NVIDIA GeForce GTX 1650 SUPER vs NVIDIA Tesla M2090 Comparison

The recorded data frames this matchup as a generational collision rather than a conventional rivalry: a 2011 Fermi-based compute card against a 2019 Turing-based gaming card, both now end-of-life. On the only benchmark the database holds for both parts, Geekbench OpenCL, the GeForce GTX 1650 SUPER scores 43875 against the Tesla M2090's 13075, a gap of 70.2 percent in favor of the newer card. That single shared result, combined with the architectural record, makes the split straightforward to call.

The Verdict

For any workload represented in the database, the GTX 1650 SUPER is the pick. It wins the only common benchmark outright, and it does so while drawing 100 W against the Tesla's 250 W, needing only a single 6-pin connector instead of a 6-pin plus an 8-pin, and running on a 300 W suggested PSU versus the Tesla's 600 W. It is also the only one of the two that can drive a display at all, since the M2090 ships with no display outputs.

The M2090's case rests on its compute heritage: a wide 384-bit memory bus, 6 GB of GDDR5, and a design purpose-built for datacenter workloads as part of the Tesla Fermi line. But nothing in the recorded benchmark data supports choosing it over the 1650 SUPER today. Its OpenCL score sits in the 53rd percentile of all GPUs in the database, and its closest rivals by average score are the GTX 1660 SUPER, GTX 950, RTX 3050 Ti Mobile, and Radeon RX 580, all clustered within roughly one percent of it. The 1650 SUPER, in the 50th percentile, is surrounded by the Radeon RX 550, two RTX PRO 6000 Blackwell Max-Q variants, and the GeForce MX350, again all within about 1.5 percent. Percentile placement is nearly identical, yet the head-to-head Geekbench result breaks decisively toward the Turing card.

Architecture Differences

The two cards come from opposite ends of a decade of NVIDIA silicon. The M2090 uses the GF110 chip on the Fermi 2.0 architecture, fabricated at TSMC on a 40 nm process. It packs 3,000 million transistors into a 520 mm² die, yielding a density of 5.8 million transistors per square millimeter. The GTX 1650 SUPER uses the TU116 chip on the Turing architecture, built at TSMC on a 12 nm process: 6,600 million transistors in a 284 mm² die at 23.2 million transistors per square millimeter. That is roughly four times the transistor density on a die less than half the area, with more than double the total transistor count, and it is the clearest single illustration of the generational gap in this pairing.

Core counts tell a subtler story. The M2090 carries 512 shading units, 64 TMUs, and 48 ROPs. The 1650 SUPER has 1280 shading units, 80 TMUs, and 32 ROPs. The Tesla's ROP lead is modest, but its throughput deficit elsewhere is not: the 1650 SUPER delivers a pixel rate of 55.20 GPixel/s versus 20.83 for the M2090, and a texture rate of 138.0 GTexel/s versus 41.66, advantages of roughly 2.6x and 3.3x respectively. Floating point follows the same pattern. The M2090 produces 1332.2 GFLOPS of FP32; the 1650 SUPER produces 4.416 TFLOPS of FP32 and 8.832 TFLOPS of FP16 at a 2:1 rate, a figure the Fermi card simply has no recorded equivalent for.

Memory design is where the M2090 holds its ground. It pairs 6 GB of GDDR5 with a 384-bit bus for 177.4 GB/s of bandwidth, with memory running at 924 MHz (3.7 Gbps effective). The 1650 SUPER counters with 4 GB of faster GDDR6 on a narrower 128-bit bus, clocked at 1500 MHz (12 Gbps effective), for 192.0 GB/s. The Turing card wins on raw bandwidth, but the Tesla wins on capacity, which mattered for its original compute role. Neither card has RT cores or tensor cores.

The platform and feature split is equally lopsided. The M2090 connects over PCIe 2.0 x16 and reports no Vulkan support, listing DirectX 12 (11_0) feature level and OpenGL 4.6. The 1650 SUPER uses PCIe 3.0 x16, supports Vulkan 1.4 and DirectX 12 (12_1), and offers DVI, HDMI 2.0, and DisplayPort 1.4a outputs. Physically, both are dual-slot cards, but the Tesla is longer at 248 mm against 229 mm, and only the 1650 SUPER has recorded height and width figures (111 mm and 35 mm).

