NVIDIA GeForce GTX 660 vs NVIDIA GeForce GTX 850M Comparison

The NVIDIA GeForce GTX 850M and the NVIDIA GeForce GTX 660 occupy different corners of the same 28 nm generation, but their benchmark results reveal a clear performance hierarchy. In the head-to-head data, the desktop-oriented GTX 660 dominates both available tests. The GTX 660 scores 11,347 in Geekbench OpenCL against the GTX 850M’s 9,821, a delta of -13.4% from the mobile part’s perspective. The gap widens significantly in Geekbench Vulkan, where the GTX 660 reaches 11,415 while the GTX 850M manages 8,782, a -23.1% deficit. The GTX 660 claims two wins out of two head-to-head benchmarks, leaving the GTX 850M with zero.

Head-to-Head Benchmarks

The OpenCL result places the GTX 660 roughly 15.5% ahead of the GTX 850M in raw compute throughput. This aligns with the underlying specifications: the GTX 660 carries 960 shading units versus 640, and its FP32 output is listed at 1.981 TFLOPS compared to 1,154.6 GFLOPS for the GTX 850M. The GTX 660’s memory subsystem also contributes — its GDDR5 memory operates at 6 Gbps effective across a 192-bit bus, yielding 144.2 GB/s of bandwidth, while the GTX 850M uses DDR3 at 2 Gbps effective on a 128-bit bus for 32.03 GB/s. That 4.5x bandwidth advantage explains why the GTX 660’s OpenCL lead, while substantial, is not even larger.

The Vulkan test tells a more pronounced story. The GTX 660’s 11,415 score versus the GTX 850M’s 8,782 represents a 30% advantage for the desktop card. This delta likely reflects the combination of higher texture rate (82.56 GTexel/s versus 36.08 GTexel/s) and pixel rate (20.64 GPixel/s versus 14.43 GPixel/s). The GTX 660 also benefits from 80 texture mapping units and 24 ROPs, compared to 40 TMUs and 16 ROPs on the GTX 850M. In graphics-API workloads, these resources translate directly into visible frame-rate differences.

Contextualizing the averages, the GTX 850M’s average benchmark score is 9,302, placing it at the 46th percentile among all GPUs. Its nearest rivals include the NVIDIA GeForce GTX 465 at 9,294 (0.1% ahead), the AMD Radeon R7 M380 at 9,313 (0.1% behind), and the NVIDIA GeForce GTX 960 at 9,273 (0.3% ahead). The GTX 660, by contrast, averages 9,022 and sits at the 45th percentile, with rivals like the NVIDIA TITAN V CEO Edition at 9,037 (-0.2% delta) and the NVIDIA GeForce GTX 560 at 9,058 (-0.4% delta). The striking takeaway is that despite losing both head-to-head tests, the GTX 850M actually posts a higher average benchmark score (9,302 vs 9,022). This apparent contradiction stems from the test mix: the GTX 850M has no Metal benchmark in its record, while the GTX 660’s Metal score of 4,305 drags its average down. In OpenCL and Vulkan, the GTX 660 wins outright.

The Verdict

From the data alone, the NVIDIA GeForce GTX 660 is the stronger GPU for compute and graphics workloads. It wins both head-to-head benchmarks with margins of 13.4% in OpenCL and 23.1% in Vulkan. Users prioritizing raw throughput in OpenCL or Vulkan applications should select the GTX 660 without hesitation. The GTX 850M, however, is not without merit. Its average benchmark score of 9,302 exceeds the GTX 660’s 9,022, and it achieves this with a TDP of 45 W versus 140 W for the GTX 660. The GTX 850M is an integrated-mobile-part (IGP slot width) with no power connectors, while the GTX 660 is a dual-slot card requiring a 6-pin connector and a 300 W suggested PSU. For a laptop or compact system where power draw and physical space are constraints, the GTX 850M is the only viable choice between the two. But for a desktop with adequate cooling and power delivery, the GTX 660’s benchmark dominance makes it the clear performance pick. The GTX 660’s launch MSRP was 229 USD, reflecting its original desktop positioning.

Architecture Differences

The two GPUs hail from different NVIDIA architectures. The GTX 850M uses the GM107 chip built on the Maxwell architecture, while the GTX 660 uses the GK106 chip on the older Kepler architecture. Both are fabricated by TSMC on a 28 nm process, but the transistor counts diverge: the GM107 packs 1,870 million transistors on a 148 mm² die, yielding a transistor density of 12.6 million per mm². The GK106 contains 2,540 million transistors across a 221 mm² die, for a density of 11.5 million per mm². Despite the GTX 660’s larger die and higher transistor count, the Maxwell design gives the GTX 850M a density advantage.

