NVIDIA GeForce GTX 1660 vs NVIDIA Tesla M2090 Comparison

The NVIDIA Tesla M2090 and the NVIDIA GeForce GTX 1660 represent two very different eras of GPU design. The M2090 is a compute-oriented accelerator built on the Fermi architecture, while the GTX 1660 is a consumer gaming card based on the Turing architecture. Their recorded benchmark data shows a clear performance gap, but each card has specific strengths that matter depending on the workload. Below is a breakdown of how they compare using only the measured results from the database.

Head-to-Head Benchmarks

The database contains a single direct comparison between these two cards: the Geekbench OpenCL test. In that test, the GTX 1660 scored 47,850 points, while the Tesla M2090 scored 13,075 points. The GTX 1660 is roughly 72.7% ahead of the M2090 in this OpenCL workload. That is a massive margin, indicating the GTX 1660 delivers more than three times the compute throughput in this specific benchmark.

Looking at the M2090’s rivals in the database, its OpenCL score of 13,075 places it very close to several other cards. The GeForce GTX 1660 SUPER scores 12,986, which is only 0.7% slower than the M2090. The GeForce GTX 950 scores 13,189, about 0.9% faster. The RTX 3050 Ti Mobile scores 12,940, about 1% slower, and the AMD Radeon RX 580 scores 12,928, about 1.1% slower. So the M2090’s OpenCL result sits right in the middle of a group of mid-range consumer cards from several generations. It is not an outlier on either side; it is competitive with those specific GPUs in this one test.

The GTX 1660’s average benchmark score across all recorded tests is 11,680. Its nearest rival in the database is the AMD Radeon RX 7800 XT, which scores 11,627, just 0.5% slower. The AMD Radeon Pro 5500M scores 11,528, about 1.3% slower, and the NVIDIA Tesla K20c scores 11,479, about 1.8% slower. On the other side, the AMD Radeon RX 6500 XT scores 11,842, about 1.4% faster. The GTX 1660’s average is thus tightly clustered with these other cards, meaning its overall performance level is comparable to a range of modern and semi-modern GPUs.

The M2090’s only recorded benchmark is the OpenCL test, so its average score equals that single result: 13,075. The GTX 1660 has ten recorded benchmarks, covering DirectX 9 through 12, OpenCL, Vulkan, 2D graphics, 3D graphics, and compute. That broader set of measurements means the GTX 1660 is validated across a wider range of application types, while the M2090 is only characterized for OpenCL compute.

Architecture Differences

The M2090 uses the GF110 chip, built on the Fermi 2.0 architecture, fabricated on a 40 nm process at TSMC. The chip contains 3,000 million transistors on a die size of 520 mm², giving a transistor density of 5.8 million per square millimeter. The GTX 1660 uses the TU116 chip, built on the Turing architecture, fabricated on a 12 nm process, also at TSMC. That chip contains 6,600 million transistors on a die size of 284 mm², resulting in a density of 23.2 million per square millimeter. The GTX 1660 packs more than twice the transistors into roughly half the die area, which reflects the much smaller manufacturing node.

The M2090 has 512 shading units, 64 texture mapping units, and 48 raster operation pipelines. The GTX 1660 has 1,408 shading units, 88 texture mapping units, and the same 48 ROPs. So the GTX 1660 has significantly more shader and texture hardware, while both cards have identical ROP counts. Neither card has ray tracing cores or tensor cores.

Memory configurations differ as well. Both cards have 6 GB of GDDR5 memory. The M2090 uses a 384-bit memory bus, giving it a bandwidth of 177.4 GB/s. The GTX 1660 uses a 192-bit bus, but its memory runs at a higher effective speed, yielding a bandwidth of 192.1 GB/s. Despite a narrower bus, the GTX 1660 achieves slightly higher bandwidth due to faster memory clocks.

The M2090’s memory clock is listed as 924 MHz, with an effective data rate of 3.7 Gbps. The GTX 1660’s memory clock is 2001 MHz, with an effective data rate of 8 Gbps. That is more than double the effective memory rate.

Pixel and texture rates tell a similar story. The M2090 reaches 20.83 gigapixels per second and 41.66 gigatexels per second. The GTX 1660 reaches 85.68 gigapixels per second and 157.1 gigatexels per second. The GTX 1660 is roughly four times faster in pixel fill and nearly four times faster in texture fill.

