 0.9% of the RX 7900 XT—a flagship card that scores 13,745.

Where Each One Wins

The RX 570X wins decisively in raw compute and memory-bound workloads. Its Geekbench OpenCL performance is in a different league, and its specification sheet supports this: 256-bit memory bus, 128 texture mapping units, 32 ROPs, and a texture rate of 159.2 GTexel/s. For applications that hammer on FP32 arithmetic—scientific computing, rendering, or any OpenCL-accelerated task—the RX 570X is the clear choice, as evidenced by its 202.4% lead in the head-to-head test. The Passmark G3D score of 1,923, while modest, still indicates solid direct3D rasterization capability for its era.

The 660M wins in efficiency and modern API support. Despite its lower raw throughput, its 2.918 TFLOPS FP16 performance (at 2:1 ratio) doubles its FP32 rate, a feature the RX 570X lacks entirely (its FP16 equals FP32 at 1:1). The 660M also supports DirectX 12 Ultimate (12_2) versus the RX 570X’s DirectX 12 (12_0), and Vulkan 1.4 versus 1.3. Its Geekbench Vulkan score of 14,748 suggests that in games and applications using modern APIs, the 660M can punch well above its compute-weight class. The 660M’s 40 W TDP versus the RX 570X’s 150 W also makes it the only viable option for thin-and-light laptops, where the RX 570X’s dual-slot footprint and 1x 6-pin power connector cannot fit. The 660M’s ray tracing cores (6 of them) provide hardware-accelerated ray tracing, a feature entirely absent from the RX 570X.

Architecture Differences

The two GPUs represent two distinct architectural eras from AMD. The RX 570X is built on GCN 4.0, using the Polaris 20 chip manufactured on GlobalFoundries’ 14 nm process. It packs 5,700 million transistors into a 232 mm² die, yielding a transistor density of 24.6 million per square millimeter. The 660M, by contrast, uses RDNA 2.0 on the Rembrandt chip, fabricated by TSMC on a 6 nm node. It has 13,100 million transistors in a slightly smaller 208 mm² die, achieving a much higher density of 63.0 million per square millimeter—2.56x more transistors per area. This density advantage is the core of the 660M’s efficiency story: it delivers comparable aggregate performance at 40 W versus 150 W (a 73% power reduction) with a smaller physical footprint.

Memory architecture is another fundamental split. The RX 570X has 8 GB of GDDR5 on a 256-bit bus, providing 224.0 GB/s of dedicated bandwidth. The 660M uses system-shared memory, with the bus width and bandwidth listed as “System Shared” and “System Dependent,” respectively. This means the 660M’s performance scales with the host system’s memory configuration—a factor that introduces variability the RX 570X never faces. The clock speeds also differ: the RX 570X runs at 1,168 MHz base and 1,244 MHz boost, while the 660M runs at 1,500 MHz base and 1,900 MHz boost. Despite the 660M’s higher clocks, its 384 shading units simply cannot match the 2,048 units of the RX 570X in parallel throughput. The 660M compensates with 6 dedicated ray tracing cores and support for FP16 at double rate, features that are entirely absent from the older GCN design.

FAQ

Q: Which GPU has a higher average benchmark score?

A: The AMD Radeon RX 570X has an average benchmark score of 13,871, which is 0.4% higher than the AMD Radeon 660M’s 13,812.

Q: How large is the performance gap in the Geekbench OpenCL test?

A: The RX 570X scores 38,939 versus the 660M’s 12,876, giving the RX 570X a 202.4% advantage in that specific benchmark.

Q: Does the AMD Radeon 660M support ray tracing?

A: Yes, the 660M includes 6 ray tracing cores, while the RX 570X has no ray tracing cores listed.

Q: What are the power consumption figures for each card?

A: The RX 570X has a TDP of 150 W, while the 660M has a TDP of 40 W.

Q: Which card has more shading units?