Where Each One Wins

The 1650 SUPER wins every recorded measurable contest. It takes the lone head-to-head benchmark, Geekbench OpenCL, 43875 to 13075. Beyond that shared test, the database holds ten benchmark results for the 1650 SUPER and only one for the M2090, so the newer card dominates by sheer coverage: 3DMark Steel Nomad DX12 at 352, Geekbench Vulkan at 50519, Passmark G3D at 10179, Passmark GPU Compute at 4477, Passmark G2D at 749, and Passmark DirectX results of 148 (9), 50 (10), 73 (11), and 45 (12). The M2090 contributes nothing comparable outside OpenCL.

Where the M2090 wins is on paper, not in measurements: memory capacity, at 6 GB versus 4 GB, and bus width, at 384 bit versus 128 bit. These attributes defined its role as a Tesla-series accelerator, the successor to the original Tesla line and the predecessor to Tesla Kepler. For tasks whose only recorded proxy is Geekbench OpenCL, even that theoretical positioning collapses: the Fermi card trails by 70.2 percent.

Power is the other axis of "winning." The 1650 SUPER's 100 W board power against the M2090's 250 W means that in any efficiency-sensitive deployment, the newer card delivers substantially more measured compute per watt. Its single 6-pin connector and 300 W PSU recommendation make it installable in far more systems than the Tesla, which requires both a 6-pin and an 8-pin feed and a 600 W PSU.

FAQ

Q: Which card is faster in the only benchmark both have recorded?

A: The GTX 1650 SUPER, by a wide margin. It scores 43875 in Geekbench OpenCL versus 13075 for the Tesla M2090, a 70.2 percent advantage.

Q: Which card has more memory?

A: The Tesla M2090, with 6 GB of GDDR5 on a 384-bit bus. The 1650 SUPER has 4 GB of GDDR6 on a 128-bit bus.

Q: Which card has higher memory bandwidth?

A: The GTX 1650 SUPER, at 192.0 GB/s versus 177.4 GB/s, thanks to 12 Gbps effective GDDR6 despite the narrower bus.

Q: Can either card drive a monitor?

A: Only the GTX 1650 SUPER. It provides one DVI, one HDMI 2.0, and one DisplayPort 1.4a output. The M2090 has no display outputs, consistent with its compute-focused design.

Q: How do their power requirements compare?

A: The M2090 is rated at 250 W with a 6-pin plus 8-pin connector pair and a 600 W suggested PSU. The 1650 SUPER is rated at 100 W, uses one 6-pin connector, and suggests a 300 W PSU.

Q: How do the two compare in FP32 compute?

A: The 1650 SUPER reaches 4.416 TFLOPS FP32 and 8.832 TFLOPS FP16 (2:1). The M2090 records 1332.2 GFLOPS FP32, with no FP16 figure listed.

Head-to-Head Benchmarks

The database contains exactly one benchmark in which both cards appear: Geekbench OpenCL. The GTX 1650 SUPER posts 43875; the Tesla M2090 posts 13075. The delta is 70.2 percent in the newer card's favor, which converts to roughly 3.4 times the OpenCL throughput. For context, that result puts the M2090 in the same measured neighborhood as the GTX 1660 SUPER, GTX 950, RTX 3050 Ti Mobile, and Radeon RX 580, each within about one percent of its score, while the 1650 SUPER's average score of 11047 across its full benchmark set places it alongside the Radeon RX 550, the RTX PRO 6000 Blackwell Max-Q variants, and the GeForce MX350.

The OpenCL gap tracks the raw hardware numbers almost perfectly. The 1650 SUPER's FP32 output is about 3.3 times the M2090's, its texture rate is 3.3 times higher, and its pixel rate is 2.6 times higher, so a roughly 3.4x OpenCL result should surprise no one who reads the spec sheet. The M2090's structural advantages, capacity and bus width, do not translate into bandwidth leadership, since its 177.4 GB/s still trails the 1650 SUPER's 192.0 GB/s.

Texture fill comparisons underscore the same story: 80 TMUs running at Turing clock speeds deliver 138.0 GTexel/s against 41.66 GTexel/s from 64 Fermi TMUs. ROPs are the single count the M2090 wins, 48 to 32, yet its pixel rate of 20.83 GPixel/s falls well short of the 1650 SUPER's 55.20 GPixel/s because clock speeds dominate: the 1650 SUPER lists a 1530 MHz base and 1725 MHz boost, while no core clock figures are recorded for the M2090.

The final tally in the recorded data is one benchmark, one winner. The GTX 1650 SUPER takes Geekbench OpenCL, and the Tesla M2090 takes none. Combined with lower power draw, modern API support including Vulkan 1.4 and DirectX 12 (12_1), PCIe 3.0 connectivity, and working display outputs, the data presents a clean sweep for the Turing card, with the M2090's 6 GB frame buffer standing as its only unresolved edge in the record.