AMD Radeon 660M vs AMD Radeon HD 8950M Comparison

AMD Radeon 660M and AMD Radeon HD 8950M represent two very different eras of mobile graphics, yet their average benchmark scores land surprisingly close together. The 660M averages 13,812 across tested workloads, placing it in the 55th percentile of all GPUs, while the HD 8950M averages 13,376 and sits in the 54th percentile. The data shows a near-tie in overall compute, but the split between OpenCL and Vulkan results tells a clear story about which GPU suits which task. This comparison is not about raw dominance; it is about matching the right silicon to the right workload.

The Verdict

Pick the AMD Radeon 660M if your priority is modern API support and forward-looking compute performance. Its Vulkan score of 14,748 beats the HD 8950M by 13.6%, and it supports DirectX 12 Ultimate (12_2) and Vulkan 1.4, making it the better choice for newer games and compute applications that leverage these APIs. The 660M also does this at a 40 W TDP, which is 60 W lower than the HD 8950M’s 100 W, a massive efficiency advantage for thin-and-light laptops.

Pick the AMD Radeon HD 8950M if your workload is heavily OpenCL-based and you need dedicated memory. Its OpenCL score of 13,772 is 6.5% ahead of the 660M’s 12,876, and it comes with 2 GB of GDDR5 on a 128-bit bus with 88.00 GB/s bandwidth. That dedicated memory pool avoids the system-shared bottleneck of the 660M, which relies on system RAM and has system-dependent bandwidth. The HD 8950M also has double the shading units (768 vs 384) and TMUs (48 vs 24), giving it raw shader throughput that the 660M cannot match in legacy compute paths.

In short: the 660M is the modern all-rounder with better efficiency and API readiness; the HD 8950M is the old-school compute workhorse with dedicated VRAM and a clear OpenCL edge. Benchmark results indicate that neither is a universal winner, which is why the head-to-head tally sits at 1 win each.

Architecture Differences

The architectural gap between these two is generational. The AMD Radeon 660M uses the Rembrandt chip built on RDNA 2.0 architecture, fabricated on a 6 nm process at TSMC. It packs 13,100 million transistors into a 208 mm² die, yielding a transistor density of 63.0M per mm². This is an integrated graphics processor (IGP) with 384 shading units, 24 TMUs, 16 ROPs, and 6 ray tracing cores. It supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

The AMD Radeon HD 8950M uses the Saturn chip based on GCN 2.0 architecture, built on a 28 nm process, also at TSMC. It has 2,080 million transistors on a 160 mm² die, with a density of 13.0M per mm² — roughly five times lower than the 660M. It is a discrete MXM module with 768 shading units, 48 TMUs, and 16 ROPs, but no ray tracing cores. It supports DirectX 12 (12_0), OpenGL 4.6, and Vulkan 1.2.170.

Key feature differences: the 660M has ray tracing hardware (6 RT cores) and a newer DirectX feature level (12_2 vs 12_0), while the HD 8950M lacks RT cores entirely. The 660M also has a higher pixel rate (30.40 GPixel/s vs 17.20 GPixel/s) despite fewer ROPs, due to its much higher base clock of 1500 MHz and boost of 1900 MHz; the HD 8950M has no listed base or boost clock, only a memory clock of 1375 MHz (5.5 Gbps effective). The 660M’s FP16 throughput is listed at 2.918 TFLOPS (2:1), whereas the HD 8950M has no FP16 data. In FP32, the HD 8950M leads slightly at 1.651 TFLOPS vs 1,459.2 GFLOPS for the 660M.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon 660M averages 13,812, which is 436 points higher than the HD 8950M’s 13,376. The 660M also sits at the 55th percentile of all GPUs, one point above the HD 8950M’s 54th percentile.

Q: How do they compare in OpenCL performance?

A: The HD 8950M wins OpenCL with a score of 13,772, which is 6.5% higher than the 660M’s 12,876. This is likely due to its 768 shading units and dedicated 2 GB GDDR5 memory.

Q: What about Vulkan performance?

A: The 660M wins Vulkan decisively, scoring 14,748 against the HD 8950M’s 12,979 — a 13.6% advantage. This aligns with the 660M’s newer RDNA 2.0 architecture and Vulkan 1.4 support.

Q: Which GPU is more power-efficient?

A: The 660M has a 40 W TDP, while the HD 8950M is rated at 100 W. The 660M delivers comparable average performance at 60 W less power draw, making it far more suitable for compact laptops.

Q: Does the HD 8950M have dedicated memory?

A: Yes, it has 2 GB of GDDR5 on a 128-bit bus with 88.00 GB/s bandwidth. The 660M uses system-shared memory with system-dependent bandwidth, which can be a bottleneck depending on the laptop’s RAM.

