NVIDIA GeForce GTX 1660 SUPER vs NVIDIA Tesla K10 Comparison

The NVIDIA Tesla K10 and the NVIDIA GeForce GTX 1660 SUPER represent two distinct eras of GPU design, with the former serving as a compute-oriented accelerator from the Kepler generation and the latter as a consumer gaming card built on the Turing architecture. While the Tesla K10 was a dual-GPU workhorse for scientific computing in 2012, the GTX 1660 SUPER is a more modern, efficient, and significantly faster part in general-purpose benchmarks. The data shows a decisive victory for the newer card, but the comparison reveals fundamental shifts in GPU architecture, memory technology, and feature support that go beyond simple performance deltas.

FAQ

Q: Which GPU is faster in the Geekbench OpenCL benchmark?

A: The NVIDIA GeForce GTX 1660 SUPER is dramatically faster, scoring 52,490 points compared to the Tesla K10's 14,029 points. This represents a 73.3% lead for the GTX 1660 SUPER in this compute test.

Q: How does the Tesla K10's performance compare to its nearest rivals?

A: The Tesla K10's score of 14,029 places it 0.9% behind the NVIDIA GeForce GTX 680 (14,150) and 1.1% ahead of the AMD Radeon RX 570X (13,871). It also edges out the NVIDIA RTX A2000 Mobile and AMD Radeon 660M by 1.5% and 1.6%, respectively.

Q: What are the memory specifications of each card?

A: The Tesla K10 features 4 GB of GDDR5 memory on a 256-bit bus, delivering 160.0 GB/s of bandwidth. The GTX 1660 SUPER comes with 6 GB of GDDR6 memory on a 192-bit bus, providing more than double the bandwidth at 336.0 GB/s.

Q: Do both cards support the same DirectX version?

A: No. The Tesla K10 supports DirectX 12 (11_0), while the GTX 1660 SUPER supports the full DirectX 12 (12_1) feature set. This indicates the newer card has access to more advanced rendering features.

Q: Which GPU has a higher pixel fill rate?

A: The GTX 1660 SUPER has a significantly higher pixel rate of 85.68 GPixel/s, compared to the Tesla K10's 23.84 GPixel/s. This suggests a major advantage in rasterization throughput for the newer card.

Q: What is the transistor density difference between the two chips?

A: The GTX 1660 SUPER's TU116 chip packs 23.2 million transistors per mm², while the Tesla K10's GK104 chip has a density of 12.0 million per mm². This reflects the move from a 28 nm to a 12 nm manufacturing process.

Architecture Differences

The Tesla K10 and GTX 1660 SUPER are built on fundamentally different architectures that reflect a five-generation gap in NVIDIA's design philosophy. The Tesla K10 uses the Kepler architecture, fabricated on TSMC's 28 nm process, and features a chip codenamed GK104 with 3,540 million transistors on a 294 mm² die. In contrast, the GTX 1660 SUPER uses the Turing architecture, built on a 12 nm process, with a TU116 chip containing 6,600 million transistors on a slightly smaller 284 mm² die. This shift allows the Turing chip to achieve a transistor density of 23.2M per mm², nearly double the Kepler part's 12.0M per mm².

The compute capabilities differ starkly. The Tesla K10 has 1,536 shading units, 128 texture mapping units, and 32 ROPs, while the GTX 1660 SUPER has fewer shading units (1,408) and TMUs (88) but more ROPs (48). Despite having fewer cores, the GTX 1660 SUPER's Turing architecture delivers much higher raw throughput, with a FP32 performance of 5.027 TFLOPS versus the K10's 2.289 TFLOPS. The newer card also supports FP16 compute at 10.05 TFLOPS (2:1), a feature entirely absent from the Kepler-based K10.

Memory technology marks another major divergence. The Tesla K10 relies on GDDR5 memory running at 1250 MHz (5 Gbps effective), while the GTX 1660 SUPER uses faster GDDR6 at 1750 MHz (14 Gbps effective). This contributes to the GTX 1660 SUPER's 336.0 GB/s bandwidth, more than double the K10's 160.0 GB/s. The API support also differs, with the GTX 1660 SUPER supporting DirectX 12 (12_1) and Vulkan 1.4, whereas the Tesla K10 is limited to DirectX 12 (11_0) and Vulkan 1.2.175.

Head-to-Head Benchmarks

The only shared benchmark between these two GPUs is Geekbench OpenCL, and the results are lopsided. The GTX 1660 SUPER scores 52,490 points, while the Tesla K10 manages just 14,029 points. This yields a delta percentage of -73.3% for the older card, meaning the GTX 1660 SUPER is roughly 3.7 times faster in this compute workload. This massive gap can be attributed to the combination of higher clock speeds, superior memory bandwidth, and architectural efficiency of the Turing design.