Floating point performance follows the same pattern. The M2090’s FP32 compute is 1,332.2 GFLOPS, while the GTX 1660 delivers 5.027 TFLOPS, which is 5,027 GFLOPS. The GTX 1660 also supports FP16 at 10.05 TFLOPS with a 2:1 ratio, while the M2090 has no recorded FP16 capability. That means the GTX 1660 can handle half-precision workloads at twice the rate of its FP32, something the M2090 cannot do.

Power consumption differs substantially. The M2090 has a TDP of 250 W, while the GTX 1660 has a TDP of 120 W. The M2090 requires a 600 W suggested power supply and uses one 6-pin plus one 8-pin connector. The GTX 1660 needs only a 300 W power supply and a single 8-pin connector. That is a major efficiency advantage for the GTX 1660, as it delivers far more performance while drawing less than half the power.

Bus interface also differs. The M2090 uses PCIe 2.0 x16, while the GTX 1660 uses PCIe 3.0 x16. The newer PCIe standard provides double the bandwidth per lane, which matters for data transfer to and from the CPU.

Display outputs are absent on the M2090; it has no video outputs at all. The GTX 1660 has one DVI, one HDMI 2.0, and one DisplayPort 1.4a. This is a fundamental difference: the M2090 is a compute accelerator designed for servers, while the GTX 1660 is a consumer graphics card meant to drive monitors.

API support also differs. The M2090 supports DirectX 12 (feature level 11_0) and OpenGL 4.6, but has no Vulkan support. The GTX 1660 supports DirectX 12 (feature level 12_1), OpenGL 4.6, and Vulkan 1.4. So the GTX 1660 has a more modern feature set for games and applications that rely on Vulkan.

Where Each One Wins

The GTX 1660 wins the only head-to-head benchmark available, and it wins by a very large margin. In OpenCL compute, it is about 72.7% faster than the M2090. That is the clear and unambiguous outcome of the direct comparison.

The M2090’s competitive position is best understood through its nearest rivals in the database. It is within 1.1% of the GeForce GTX 1660 SUPER, the GeForce GTX 950, the RTX 3050 Ti Mobile, and the AMD Radeon RX 580 in OpenCL. That means the M2090, despite being a much older architecture, holds its own against those specific GPUs in that one test. However, none of those rivals are modern high-end cards, and OpenCL is only one measure of compute capability.

For gaming and general graphics workloads, the GTX 1660 has a decisive advantage. It has higher pixel and texture rates, more shading units, faster memory, and support for modern APIs like Vulkan. The M2090 has no display outputs, so it cannot be used for gaming at all in a conventional setup. Its purpose is compute acceleration, not rendering frames to a screen.

For compute-heavy tasks that rely on OpenCL, the GTX 1660 still comes out ahead based on the direct benchmark. The M2090’s lack of FP16 support also limits its usefulness in workloads that can leverage half-precision arithmetic. The GTX 1660’s FP16 capability at 10.05 TFLOPS gives it an extra tool for certain scientific and machine learning tasks.

The M2090 does have one advantage in terms of memory bus width. Its 384-bit bus is wider than the GTX 1660’s 192-bit bus, but the GTX 1660 compensates with higher memory clocks. The net effect is that the GTX 1660 still has slightly higher bandwidth. So even in memory throughput, the newer card wins.

In terms of power efficiency, the GTX 1660 is clearly superior. It delivers roughly 3.7 times the FP32 compute while drawing less than half the power. The M2090’s 250 W TDP is high for its performance level, and the GTX 1660’s 120 W TDP makes it far easier to cool and integrate into a system.

The Verdict

The data points to a simple conclusion: the GTX 1660 is the stronger card in nearly every measurable way. It wins the only direct benchmark, has more shader and texture hardware, faster memory, higher fill rates, more compute throughput, lower power consumption, and modern API support. The M2090 is an older compute accelerator with no display outputs, a higher power draw, and a much lower OpenCL score.

For anyone building a system that needs to output video to a monitor, the M2090 is not an option. It has no video outputs. The GTX 1660, with its DVI, HDMI, and DisplayPort connections, is the only choice of these two for a standard desktop PC.

For compute workloads, the GTX 1660’s OpenCL result of 47,850 versus the M2090’s 13,075 is decisive. The GTX 1660 is more than three times faster in that test. Even considering that the M2090 is competitive with some mid-range consumer cards in OpenCL, it cannot match the GTX 1660.

