NVIDIA GeForce GTX 1070 vs NVIDIA GeForce GTX 660 Comparison

The NVIDIA GeForce GTX 1070 and the NVIDIA GeForce GTX 660 occupy different ends of the performance spectrum within their respective generations. Based on the benchmark data, the GTX 1070 is decisively faster, winning all three head-to-head tests with substantial margins. The GTX 1070’s average benchmark score of 9780 places it in the 47th percentile of all GPUs, while the GTX 660’s average score of 9022 lands in the 45th percentile. Despite the raw score gap, their percentile ranks are surprisingly close, suggesting that the GTX 660’s scores are relatively high for its era, while the GTX 1070’s are more modest relative to the modern field. The verdict is clear: the GTX 1070 is the superior card for any compute-heavy or modern API workload, but the GTX 660 remains a competent legacy option for older software.

The Verdict

The data points to a single conclusion: the GTX 1070 is the card to choose for anyone running contemporary applications or demanding resolutions. Its 8 GB of GDDR5 memory, compared to the GTX 660’s 2 GB, provides a significant buffer for modern game textures and compute tasks. The GTX 1070’s lead in the Geekbench tests—336.7% ahead in Metal, 293.9% ahead in OpenCL, and 93.8% ahead in Vulkan—is not incremental; it is transformative. For users with a GTX 660, upgrading to a GTX 1070 represents a leap of roughly three to four generations of performance improvement in raw compute.

However, the GTX 660 is not without a niche. Its 45th percentile rank, while lower, is not dramatically far from the GTX 1070’s 47th percentile, indicating that its performance is still respectable in a historical context. The GTX 660’s lower power draw (140 W vs. 150 W) and smaller physical footprint (241 mm length vs. 267 mm) make it a more manageable card for small form-factor systems or older power supplies. For a user with a legacy system running DirectX 11 or older titles, the GTX 660 remains a functional option, provided its 2 GB memory is sufficient. The GTX 660’s nearest rivals include the NVIDIA GeForce GTX 560 (0.4% behind) and the AMD Radeon 550X (1.2% behind), confirming that it sits in a competitive mid-range tier for its time.

The GTX 1070, by contrast, counts the AMD FirePro W5000 (0.2% behind) and the NVIDIA Quadro M2000M (0.5% behind) as its closest competitors, which are professional-grade cards. This suggests that the GTX 1070’s performance profile is closer to workstation hardware than to consumer mid-range cards, further underscoring its superiority. In sum, the GTX 1070 is the pick for anyone prioritizing performance; the GTX 660 is a fallback for those with strict power, size, or budget constraints who cannot accommodate a newer card.

Architecture Differences

The architectural gap between these two cards is a generational chasm. The GTX 1070 is built on the Pascal architecture, using the GP104 chip fabricated on a 16 nm process at TSMC. It contains 7,200 million transistors on a 314 mm² die, yielding a transistor density of 22.9M per mm². In contrast, the GTX 660 employs the Kepler architecture with the GK106 chip on a 28 nm process, also from TSMC. It packs 2,540 million transistors into a 221 mm² die, with a density of 11.5M per mm². The Pascal node advantage is stark: not only does the GTX 1070 have nearly three times the transistor count, but it does so on a significantly more advanced manufacturing process.

The memory subsystems differ equally. The GTX 1070 features 8 GB of GDDR5 memory on a 256-bit bus, delivering 256.3 GB/s of bandwidth. The GTX 660 offers 2 GB of GDDR5 on a 192-bit bus, with 144.2 GB/s of bandwidth. This nearly 78% bandwidth advantage for the GTX 1070 is critical for high-resolution textures and compute workloads. The compute resources are also dramatically skewed: the GTX 1070 has 1920 shading units, 120 texture mapping units, and 64 ROPs, versus the GTX 660’s 960 shading units, 80 TMUs, and 24 ROPs. The pixel rate of the GTX 1070 is 107.7 GPixel/s compared to 20.64 GPixel/s on the GTX 660, and the texture rate is 202.0 GTexel/s versus 82.56 GTexel/s. Floating-point performance tells the same story: the GTX 1070 delivers 6.463 TFLOPS of FP32, while the GTX 660 manages 1.981 TFLOPS.

