AMD Radeon RX 580 vs NVIDIA GeForce GTX 660 Ti Comparison

Head-to-Head Benchmarks

The recorded data shows a decisive outcome in the two shared benchmark tests. In Geekbench OpenCL, the AMD Radeon RX 580 scores 37,453, while the NVIDIA GeForce GTX 660 Ti manages 15,113. That is a 59.6% deficit for the NVIDIA card, meaning the RX 580 delivers well over double the compute throughput in this API. The gap is even larger in Geekbench Vulkan: the RX 580 posts 45,173 against 15,012 for the GTX 660 Ti, a 66.8% difference. These are not marginal gains; the AMD part essentially triples the NVIDIA card's Vulkan score.

The average benchmark score tells a similar story, though the margin narrows because the RX 580's average includes a wider set of tests. The GTX 660 Ti has an average of 15,063 across its two recorded benchmarks, placing it in the 57th percentile of all GPUs in the database. The RX 580 averages 12,928 across eleven tests, putting it in the 53rd percentile. That comparison is slightly misleading, however, because the GTX 660 Ti's average is drawn only from OpenCL and Vulkan, where it performs at its relative best. The RX 580's average includes DirectX 9, 10, 11, 12, compute, and 2D tests, which drag its mean down even while it dominates the two compute-oriented workloads shared with the older card.

Look at the individual RX 580 scores to see where the strength lies. Its Geekbench Metal result is 45,235, essentially tied with its Vulkan score. Passmark G3D gives it 8,813, while Passmark DirectX 11 reaches 60, DirectX 12 drops to 43, and DirectX 9 hits 124. The GPU compute score is 3,488. These numbers indicate a card that is far more capable in modern, low-level APIs and raw compute than in legacy DirectX workloads. The GTX 660 Ti has no comparable DirectX or Metal data in the database, so the head-to-head comparison is limited to OpenCL and Vulkan, but those two tests are sufficient to establish a clear hierarchy.

For context on the GTX 660 Ti, its nearest rivals in the database are the AMD Radeon Pro 560X at 15,082 (0.1% ahead), the AMD Radeon RX 7600 at 15,171 (0.7% ahead), the NVIDIA GeForce RTX 3050 OEM at 15,199 (0.9% ahead), and the AMD Radeon 680M at 15,270 (1.4% ahead). This places the GTX 660 Ti right at the edge of a cluster of modern entry-level and integrated parts. It is competitive with those, but that cluster sits far below the RX 580's compute peak.

The RX 580's nearest rivals tell a different story. The NVIDIA GeForce RTX 3050 Ti Mobile averages 12,940 (0.1% ahead), the GTX 1660 SUPER averages 12,986 (0.4% ahead), the AMD Radeon 740M averages 12,870 (0.5% behind), and the AMD FirePro W5100 averages 12,847 (0.6% behind). So the RX 580 sits in the middle of a group of cards that are all within 1% of each other, but that group is anchored around the 12,900 score range, not the 15,000 range. The GTX 660 Ti's average is about 16.5% higher than the RX 580's average, which seems contradictory given the head-to-head results. The resolution is that the RX 580's broader test suite pulls its average down, while the GTX 660 Ti benefits from having only two favorable tests recorded.

The Verdict

From the data, the AMD Radeon RX 580 is the clear winner for anyone running OpenCL or Vulkan workloads. It beats the GTX 660 Ti by 59.6% and 66.8% respectively in those two APIs. If your primary use case involves compute, GPU-accelerated rendering, or modern game engines built around Vulkan, the RX 580 is the only rational choice here. The GTX 660 Ti cannot compete in those scenarios, and the margin is so large that no driver tweak or overclock would close it.

However, the GTX 660 Ti does have one statistical advantage: its average benchmark score sits at 15,063, which is 16.5% higher than the RX 580's 12,928 average. This is because the RX 580's average includes several weak legacy DirectX scores (Passmark DirectX 12 at 43, DirectX 10 at 46, DirectX 11 at 60) that drag it down. If you only care about the two tests both cards have in common, the RX 580 wins outright. If you want a balanced view across many workloads, the GTX 660 Ti looks better on paper, but that is an artifact of missing data for the NVIDIA card, not a real performance advantage.

Who should pick which? Choose the RX 580 if you need 8 GB of VRAM, a 256-bit memory bus, and modern API support including Vulkan 1.3 and DirectX 12 (12_0). It also has a much higher pixel rate (42.88 GPixel/s versus 27.44), texture rate (193.0 GTexel/s versus 109.8), and FP32 throughput (6.175 TFLOPS versus 2.634). The GTX 660 Ti is only preferable if you are building a retro system around a 2012-era card and do not care about compute performance. But even then, the RX 580's launch MSRP was 229 USD, compared to 299 USD for the GTX 660 Ti, and it offers more memory, higher clocks, and better efficiency per transistor.

Architecture Differences

The GTX 660 Ti uses NVIDIA's Kepler architecture on the GK104 chip, fabricated on a 28 nm process at TSMC. It packs 3,540 million transistors into a 294 mm² die, giving a transistor density of 12.0 million per square millimeter. The RX 580 uses AMD's GCN 4.0 architecture on the Polaris 20 chip, built on a 14 nm process at GlobalFoundries. It has 5,700 million transistors on a smaller 232 mm² die, resulting in a density of 24.6 million per square millimeter. That is more than double the density, which explains how AMD fits nearly double the transistors into a smaller area.

