NVIDIA GeForce GTX 660 vs NVIDIA Tesla M10 Comparison

The NVIDIA Tesla M10 and NVIDIA GeForce GTX 660 are both end-of-life products built on a 28 nm TSMC process, but they serve entirely different purposes. The Tesla M10 is a Maxwell-based compute accelerator with no display outputs, designed for datacenter workloads, while the GTX 660 is a Kepler-based consumer graphics card with full display connectivity. Benchmark data shows the GTX 660 winning both available head-to-head tests, yet the Tesla M10 holds a higher average score and better percentile ranking. The following analysis breaks down those results using only the provided facts.

Head-to-Head Benchmarks

The GeForce GTX 660 dominates the two shared benchmark tests. In Geekbench OpenCL, the GTX 660 scores 11,347 against the Tesla M10’s 10,318, a delta of -9.1% from the Tesla’s perspective. That translates to the GTX 660 being roughly 10% faster in this compute test. The gap widens considerably in Geekbench Vulkan, where the GTX 660 records 11,415 versus 9,130 for the Tesla M10, a -20% delta. The GTX 660 is a full 25% faster in Vulkan, which is a substantial margin for two cards from the same vendor.

However, the average benchmark score tells a different story. The Tesla M10 averages 9,724 across all recorded benchmarks, while the GTX 660 averages 9,022. This means the Tesla M10 has a higher overall standing despite losing both head-to-head tests. The explanation lies in the benchmark sets: the Tesla M10 only has OpenCL and Vulkan results, while the GTX 660 adds a Metal score of 4,305, which drags down its average. If Metal were excluded, the GTX 660’s average would be higher, but the data as presented shows the Tesla M10 with a 7.8% better average score.

The percentile rankings reinforce this split. The Tesla M10 sits at the 47th percentile of all GPUs, while the GTX 660 sits at the 45th. That two-percentile gap is modest but consistent with the average score difference. The nearest rivals for each card further contextualize the numbers. The Tesla M10’s closest competitor is the NVIDIA Tesla C2070, with an average score of 9,716 and a delta of 0.1%, meaning the M10 is essentially tied with that older compute card. The GTX 660’s nearest rival is the NVIDIA TITAN V CEO Edition, which scores 9,037, a delta of -0.2%, again a near tie. Both cards sit at the edge of their respective performance clusters, with neither showing a decisive advantage over its closest peers.

The Verdict

The data clearly shows the GeForce GTX 660 as the faster card in raw compute benchmarks. It wins both head-to-head tests, with a 9.1% advantage in OpenCL and a 20% advantage in Vulkan. For any workload that relies on those APIs, the GTX 660 is the superior choice. Its higher shading unit count (960 versus 640), texture mapping units (80 versus 40), and FP32 throughput (1.981 TFLOPS versus 1.672 TFLOPS) all point in the same direction.

Yet the Tesla M10 is not without merit. Its 8 GB of GDDR5 memory is four times the GTX 660’s 2 GB, and its memory bandwidth of 83.20 GB/s is lower than the GTX 660’s 144.2 GB/s, but the capacity advantage matters for large datasets. The M10 also has a higher average benchmark score (9,724 versus 9,022) and a better percentile ranking (47th versus 45th). This suggests that in a broader benchmark suite, the M10 holds up better than its head-to-head losses would imply.

The verdict depends on the use case. For general compute tasks that fit within 2 GB of memory, the GTX 660 is the clear winner. For workloads that require more than 2 GB of VRAM or that benefit from the Tesla’s datacenter positioning, the M10 is the better pick. The GTX 660 also carries a launch MSRP of 229 USD, while the Tesla M10 has no launch MSRP listed, which indicates a different market tier. The GTX 660 is a consumer card with display outputs; the M10 has no outputs and is strictly a compute accelerator. Choose the GTX 660 for performance per benchmark, the M10 for memory capacity.

Where Each One Wins

The GeForce GTX 660 wins in every direct benchmark comparison. Its OpenCL score of 11,347 exceeds the M10’s 10,318, and its Vulkan score of 11,415 more than doubles the gap. The GTX 660 also has higher texture rate (82.56 GTexel/s versus 52.24 GTexel/s) and higher pixel rate (20.64 GPixel/s versus 20.90 GPixel/s, though the M10 is marginally ahead here). The GTX 660’s 960 shading units and 80 TMUs give it a structural advantage in shader-heavy workloads. Its 192-bit memory bus provides 144.2 GB/s of bandwidth, which is 73% higher than the M10’s 83.20 GB/s. For gaming or any graphics-oriented task, the GTX 660 is categorically superior.