API support also separates them. The GTX 1070 supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4. The GTX 660 supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. The higher DirectX feature level and newer Vulkan version on the GTX 1070 are significant for modern titles that leverage advanced rendering features. Both cards are end-of-life, but the GTX 1070’s successor is the GeForce 20 series, while the GTX 660’s is the GeForce 700 series, confirming their respective positions in the product timeline.

Where Each One Wins

The GTX 1070 wins everywhere that matters in the head-to-head data. In the Geekbench Metal test, the GTX 1070 scores 18801 against the GTX 660’s 4305, a delta of 336.7%. This test is particularly relevant for macOS users or applications leveraging Apple’s Metal API, where the GTX 1070’s modern architecture provides a massive advantage. In Geekbench OpenCL, the GTX 1070 scores 44700 versus 11347, a 293.9% lead, indicating superior general-purpose compute performance for tasks like video encoding, physics simulations, or cryptocurrency mining. The Vulkan test is the closest of the three, with the GTX 1070 scoring 22121 to the GTX 660’s 11415, a 93.8% margin. Even in this best-case scenario for the GTX 660, the GTX 1070 is nearly twice as fast.

The GTX 660’s only qualitative wins are practical rather than performance-based. Its 140 W TDP is 10 W lower than the GTX 1070’s 150 W, and its suggested PSU of 300 W is significantly less than the GTX 1070’s 450 W requirement. The GTX 660 also uses a single 6-pin power connector versus the GTX 1070’s 8-pin, making it easier to install in systems with older power supplies. Its 241 mm length is shorter than the GTX 1070’s 267 mm, which may fit better in compact cases. The GTX 660 also has two DVI outputs, while the GTX 1070 has only one, which could matter for users with legacy DVI-only monitors. However, the GTX 1070 counters with three DisplayPort 1.4a outputs and one HDMI 2.0 port, supporting modern high-refresh-rate and high-resolution displays.

In terms of benchmark coverage, the GTX 1070 has a much broader test suite, including PassMark DirectX 9 through 12 tests and a G2D score of 846. The GTX 660 only has Geekbench scores in the data, which limits direct comparisons. The GTX 1070’s PassMark G3D score of 13498 and GPU compute score of 6102 provide additional context for its dominance, though no comparable figures exist for the GTX 660.

FAQ

Q: Is the GTX 1070 worth upgrading to from a GTX 660?

A: The data strongly suggests yes. The GTX 1070 leads by 336.7% in Metal, 293.9% in OpenCL, and 93.8% in Vulkan, making it a transformative upgrade for any compute or modern gaming workload.

Q: Which card has better memory capacity?

A: The GTX 1070 has 8 GB of GDDR5 memory, while the GTX 660 has 2 GB. This fourfold difference is crucial for high-resolution textures and large data sets.

Q: Are both cards still supported by modern APIs?

A: Both support DirectX 12 and OpenGL 4.6, but the GTX 1070 supports DirectX 12 (12_1) and Vulkan 1.4, while the GTX 660 supports DirectX 12 (11_0) and Vulkan 1.2.175, with the GTX 1070 having the newer feature sets.

Q: What is the power consumption difference?

A: The GTX 1070 has a TDP of 150 W and requires a 450 W PSU, while the GTX 660 has a TDP of 140 W and requires a 300 W PSU. The GTX 660 is more power-friendly for older systems.

Q: Which card has higher bandwidth?

A: The GTX 1070’s memory bandwidth is 256.3 GB/s, compared to the GTX 660’s 144.2 GB/s, a difference of approximately 78% in favor of the GTX 1070.