The GTX 1660 also uses a 12 nm process with 6,600 million transistors, compared to the M2090’s 40 nm process with 3,000 million transistors. That process advantage translates into higher density and lower power draw. The GTX 1660’s 120 W TDP versus the M2090’s 250 W TDP is a major practical difference for system builders.

Who should pick the M2090? Only someone with a legacy compute environment that specifically requires a Fermi-based accelerator with a 384-bit memory bus and no need for video output. Its performance is in line with several older consumer GPUs, but it is far behind the GTX 1660.

Who should pick the GTX 1660? Anyone who wants a functional graphics card for gaming, general compute, or any workload that benefits from OpenCL, Vulkan, or DirectX 12. The GTX 1660 also supports FP16, which the M2090 does not. It is more power efficient, smaller, and has a much more recent release date.

The recorded data does not support any scenario where the M2090 outperforms the GTX 1660 in a head-to-head test. The only benchmark that pits them directly shows a 72.7% advantage for the GTX 1660. Unless a specific application is somehow optimized for Fermi and completely incompatible with Turing, the GTX 1660 is the superior choice.

FAQ

Q: Which card has a higher OpenCL benchmark score?

A: The GTX 1660 scored 47,850 in Geekbench OpenCL, while the M2090 scored 13,075. That is a 72.7% difference in favor of the GTX 1660.

Q: Do both cards support the same amount of memory?

A: Yes, both have 6 GB of GDDR5 memory. However, the GTX 1660 uses a 192-bit bus with 192.1 GB/s bandwidth, while the M2090 uses a 384-bit bus with 177.4 GB/s bandwidth.

Q: Can the M2090 be used for gaming with a monitor?

A: No. The M2090 has no display outputs. The GTX 1660 has one DVI, one HDMI 2.0, and one DisplayPort 1.4a.

Q: Which card has better compute performance in FP32?

A: The GTX 1660 delivers 5.027 TFLOPS of FP32, while the M2090 delivers 1,332.2 GFLOPS. The GTX 1660 is roughly four times faster in FP32.

Q: Does the M2090 support FP16 arithmetic?

A: No. The M2090 has no recorded FP16 capability. The GTX 1660 supports FP16 at 10.05 TFLOPS with a 2:1 ratio.

Q: Which card requires less power?

A: The GTX 1660 has a TDP of 120 W and a suggested power supply of 300 W. The M2090 has a TDP of 250 W and a suggested power supply of 600 W.

Specification Differences

| Specification | NVIDIA Tesla M2090 | NVIDIA GeForce GTX 1660 |

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

| Architecture | Fermi 2.0 | Turing |

| Chip | GF110 | TU116 |

| Process node | 40 nm | 12 nm |

| Transistors | 3,000 million | 6,600 million |

| Die size | 520 mm² | 284 mm² |

| Transistor density | 5.8M / mm² | 23.2M / mm² |

| Base clock | Not specified | 1530 MHz |

| Boost clock | Not specified | 1785 MHz |

| Memory clock | 924 MHz, 3.7 Gbps effective | 2001 MHz, 8 Gbps effective |

| Memory bus width | 384 bit | 192 bit |

| Memory bandwidth | 177.4 GB/s | 192.1 GB/s |

| Shading units | 512 | 1408 |

| Texture mapping units | 64 | 88 |

| Raster operation pipelines | 48 | 48 |

| FP32 performance | 1,332.2 GFLOPS | 5.027 TFLOPS |

| FP16 performance | Not specified | 10.05 TFLOPS (2:1) |

| Pixel rate | 20.83 GPixel/s | 85.68 GPixel/s |

| Texture rate | 41.66 GTexel/s | 157.1 GTexel/s |

| TDP | 250 W | 120 W |

| Suggested PSU | 600 W | 300 W |

| Power connectors | 1x 6-pin + 1x 8-pin | 1x 8-pin |

| Bus interface | PCIe 2.0 x16 | PCIe 3.0 x16 |

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

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

| Vulkan support | Not specified | 1.4 |

| Release date | July 2011 | March 2019 |

| Launch MSRP | Not specified | 219 USD |

| Length | 248 mm (9.8 inches) | 229 mm (9 inches) |

| Height | Not specified | 111 mm (4.4 inches) |

| Width | Not specified | 35 mm (1.4 inches) |