NVIDIA GeForce GTX 660 Ti vs NVIDIA Tesla K20Xm Comparison

Head-to-Head Benchmarks

The database records a single direct comparison between these two NVIDIA Kepler cards: Geekbench OpenCL. In that test, the Tesla K20Xm scores 17,215 points, while the GeForce GTX 660 Ti scores 15,113 points. That gives the Tesla a 13.9% lead in raw compute throughput, a decisive margin for a workload that scales across the full GPU.

Looking at overall average benchmark scores, the GTX 660 Ti actually lands higher in the database: 15,063 versus 12,625 for the Tesla K20Xm. This difference of 2,438 points, roughly 19.3%, reflects how each card is positioned. The GTX 660 Ti benefits from a wider set of tested scenarios, many of which favor its driver optimization and consumer-oriented feature set. The Tesla K20Xm, meanwhile, scores 8,035 in Geekbench Metal, a test the GTX 660 Ti does not appear in, and 17,215 in OpenCL, where it beats the GTX 660 Ti by 2,102 points.

In percentile terms, the GTX 660 Ti sits at the 57th percentile of all GPUs in the database, while the Tesla K20Xm sits at the 52nd percentile. That is a modest but consistent gap. The nearest rival to the Tesla K20Xm, the AMD Radeon RX 7600M XT, averages 12,710 points, just 0.7% higher. The Tesla is also 1.2% behind the GeForce GTX 670 and 1.6% behind both the GeForce GTX 590 and the AMD Radeon Pro 455. For the GTX 660 Ti, its closest competitor is the AMD Radeon Pro 560X at 15,082 points, only 0.1% ahead. The AMD Radeon RX 7600 trails by 0.7%, the RTX 3050 OEM by 0.9%, and the AMD Radeon 680M by 1.4%.

The head-to-head record is one win for the Tesla K20Xm and zero for the GTX 660 Ti, but the broader average score tells a more nuanced story. The Tesla wins the single compute-heavy benchmark, while the GTX 660 Ti holds a higher aggregate across all recorded tests. Neither card dominates outright; the data indicates a split depending on workload type.

Architecture Differences

Both GPUs use NVIDIA's Kepler architecture, fabricated on TSMC's 28 nm process. The similarity ends there. The Tesla K20Xm uses the GK110 chip, a large compute-oriented design with 7,080 million transistors on a 561 mm² die. The GeForce GTX 660 Ti uses the GK104 chip, a smaller graphics-focused design with 3,540 million transistors on a 294 mm² die. The transistor density is nearly identical, 12.6M per mm² for the Tesla and 12.0M per mm² for the GTX 660 Ti, reflecting the same manufacturing generation.

The compute resources differ substantially. The Tesla K20Xm packs 2,688 shading units, 224 texture mapping units, and 48 ROPs. The GTX 660 Ti has 1,344 shading units, 112 TMUs, and 24 ROPs, exactly half in each category. This doubling of execution resources is the primary reason the Tesla leads in OpenCL. Pixel rate measures 40.99 GPixel/s for the Tesla versus 27.44 GPixel/s for the GTX 660 Ti. Texture rate is 164.0 GTexel/s versus 109.8 GTexel/s. Floating-point performance, FP32, reaches 3.935 TFLOPS on the Tesla and 2.634 TFLOPS on the GTX 660 Ti, a 49.4% advantage for the Tesla.

Memory configurations also diverge. The Tesla K20Xm carries 6 GB of GDDR5 on a 384-bit bus, yielding 249.6 GB/s of bandwidth. The GTX 660 Ti has 2 GB of GDDR5 on a 192-bit bus, producing 144.2 GB/s. The Tesla's memory clock is listed at 1300 MHz with 5.2 Gbps effective, while the GTX 660 Ti runs at 1502 MHz with 6.0 Gbps effective. Despite the higher clock on the GTX 660 Ti, the Tesla's wider bus delivers 73.1% more bandwidth.

The GTX 660 Ti includes base and boost clocks of 915 MHz and 980 MHz, respectively. The Tesla K20Xm's base and boost clocks are not recorded in the database, so no direct clock comparison is possible. The Tesla compensates with its larger chip and higher shader count.

Power draw differs as well. The Tesla K20Xm has a 235 W TDP, while the GTX 660 Ti uses 150 W. The suggested power supply is 550 W for the Tesla and 450 W for the GTX 660 Ti. Physical size also varies: the Tesla is 267 mm long (10.5 inches), and the GTX 660 Ti is 241 mm (9.5 inches). Both are dual-slot cards. The GTX 660 Ti has two 6-pin power connectors, while the Tesla's power connector configuration is not recorded. Display outputs are absent on the Tesla K20Xm, as it is a compute accelerator, whereas the GTX 660 Ti provides 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2.