Q: Which GPU supports ray tracing?

A: Only the 660M supports ray tracing, with 6 dedicated RT cores. The HD 8950M has no RT cores and is limited to DirectX 12 (12_0), while the 660M supports DirectX 12 Ultimate (12_2).

Specification Differences

| Specification | AMD Radeon 660M | AMD Radeon HD 8950M |

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

| Architecture | RDNA 2.0 | GCN 2.0 |

| Process Node | 6 nm | 28 nm |

| Transistors | 13,100 million | 2,080 million |

| Die Size | 208 mm² | 160 mm² |

| Transistor Density | 63.0M / mm² | 13.0M / mm² |

| Base Clock | 1500 MHz | Not specified |

| Boost Clock | 1900 MHz | Not specified |

| Memory Size | System Shared | 2 GB |

| Memory Type | System Shared | GDDR5 |

| Memory Bus Width | System Shared | 128 bit |

| Memory Bandwidth | System Dependent | 88.00 GB/s |

| Shading Units | 384 | 768 |

| TMUs | 24 | 48 |

| ROPs | 16 | 16 |

| RT Cores | 6 | None |

| Pixel Rate | 30.40 GPixel/s | 17.20 GPixel/s |

| Texture Rate | 45.60 GTexel/s | 51.60 GTexel/s |

| FP32 | 1,459.2 GFLOPS | 1.651 TFLOPS |

| FP16 | 2.918 TFLOPS (2:1) | Not specified |

| TDP | 40 W | 100 W |

| Slot Width | IGP | MXM Module |

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

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

| Vulkan | 1.4 | 1.2.170 |

Head-to-Head Benchmarks

The two benchmark results split cleanly along API lines. In Geekbench OpenCL, the HD 8950M takes the win with 13,772 points against the 660M’s 12,876, a delta of -6.5% for the 660M. That 896-point gap comes from the HD 8950M’s double shading unit count (768 vs 384) and its dedicated 2 GB GDDR5 memory with 88.00 GB/s bandwidth, which eliminates the shared-memory latency that can hamper integrated graphics.

In Geekbench Vulkan, the 660M flips the script with a commanding 14,748 score versus 12,979 for the HD 8950M, a 13.6% advantage. This 1,769-point win is a direct reflection of the 660M’s modern RDNA 2.0 architecture, which is designed to excel in Vulkan’s low-overhead model. The HD 8950M’s older GCN 2.0 design and Vulkan 1.2.170 support simply cannot keep pace in this workload.

The average benchmark scores reinforce the split: the 660M averages 13,812 (55th percentile), while the HD 8950M averages 13,376 (54th percentile). The 660M’s nearest rival is the NVIDIA RTX A2000 Mobile at 13,821 (a -0.1% delta), showing it hangs with professional mobile GPUs. The HD 8950M’s closest competitor is the AMD Radeon Pro 555 at 13,407 (a -0.2% delta), indicating it sits in a similar performance class.

Where Each One Wins

The AMD Radeon 660M wins in Vulkan-based workloads, modern game titles that leverage DirectX 12 Ultimate features, and any scenario where power consumption matters. Its 13.6% Vulkan lead over the HD 8950M makes it the obvious pick for newer game engines and compute frameworks that use Vulkan. The 6 RT cores provide ray tracing capability that the HD 8950M simply does not have. At 40 W, it draws 60% less power than the HD 8950M, making it the only sensible choice for ultraportable laptops where battery life and thermals are critical. Its FP16 throughput of 2.918 TFLOPS also gives it a niche advantage in AI and machine learning inference tasks that use half-precision math.

The AMD Radeon HD 8950M wins in OpenCL-heavy applications, legacy compute workloads, and scenarios that demand dedicated VRAM. Its 6.5% OpenCL advantage over the 660M means it will handle OpenCL-based rendering, scientific computing, and older professional applications with better raw throughput. The 2 GB GDDR5 pool is a hard advantage — the 660M’s system-shared memory is system-dependent, meaning performance can vary wildly based on the laptop’s RAM speed and capacity. The HD 8950M also has a higher texture rate (51.60 GTexel/s vs 45.60 GTexel/s) and double the TMUs, which helps in texture-bound OpenCL workloads. Its PCIe 3.0 x16 interface provides more bandwidth to the host system than the 660M’s PCIe 4.0 x8, though the 660M’s newer standard may compensate in practice.

The data shows a clear use-case split. Choose the 660M for modern gaming, Vulkan compute, and efficiency. Choose the HD 8950M for legacy OpenCL software, dedicated memory needs, and workloads that favor raw shader count. The 1-1 head-to-head tally is honest: neither GPU is objectively better across the board, and the right choice depends entirely on which benchmark — and which API — matches your actual workload.