NVIDIA GeForce GTX 1660 SUPER vs NVIDIA Tesla M2090 Comparison

The NVIDIA Tesla M2090 and the NVIDIA GeForce GTX 1660 SUPER are both end-of-life products, but they represent opposite ends of the GPU design spectrum. The M2090 is a 2011-era compute accelerator built on the Fermi architecture, while the 1660 SUPER is a 2019 consumer card based on Turing. The data shows that despite being separated by eight years of architectural evolution, their average benchmark scores are remarkably close, with the M2090 scoring 13,075 and the 1660 SUPER scoring 12,986. That is a difference of only 0.7%, placing them in the same performance tier in the 53rd percentile of all GPUs. However, this surface-level similarity masks deep differences in features, power efficiency, and raw compute capabilities. The GTX 1660 SUPER is a far more practical choice for any modern workload, while the Tesla M2090 is a historical artifact with niche appeal.

The Verdict

The GTX 1660 SUPER is the clear winner for virtually every use case. Its Geekbench OpenCL score of 52,490 is 301.6% higher than the M2090’s 13,075, and it delivers this performance with a 125 W TDP compared to the M2090’s 250 W. The 1660 SUPER also brings modern features like Vulkan 1.4 support, DirectX 12 (12_1), and display outputs, making it a functional graphics card. The M2090, by contrast, has no display outputs, no Vulkan support, and a DirectX 12 (11_0) feature level, rendering it useless for gaming or general desktop use. The only scenario where the M2090 makes sense is if you are building a period-correct compute server for legacy Fermi software, but even then, its single benchmark win (0 vs. 1) shows it is outclassed. The verdict is simple: pick the 1660 SUPER for anything that requires actual graphics output, gaming, or modern API support. Pick the M2090 only if you are a collector or need a specific Fermi compute feature that no other card offers.

FAQ

Q: Which GPU has a higher average benchmark score?

A: The NVIDIA Tesla M2090 has an average score of 13,075, which is 0.7% higher than the GTX 1660 SUPER’s 12,986. However, this is based on a single Geekbench OpenCL test for the M2090, while the 1660 SUPER’s average is derived from ten different benchmarks.

Q: How does the GTX 1660 SUPER perform in raw compute compared to the M2090?

A: The data shows a massive gap. The GTX 1660 SUPER scores 52,490 in Geekbench OpenCL, while the M2090 scores 13,075. This represents a 75.1% deficit for the M2090 in this specific head-to-head test.

Q: What is the power consumption difference between the two?

A: The GTX 1660 SUPER has a TDP of 125 W and requires a 300 W power supply, while the Tesla M2090 has a TDP of 250 W and requires a 600 W power supply. The M2090 uses double the power for significantly less performance.

Q: Do both cards support the same graphics APIs?

A: No. The GTX 1660 SUPER supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4. The Tesla M2090 supports DirectX 12 (11_0) and OpenGL 4.6 but has no Vulkan support.

Q: Which card has more memory bandwidth?

A: The GTX 1660 SUPER has a memory bandwidth of 336.0 GB/s, which is nearly double the M2090’s 177.4 GB/s. Both have 6 GB of memory, but the 1660 SUPER uses GDDR6 on a 192-bit bus, while the M2090 uses GDDR5 on a 384-bit bus.

Q: Are there any physical size differences?

A: The M2090 is longer at 248 mm (9.8 inches) compared to the 1660 SUPER’s 229 mm (9 inches). The 1660 SUPER also has defined height (111 mm) and width (35 mm) dimensions, while the M2090’s height and width are not listed. Both are dual-slot cards.

Architecture Differences

The architectural divide between these two GPUs is enormous. The Tesla M2090 uses the GF110 chip based on the Fermi 2.0 architecture, fabricated on a 40 nm process at TSMC. This process node is a full generation behind, resulting in a die size of 520 mm² and a transistor count of 3,000 million, yielding a transistor density of only 5.8 million per mm². The GTX 1660 SUPER, in contrast, uses the TU116 chip based on the Turing architecture, fabricated on a 12 nm process at the same foundry. This modern node allows for 6,600 million transistors on a smaller 284 mm² die, achieving a density of 23.2 million per mm². That is a fourfold increase in transistor density. The M2090 is a compute-oriented card with no display outputs, while the 1660 SUPER is a full-featured consumer card with 1x DVI, 1x HDMI 2.0, and 1x DisplayPort 1.4a outputs. The M2090 also lacks Vulkan support entirely, whereas the 1660 SUPER has Vulkan 1.4. The feature set, node, and design philosophy are completely different, with the Turing architecture being far more advanced in every measurable way.

