NVIDIA GeForce GTX 1660 SUPER vs NVIDIA Tesla K20Xm Comparison

The NVIDIA GeForce GTX 1660 SUPER and NVIDIA Tesla K20Xm represent two distinct eras of GPU design, separated by seven years of architecture evolution. The benchmark data shows a decisive victory for the newer card, with the GTX 1660 SUPER winning the only shared test by a massive margin.

Head-to-Head Benchmarks

The sole direct comparison available is the Geekbench OpenCL test, and the results are stark. The GeForce GTX 1660 SUPER scores 52,490 points, while the Tesla K20Xm manages only 17,215 points. This translates to a 204.9% advantage for the GTX 1660 SUPER, meaning it delivers more than three times the compute performance in this workload.

This result is particularly striking given the architectural gulf between the two cards. The Tesla K20Xm was a high-performance computing accelerator designed for scientific workloads, yet it is utterly eclipsed in this general-purpose compute benchmark. The GTX 1660 SUPER's score places it in the 53rd percentile of all GPUs, while the K20Xm sits just one point lower at the 52nd percentile, highlighting how close their overall standing is despite the massive OpenCL gap.

Looking at the broader benchmark landscape, the GTX 1660 SUPER's average benchmark score of 12,986 places it just 0.4% above the NVIDIA GeForce RTX 3050 Ti Mobile and AMD Radeon RX 580, and 0.9% above the AMD Radeon 740M. Its nearest rival below is the NVIDIA Tesla M2090, which sits 0.7% lower. The Tesla K20Xm, with an average score of 12,625, trails the AMD Radeon RX 7600M XT by 0.7% and the NVIDIA GeForce GTX 670 by 1.2%. This puts the two cards within 2.9% of each other on average, yet the OpenCL test reveals a fundamentally different performance character.

The GTX 1660 SUPER also shows strong results in other tests not shared with the K20Xm. It scores 12,699 in Passmark G3D and 5,076 in Passmark GPU Compute, alongside 52,490 in Geekbench OpenCL and 57,102 in Geekbench Vulkan. The Tesla K20Xm's only other benchmark is Geekbench Metal, where it scores 8,035, a figure that has no direct counterpart on the GeForce card.

Architecture Differences

The two GPUs are built on completely different architectures and process nodes. The GeForce GTX 1660 SUPER uses the TU116 chip based on the Turing architecture, manufactured on a 12 nm process at TSMC. The Tesla K20Xm uses the GK110 chip based on the older Kepler architecture, built on a 28 nm process, also at TSMC. This process difference is critical: the 12 nm node allows for significantly higher transistor density at 23.2 million transistors per square millimeter, compared to just 12.6 million for the 28 nm Kepler chip.

Despite the newer process, the TU116 packs fewer transistors overall — 6,600 million versus 7,080 million on the GK110. However, the die size tells a different story: the TU116 measures just 284 mm², while the GK110 is a massive 561 mm². This means the GTX 1660 SUPER achieves its performance in less than half the silicon area, a signal of architectural efficiency gains over the seven-year gap.

The shading unit counts illustrate the design philosophy shift. The Tesla K20Xm features 2,688 shading units and 224 texture mapping units, far more than the GTX 1660 SUPER's 1,408 shading units and 88 TMUs. Both have 48 ROPs. The K20Xm's higher raw unit count was typical of the Kepler generation, which relied on massive parallelism. The Turing architecture instead uses fewer, more efficient units with higher clock speeds and better instruction-level efficiency.

Clock speeds are a major differentiator. The GTX 1660 SUPER has a base clock of 1530 MHz and a boost clock of 1785 MHz, while the Tesla K20Xm has no listed base or boost clocks. This clock advantage, combined with the architectural improvements, explains why the newer card dominates in compute despite having roughly half the shading units.

FAQ

Q: Which GPU is faster in OpenCL compute performance?

A: The NVIDIA GeForce GTX 1660 SUPER is dramatically faster, scoring 52,490 points compared to the Tesla K20Xm's 17,215 points in Geekbench OpenCL, a 204.9% difference.

Q: Do both cards support the same DirectX version?

A: No. The GTX 1660 SUPER supports DirectX 12 (12_1), while the Tesla K20Xm is limited to DirectX 12 (11_0), a lower feature level.

Q: What is the memory configuration difference?

