NVIDIA GeForce GTX 660

The NVIDIA GeForce GTX 660 is a Kepler-architecture graphics card from the GeForce 600 generation, built on TSMC's 28 nm process with 2,540 million transistors on a 221 mm² die. It was released on September 5, 2012, and has since reached end-of-life status. With a 44th percentile ranking among all GPUs and an average benchmark score of 9000, this card occupies a distinctly mid-range position in the historical performance hierarchy. The following analysis draws exclusively from benchmark data and technical specifications to assess its standing against immediate rivals.

Benchmark Performance

The GTX 660's average benchmark score of 9000 places it in a tightly contested performance cluster. Against its nearest rivals, the data shows a razor-thin margin of separation. The card trails the NVIDIA TITAN V CEO Edition by just 0.4%, with the latter scoring 9037. This negligible delta is within run-to-run variance and effectively places the two cards at parity in synthetic workloads. Similarly, the GTX 660 sits 1.2% behind the NVIDIA GeForce MX330, which scores 9108 — again a marginal difference that offers no practical performance advantage to either card in real-world scenarios.

Looking at the other direction, the GTX 660 holds a 1.1% lead over the NVIDIA GeForce GTX 560, which averages 8899. The gap widens slightly against the AMD Radeon R9 M265X, where the GTX 660 is 1.7% ahead of that card's 8851 average. In all four comparisons, the deltas remain under two percentage points, indicating that this performance tier is defined by incremental differences rather than generational leaps. The Geekbench results paint a more detailed picture: the card scores 11347 in OpenCL and 11368 in Vulkan, showing strong consistency across compute APIs. The Metal score of 4284 is considerably lower, reflecting the card's age relative to Apple's modern graphics stack rather than a fundamental compute weakness.

What these numbers reveal is a GPU that was competitive at launch but has aged into a narrow performance band. The 1.981 TFLOPS of FP32 compute power and 82.56 GTexel/s texture fill rate are modest by contemporary standards, yet the benchmark deltas suggest that the GTX 660 still trades blows with much newer entry-level parts. The pixel rate of 20.64 GPixel/s further contextualizes its capabilities — adequate for 1080p gaming at medium settings in its heyday, but increasingly strained by modern workloads.

Ray Tracing and Feature Set

The GTX 660 does not include dedicated ray tracing cores or tensor cores, as these hardware units were not part of the Kepler architecture. Instead, the card relies on 960 shading units, 80 texture mapping units, and 24 ROPs to handle all rendering tasks. The absence of RT and tensor hardware means the card cannot accelerate ray-traced effects or AI-based features like DLSS, limiting its relevance for contemporary gaming features that depend on such silicon.

API support is surprisingly robust for a card of this vintage. The GTX 660 supports DirectX 12 with a feature level of 11_0, meaning it can run DirectX 12 titles but lacks the higher-tier features like mesh shaders or variable rate shading that newer architectures provide. OpenGL 4.6 and Vulkan 1.2.175 support are both present, giving the card access to modern cross-platform graphics APIs. The Vulkan score of 11368 in Geekbench confirms that the card can leverage this API effectively, though the absence of hardware-accelerated ray tracing remains a definitive limitation. Display output is handled through 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2, which covers the standard connectivity options of its era.

Memory Subsystem

The memory configuration consists of 2 GB of GDDR5 running at an effective 6 Gbps across a 192-bit bus. This combination yields a bandwidth of 144.2 GB/s. For a card in this performance class, the 192-bit bus width is a notable compromise — it provides sufficient bandwidth for 1080p gaming but becomes a bottleneck at higher resolutions where texture data demands exceed what the memory subsystem can deliver.

The 2 GB VRAM capacity was generous for 2012 but has become a limiting factor in modern titles, particularly at 1440p or 4K where texture packs routinely exceed this allocation. The 144.2 GB/s bandwidth, while competitive at launch, is roughly half of what contemporary mid-range cards offer, and this constraint manifests in scenarios with heavy anti-aliasing or high-resolution texture streaming. The card's memory clock of 1502 MHz (6 Gbps effective) is respectable for its generation, yet the narrow bus width means the card cannot fully compensate for its modest capacity through raw throughput. For users targeting 1080p with reduced texture quality, this subsystem remains workable; for anything beyond that, the memory becomes the primary performance limiter.

FAQ

Q: Does the GTX 660 support hardware ray tracing?

A: No, the GTX 660 does not include dedicated ray tracing cores, as these were not part of the Kepler architecture. It relies on its 960 shading units for all rendering tasks.

Q: What is the card's average benchmark score and how does it compare to the NVIDIA GeForce GTX 560?

A: The GTX 660 averages 9000 in benchmark scores, which is 1.1% higher than the GTX 560's 8899 average.

Q: What DirectX version can the GTX 660 run?

A: The card supports DirectX 12 with a feature level of 11_0, meaning it can run DirectX 12 titles but without the highest-tier feature set.

Q: How much memory bandwidth does the GTX 660 have?

A: The memory subsystem provides 144.2 GB/s of bandwidth, achieved with 2 GB of GDDR5 memory on a 192-bit bus running at 6 Gbps effective.

Q: What is the card's thermal design power and power connector requirement?

A: The GTX 660 has a TDP of 140 W and requires a single 6-pin power connector, with a suggested power supply rating of 300 W.

Q: How does the GTX 660 perform in Vulkan compute workloads?

A: The card scores 11368 in Geekbench's Vulkan test, which is nearly identical to its OpenCL score of 11347, indicating consistent compute performance across both APIs.

How It Compares

NVIDIA TITAN V CEO Edition: The GTX 660 trails this card by a mere 0.4% in average benchmark scores, with 9000 versus 9037. This effectively puts the two at performance parity despite the TITAN V being a far more advanced product. The delta is so small that it would not be perceptible in any practical workload, though the TITAN V's additional features and modern architecture make it the more capable card overall.

NVIDIA GeForce GTX 560: The GTX 660 leads its predecessor by 1.1%, scoring 9000 against 8899. This modest improvement reflects the Kepler architecture's efficiency gains over the older Fermi design. The 1.1% delta is underwhelming as a generational uplift, suggesting that the GTX 660 was an incremental step rather than a transformative upgrade over the GTX 560.

NVIDIA GeForce MX330: The MX330 holds a 1.2% advantage over the GTX 660, scoring 9108 versus 9000. This modern entry-level laptop GPU edges out the older desktop card by a hair, demonstrating how far integrated and low-power designs have come. The near-tie means that the GTX 660 remains competitive with contemporary budget offerings despite being nearly a decade older.

AMD Radeon R9 M265X: The GTX 660 leads this AMD mobile GPU by 1.7%, with scores of 9000 and 8851 respectively. This is the largest margin in the rival group, yet it remains under two percentage points. The R9 M265X's lower score reflects its mobile positioning, but the tight delta underscores how closely these disparate products cluster in synthetic benchmark performance.