NVIDIA GeForce GTX 960A vs NVIDIA Tesla M2090 Comparison

The NVIDIA Tesla M2090 and the NVIDIA GeForce GTX 960A occupy vastly different corners of the GPU landscape, separated by four years of architectural evolution and distinct design goals. The M2090 is a dual-slot, compute-focused accelerator from the Fermi era, demanding 250 W and external power connectors, while the GTX 960A is a 75 W MXM module built for portable systems. Despite this generational gap, benchmark results show the older card holding a 9% lead in the sole OpenCL test, a data point that frames a nuanced comparison of raw compute throughput against architectural efficiency.

FAQ

Q: Which GPU has the higher benchmark score in the head-to-head comparison?

A: The NVIDIA Tesla M2090 wins the only head-to-head benchmark, Geekbench OpenCL, with a score of 13075 compared to the GTX 960A’s 11998. This represents a 9% advantage for the Tesla M2090.

Q: How does the Tesla M2090 perform relative to its nearest rivals?

A: The M2090’s average benchmark score of 13075 places it 0.7% ahead of the GeForce GTX 1660 SUPER (12986) and 1% ahead of the RTX 3050 Ti Mobile (12940). It trails the GeForce GTX 950 (13189) by 0.9% and sits 1.1% ahead of the AMD Radeon RX 580 (12928).

Q: What is the GTX 960A’s standing among its closest competitors?

A: With an average score of 11998, the GTX 960A is 0.3% ahead of the GeForce GTX 1080 (11960) and 1.3% ahead of the AMD Radeon RX 6500 XT (11842). It shows a more decisive 2.7% lead over the GeForce GTX 1660 (11680) and a 3.2% lead over the AMD Radeon RX 7800 XT (11627).

Q: What are the key memory specifications that differ between the two cards?

A: The Tesla M2090 features 6 GB of GDDR5 memory on a 384-bit bus, yielding 177.4 GB/s of bandwidth. The GTX 960A has 2 GB of GDDR5 on a 128-bit bus, providing 80.19 GB/s—less than half the bandwidth of the M2090.

Q: What is the difference in power consumption and physical design?

A: The Tesla M2090 has a 250 W TDP and requires a dual-slot cooler with 1x 6-pin and 1x 8-pin power connectors. The GTX 960A has a 75 W TDP, uses an MXM Module form factor, and requires no external power connectors.

Q: How do the shading unit counts compare between the two architectures?

A: The GTX 960A has 640 shading units, which is greater than the Tesla M2090’s 512. However, the M2090 has higher pixel rate (20.83 GPixel/s vs 18.82 GPixel/s) and a larger memory bus.

The Verdict

The data presents a clear split: the Tesla M2090 is the compute-oriented choice, while the GTX 960A is a more efficient, mobile-friendly part. In the single benchmark available, the M2090 delivers 9% higher performance, a margin that aligns with its workstation heritage. Its 6 GB frame buffer and 384-bit bus are designed for memory-intensive workloads, and its 53rd percentile ranking among all GPUs is slightly above the GTX 960A’s 51st percentile.

However, the GTX 960A should not be dismissed. Its 640 shading units outnumber the M2090’s 512, and its Maxwell architecture supports Vulkan 1.4, a feature the Fermi-based M2090 lacks entirely. The 960A’s 75 W TDP and MXM form factor make it suitable for portable devices, where the M2090’s dual-slot, 250 W design would be impractical. For users prioritizing raw compute and memory capacity in a stationary chassis, the M2090 is the data-backed winner. For those needing a low-power, Vulkan-capable module in a laptop or compact system, the GTX 960A is the sensible pick.

Head-to-Head Benchmarks

The only direct comparison available is the Geekbench OpenCL test, where the Tesla M2090 scores 13075 against the GTX 960A’s 11998. This 9% delta is significant, showing that the older Fermi architecture still holds an edge in this compute workload. The M2090’s victory is likely attributable to its 177.4 GB/s memory bandwidth, which is more than double the 960A’s 80.19 GB/s, and its 48 ROPs versus the 960A’s 16. These factors allow the M2090 to move data faster and rasterize more efficiently, despite having fewer shading units.

The 960A’s counterpoints are its higher texture rate (47.04 GTexel/s vs 41.66 GTexel/s) and its 1.505 TFLOPS of FP32 compute, which exceeds the M2090’s 1,332.2 GFLOPS. Yet these advantages do not translate to a benchmark win, as the OpenCL test favors the M2090’s memory subsystem. The result suggests that raw shading throughput is not the sole determinant of compute performance; memory bandwidth and ROP count play critical roles. The 960A’s nearest rival scores—0.3% ahead of the GTX 1080 and 3.2% ahead of the RX 7800 XT—indicate it is competitive within its power class, but the M2090’s 9% lead in this test underscores the value of a wider memory interface in certain workloads.

Specification Differences

The two cards diverge on nearly every major specification. The M2090 uses a 40 nm process node, while the GTX 960A is built on 28 nm, resulting in a transistor density of 5.8M per mm² for the former versus 12.6M per mm² for the latter. The M2090’s die size is 520 mm², housing 3,000 million transistors; the 960A’s die is 148 mm² with 1,870 million transistors. Clock speeds differ fundamentally: the M2090 has no listed base or boost clock, while the 960A runs at 1097 MHz base and 1176 MHz boost. Memory clocks also vary, with the M2090 at 924 MHz (3.7 Gbps effective) and the 960A at 1253 MHz (5 Gbps effective).

The memory configuration is starkly different: 6 GB on a 384-bit bus for the M2090 versus 2 GB on a 128-bit bus for the 960A. This yields bandwidth figures of 177.4 GB/s and 80.19 GB/s, respectively. The M2090 has 512 shading units, 64 TMUs, and 48 ROPs, while the 960A has 640 shading units, 40 TMUs, and 16 ROPs. Pixel rates are 20.83 GPixel/s (M2090) and 18.82 GPixel/s (960A), while texture rates are 41.66 GTexel/s and 47.04 GTexel/s. Power consumption is a major divide: 250 W with dual-slot cooling and two power connectors for the M2090, versus 75 W with an MXM Module and no connectors for the 960A. The bus interface is PCIe 2.0 x16 for the former and MXM-B (3.0) for the latter.

Architecture Differences

The M2090 is built on the Fermi 2.0 architecture (chip GF110), a design that prioritized compute throughput and memory bandwidth over power efficiency. The GTX 960A uses the Maxwell architecture (chip GM107), which focused on improving performance-per-watt and feature support. This generational shift is evident in the process node: 40 nm for Fermi versus 28 nm for Maxwell, allowing the latter to pack 12.6M transistors per mm² compared to 5.8M for the former.

Feature support differs notably. The M2090 supports DirectX 12 (11_0) and OpenGL 4.6, but has no Vulkan support. The GTX 960A also supports DirectX 12 (11_0) and OpenGL 4.6, but adds Vulkan 1.4, a significant advantage for modern applications. Display outputs also diverge: the M2090 has no outputs, marking it as a pure compute accelerator, while the 960A’s outputs are portable device dependent, reflecting its mobile design intent. The M2090’s production status is end-of-life, with a release date in July 2011 and a successor named Tesla Kepler; the 960A is also end-of-life, released in March 2015, with its predecessor listed as GeForce 800A and no successor. The M2090's predecessor is simply "Tesla," while the 960A belongs to the GeForce 900A generation. These architectural differences explain why the older M2090 still wins a compute benchmark: its Fermi design was built for heavy, sustained workloads, whereas Maxwell was optimized for balanced performance in constrained power envelopes.