A: Both cards have 6 GB of memory, but the GTX 1660 SUPER uses GDDR6 on a 192-bit bus with 336.0 GB/s bandwidth, while the K20Xm uses GDDR5 on a 384-bit bus with 249.6 GB/s bandwidth.

Q: Which card has a higher transistor density?

A: The GTX 1660 SUPER's TU116 chip has a transistor density of 23.2 million per square millimeter, nearly double the K20Xm's GK110 at 12.6 million per square millimeter.

Q: Are both cards still in production?

A: No, both are end-of-life products. The GTX 1660 SUPER was released on October 28, 2019, and the Tesla K20Xm on November 11, 2012.

Q: What is the pixel fillrate difference?

A: The GTX 1660 SUPER achieves 85.68 GPixel/s, more than double the Tesla K20Xm's 40.99 GPixel/s, despite both having 48 ROPs.

Specification Differences

| Specification | NVIDIA GeForce GTX 1660 SUPER | NVIDIA Tesla K20Xm |

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

| Architecture | Turing | Kepler |

| Process Node | 12 nm | 28 nm |

| Transistors | 6,600 million | 7,080 million |

| Die Size | 284 mm² | 561 mm² |

| Transistor Density | 23.2M / mm² | 12.6M / mm² |

| Base Clock | 1530 MHz | N/A |

| Boost Clock | 1785 MHz | N/A |

| Memory Type | GDDR6 | GDDR5 |

| Memory Bus Width | 192 bit | 384 bit |

| Memory Bandwidth | 336.0 GB/s | 249.6 GB/s |

| Memory Clock | 1750 MHz (14 Gbps effective) | 1300 MHz (5.2 Gbps effective) |

| Shading Units | 1408 | 2688 |

| Texture Mapping Units | 88 | 224 |

| Pixel Rate | 85.68 GPixel/s | 40.99 GPixel/s |

| Texture Rate | 157.1 GTexel/s | 164.0 GTexel/s |

| FP32 Performance | 5.027 TFLOPS | 3.935 TFLOPS |

| FP16 Performance | 10.05 TFLOPS (2:1) | N/A |

| TDP | 125 W | 235 W |

| Suggested PSU | 300 W | 550 W |

| Power Connectors | 1x 8-pin | N/A |

| Display Outputs | 1x DVI, 1x HDMI 2.0, 1x DisplayPort 1.4a | No outputs |

| DirectX Support | 12 (12_1) | 12 (11_0) |

| Vulkan Support | 1.4 | 1.2.175 |

| Dimensions | 229 mm x 111 mm x 35 mm | 267 mm length |

| Launch MSRP | 229 USD | 7,699 USD |

Where Each One Wins

The GeForce GTX 1660 SUPER is the clear winner for any consumer-facing workload. Its display outputs allow it to drive monitors directly, while the Tesla K20Xm has no outputs at all, making it unsuitable for gaming or desktop use. The GTX 1660 SUPER's FP32 performance of 5.027 TFLOPS exceeds the K20Xm's 3.935 TFLOPS, and its FP16 capability at 10.05 TFLOPS gives it an edge in mixed-precision workloads. The pixel rate advantage of 85.68 GPixel/s versus 40.99 GPixel/s makes it better suited for rasterization-heavy tasks. Its lower TDP of 125 W versus 235 W, combined with a smaller 229 mm length, makes it far easier to integrate into standard systems.

The Tesla K20Xm retains a few niche advantages. Its 384-bit memory bus provides a wider interface, though the GDDR5 memory runs at lower effective speeds, resulting in lower overall bandwidth. The higher shading unit and TMU counts — 2,688 and 224 respectively — suggest that certain highly parallel compute workloads might use its resources more efficiently, though the benchmark data does not confirm this. Its texture rate of 164.0 GTexel/s is slightly higher than the GTX 1660 SUPER's 157.1 GTexel/s, indicating a marginal edge in texture-heavy operations. For legacy compute environments that require Kepler architecture support, the K20Xm remains a functional option, but its 2012 release date and lack of modern API support — Vulkan 1.2.175 versus 1.4 — limit its future-proofing.

The data overwhelmingly favors the GTX 1660 SUPER for modern applications. Its superior OpenCL score, higher clock speeds, newer architecture, and vastly better power efficiency make it the more capable card in nearly every measurable way. The K20Xm's only theoretical advantages lie in raw unit counts that do not translate to benchmark wins, leaving it as a historical artifact rather than a competitive option.