The Tesla M10 wins on memory capacity and average score. Its 8 GB frame buffer is four times larger, which is critical for datasets that exceed 2 GB. The M10 also has a higher boost clock (1306 MHz versus 1032 MHz), which partially compensates for its fewer cores. Its average benchmark score of 9,724 is 7.8% higher than the GTX 660’s 9,022, and its 47th percentile ranking beats the 45th percentile of the GTX 660. The M10 also supports Vulkan 1.4, while the GTX 660 only supports Vulkan 1.2.175, giving the M10 a software feature advantage. The M10’s transistor density of 12.6M per mm² is higher than the GTX 660’s 11.5M per mm², indicating a more efficient design despite fewer total transistors (1,870 million versus 2,540 million).

FAQ

Q: Which card is faster in OpenCL?

A: The GeForce GTX 660 scores 11,347 in Geekbench OpenCL, compared to the Tesla M10’s 10,318, making the GTX 660 9.1% faster in that test.

Q: What is the Tesla M10’s advantage in memory size?

A: The Tesla M10 has 8 GB of GDDR5 memory, which is four times the 2 GB found on the GeForce GTX 660.

Q: How do the average benchmark scores compare?

A: The Tesla M10 has an average benchmark score of 9,724, while the GeForce GTX 660 averages 9,022. The M10 is 7.8% higher on average, despite losing both head-to-head tests.

Q: Which card has a better Vulkan API support?

A: The Tesla M10 supports Vulkan 1.4, whereas the GeForce GTX 660 supports Vulkan 1.2.175. However, the GTX 660 scores 11,415 in Geekbench Vulkan versus 9,130 for the M10, a 20% performance gap.

Q: What is the GTX 660’s launch MSRP?

A: The GeForce GTX 660 has a launch MSRP of 229 USD. The Tesla M10 has no launch MSRP listed.

Q: Which card has more shading units?

A: The GeForce GTX 660 has 960 shading units, compared to 640 for the Tesla M10, a 50% advantage for the GTX 660.

Architecture Differences

The two cards come from different NVIDIA architectures. The Tesla M10 uses the GM107 chip based on Maxwell, while the GTX 660 uses the GK106 chip based on Kepler. Both are fabricated on a 28 nm process at TSMC, but the transistor counts differ significantly: the M10 has 1,870 million transistors on a 148 mm² die, giving a density of 12.6M per mm², while the GTX 660 packs 2,540 million transistors onto a 221 mm² die, with a density of 11.5M per mm². The Maxwell design achieves higher density despite fewer transistors, reflecting a more compact architecture.

The memory subsystems are fundamentally different. The M10 uses a 128-bit bus with 8 GB of GDDR5 at 1300 MHz (5.2 Gbps effective), yielding 83.20 GB/s of bandwidth. The GTX 660 uses a 192-bit bus with 2 GB of GDDR5 at 1502 MHz (6 Gbps effective), yielding 144.2 GB/s. The GTX 660 has 73% more bandwidth but only a quarter of the capacity. The M10’s base clock is 1033 MHz with a boost of 1306 MHz, while the GTX 660 runs at 980 MHz base and 1032 MHz boost. The M10 has a higher boost clock, but the GTX 660 compensates with more execution units.

Compute resources favor the GTX 660. It has 960 shading units, 80 TMUs, and 24 ROPs, versus 640 shading units, 40 TMUs, and 16 ROPs for the M10. This results in the GTX 660 delivering 1.981 TFLOPS of FP32 throughput and 82.56 GTexel/s, while the M10 delivers 1.672 TFLOPS and 52.24 GTexel/s. Pixel rates are nearly identical: 20.64 GPixel/s for the GTX 660 and 20.90 GPixel/s for the M10. Power draw favors the GTX 660 at 140 W with a 300 W suggested PSU and a single 6-pin connector, versus the M10’s 225 W with a 550 W suggested PSU and an 8-pin connector.

Physical and connectivity differences are stark. The Tesla M10 has no display outputs, reflecting its server role, while the GTX 660 offers 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2. The M10 is 267 mm long, the GTX 660 is 241 mm. Both are dual-slot cards with PCIe 3.0 x16 interfaces. API support is mostly similar: both support DirectX 12 (11_0) and OpenGL 4.6, but the M10 supports Vulkan 1.4 while the GTX 660 is limited to Vulkan 1.2.175. The release dates are separated by nearly four years, with the M10 launching in May 2016 and the GTX 660 in September 2012, placing them in different NVIDIA generations—Tesla Maxwell versus GeForce 600.