NVIDIA GeForce GTX 580 vs NVIDIA GeForce GTX 660 Ti Comparison

The GeForce GTX 580 and GeForce GTX 660 Ti are both end-of-life NVIDIA parts, but they represent two very different design philosophies from successive generations. The data shows the GTX 580 edges out the GTX 660 Ti in raw OpenCL compute, scoring 15,283 to 15,113, a 1.1% margin. However, the GTX 660 Ti counters with a broader feature set, including Vulkan support and a lower power draw. The verdict is clear: the GTX 580 is the pick for pure legacy compute performance, while the GTX 660 Ti is the more balanced, modern-feeling card for those who need API versatility and efficiency.

The Verdict

Based strictly on the benchmark data, the GTX 580 is the faster card. It wins the only head-to-head benchmark available, the Geekbench OpenCL test, with a score of 15,283 against the GTX 660 Ti’s 15,113. That 1.1% lead places the GTX 580 in the 58th percentile of all GPUs, one point higher than the GTX 660 Ti’s 57th percentile. For users whose primary concern is sheer compute throughput in OpenCL workloads, the GTX 580 is the definitive choice.

Yet, the GTX 660 Ti is not without its own merits. Its average benchmark score of 15,063 is only 1.4% behind the GTX 580’s 15,283, making the performance gap nearly negligible in real-world terms. The GTX 660 Ti’s advantage lies elsewhere: it supports Vulkan 1.2.175, whereas the GTX 580 has no Vulkan support listed. It also draws significantly less power (150 W vs. 244 W) and requires a less demanding 450 W power supply versus the GTX 580’s 550 W suggestion. For a system builder prioritizing modern API compatibility and lower system strain, the GTX 660 Ti is the more sensible pick, even if it loses the single compute benchmark.

Architecture Differences

The two cards are built on fundamentally different architectures and process nodes. The GTX 580 uses the GF110 chip, based on the Fermi 2.0 architecture, manufactured on TSMC’s 40 nm process. This is an older, larger design with 3,000 million transistors packed into a 520 mm² die, yielding a transistor density of just 5.8 million transistors per square millimeter. In contrast, the GTX 660 Ti uses the GK104 chip, based on the Kepler architecture, built on a more advanced 28 nm process. It fits 3,540 million transistors into a much smaller 294 mm² die, achieving a density of 12.0 million transistors per square millimeter, more than double that of the GTX 580.

These architectural differences manifest in the core configurations. The GTX 580 has 512 shading units, 64 texture mapping units (TMUs), and 48 render output units (ROPs). The GTX 660 Ti, despite being a smaller die, packs more shading units (1,344) and more TMUs (112), but far fewer ROPs (24). The GTX 660 Ti also has explicit base and boost clocks of 915 MHz and 980 MHz, respectively, while the GTX 580’s base and boost clocks are not listed. The GTX 580 relies on its memory subsystem: a 384-bit bus with 1536 MB of GDDR5 delivering 192.4 GB/s of bandwidth. The GTX 660 Ti uses a narrower 192-bit bus but pairs it with 2 GB of GDDR5, running at a higher effective speed of 6 Gbps, yet achieves lower overall bandwidth at 144.2 GB/s.

FAQ

Q: Which card has higher raw compute performance?

A: The GTX 580. It scores 15,283 in Geekbench OpenCL, which is 1.1% higher than the GTX 660 Ti’s 15,113. Its FP32 throughput is 1.581 TFLOPS, though the GTX 660 Ti actually has a higher listed FP32 figure at 2.634 TFLOPS.

Q: Does the GTX 660 Ti support newer graphics APIs?

A: Yes. The GTX 660 Ti supports Vulkan 1.2.175, while the GTX 580 has no Vulkan support listed. Both cards support DirectX 12 (11_0) and OpenGL 4.6.

Q: Which card consumes less power?

A: The GTX 660 Ti, by a substantial margin. Its TDP is 150 W, compared to the GTX 580’s 244 W. The GTX 660 Ti also suggests a 450 W power supply, whereas the GTX 580 suggests a 550 W unit.

Q: How do their memory configurations differ?

A: The GTX 580 has 1536 MB of GDDR5 on a 384-bit bus, yielding 192.4 GB/s bandwidth. The GTX 660 Ti has 2 GB of GDDR5 on a 192-bit bus, yielding 144.2 GB/s bandwidth. The GTX 660 Ti’s memory runs at 6 Gbps effective, compared to the GTX 580’s 4 Gbps.

Q: Which card is physically smaller?

A: The GTX 660 Ti is shorter at 241 mm (9.5 inches) in length, while the GTX 580 is 267 mm (10.5 inches) long. Both are dual-slot cards.

Q: What are their launch prices?

