NVIDIA GeForce GTX 660 Ti vs NVIDIA Tesla M2090 Comparison

FAQ

Q: Which GPU has the higher average benchmark score?

A: The NVIDIA GeForce GTX 660 Ti records an average benchmark score of 15,063, while the NVIDIA Tesla M2090 records 13,075. The GTX 660 Ti leads by a margin of 15.6% in the single shared head-to-head test (Geekbench OpenCL).

Q: How does the GTX 660 Ti compare to its nearest rivals?

A: The GTX 660 Ti sits within 1.4% of four rivals. It is 0.1% behind the AMD Radeon Pro 560X (15,082), 0.7% behind the AMD Radeon RX 7600 (15,171), 0.9% behind the NVIDIA GeForce RTX 3050 OEM (15,199), and 1.4% behind the AMD Radeon 680M (15,270).

Q: Where does the Tesla M2090 rank among its closest competitors?

A: The Tesla M2090 is 0.7% ahead of the NVIDIA GeForce GTX 1660 SUPER (12,986), 0.9% behind the NVIDIA GeForce GTX 950 (13,189), 1.0% ahead of the NVIDIA GeForce RTX 3050 Ti Mobile (12,940), and 1.1% ahead of the AMD Radeon RX 580 (12,928).

Q: What is the memory capacity difference between the two cards?

A: The Tesla M2090 carries 6 GB of GDDR5 memory, triple the 2 GB found on the GTX 660 Ti. The Tesla also has a wider 384-bit bus versus 192-bit, yielding 177.4 GB/s bandwidth compared to 144.2 GB/s.

Q: Which card has more shading units and texture mapping units?

A: The GTX 660 Ti has 1,344 shading units and 112 TMUs. The Tesla M2090 has 512 shading units and 64 TMUs. The GTX 660 Ti more than doubles the shading unit count.

Q: Do both cards support the same API levels?

A: Both support DirectX 12 (11_0) and OpenGL 4.6. The GTX 660 Ti adds Vulkan 1.2.175 support, while the Tesla M2090 has no Vulkan entry in the database.

Architecture Differences

The two NVIDIA cards come from different architectural generations and process nodes. The GeForce GTX 660 Ti is built on the GK104 chip using the Kepler architecture, fabricated on a 28 nm process at TSMC. The Tesla M2090 uses the GF110 chip with the older Fermi 2.0 architecture, also from TSMC but on a 40 nm node. This process difference explains part of the performance and efficiency gap.

Transistor counts are close in absolute terms: the GK104 packs 3,540 million transistors on a 294 mm² die, while the GF110 houses 3,000 million transistors on a much larger 520 mm² die. Transistor density tells the story: 12.0M per mm² for Kepler versus 5.8M per mm² for Fermi 2.0. The newer process allows nearly double the density.

The compute resource allocation differs sharply. The GTX 660 Ti fields 1,344 shading units, 112 TMUs, and 24 ROPs. The Tesla M2090 has 512 shading units, 64 TMUs, and 48 ROPs. While the GTX 660 Ti dominates in shader and texture throughput, the Tesla has twice the ROP count, which affects pixel processing capabilities.

Clock behavior also diverges. The GTX 660 Ti has a base clock of 915 MHz and a boost clock of 980 MHz. The Tesla M2090 has no base or boost clock listed in the database, only a memory clock of 924 MHz (3.7 Gbps effective). The GTX 660 Ti memory runs at 1502 MHz (6 Gbps effective).

Memory subsystems reflect different design goals. The GTX 660 Ti uses a 192-bit bus with 2 GB GDDR5, achieving 144.2 GB/s. The Tesla M2090 uses a 384-bit bus with 6 GB GDDR5, achieving 177.4 GB/s. The Tesla prioritizes capacity and bandwidth for compute workloads, while the GTX 660 Ti balances capacity with lower power.

Power and interface specifications also differ. The GTX 660 Ti has a 150 W TDP, dual-slot cooler, and requires 2x 6-pin power connectors with a suggested 450 W PSU. The Tesla M2090 has a 250 W TDP, dual-slot cooler, requires 1x 6-pin plus 1x 8-pin connectors, and suggests a 600 W PSU. Bus interfaces differ as well: PCIe 3.0 x16 for the GTX 660 Ti versus PCIe 2.0 x16 for the Tesla.

Display outputs separate the two completely. The GTX 660 Ti offers 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2. The Tesla M2090 has no display outputs at all, reflecting its compute-only role. Physical dimensions are similar: 241 mm (9.5 inches) for the GTX 660 Ti versus 248 mm (9.8 inches) for the Tesla.

Head-to-Head Benchmarks

The database includes one direct comparison: Geekbench OpenCL. The GTX 660 Ti scores 15,113, while the Tesla M2090 scores 13,075. The GTX 660 Ti wins by 15.6%, a substantial margin that reflects the architectural advantages of Kepler over Fermi 2.0 in this workload.

Breaking down what drives this result, the GTX 660 Ti's shading unit advantage (1,344 versus 512) is the most obvious factor. OpenCL workloads often scale with shader count, and the 2.6x difference in shading units gives the GTX 660 Ti a structural edge. The GTX 660 Ti also posts higher FP32 throughput at 2.634 TFLOPS versus 1,332.2 GFLOPS for the Tesla, nearly double the raw floating-point rate.

Texture and pixel rates tell a mixed story. The GTX 660 Ti reaches 109.8 GTexel/s versus 41.66 GTexel/s for the Tesla, a 2.6x advantage. But the Tesla's pixel rate of 20.83 GPixel/s trails the GTX 660 Ti's 27.44 GPixel/s by a smaller margin, thanks to its 48 ROPs versus 24.

