NVIDIA GeForce GTX 670 vs NVIDIA Tesla M2090 Comparison

NVIDIA’s Tesla M2090 and GeForce GTX 670 are both end-of-life products, but they represent two very different philosophies from the same company. The M2090 is a compute-oriented accelerator with no display outputs, while the GTX 670 is a mainstream graphics card designed for gaming and general use. Benchmark data shows a clear split: the GTX 670 holds the only head-to-head win, but the M2090’s specifications tell a story of raw compute capacity versus efficiency and feature support.

Where Each One Wins

The single direct benchmark comparison favors the GeForce GTX 670 decisively. In the Geekbench OpenCL test, the GTX 670 scores 15,341 points, while the Tesla M2090 trails at 13,075 points. That is a 14.8% advantage for the GTX 670, a substantial margin in a compute-oriented workload. This is not a close contest; the data indicates the GTX 670 outperforms the M2090 by nearly fifteen percent in this specific test.

However, the M2090’s strengths lie elsewhere. Its 6 GB of GDDR5 memory is triple the GTX 670’s 2 GB, and its 384-bit memory bus is wider than the 256-bit interface on the GTX 670. While the GTX 670 has higher raw memory bandwidth (192.3 GB/s vs. 177.4 GB/s), the M2090’s larger memory pool and wider bus suggest it is better suited for workloads that require large datasets resident in VRAM, such as scientific simulations or complex data processing. The M2090 also has 48 ROPs, versus 32 on the GTX 670, which benefits fill-rate-heavy operations.

The GTX 670 wins on efficiency and feature completeness. Its 170 W TDP is 80 W lower than the M2090’s 250 W, and it requires a less demanding power supply (450 W suggested versus 600 W). It also supports PCIe 3.0, while the M2090 is limited to PCIe 2.0. The GTX 670 has Vulkan API support (version 1.2.175), whereas the M2090 lists no Vulkan support at all. For any user needing display outputs, the GTX 670 is the only option, as it provides dual DVI, HDMI 1.4a, and DisplayPort 1.2 outputs, while the M2090 has none.

Architecture Differences

The architectural divide is stark. The Tesla M2090 uses the GF110 chip, based on the Fermi 2.0 architecture, and is manufactured on TSMC’s 40 nm process. It packs 3,000 million transistors onto a 520 mm² die, yielding a transistor density of 5.8 million per square millimeter. The GeForce GTX 670 uses the GK104 chip, based on the newer Kepler architecture, built on a 28 nm process. It fits 3,540 million transistors into a much smaller 294 mm² die, achieving a density of 12.0 million per square millimeter — more than double the M2090’s density.

The Kepler chip has significantly more shading units: 1,344 versus 512 on Fermi. It also has more texture mapping units (112 vs. 64). However, the Fermi chip has more ROPs (48 vs. 32). Clock behavior differs as well: the GTX 670 has a base clock of 915 MHz and a boost clock of 980 MHz, while the M2090 has no listed base or boost clocks, only a memory clock of 924 MHz (3.7 Gbps effective). The GTX 670’s memory runs at 1502 MHz (6 Gbps effective).

These architectural differences lead to divergent theoretical performance. The GTX 670 delivers 2.634 TFLOPS of FP32 compute, while the M2090 provides 1,332.2 GFLOPS (approximately 1.33 TFLOPS). That is nearly double the floating-point throughput on the GTX 670. Pixel fill rates also favor the GTX 670 at 27.44 GPixel/s versus 20.83 GPixel/s, and texture fill rates are dramatically higher: 109.8 GTexel/s versus 41.66 GTexel/s. The M2090’s only specification advantage is its 6 GB memory capacity and 48 ROPs.

Both cards support DirectX 12 (11_0) and OpenGL 4.6, but the GTX 670 adds Vulkan 1.2.175, while the M2090 has no Vulkan entry. The GTX 670 also uses a different power connector configuration (2x 6-pin) versus the M2090’s 1x 6-pin + 1x 8-pin setup. Physical dimensions are similar: the M2090 is 248 mm long (9.8 inches), and the GTX 670 is 241 mm long (9.5 inches), with the GTX 670 also having listed height and width measurements of 111 mm and 38 mm.