Kepler is a graphics-centric design from 2012, oriented around the GeForce 600 generation. It supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. GCN 4.0 is a compute-oriented architecture from 2017, part of the Polaris (RX 500) generation. It supports DirectX 12 (12_0), OpenGL 4.6, and Vulkan 1.3. The newer Vulkan revision and the full DirectX 12 feature level give the RX 580 a software advantage that shows up directly in the Vulkan benchmark gap.

The RX 580 also has explicit FP16 support at a 1:1 ratio with FP32, meaning it can process half-precision floats at the same rate as full precision. The GTX 660 Ti lists no FP16 capability, so it likely relies on emulation or lacks it entirely. This matters for machine learning inference and some compute workloads that use FP16. Additionally, the RX 580's shading unit count is 2,304 versus 1,344 for the GTX 660 Ti, and its TMU count is 144 versus 112. The ROP count is 32 versus 24. These are not small differences; they compound to explain the massive compute gap.

Specification Differences

The two cards differ on nearly every specification that matters. The GTX 660 Ti has 2 GB of GDDR5 memory on a 192-bit bus, yielding 144.2 GB/s bandwidth. The RX 580 has 8 GB of GDDR5 on a 256-bit bus, yielding 256.0 GB/s bandwidth. That is 4 times the capacity and 77.5% more bandwidth. Clock speeds favor the RX 580 as well: 1257 MHz base and 1340 MHz boost versus 915 MHz base and 980 MHz boost. Memory clock is 2000 MHz (8 Gbps effective) versus 1502 MHz (6 Gbps effective).

Power draw is higher on the RX 580 at 185 W TDP versus 150 W for the GTX 660 Ti, though both recommend the same 450 W PSU. The power connector differs: the RX 580 uses a single 8-pin, while the GTX 660 Ti requires two 6-pin connectors. Display outputs favor the RX 580 with one HDMI 2.0b and three DisplayPort 1.4a, versus two DVI, one HDMI 1.4a, and one DisplayPort 1.2 on the GTX 660 Ti. Both are dual-slot cards at 241 mm length. The RX 580 is newer by almost five years, releasing in April 2017 versus August 2012, and its predecessor is Arctic Islands while the successor is Vega. The GTX 660 Ti's predecessor is GeForce 500 and successor is GeForce 700.

FAQ

Q: Which card is faster in Vulkan?

A: The AMD Radeon RX 580 scores 45,173 in Geekbench Vulkan, while the GTX 660 Ti scores 15,012. The RX 580 is 66.8% faster.

Q: How much memory do these cards have?

A: The GTX 660 Ti has 2 GB of GDDR5 on a 192-bit bus. The RX 580 has 8 GB of GDDR5 on a 256-bit bus.

Q: What is the FP32 compute throughput difference?

A: The RX 580 delivers 6.175 TFLOPS FP32, while the GTX 660 Ti delivers 2.634 TFLOPS. The RX 580 is roughly 2.3 times higher.

Q: Which card has better API support?

A: The RX 580 supports DirectX 12 (12_0) and Vulkan 1.3. The GTX 660 Ti supports DirectX 12 (11_0) and Vulkan 1.2.175.

Q: What is the power draw of each card?

A: The RX 580 has a 185 W TDP and uses one 8-pin connector. The GTX 660 Ti has a 150 W TDP and uses two 6-pin connectors.

Q: How do their average benchmark scores compare?

A: The GTX 660 Ti averages 15,063, while the RX 580 averages 12,928. However, the RX 580's average includes many legacy DirectX tests that score low, while the GTX 660 Ti only has two compute tests recorded.

Where Each One Wins

The AMD Radeon RX 580 wins in compute-heavy scenarios. Its OpenCL score of 37,453 is 2.5 times the GTX 660 Ti's 15,113. Its Vulkan score of 45,173 is 3 times the GTX 660 Ti's 15,012. For any workload that uses these APIs, such as GPGPU compute, video encoding acceleration, or modern game engines built on Vulkan, the RX 580 is the superior card by a wide margin. It also wins on memory capacity and bandwidth, which matters for texture-heavy workloads and larger datasets. The 8 GB framebuffer is especially relevant for modded games or high-resolution textures that would exceed 2 GB.

The NVIDIA GeForce GTX 660 Ti wins in one narrow statistical sense: its average benchmark score is higher because the database only has two tests for it, both of which are compute-oriented and relatively favorable. It also has a lower TDP at 150 W versus 185 W, so it draws less power under load. If you are building a low-power system or have a PSU with limited connectors, the GTX 660 Ti's two 6-pin requirement might actually be easier to accommodate than a single 8-pin, depending on your PSU. But the performance gap is so large that power savings are the only practical reason to choose it.

For legacy DirectX workloads, the RX 580 has recorded scores in DirectX 9 (124), DirectX 10 (46), DirectX 11 (60), and DirectX 12 (43) via Passmark. These are low numbers, but the GTX 660 Ti has no comparable data, so it is impossible to say the NVIDIA card wins there. The RX 580's DirectX 12 support at feature level 12_0 gives it a structural advantage for newer titles, even if the Passmark numbers look weak. In practice, the RX 580 is the card to pick for any modern use case, while the GTX 660 Ti is only relevant for a period-correct build where power draw is the priority.