The Tesla M2090 does not lack merit. Its memory bandwidth of 177.4 GB/s exceeds the GTX 660 Ti's 144.2 GB/s by 23%. Its 6 GB frame buffer is three times larger, which matters for memory-bound compute tasks. Yet in the recorded Geekbench OpenCL test, these advantages did not overcome the GTX 660 Ti's compute throughput lead.

Percentile rankings align with the head-to-head result. The GTX 660 Ti sits at the 57th percentile of all GPUs in the database, while the Tesla M2090 sits at the 53rd percentile. The four-point gap is modest, but the direct benchmark comparison shows a clearer separation.

The nearest rival data provides context for each card's standing. The GTX 660 Ti's closest competitor, the AMD Radeon Pro 560X, scores 15,082, just 0.1% higher. The AMD Radeon RX 7600 at 15,171 is 0.7% higher, and the NVIDIA GeForce RTX 3050 OEM at 15,199 is 0.9% higher. The AMD Radeon 680M leads the group at 15,270, 1.4% above the GTX 660 Ti. These are all narrow gaps, suggesting the GTX 660 Ti performs in a tight cluster.

The Tesla M2090's rivals cluster similarly. The NVIDIA GeForce GTX 1660 SUPER scores 12,986, 0.7% below. The NVIDIA GeForce GTX 950 scores 13,189, 0.9% above. The NVIDIA GeForce RTX 3050 Ti Mobile scores 12,940, 1.0% below. The AMD Radeon RX 580 scores 12,928, 1.1% below. The Tesla sits near the middle of this group, slightly ahead of most but behind the GTX 950.

The Verdict

The data points to the GTX 660 Ti as the stronger performer in the shared benchmark. It wins the only head-to-head test by 15.6%, holds a higher average benchmark score (15,063 versus 13,075), and ranks higher in the overall GPU percentile (57th versus 53rd). For any workload measured by Geekbench OpenCL, the GTX 660 Ti is the clear choice.

The GTX 660 Ti's advantages are rooted in compute density. More shading units, more TMUs, higher clocks, and nearly double the FP32 throughput make it fundamentally faster for parallel compute tasks. Its smaller memory bus and lower bandwidth become secondary when the compute engine is this much faster.

The Tesla M2090 is not without purpose, but its strengths lie elsewhere. With 6 GB of memory and 177.4 GB/s of bandwidth, it offers more capacity and higher throughput for memory-intensive workloads. Its 48 ROPs also give it a pixel processing advantage relative to its compute throughput. However, in the recorded benchmark, these features did not translate into a higher score.

For users selecting a GPU strictly from this data, the GTX 660 Ti suits general compute and graphics workloads. It also provides display outputs, making it usable in a standard desktop setup. The Tesla M2090, with no display outputs and a higher 250 W TDP, targets compute-only environments where memory capacity is the priority.

The specification differences reinforce this split. The GTX 660 Ti is newer (August 2012 release versus July 2011), uses a more advanced 28 nm process, and supports PCIe 3.0. The Tesla M2090 uses the older 40 nm node, PCIe 2.0, and a larger die. The GTX 660 Ti's launch MSRP was 299 USD, while the Tesla M2090 has no recorded launch MSRP.

In the end, the GTX 660 Ti wins on measured performance. The Tesla M2090 offers more memory and bandwidth but loses decisively in the benchmark that matters here. Buyers who need compute throughput should choose the GTX 660 Ti. Those who specifically need 6 GB of memory and can accept lower compute performance may consider the Tesla M2090, but the data does not favor it.

Specification Differences

| Specification | NVIDIA GeForce GTX 660 Ti | NVIDIA Tesla M2090 |

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

| Architecture | Kepler | Fermi 2.0 |

| Chip | GK104 | GF110 |

| Process Node | 28 nm | 40 nm |

| Transistors | 3,540 million | 3,000 million |

| Die Size | 294 mm² | 520 mm² |

| Transistor Density | 12.0M / mm² | 5.8M / mm² |

| Base Clock | 915 MHz | Not listed |

| Boost Clock | 980 MHz | Not listed |

| Memory Clock | 1502 MHz, 6 Gbps effective | 924 MHz, 3.7 Gbps effective |

| Memory Size | 2 GB | 6 GB |

| Memory Bus Width | 192 bit | 384 bit |

| Memory Bandwidth | 144.2 GB/s | 177.4 GB/s |

| Shading Units | 1,344 | 512 |

| TMUs | 112 | 64 |

| ROPs | 24 | 48 |

| Pixel Rate | 27.44 GPixel/s | 20.83 GPixel/s |

| Texture Rate | 109.8 GTexel/s | 41.66 GTexel/s |

| FP32 | 2.634 TFLOPS | 1,332.2 GFLOPS |

| TDP | 150 W | 250 W |

| Power Connectors | 2x 6-pin | 1x 6-pin + 1x 8-pin |

| Suggested PSU | 450 W | 600 W |

| Bus Interface | PCIe 3.0 x16 | PCIe 2.0 x16 |

| Display Outputs | 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2 | No outputs |

| Vulkan Support | 1.2.175 | Not listed |

| Release Date | 2012-08-15 | 2011-07-24 |

| Predecessor | GeForce 500 | Tesla |

| Successor | GeForce 700 | Tesla Kepler |

| Launch MSRP | 299 USD | Not listed |

| Average Benchmark Score | 15,063 | 13,075 |

| Percentile vs All GPUs | 57 | 53 |