AMD Radeon 660M vs NVIDIA Tesla K20c Comparison

Head-to-Head Benchmarks

The only direct benchmark comparison recorded in the database is Geekbench OpenCL, and it shows a clear, if not overwhelming, advantage for the AMD Radeon 660M. The integrated part scores 12,876 points, while the NVIDIA Tesla K20c manages 11,479 points. That is a 12.2% delta in favor of the AMD chip, a substantial margin for a single test.

This result is interesting because of what it implies about the two architectures. The Tesla K20c has far more raw shading units (2,496 versus 384) and a much higher rated FP32 throughput (3.524 TFLOPS versus 1,459.2 GFLOPS), yet it falls behind in this particular workload. The database suggests that OpenCL performance is not purely a function of peak compute, but also of driver optimization, memory subsystem behavior, and the efficiency of the architecture itself. The Radeon 660M's 12.2% lead here is the only recorded head-to-head data point, so it carries significant weight in the overall comparison.

When looking at the broader context of the database, the AMD Radeon 660M sits at the 55th percentile of all GPUs, with an average benchmark score of 13,812. Its nearest rivals include the NVIDIA RTX A2000 Mobile (average score 13,821, delta -0.1%) and the AMD Radeon RX 570X (average score 13,871, delta -0.4%). It also edges out the AMD Radeon RX 7900 XT in the database (average score 13,745, delta +0.5%), which is a surprising result given the latter's enormous performance class in other contexts. The 660M effectively trades blows with these parts, landing in a tight cluster around 13.7k to 13.9k points.

The NVIDIA Tesla K20c, by contrast, is at the 51st percentile with an average score of 11,479. Its nearest rivals include the AMD Radeon Pro 5500M (11,528, delta -0.4%) and the Radeon RX 7800 XT (11,627, delta -1.3%). The K20c is also 1.9% ahead of the GeForce GTX 780M (11,261). This places the Tesla in a lower performance bracket than the Radeon 660M, roughly 20% behind the AMD part on average (11,479 vs 13,812). The data shows that despite its age, the K20c can still hang around modern midrange parts, but it does not lead any of its nearest rivals by more than 1.9%.

Architecture Differences

The two GPUs represent radically different design eras. The AMD Radeon 660M uses the Rembrandt chip built on TSMC's 6 nm process, featuring the RDNA 2.0 architecture. It is part of the Navi II IGP generation for Rembrandt Mobile. The NVIDIA Tesla K20c uses the GK110 chip, built on a 28 nm process, using the Kepler architecture from the Tesla Kepler (Kxx) generation. The process node difference is stark: 6 nm versus 28 nm, which impacts transistor density and power efficiency.

The transistor counts and die sizes tell a story of design philosophy. The Radeon 660M packs 13,100 million transistors into a 208 mm² die, yielding a density of 63.0 million transistors per square millimeter. The Tesla K20c has only 7,080 million transistors on a massive 561 mm² die, giving it a density of just 12.6 million per square millimeter. The modern chip is far more efficient in terms of packing logic into a small space, even though it has a smaller absolute transistor count.

The memory configuration differences are fundamental. The Radeon 660M uses System Shared memory, meaning it has no dedicated VRAM and relies on the host system's memory, with bandwidth described as System Dependent. The Tesla K20c has 5 GB of GDDR5 memory on a 320-bit bus, delivering 208.0 GB/s of bandwidth. This dedicated memory is a huge advantage for the older card in bandwidth-bound workloads, but the 660M's shared memory might benefit from faster system memory in a modern platform.

The compute resources are also completely different. The Radeon 660M has 384 shading units, 24 texture mapping units, 16 ROPs, and 6 ray tracing cores. The Tesla K20c has 2,496 shading units, 208 TMUs, and 40 ROPs, but no ray tracing cores. The NVIDIA card has far more raw execution units, but the AMD part has a much higher peak clock speed (1,900 MHz boost versus no listed boost clock for the Tesla, with memory clock at 1,300 MHz). The 660M's clock speed is the key to its competitive showing in OpenCL, as it helps offset the lower unit count.

The power envelopes are also wildly different. The Radeon 660M is an integrated GPU with a 40 W TDP, and it requires no power connectors. It is an IGP, meaning it occupies no slot width. The Tesla K20c is a dual-slot card with a 225 W TDP, requiring a 6-pin and 8-pin power connector, plus a suggested 550 W power supply. The Tesla also has a physical length of 267 mm (10.5 inches), and it has no display outputs, which is typical for a compute accelerator. The AMD part is portable-device dependent for its display outputs.

The PCIe interface also differs: the 660M uses PCIe 4.0 x8, while the Tesla uses the older PCIe 2.0 x16. The newer, faster bus does not fully compensate for the lack of dedicated memory, but it helps with data transfer in shared memory scenarios.

Where Each One Wins

The data suggests a clear split. The AMD Radeon 660M wins the only recorded head-to-head benchmark, and it does so by a substantial margin. Its average performance is also higher, and it ranks higher in the overall percentile distribution. If you are looking at a general compute workload as measured by Geekbench OpenCL, the Radeon 660M is the superior part.