The Tesla K10's score of 14,029 places it in the 55th percentile of all GPUs, while the GTX 1660 SUPER's average benchmark score of 12,986 sits slightly lower at the 53rd percentile. It's important to note that the GTX 1660 SUPER's average score is dragged down by its inclusion of multiple other benchmarks, including 3DMark Steel Nomad (1,278), Passmark DirectX 10 (65), and Passmark DirectX 9 (189). In the OpenCL test specifically, the GTX 1660 SUPER's advantage is overwhelming.

The nearest rivals for each card highlight their respective competitive positions. The Tesla K10 trades blows with the GeForce GTX 680 (0.9% behind) and the Radeon RX 570X (1.1% ahead), showing it was competitive with high-end cards of its generation. The GTX 1660 SUPER, meanwhile, sits in a similar performance tier relative to its own contemporaries, with the RTX 3050 Ti Mobile just 0.4% ahead and the Radeon RX 580 also 0.4% behind.

Specification Differences

The two cards differ across nearly every major specification category. The most striking difference is in memory: the Tesla K10 offers 4 GB of GDDR5 with a 256-bit bus, while the GTX 1660 SUPER provides 6 GB of GDDR6 on a 192-bit bus. The bandwidth advantage is substantial, with the newer card delivering 336.0 GB/s compared to 160.0 GB/s. Clock speeds also differ, with the GTX 1660 SUPER having a base clock of 1530 MHz and a boost clock of 1785 MHz, while the Tesla K10 has no listed base or boost clocks, only a memory clock of 1250 MHz.

The compute resources vary as well. The Tesla K10 has more shading units (1,536 vs. 1,408) and TMUs (128 vs. 88), but the GTX 1660 SUPER has more ROPs (48 vs. 32). This results in significantly higher fill rates for the newer card: 85.68 GPixel/s pixel rate and 157.1 GTexel/s texture rate, versus 23.84 GPixel/s and 95.36 GTexel/s for the K10. The FP32 throughput is also much higher on the GTX 1660 SUPER at 5.027 TFLOPS versus 2.289 TFLOPS.

Power and physical specifications show a clear generational efficiency gain. The Tesla K10 has a TDP of 225 W and requires a 550 W power supply, while the GTX 1660 SUPER uses only 125 W with a 300 W suggested PSU. The older card also needs two power connectors (1x 6-pin + 1x 8-pin), whereas the newer one uses a single 8-pin. Physically, the Tesla K10 is longer at 272 mm (10.7 inches) compared to the GTX 1660 SUPER's 229 mm (9 inches), though the newer card has defined height (111 mm) and width (35 mm) dimensions that the K10 lacks.

Display output is another differentiator: the Tesla K10 has no display outputs, being a compute-only accelerator, while the GTX 1660 SUPER includes 1x DVI, 1x HDMI 2.0, and 1x DisplayPort 1.4a. The Tesla K10's launch MSRP was 5,099 USD, while the GTX 1660 SUPER launched at 229 USD. The production status for both is end-of-life, with the Tesla K10 releasing on 2012-04-30 and the GTX 1660 SUPER on 2019-10-28.

Where Each One Wins

The GTX 1660 SUPER wins decisively in every measurable performance category from the data. Its OpenCL score is 73.3% higher than the Tesla K10's, and it offers more than double the memory bandwidth, nearly triple the pixel fill rate, and over double the FP32 compute throughput. This makes it the clear choice for general-purpose computing, gaming, or any workload that benefits from modern API support like DirectX 12 (12_1) and Vulkan 1.4. Its lower TDP of 125 W also makes it far more practical for a desktop system, requiring only a 300 W PSU and a single 8-pin connector.

The Tesla K10's only advantages are in areas that are largely obsolete for modern use. It has more shading units and TMUs, which could theoretically benefit certain legacy compute workloads that scale with core count rather than clock speed or memory bandwidth. Its 256-bit memory bus is wider than the GTX 1660 SUPER's 192-bit bus, though the newer card's GDDR6 memory more than compensates with over double the effective bandwidth. The K10's position in the 55th percentile of all GPUs versus the GTX 1660 SUPER's 53rd percentile is a statistical artifact of the different benchmark suites used, not an indication of real-world superiority.

For a system builder or researcher looking at these two cards today, the GTX 1660 SUPER is the only rational choice. It delivers superior performance in the one benchmark they share, supports modern rendering APIs, includes display outputs, and consumes significantly less power. The Tesla K10, with its lack of display outputs, higher power draw, and legacy architecture, is best suited for niche scenarios where its specific Kepler-era compute characteristics are required, such as maintaining compatibility with older scientific software stacks that were optimized for that generation.