Specification Differences

The specifications diverge significantly across nearly every field. The M2090 has 512 shading units, 64 texture mapping units (TMUs), and 48 raster operation units (ROPs). The 1660 SUPER has 1408 shading units, 88 TMUs, and 48 ROPs. The shading unit count is nearly triple on the newer card. Memory configurations show both have 6 GB, but the M2090 uses GDDR5 on a 384-bit bus with 177.4 GB/s bandwidth, while the 1660 SUPER uses GDDR6 on a 192-bit bus with 336.0 GB/s bandwidth. The memory clock is also different: the M2090 runs at 924 MHz (3.7 Gbps effective), while the 1660 SUPER runs at 1750 MHz (14 Gbps effective). The M2090 has no base or boost clock listed, while the 1660 SUPER has a base clock of 1530 MHz and a boost clock of 1785 MHz. Pixel rate is 20.83 GPixel/s for the M2090 versus 85.68 GPixel/s for the 1660 SUPER. Texture rate is 41.66 GTexel/s versus 157.1 GTexel/s. FP32 performance is 1,332.2 GFLOPS for the M2090 versus 5.027 TFLOPS for the 1660 SUPER. The 1660 SUPER also lists FP16 performance of 10.05 TFLOPS (2:1), which the M2090 does not have. Power requirements are starkly different: 250 W TDP with 1x 6-pin + 1x 8-pin connectors and a 600 W PSU recommendation for the M2090, versus 125 W TDP with a single 8-pin connector and a 300 W PSU for the 1660 SUPER. The bus interface is PCIe 2.0 x16 on the M2090 and PCIe 3.0 x16 on the 1660 SUPER. The M2090 is longer at 248 mm versus 229 mm for the 1660 SUPER.

Head-to-Head Benchmarks

The head-to-head data is sparse but decisive. The only direct comparison available is the Geekbench OpenCL test, where the GTX 1660 SUPER scores 52,490 and the Tesla M2090 scores 13,075. This gives the 1660 SUPER a win with a delta of -75.1% from the M2090’s perspective, meaning the M2090 is 75.1% slower. In practical terms, the 1660 SUPER is performing at roughly 4x the level of the M2090 in this compute workload. When looking at the broader benchmark suite, the 1660 SUPER has a diverse set of scores: 12,699 in Passmark G3D, 5,076 in Passmark GPU Compute, and 57,102 in Geekbench Vulkan. The M2090 only has the single OpenCL score of 13,075. This lack of data for the M2090 in other tests like DirectX 11 or Vulkan is telling—it simply cannot run those modern workloads. The 1660 SUPER’s Passmark DirectX 11 score of 104 and DirectX 12 score of 50 further demonstrate its capability in graphics APIs that the M2090 cannot handle. The data clearly shows that in the one test where both can be measured, the 1660 SUPER dominates, and in all other metrics, the M2090 is absent or inferior.

Where Each One Wins

The GTX 1660 SUPER wins in every category where a comparison is possible. It wins the only head-to-head benchmark by a massive margin. It wins on power efficiency, using half the TDP (125 W vs. 250 W). It wins on memory bandwidth (336.0 GB/s vs. 177.4 GB/s). It wins on raw compute with 5.027 TFLOPS FP32 versus 1,332.2 GFLOPS. It wins on pixel rate (85.68 GPixel/s vs. 20.83 GPixel/s) and texture rate (157.1 GTexel/s vs. 41.66 GTexel/s). It wins on API support with Vulkan 1.4 and DirectX 12 (12_1) versus no Vulkan and DirectX 12 (11_0). It also has display outputs, making it a functional graphics card, whereas the M2090 has none. The M2090’s only theoretical advantages are its wider 384-bit memory bus and larger die size, but these do not translate to performance wins. In terms of percentile, both sit at 53, but the 1660 SUPER achieves this with a modern feature set. The M2090’s single benchmark score of 13,075 places it near rivals like the GTX 950 (13,189) and the RTX 3050 Ti Mobile (12,940), but it lacks the versatility of those cards. The 1660 SUPER, with its nearest rivals being the RTX 3050 Ti Mobile (12,940) and RX 580 (12,928), is a far more future-proof option. For any workload involving modern graphics APIs, gaming, or general compute, the 1660 SUPER is the only rational choice. The M2090 wins only in the niche sense of being a historical Fermi compute accelerator with a unique architectural profile that no modern card replicates.