Memory configurations differ fundamentally. The GTX 850M uses 2 GB of DDR3 with a 128-bit bus and 32.03 GB/s bandwidth. The GTX 660 uses 2 GB of GDDR5 with a 192-bit bus and 144.2 GB/s bandwidth. Memory clocks also differ: the GTX 850M runs at 1,001 MHz (2 Gbps effective), while the GTX 660 runs at 1,502 MHz (6 Gbps effective). The GTX 660’s base clock is 980 MHz with a boost of 1,032 MHz; the GTX 850M’s base and boost clocks are not specified in the data.

Compute resources scale similarly. The GTX 660 has 960 shading units, 80 TMUs, and 24 ROPs, versus 640 shading units, 40 TMUs, and 16 ROPs for the GTX 850M. Pixel and texture rates follow: 20.64 GPixel/s and 82.56 GTexel/s for the GTX 660, versus 14.43 GPixel/s and 36.08 GTexel/s for the GTX 850M. FP32 performance is 1.981 TFLOPS for the GTX 660 and 1,154.6 GFLOPS for the GTX 850M. Neither GPU has ray tracing or tensor cores. API support is similar for DirectX (both 12 with 11_0 feature level) and OpenGL (both 4.6), but Vulkan support differs: the GTX 850M supports Vulkan 1.4, while the GTX 660 supports Vulkan 1.2.175.

Physical and power characteristics are polar opposites. The GTX 850M is an IGP (integrated graphics processor) with no power connectors and a 45 W TDP, designed for portable devices with display outputs that are device-dependent. The GTX 660 is a dual-slot card measuring 241 mm (9.5 inches) in length, with one 6-pin power connector, a 140 W TDP, and a suggested PSU of 300 W. Its display outputs are 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2. The GTX 850M belongs to the GeForce 800M generation, released March 2014, while the GTX 660 is part of the GeForce 600 generation, released September 2012. Both are end-of-life products.

FAQ

Q: Which GPU wins in Geekbench OpenCL?

A: The NVIDIA GeForce GTX 660 wins with a score of 11,347 versus the GTX 850M’s 9,821, a delta of -13.4% from the GTX 850M’s perspective.

Q: How large is the Vulkan performance gap?

A: The GTX 660 scores 11,415 in Geekbench Vulkan, while the GTX 850M scores 8,782, giving the GTX 660 a 23.1% advantage.

Q: Does the GTX 850M have any benchmark where it wins?

A: No. In the head-to-head data, the GTX 850M records zero wins across the two tests (OpenCL and Vulkan), both of which go to the GTX 660.

Q: What explains the GTX 850M’s higher average benchmark score?

A: The GTX 850M’s average is 9,302 versus 9,022 for the GTX 660. The GTX 660’s average is pulled down by its Geekbench Metal score of 4,305, a test that is not present in the GTX 850M’s benchmark record.

Q: How do their memory bandwidths compare?

A: The GTX 660 offers 144.2 GB/s of bandwidth via GDDR5 on a 192-bit bus, while the GTX 850M provides 32.03 GB/s via DDR3 on a 128-bit bus.

Q: What are the power requirements for each card?

A: The GTX 850M has a 45 W TDP and requires no power connectors. The GTX 660 has a 140 W TDP, requires a single 6-pin connector, and has a suggested PSU of 300 W.

Where Each One Wins

The GTX 660 wins decisively in raw compute and graphics throughput. In OpenCL, its 11,347 score exceeds the GTX 850M’s 9,821 by 1,526 points. In Vulkan, the margin grows to 2,633 points (11,415 versus 8,782). This makes the GTX 660 the preferred choice for desktop workloads involving GPU compute, modern API gaming, or any task that stresses texture fill and pixel output. Its 82.56 GTexel/s texture rate and 20.64 GPixel/s pixel rate are more than double the GTX 850M’s figures, so games with heavy texture detail or high-resolution rendering will see pronounced benefits. The GTX 660’s 1.981 TFLOPS FP32 output also positions it well for general-purpose compute tasks that scale with shader count.

The GTX 850M wins in efficiency and integration. Its 45 W TDP is less than one-third of the GTX 660’s 140 W, and it needs no external power connector. For mobile platforms, this is the defining advantage. The GTX 850M’s average benchmark score of 9,302 is 280 points higher than the GTX 660’s 9,022, indicating that in a mixed benchmark suite (one that excludes Metal), the GTX 850M can hold its own. Its Maxwell architecture delivers a higher transistor density (12.6M/mm² versus 11.5M/mm²), and its Vulkan 1.4 support is more recent than the GTX 660’s Vulkan 1.2.175. For users constrained by power budgets, thermal limits, or physical space — such as laptop owners or small-form-factor builders — the GTX 850M is the rational choice despite losing every head-to-head test. The GTX 660, conversely, is the clear winner for anyone with a desktop chassis, a 300 W PSU, and a need for maximum benchmark performance. The data does not support choosing the GTX 850M for performance reasons; it only supports choosing it for form factor and power efficiency.