The Tesla K20c, however, has a specific niche: any workload that benefits from its dedicated 5 GB of GDDR5 memory and 208 GB/s of bandwidth. The Radeon 660M's System Shared memory is a bottleneck for certain tasks, and the Tesla's dedicated VRAM is a clear advantage for large datasets that fit within its memory pool. The Tesla also has a higher texture rate (146.8 GTexel/s) and a higher pixel rate (36.71 GPixel/s) than the AMD part, which has 45.60 GTexel/s and 30.40 GPixel/s respectively. So in fill-rate-sensitive tasks, the NVIDIA card should lead.

The Radeon 660M wins on power efficiency by a huge margin. Its 40 W TDP versus the Tesla's 225 W means it can be used in thin-and-light laptops, while the Tesla is a desktop compute card. The Radeon also has modern feature support: DirectX 12 Ultimate (12_2), Vulkan 1.4, and OpenGL 4.6. The Tesla has DirectX 12 (11_0), Vulkan 1.2.175, and OpenGL 4.6. The Radeon's ray tracing cores also give it a feature that the Tesla lacks entirely.

For a modern gaming or content creation workload on a portable device, the Radeon 660M is the clear choice. For a legacy compute server with a specific need for dedicated memory and no power constraints, the Tesla K20c might still be viable, especially if the workload is not well-served by the shared memory architecture.

The Verdict

Based purely on the recorded data, the AMD Radeon 660M is the superior product. It wins the head-to-head OpenCL benchmark by 12.2%, has a higher average score (13,812 vs 11,479), and sits at a higher percentile (55th vs 51st). It achieves this with a fraction of the power draw (40 W vs 225 W) and a much smaller footprint. The data strongly indicates that the modern IGP outperforms the older dedicated compute card in the only test that was recorded.

However, the Tesla K20c is not without a role. Its 5 GB of dedicated memory and 208 GB/s bandwidth are not matched by the Radeon 660M, which relies on system memory. In scenarios where memory capacity and bandwidth are the primary bottlenecks, the Tesla's dedicated VRAM could give it an edge that is not reflected in the Geekbench OpenCL score. The Tesla's higher pixel rate and texture rate also suggest it might excel in certain rasterization tasks, though these are not tested in the head-to-head data.

For a user who needs a low-power, modern GPU for a mobile device, the Radeon 660M is the only choice. For a user who needs a dedicated compute card for a legacy system with a tight memory budget, the Tesla K20c is the only one with discrete memory. If the workload is compatible with the Radeon's shared memory model, the 660M is a clear winner. If the workload requires a dedicated memory, the Tesla is the data-supported pick.

FAQ

Q: Which GPU has a higher Geekbench OpenCL score?

A: The AMD Radeon 660M scores 12,876, which is 12.2% higher than the NVIDIA Tesla K20c's score of 11,479.

Q: How much dedicated memory does the NVIDIA Tesla K20c have?

A: The Tesla K20c has 5 GB of GDDR5 memory on a 320-bit bus, providing 208.0 GB/s of bandwidth.

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

A: Yes, the Radeon 660M includes 6 ray tracing cores, while the Tesla K20c has no ray tracing cores.

Q: Which GPU has a higher TDP?

A: The NVIDIA Tesla K20c has a 225 W TDP, while the AMD Radeon 660M has a 40 W TDP.

Q: What is the average benchmark score for each GPU?

A: The AMD Radeon 660M has an average score of 13,812, while the NVIDIA Tesla K20c has an average score of 11,479.

Q: Which GPU is newer in release date?

A: The AMD Radeon 660M was released on January 3, 2022, while the NVIDIA Tesla K20c was released on November 11, 2012.

Specification Differences

| Field | AMD Radeon 660M | NVIDIA Tesla K20c |

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

| Manufacturer | AMD | NVIDIA |

| Chip | Rembrandt | GK110 |

| Architecture | RDNA 2.0 | Kepler |

| Process Node | 6 nm | 28 nm |

| Transistors | 13,100 million | 7,080 million |

| Die Size | 208 mm² | 561 mm² |

| Transistor Density | 63.0M / mm² | 12.6M / mm² |

| Base Clock | 1500 MHz | Not listed |

| Boost Clock | 1900 MHz | Not listed |

| Memory Size | System Shared | 5 GB |

| Memory Type | System Shared | GDDR5 |

| Memory Bus Width | System Shared | 320 bit |

| Memory Bandwidth | System Dependent | 208.0 GB/s |

| Shading Units | 384 | 2,496 |

| TMUs | 24 | 208 |

| ROPs | 16 | 40 |

| RT Cores | 6 | None |

| Pixel Rate | 30.40 GPixel/s | 36.71 GPixel/s |

| Texture Rate | 45.60 GTexel/s | 146.8 GTexel/s |

| FP32 | 1,459.2 GFLOPS | 3.524 TFLOPS |

| TDP | 40 W | 225 W |

| Slot Width | IGP | Dual-slot |

| Power Connectors | None | 1x 6-pin + 1x 8-pin |

| Suggested PSU | Not listed | 550 W |

| Bus Interface | PCIe 4.0 x8 | PCIe 2.0 x16 |

| Display Outputs | Portable Device Dependent | No outputs |

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

| Vulkan | 1.4 | 1.2.175 |

| OpenGL | 4.6 | 4.6 |

| Dimensions (Length) | Not listed | 267 mm (10.5 inches) |

| Release Date | 2022-01-03 | 2012-11-11 |

| Successor | Navi III IGP | Tesla Maxwell |

| Predecessor | Vega II IGP | Tesla Fermi |

| Launch MSRP | Not listed | 3,199 USD |