NVIDIA GeForce GTX 660 vs NVIDIA GeForce GTX 675MX Comparison

# NVIDIA GeForce GTX 660 vs NVIDIA GeForce GTX 675MX

The NVIDIA GeForce GTX 660 and NVIDIA GeForce GTX 675MX represent two distinct approaches to Kepler graphics from late 2012, with the desktop GTX 660 trading blows with the mobile 675MX across the two available benchmark suites. The GTX 675MX wins the Geekbench Metal test by a substantial margin (6131 vs 4305, a 29.8% advantage), while the GTX 660 takes the Geekbench OpenCL test with 11347 vs 10723, a 5.8% lead. This split result means the "winner" depends entirely on which workload you prioritize, and the underlying hardware differences explain why each card excels in different scenarios.

Head-to-Head Benchmarks

The Geekbench Metal benchmark delivers the most decisive verdict in this comparison, and it goes to the GTX 675MX. Scoring 6131 against the GTX 660's 4305, the mobile card outperforms its desktop sibling by 29.8%. This is not a marginal difference; it is a commanding lead that suggests the 675MX's architecture handles Metal-accelerated workloads with significantly greater efficiency. The 675MX's position in the overall GPU landscape reinforces this strength, as its nearest rivals include the AMD Radeon 880M (8436 avg score, only 0.1% ahead) and the NVIDIA GeForce MX330 (8458, 0.4% ahead). Notably, the 675MX's Metal score of 6131 is well above its average benchmark score of 8427, indicating that Metal is a particularly favorable workload for this card.

The Geekbench OpenCL benchmark flips the script, favoring the GTX 660 with a score of 11347 versus 10723 for the 675MX. The 5.8% delta is solid but not overwhelming, and it places the GTX 660 ahead of several notable rivals. The GTX 660's average benchmark score of 9022 puts it in the 45th percentile of all GPUs, while the 675MX sits at the 43rd percentile with an average of 8427. This 7% difference in average scores aligns with the OpenCL result, suggesting that the GTX 660 is the more consistent performer across general compute tasks, even if it loses decisively in Metal. The GTX 660's nearest rivals include the NVIDIA TITAN V CEO Edition (9037, just 0.2% ahead) and the NVIDIA GeForce GTX 560 (9058, 0.4% ahead), which underscores how competitive its average score is relative to a broad range of hardware.

Architecture Differences

Both GPUs are built on NVIDIA's Kepler architecture, manufactured by TSMC on a 28 nm process, but they diverge significantly in their physical design. The GTX 660 uses the GK106 chip with 2,540 million transistors on a 221 mm² die, achieving a transistor density of 11.5M per mm². The GTX 675MX, by contrast, employs the larger GK104 chip with 3,540 million transistors on a 294 mm² die, yielding a slightly higher density of 12.0M per mm². This 1,000 million transistor difference gives the 675MX a more complex chip, but the GTX 660 compensates with higher clock speeds — 980 MHz base and 1032 MHz boost versus the 675MX's unspecified base and boost clocks.

The memory subsystems tell a more nuanced story. The GTX 660 has a 192-bit memory bus with a bandwidth of 144.2 GB/s, while the 675MX uses a wider 256-bit bus but achieves only 115.2 GB/s due to its slower 900 MHz memory clock (3.6 Gbps effective) compared to the GTX 660's 1502 MHz (6 Gbps effective). Both cards feature 2 GB of GDDR5 memory, but the GTX 660's higher memory clock gives it a 25% bandwidth advantage. The shading unit count is identical at 960, as are the texture mapping units at 80, but the ROP count differs: the GTX 660 has 24 ROPs, while the 675MX has 32 ROPs. This ROP advantage likely contributes to the 675MX's Metal performance, while the GTX 660's higher clocks help its OpenCL showing.

The GTX 660's pixel rate of 20.64 GPixel/s and texture rate of 82.56 GTexel/s both exceed the 675MX's 13.08 GPixel/s and 52.32 GTexel/s, reflecting the desktop card's higher clock speeds. The FP32 compute is similarly lopsided: the GTX 660 delivers 1.981 TFLOPS versus 1,255.7 GFLOPS for the 675MX. Power consumption also differs markedly, with the GTX 660 rated at 140 W TDP and the 675MX at just 100 W TDP, though the desktop card requires a 6-pin power connector and a 300 W suggested PSU, while the 675MX is an MXM module with no dedicated power connectors.

FAQ

Q: Which card has more raw compute power?

A: The GTX 660 is clearly ahead in FP32 performance, delivering 1.981 TFLOPS versus 1,255.7 GFLOPS for the 675MX. This aligns with the Geekbench OpenCL result where the GTX 660 scored 11347 against 10723, a 5.8% advantage.

Q: Why does the GTX 675MX win the Metal benchmark despite lower clock speeds?

A: The 675MX has a wider 256-bit memory bus and 32 ROPs compared to the GTX 660's 192-bit bus and 24 ROPs. While the GTX 660 has higher memory bandwidth (144.2 GB/s vs 115.2 GB/s), the 675MX's ROP count and chip design (GK104 with 3,540 million transistors) appear to favor Metal workloads, resulting in a 29.8% higher Metal score.

Q: Are both cards the same generation?

A: Yes, both belong to the GeForce 600 generation, but they target different segments. The GTX 660 is part of the desktop GeForce 600 lineup, while the 675MX is from the GeForce 600M mobile series. Both use the Kepler architecture and are manufactured on a 28 nm TSMC process.