A: The RX 570X has 2,048 shading units, compared to the 660M’s 384 shading units.

Q: How do their memory subsystems differ?

A: The RX 570X uses 8 GB of GDDR5 on a 256-bit bus with 224.0 GB/s bandwidth, while the 660M uses system-shared memory with bus width and bandwidth listed as “System Shared” and “System Dependent.”

Specification Differences

| Specification | AMD Radeon RX 570X | AMD Radeon 660M |

|---|---|---|

| Architecture | GCN 4.0 | RDNA 2.0 |

| Process Node | 14 nm | 6 nm |

| Foundry | GlobalFoundries | TSMC |

| Transistors | 5,700 million | 13,100 million |

| Die Size | 232 mm² | 208 mm² |

| Transistor Density | 24.6M / mm² | 63.0M / mm² |

| Base Clock | 1168 MHz | 1500 MHz |

| Boost Clock | 1244 MHz | 1900 MHz |

| Memory Size | 8 GB | System Shared |

| Memory Type | GDDR5 | System Shared |

| Memory Bus Width | 256 bit | System Shared |

| Memory Bandwidth | 224.0 GB/s | System Dependent |

| Shading Units | 2048 | 384 |

| TMUs | 128 | 24 |

| ROPs | 32 | 16 |

| RT Cores | None | 6 |

| Pixel Rate | 39.81 GPixel/s | 30.40 GPixel/s |

| Texture Rate | 159.2 GTexel/s | 45.60 GTexel/s |

| FP32 | 5.095 TFLOPS | 1,459.2 GFLOPS |

| FP16 | 5.095 TFLOPS (1:1) | 2.918 TFLOPS (2:1) |

| TDP | 150 W | 40 W |

| Slot Width | Dual-slot | IGP |

| Power Connectors | 1x 6-pin | None |

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

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

| Vulkan Support | 1.3 | 1.4 |

| Release Date | 2018-04-10 | 2022-01-03 |

The Verdict

The data presents a clear split based on workload and form factor. For compute-heavy tasks that leverage OpenCL, the RX 570X is the overwhelming choice—its 202.4% lead in the head-to-head benchmark is not a marginal edge but a categorical dominance. Anyone running OpenCL-accelerated applications, from rendering pipelines to scientific simulations, should favor the RX 570X’s dedicated 8 GB GDDR5 memory and 5.095 TFLOPS FP32 performance without hesitation. The 224.0 GB/s of bandwidth versus the 660M’s system-dependent throughput is a structural advantage that no driver optimization can overcome.

For modern gaming and efficiency-sensitive scenarios, the 660M is the more sensible pick. Its DirectX 12 Ultimate support, Vulkan 1.4 compatibility, and 6 ray tracing cores make it future-proof in ways the RX 570X cannot match. The 14,748 Geekbench Vulkan score, while only 10.8% behind the RX 570X’s OpenCL number, suggests that in contemporary graphics APIs, the 660M closes much of the gap. The 40 W TDP versus 150 W also means the 660M can operate in systems where a discrete dual-slot card with a 6-pin connector is physically impossible. The 660M’s 2,918 TFLOPS FP16 output also gives it a niche in AI inference workloads that the RX 570X simply cannot accelerate efficiently.

The 55th percentile ranking for both cards indicates they are mid-pack performers, but the journey matters. The RX 570X is a 2018-era discrete card from the end of the Polaris line, while the 660M is a 2022 integrated solution from the Rembrandt generation. The fact that they land within 0.4% of each other on average speaks to the massive efficiency gains of RDNA 2.0 over GCN 4.0—the 660M achieves parity with 73% less power and a 59% smaller die area. Users with existing desktop slots and power budgets should pick the RX 570X for raw compute; users building or buying laptops with no expansion options should pick the 660M as the only viable choice between these two, while gaining modern API features in the bargain. The verdict is not a matter of which is faster overall, but which kind of fast fits the use case.