A: The GTX 580 had a launch MSRP of 499 USD, while the GTX 660 Ti had a launch MSRP of 299 USD.

Specification Differences

The following specifications differ between the two cards, based solely on the data provided:

• Chip: GF110 (GTX 580) vs. GK104 (GTX 660 Ti)
• Architecture: Fermi 2.0 vs. Kepler
• Generation: GeForce 500 vs. GeForce 600
• Process Node: 40 nm vs. 28 nm
• Transistors: 3,000 million vs. 3,540 million
• Die Size: 520 mm² vs. 294 mm²
• Transistor Density: 5.8M / mm² vs. 12.0M / mm²
• Memory Clock: 1002 MHz / 4 Gbps effective vs. 1502 MHz / 6 Gbps effective
• Memory Size: 1536 MB vs. 2 GB
• Memory Bus Width: 384 bit vs. 192 bit
• Memory Bandwidth: 192.4 GB/s vs. 144.2 GB/s
• Shading Units: 512 vs. 1,344
• TMUs: 64 vs. 112
• ROPs: 48 vs. 24
• Pixel Rate: 24.70 GPixel/s vs. 27.44 GPixel/s
• Texture Rate: 49.41 GTexel/s vs. 109.8 GTexel/s
• FP32: 1.581 TFLOPS vs. 2.634 TFLOPS
• TDP: 244 W vs. 150 W
• Power Connectors: 1x 6-pin + 1x 8-pin vs. 2x 6-pin
• Suggested PSU: 550 W vs. 450 W
• Bus Interface: PCIe 2.0 x16 vs. PCIe 3.0 x16
• Display Outputs: 2x DVI, 1x mini-HDMI 1.3a vs. 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2
• Vulkan Support: None vs. 1.2.175
• Dimensions (Length): 267 mm / 10.5 inches vs. 241 mm / 9.5 inches
• Release Date: 2010-11-08 vs. 2012-08-15
• Predecessor: GeForce 400 vs. GeForce 500
• Successor: GeForce 600 vs. GeForce 700
• Launch MSRP: 499 USD vs. 299 USD

Head-to-Head Benchmarks

The only direct benchmark comparison available is the Geekbench OpenCL test. In this test, the GTX 580 scores 15,283, while the GTX 660 Ti scores 15,113. The delta is 1.1%, favoring the GTX 580. This single result gives the GTX 580 one win in the head-to-head tally, with the GTX 660 Ti earning zero wins.

Looking at the broader context, the GTX 580’s average benchmark score is 15,283, which is its only Geekbench OpenCL result. The GTX 660 Ti has two benchmark entries: a Geekbench OpenCL score of 15,113 and a Geekbench Vulkan score of 15,012. Its average across both is 15,063. This means the GTX 660 Ti’s Vulkan score is actually lower than its OpenCL score, suggesting that its API versatility does not necessarily translate into higher raw performance. However, the GTX 660 Ti’s texture rate of 109.8 GTexel/s is more than double the GTX 580’s 49.41 GTexel/s, and its pixel rate of 27.44 GPixel/s also exceeds the GTX 580’s 24.70 GPixel/s. These figures indicate that in tasks heavily reliant on texture fetching or pixel fill, the GTX 660 Ti could outperform the GTX 580, even though the OpenCL compute test shows the opposite.

Where Each One Wins

The GTX 580 wins in raw compute and memory bandwidth. Its Geekbench OpenCL score is higher, and it delivers 192.4 GB/s of memory bandwidth compared to the GTX 660 Ti’s 144.2 GB/s. The wider 384-bit bus and higher bandwidth make it better suited for memory-intensive compute tasks that are not entirely dependent on shader count. Its 48 ROPs, double the GTX 660 Ti’s 24, also suggest a stronger fill-rate capability in certain legacy workloads, despite its lower pixel rate on paper. The GTX 580 is the choice for users who prioritize raw OpenCL compute results and have a robust power supply (550 W suggested) to handle its 244 W TDP.

The GTX 660 Ti wins on efficiency, modern features, and API support. It draws 94 W less power (150 W vs. 244 W) and requires a less demanding 450 W power supply. Its support for Vulkan 1.2.175 is a significant differentiator, as the GTX 580 has no Vulkan support listed. The GTX 660 Ti also offers a more modern display output configuration, including a DisplayPort 1.2 and HDMI 1.4a, versus the GTX 580’s mini-HDMI 1.3a. Its higher texture rate (109.8 GTexel/s) and pixel rate (27.44 GPixel/s) indicate it would excel in games or applications that are heavily geometry- or texture-bound. For a system where power consumption, thermal output, or modern driver/API compatibility are primary concerns, the GTX 660 Ti is the superior option, even with its slightly lower compute benchmark score and a narrower 192-bit memory bus.