Where Each One Wins

The Tesla K20Xm wins in raw compute throughput. Its OpenCL score of 17,215 is 13.9% higher than the GTX 660 Ti's 15,113. For workloads that stress FP32 arithmetic, texture fill, or memory bandwidth, the Tesla's doubled shader count, wider memory bus, and larger frame buffer make it the stronger choice. The 6 GB of memory is particularly useful for large datasets that exceed the GTX 660 Ti's 2 GB capacity. The Tesla also has a Metal score of 8,035, a test not recorded for the GTX 660 Ti, indicating it can handle Apple's compute and graphics API when configured with the appropriate drivers.

The GeForce GTX 660 Ti wins in overall benchmark average, 15,063 versus 12,625, and in percentile rank, 57th versus 52nd. This suggests that in the database's broader set of tests, which likely includes gaming-oriented workloads, the GTX 660 Ti delivers more consistent performance. Its higher memory clock, 6 Gbps effective versus 5.2 Gbps on the Tesla, helps in latency-sensitive scenarios. The GTX 660 Ti also has display outputs, making it a functional graphics card for direct visual output, while the Tesla K20Xm has no outputs and requires a secondary GPU for display.

For a user needing compute acceleration without display capability, the Tesla K20Xm is the clear winner in the single recorded head-to-head benchmark. For a user running a mix of graphics and compute, the GTX 660 Ti's higher average score and better percentile placement indicate broader utility. The Tesla's 235 W TDP versus 150 W for the GTX 660 Ti also means the GTX 660 Ti is easier to integrate into a standard desktop, requiring only a 450 W power supply versus 550 W.

The data does not show any test where the GTX 660 Ti beats the Tesla directly, as the only head-to-head test is OpenCL. However, the average score gap implies that the GTX 660 Ti wins across multiple unrecorded tests. The Tesla's win count of 1 versus 0 for the GTX 660 Ti in the head-to-head suite is offset by the GTX 660 Ti's higher aggregate performance.

FAQ

Q: Which GPU has a higher OpenCL score in the database?

A: The NVIDIA Tesla K20Xm scores 17,215 in Geekbench OpenCL, which is 13.9% higher than the GeForce GTX 660 Ti's score of 15,113.

Q: How does the average benchmark score compare between the two cards?

A: The GeForce GTX 660 Ti averages 15,063 points, while the Tesla K20Xm averages 12,625 points. The GTX 660 Ti sits at the 57th percentile of all GPUs, and the Tesla sits at the 52nd.

Q: What is the memory size and bandwidth difference?

A: The Tesla K20Xm has 6 GB of GDDR5 memory with 249.6 GB/s bandwidth on a 384-bit bus. The GTX 660 Ti has 2 GB of GDDR5 with 144.2 GB/s bandwidth on a 192-bit bus.

Q: Which card has more shading units?

A: The Tesla K20Xm has 2,688 shading units, exactly double the 1,344 shading units found on the GeForce GTX 660 Ti.

Q: Can the Tesla K20Xm output video to a display?

A: No, the Tesla K20Xm has no display outputs. The GeForce GTX 660 Ti provides 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2.

Q: What are the power requirements listed in the database?

A: The Tesla K20Xm has a TDP of 235 W and a suggested power supply of 550 W. The GTX 660 Ti has a TDP of 150 W and a suggested power supply of 450 W.

Specification Differences

The two cards differ in every major specification category. The chip is GK110 for the Tesla K20Xm and GK104 for the GTX 660 Ti. Transistor count is 7,080 million versus 3,540 million. Die size is 561 mm² versus 294 mm². Transistor density is 12.6M per mm² versus 12.0M per mm².

Memory clock differs: the Tesla runs at 1300 MHz (5.2 Gbps effective), the GTX 660 Ti at 1502 MHz (6.0 Gbps effective). Memory size is 6 GB versus 2 GB, bus width is 384 bit versus 192 bit, and bandwidth is 249.6 GB/s versus 144.2 GB/s.

Shading units are 2,688 versus 1,344. TMUs are 224 versus 112. ROPs are 48 versus 24. Pixel rate is 40.99 GPixel/s versus 27.44 GPixel/s. Texture rate is 164.0 GTexel/s versus 109.8 GTexel/s. FP32 performance is 3.935 TFLOPS versus 2.634 TFLOPS.

The GTX 660 Ti has base and boost clocks of 915 MHz and 980 MHz, while the Tesla K20Xm has no recorded base or boost clock. TDP is 235 W versus 150 W. The Tesla has no power connector data, while the GTX 660 Ti uses 2x 6-pin. Suggested PSU is 550 W versus 450 W. Length is 267 mm versus 241 mm. Display outputs: none versus 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2.

Both cards use the same API support: DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. Both are end-of-life products. The Tesla K20Xm launched on 2012-11-11 with a launch MSRP of 7,699 USD. The GTX 660 Ti launched on 2012-08-15 with a launch MSRP of 299 USD. The Tesla's generation is "Tesla Kepler (Kxx)" and its predecessor is Tesla Fermi, successor Tesla Maxwell. The GTX 660 Ti's generation is "GeForce 600", predecessor GeForce 500, successor GeForce 700.