NVIDIA GeForce GTX 680 vs NVIDIA Tesla M2090 Comparison

The NVIDIA GeForce GTX 680 and NVIDIA Tesla M2090 represent two distinct approaches to graphics hardware from the same vendor, separated by a generation of architecture and targeting different workloads. The GTX 680 is a consumer-oriented Kepler part, while the M2090 is a compute-focused Fermi 2.0 accelerator. The data reveals a clear performance hierarchy in the one shared benchmark, but the specifications tell a story of divergent design priorities.

Head-to-Head Benchmarks

The only directly comparable benchmark result between the two cards is Geekbench OpenCL, where the GTX 680 posts a score of 16906 against the M2090’s 13075. The deltaPct of 29.3% indicates the GTX 680 is approximately 29% faster in this compute workload. This is a substantial margin, positioning the consumer card as the more capable compute performer in raw OpenCL terms. The GTX 680’s score also contributes to an average benchmark score of 14150, while the M2090’s single benchmark yields an average of 13075.

Looking at the broader competitive landscape, the GTX 680’s average score of 14150 places it in a tight cluster of rivals. It sits 0.9% above the NVIDIA Tesla K10 (14029) and 0.9% below the GeForce GTX 1070 Ti (14277). The Intel Iris Xe MAX Graphics (14315) and AMD Radeon Vega 11 (14352) are 1.2% and 1.4% ahead, respectively. The M2090’s average score of 13075 places it 0.7% above the GeForce GTX 1660 SUPER (12986) and 1.1% above the AMD Radeon RX 580 (12928), while the GeForce GTX 950 (13189) is 0.9% ahead and the RTX 3050 Ti Mobile (12940) is 1% behind.

The percentile rankings reinforce the gap: the GTX 680 sits at the 55th percentile of all GPUs, while the M2090 is at the 53rd. While both are mid-pack performers, the GTX 680’s higher percentile and the 29.3% OpenCL delta make the performance hierarchy unambiguous. In every measurable compute scenario provided, the GTX 680 wins outright, with the M2090 never claiming a single benchmark victory.

Where Each One Wins

The GTX 680 wins in the sole head-to-head compute benchmark, but its advantages extend across several key metrics that suggest different use cases. With a pixel rate of 33.86 GPixel/s and a texture rate of 135.4 GTexel/s, the GTX 680 is clearly optimized for rasterization-heavy graphics workloads. Its 1536 shading units and 128 texture mapping units provide the parallel throughput needed for gaming and real-time rendering. The card also supports Vulkan 1.2.175, giving it access to modern cross-platform graphics APIs.

The M2090, by contrast, is a compute-oriented accelerator with no display outputs. Its lower pixel rate of 20.83 GPixel/s and texture rate of 41.66 GTexel/s indicate a design focused on data-parallel tasks rather than graphics output. The card’s 512 shading units and 64 TMUs are fewer, but the 48 ROPs and 6 GB of memory suggest a configuration tailored for large datasets and memory-intensive computations. In terms of raw FP32 throughput, the GTX 680 delivers 3.250 TFLOPS, which is roughly 2.4 times the M2090’s 1,332.2 GFLOPS.

For gaming and consumer graphics, the GTX 680 is the clear choice due to its higher pixel and texture rates, plus its display outputs (2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2). For scientific computing or server-side workloads where display output is unnecessary, the M2090’s larger 6 GB memory pool and PCIe 2.0 x16 interface may be the deciding factor, despite its lower compute performance. The M2090’s higher TDP of 250 W versus 195 W also suggests it is built for sustained compute loads rather than efficiency.

Architecture Differences

The architectural divide is stark. The GTX 680 uses the GK104 chip built on a 28 nm process at TSMC, while the M2090 uses the GF110 chip on a 40 nm process, also at TSMC. This process difference is significant: the GTX 680 packs 3,540 million transistors into a 294 mm² die, yielding a transistor density of 12.0M / mm². The M2090, despite having fewer transistors (3,000 million), occupies a much larger 520 mm² die, resulting in a density of just 5.8M / mm². The newer 28 nm node allows the GTX 680 to nearly double the transistor density.

The GTX 680 belongs to the Kepler architecture and the GeForce 600 generation, while the M2090 is based on Fermi 2.0 and part of the Tesla Fermi (x20xx) generation. Clock speeds differ substantially: the GTX 680 has a base clock of 1006 MHz and a boost clock of 1058 MHz, whereas the M2090 has no listed base or boost clocks. Memory clocks also diverge, with the GTX 680 running at 1502 MHz (6 Gbps effective) and the M2090 at 924 MHz (3.7 Gbps effective).