Q: What are the memory differences between these two cards?

A: Both have 2 GB of GDDR5 memory, but the GTX 660 has a 192-bit bus with 144.2 GB/s bandwidth and a 1502 MHz memory clock (6 Gbps effective). The 675MX has a 256-bit bus but only 115.2 GB/s bandwidth due to a 900 MHz memory clock (3.6 Gbps effective).

Q: Which card is more power-efficient?

A: The GTX 675MX has a 100 W TDP, which is 40 W lower than the GTX 660's 140 W TDP. The 675MX also requires no power connectors as an MXM module, while the GTX 660 needs a single 6-pin connector and a 300 W suggested PSU.

Q: How do these cards compare to their nearest rivals?

A: The GTX 660's average score of 9022 puts it just 0.2% behind the NVIDIA TITAN V CEO Edition and 0.4% behind the GeForce GTX 560, while being 1.2% ahead of the AMD Radeon 550X. The 675MX's average of 8427 is 0.1% behind the AMD Radeon 880M and 0.4% behind the GeForce MX330, while leading the AMD Radeon R9 M375X by 1.2%.

The Verdict

The data presents a clear but split decision. For users who prioritize Metal-based workloads, the GTX 675MX is the superior choice, offering a 29.8% performance advantage that is difficult to ignore. Its 32 ROPs and wider memory bus make it particularly effective in that specific API environment. However, for OpenCL compute tasks, the GTX 660 is the stronger card, with a 5.8% lead that reflects its higher clock speeds and greater FP32 throughput. The GTX 660 also holds a higher average benchmark score (9022 vs 8427) and a better percentile ranking (45th vs 43rd), suggesting it is the more generally capable GPU across a broader range of tests. The 675MX's Metal score of 6131 is its standout achievement, and its lower TDP of 100 W makes it a more power-efficient option, but the GTX 660's 1.981 TFLOPS of compute power and higher memory bandwidth give it an edge in raw performance metrics. The choice ultimately hinges on whether your applications lean toward Metal or OpenCL, and whether the desktop form factor of the GTX 660 or the mobile MXM form factor of the 675MX fits your hardware setup.

Specification Differences

The GTX 660 and GTX 675MX differ in nearly every major specification except for their 28 nm TSMC process node, 2 GB GDDR5 memory, 960 shading units, 80 TMUs, Kepler architecture, and DirectX 12 (11_0) support. The chip design is the most fundamental divergence: the GTX 660 uses GK106 with 2,540 million transistors on a 221 mm² die, while the 675MX uses GK104 with 3,540 million transistors on a 294 mm² die. Clock speeds favor the GTX 660, with a 980 MHz base and 1032 MHz boost, whereas the 675MX has no listed base or boost clocks but a memory clock of 900 MHz (3.6 Gbps effective) against the GTX 660's 1502 MHz (6 Gbps effective). The memory bus width also differs, with the 675MX's 256-bit interface being wider than the GTX 660's 192-bit, yet the GTX 660 still achieves higher bandwidth at 144.2 GB/s versus 115.2 GB/s. ROP counts favor the 675MX at 32 versus 24, but the GTX 660 leads in pixel rate (20.64 GPixel/s vs 13.08), texture rate (82.56 GTexel/s vs 52.32), and FP32 (1.981 TFLOPS vs 1,255.7 GFLOPS). The TDP is 140 W for the GTX 660 and 100 W for the 675MX, with the former requiring a 6-pin connector and 300 W PSU while the latter uses an MXM-B (3.0) interface with no power connectors. The GTX 660 offers 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2 outputs, while the 675MX's outputs are portable-device dependent. The GTX 660 launched on 2012-09-05 with a launch MSRP of 229 USD, while the 675MX launched later on 2012-09-30 with no launch MSRP listed.

Where Each One Wins

The GTX 675MX wins decisively in Metal-based applications, as evidenced by its 6131 Geekbench Metal score versus the GTX 660's 4305 — a 29.8% advantage. This makes it the preferable choice for macOS or Metal-specific workloads, and its 32 ROPs and wider 256-bit memory bus are likely contributors. The 675MX also wins on power efficiency, with a 100 W TDP that is 40 W lower than the GTX 660, making it better suited for thermally constrained environments like laptops or compact MXM systems. Its transistor count of 3,540 million and die size of 294 mm² suggest a more complex chip that handles certain specialized tasks well, even if its raw clock speeds are lower.

The GTX 660 wins in OpenCL compute workloads, posting a 11347 score that is 5.8% higher than the 675MX's 10723. Its higher clock speeds (980 MHz base, 1032 MHz boost) and superior memory bandwidth (144.2 GB/s vs 115.2 GB/s) give it a clear edge in general-purpose GPU computing. The GTX 660 also wins on every rate metric: pixel rate (20.64 GPixel/s vs 13.08), texture rate (82.56 GTexel/s vs 52.32), and FP32 (1.981 TFLOPS vs 1,255.7 GFLOPS). Its average benchmark score of 9022 exceeds the 675MX's 8427 by 7%, and its 45th percentile ranking beats the 675MX's 43rd, indicating better overall performance across a wider range of tests. For desktop users with access to standard PCIe 3.0 x16 slots, the GTX 660's display outputs (2x DVI, HDMI 1.4a, DisplayPort 1.2) and 241 mm length make it a more versatile everyday card, provided the 140 W TDP and 300 W PSU requirement are manageable.