Q: How do their percentile ranks compare?

A: The GTX 1070 is in the 47th percentile of all GPUs, while the GTX 660 is in the 45th percentile. Despite the large performance gap, their relative standing among all GPUs is close, highlighting the different eras they represent.

Head-to-Head Benchmarks

The three head-to-head benchmarks all favor the GTX 1070, but the margins vary considerably. The largest win comes in the Geekbench Metal test, where the GTX 1070 scores 18801 against the GTX 660’s 4305. This 336.7% delta is the most extreme outlier, indicating that the GTX 1070’s Pascal architecture is exceptionally well-suited to Metal’s low-level GPU access. For users running macOS or Metal-based applications, this is the single most important metric in the comparison.

The Geekbench OpenCL test shows a 293.9% delta, with the GTX 1070 scoring 44700 and the GTX 660 scoring 11347. OpenCL is a cross-platform compute standard, and this result demonstrates that the GTX 1070’s 6.463 TFLOPS of FP32 performance, versus the GTX 660’s 1.981 TFLOPS, translates directly into compute throughput. This makes the GTX 1070 a far more capable card for scientific computing, rendering, or any GPU-accelerated workflow.

The Vulkan test is the closest margin, yet still decisive. The GTX 1070 scores 22121, while the GTX 660 scores 11415, a 93.8% delta. Vulkan is a modern graphics API that scales well across different hardware, but the GTX 1070’s newer architecture and higher pixel rate (107.7 GPixel/s vs. 20.64 GPixel/s) ensure it still nearly doubles the GTX 660’s output. This result is notable because it suggests that even in workloads where the GTX 660 is relatively efficient, it cannot overcome the fundamental hardware deficit in shading units, TMUs, and ROPs.

The GTX 1070 also has additional benchmark data not present for the GTX 660. Its PassMark G3D score of 13498 and GPU compute score of 6102 are strong indicators of its overall capability, while its PassMark DirectX 11 score of 100 (normalized) shows solid legacy performance. The GTX 660 lacks these scores in the data, which limits deeper comparison, but the available head-to-head results are sufficient to establish a clear hierarchy.

Specification Differences

The specification sheets reveal a comprehensive generational leap. The GTX 1070 uses a 16 nm process, while the GTX 660 uses 28 nm, both from TSMC. The GTX 1070’s transistor count is 7,200 million versus 2,540 million for the GTX 660, and its die size is 314 mm² versus 221 mm². Transistor density improves from 11.5M per mm² to 22.9M per mm², a doubling that reflects the advanced node.

Clock speeds differ significantly: the GTX 1070 has a base clock of 1506 MHz and a boost of 1683 MHz, while the GTX 660 runs at 980 MHz base and 1032 MHz boost. Memory clocks also rise from 1502 MHz (6 Gbps effective) on the GTX 660 to 2002 MHz (8 Gbps effective) on the GTX 1070. Memory size, bus width, and bandwidth all favor the GTX 1070, as detailed earlier. The GTX 1070’s shading units (1920), TMUs (120), and ROPs (64) are exactly double, 1.5 times, and over 2.5 times the GTX 660’s respective counts of 960, 80, and 24.

Power and physical specs show minor differences: the GTX 1070 has a 150 W TDP and requires a 450 W PSU, while the GTX 660 has a 140 W TDP and a 300 W PSU requirement. The GTX 1070 uses a single 8-pin connector, while the GTX 660 uses a single 6-pin. Both are dual-slot cards, but the GTX 1070 is longer at 267 mm versus 241 mm. Display outputs differ: the GTX 1070 offers 1x DVI, 1x HDMI 2.0, and 3x DisplayPort 1.4a, while the GTX 660 offers 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2. Both use a PCIe 3.0 x16 interface, and neither has ray tracing or tensor cores. The GTX 1070’s release date is 2016-06-09, while the GTX 660’s is 2012-09-05, a gap of nearly four years that explains the architectural disparity.