Memory configurations reflect their different missions. The GTX 680 has 2 GB of GDDR5 on a 256-bit bus, delivering 192.3 GB/s of bandwidth. The M2090 offers 6 GB of GDDR5 on a wider 384-bit bus, but achieves slightly lower bandwidth at 177.4 GB/s due to its slower memory clock. The M2090’s larger capacity is suited for datasets that exceed the GTX 680’s 2 GB limit, even though the bandwidth is marginally lower.

Feature support also separates them. The GTX 680 supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. The M2090 also supports DirectX 12 (11_0) and OpenGL 4.6, but has no Vulkan support. The GTX 680 uses a PCIe 3.0 x16 interface, while the M2090 is limited to PCIe 2.0 x16, halving the potential bus bandwidth. Power delivery differs as well: the GTX 680 uses 2x 6-pin connectors and a 450 W suggested PSU, while the M2090 requires 1x 6-pin + 1x 8-pin and a 600 W PSU.

The Verdict

From the data, the NVIDIA GeForce GTX 680 is the superior performer in every benchmark provided, with a 29.3% lead in OpenCL and a higher percentile ranking (55th vs 53rd). For users prioritizing compute performance, graphics capability, or modern API support, the GTX 680 is the clear pick. Its higher FP32 throughput, faster pixel and texture rates, and support for Vulkan make it more versatile across gaming and general-purpose GPU workloads.

The NVIDIA Tesla M2090, while slower in compute benchmarks, offers advantages in memory capacity (6 GB vs 2 GB) and a wider memory bus (384-bit vs 256-bit), albeit with lower bandwidth. Its lack of display outputs and absence of Vulkan support narrow its use case to headless compute servers or applications that specifically require large memory allocations. The M2090’s higher TDP (250 W) and larger die (520 mm²) suggest it was designed for sustained throughput in professional settings, but the benchmark data does not show any performance scenario where it outpaces the GTX 680.

For a consumer or prosumer seeking a single card for both graphics and compute, the GTX 680 is the data-backed choice. For a specialized compute node where 6 GB of memory is mandatory and Vulkan is irrelevant, the M2090 remains a viable, if slower, option. The benchmark results do not support any scenario where the M2090 wins on speed.

FAQ

Q: Which card has a higher average benchmark score?

A: The NVIDIA GeForce GTX 680 has an average benchmark score of 14150, while the NVIDIA Tesla M2090 has an average of 13075.

Q: How much faster is the GTX 680 in Geekbench OpenCL?

A: The GTX 680 scores 16906 in Geekbench OpenCL, which is 29.3% higher than the M2090’s 13075.

Q: Do both cards support Vulkan?

A: No. The GTX 680 supports Vulkan 1.2.175, while the M2090 has no Vulkan support listed.

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

A: The M2090 has 6 GB of GDDR5 memory, while the GTX 680 has 2 GB of GDDR5 memory.

Q: Which card has a higher transistor density?

A: The GTX 680 has a transistor density of 12.0M / mm², compared to the M2090’s 5.8M / mm².

Q: What are the process nodes for each card?

A: The GTX 680 is built on a 28 nm process, while the M2090 uses a 40 nm process, both at TSMC.

Specification Differences

| Specification | NVIDIA GeForce GTX 680 | NVIDIA Tesla M2090 |

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

| Architecture | Kepler | Fermi 2.0 |

| Generation | GeForce 600 | Tesla Fermi (x20xx) |

| 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 | 1006 MHz | null |

| Boost Clock | 1058 MHz | null |

| Memory Clock | 1502 MHz (6 Gbps effective) | 924 MHz (3.7 Gbps effective) |

| Memory Size | 2 GB | 6 GB |

| Memory Bus Width | 256 bit | 384 bit |

| Memory Bandwidth | 192.3 GB/s | 177.4 GB/s |

| Shading Units | 1536 | 512 |

| TMUs | 128 | 64 |

| ROPs | 32 | 48 |

| Pixel Rate | 33.86 GPixel/s | 20.83 GPixel/s |

| Texture Rate | 135.4 GTexel/s | 41.66 GTexel/s |

| FP32 Performance | 3.250 TFLOPS | 1,332.2 GFLOPS |

| TDP | 195 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 | null |

| Launch MSRP | 499 USD | null |