The Verdict

The data supports a clear recommendation for the GeForce GTX 670 in almost all scenarios. It wins the only head-to-head benchmark, delivers nearly double the FP32 throughput, has higher pixel and texture fill rates, consumes less power, supports newer PCIe and Vulkan standards, and includes display outputs. The GTX 670’s OpenCL score of 15,341 places it in the 52nd percentile of all GPUs, with its nearest rival being the AMD Radeon RX 7600M XT at just 0.5% behind.

The Tesla M2090, with its 53rd percentile ranking and OpenCL score of 13,075, sits in a similar performance tier but is strictly inferior in the measured benchmark. Its nearest rival, the NVIDIA GeForce GTX 950, is only 0.9% ahead, and the gap to the GTX 670 is substantial. The M2090’s only potential justification is its 6 GB memory capacity, which could be relevant for specific compute workloads that exceed 2 GB of VRAM. However, given that its memory bandwidth is actually lower than the GTX 670’s (177.4 GB/s vs. 192.3 GB/s), even that advantage is mitigated.

For any user choosing between these two, the GTX 670 is the logical pick. It offers superior performance, better efficiency, and broader software support. The M2090 is a niche product whose only edge is memory capacity, and that edge does not translate into benchmark wins. The GTX 670’s launch MSRP was 399 USD, which is worth noting as its official price point. The M2090 has no listed launch MSRP. If the workload fits within 2 GB of VRAM, the GTX 670 is categorically better. If the workload absolutely requires more than 2 GB, the M2090 is the only option, but its performance deficit remains a significant caveat.

FAQ

Q: Which GPU has the higher OpenCL benchmark score?

A: The NVIDIA GeForce GTX 670 scores 15,341 in Geekbench OpenCL, while the Tesla M2090 scores 13,075, making the GTX 670 14.8% faster in this test.

Q: Does the Tesla M2090 support display outputs?

A: No, the Tesla M2090 has no display outputs. The GeForce GTX 670 provides 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2.

Q: How do the memory capacities compare?

A: The Tesla M2090 has 6 GB of GDDR5 memory, while the GeForce GTX 670 has 2 GB of GDDR5 memory. The M2090 also has a wider 384-bit bus versus the GTX 670’s 256-bit bus.

Q: Which card has a higher FP32 compute throughput?

A: The GeForce GTX 670 delivers 2.634 TFLOPS, which is nearly double the Tesla M2090’s 1,332.2 GFLOPS.

Q: What are the power consumption differences?

A: The GeForce GTX 670 has a 170 W TDP and requires a 450 W suggested power supply, while the Tesla M2090 has a 250 W TDP and needs a 600 W suggested power supply.

Q: Does either GPU support Vulkan?

A: Only the GeForce GTX 670 lists Vulkan support, with version 1.2.175. The Tesla M2090 does not list any Vulkan API support.

Head-to-Head Benchmarks

The only head-to-head benchmark available is Geekbench OpenCL, and it is a decisive win for the GeForce GTX 670. The GTX 670 posts a score of 15,341, while the Tesla M2090 achieves 13,075. The delta percentage is -14.8% from the perspective of the M2090, meaning the M2090 is 14.8% slower. In absolute terms, the GTX 670 is 2,266 points ahead. This single data point is the most direct comparison available, and it aligns with the broader specification advantages of the GTX 670.

Beyond the direct benchmark, the specification sheet reinforces the GTX 670’s dominance. Its FP32 throughput of 2.634 TFLOPS is 97.7% higher than the M2090’s 1,332.2 GFLOPS. Pixel fill rate favors the GTX 670 at 27.44 GPixel/s versus 20.83 GPixel/s, a 31.7% advantage. Texture fill rate is even more lopsided: 109.8 GTexel/s versus 41.66 GTexel/s, which is more than 2.6 times higher. The GTX 670 also has more than double the shading units (1,344 vs. 512) and nearly double the TMUs (112 vs. 64).

While the M2090 does have a memory capacity advantage (6 GB vs. 2 GB) and more ROPs (48 vs. 32), those factors do not appear to influence the benchmark outcome. The GTX 670’s memory bandwidth is 8.4% higher (192.3 GB/s vs. 177.4 GB/s), which likely contributes to its OpenCL win. Additionally, the GTX 670’s boost clock of 980 MHz provides dynamic performance scaling that the M2090 lacks entirely, as it has no listed base or boost clocks. The data consistently points in one direction: the GTX 670 is the superior performer in every measurable metric that affects benchmark scores.