AMD Radeon R9 370X vs NVIDIA GeForce GTX 660 Ti Comparison

Head-to-Head Benchmarks

The data paints a sharply split picture between these two end-of-life cards. In the only two head-to-head tests available, each GPU claims one decisive victory, with the margins telling the real story. The AMD Radeon R9 370X dominates Geekbench OpenCL, scoring 25,893 against the GTX 660 Ti's 15,113 — a 71.3% advantage. That is not a marginal lead; it is a generational gap in raw compute throughput.

Flip to Geekbench Vulkan, and the tables turn completely. The NVIDIA GeForce GTX 660 Ti posts 15,012, while the R9 370X manages just 9,018. That gives NVIDIA a 39.9% edge in the newer API workload. The Vulkan result is particularly striking because the R9 370X's GCN architecture is often associated with strong low-level API support, yet the data here shows the Kepler card pulling far ahead.

Looking at the broader benchmark averages, the R9 370X holds a clear overall lead. Its average benchmark score of 15,862 compares to the GTX 660 Ti's 15,063 — roughly a 5.3% aggregate advantage. The AMD card also sits in the 58th percentile of all GPUs, one point ahead of the GTX 660 Ti's 57th percentile. These are close but consistent margins in AMD's favor across the whole test suite.

The nearest-rival data adds useful context. The R9 370X's average score lands within 0.1% of the AMD Radeon RX 7700 (15,852) and 1.2% of the Radeon Pro W5500 (15,679), while sitting 1.7% behind the NVIDIA GeForce RTX 3060 Ti (16,129). The GTX 660 Ti, meanwhile, is effectively tied with the AMD Radeon Pro 560X (15,082, -0.1%) and trails the AMD Radeon RX 7600 (15,171) by 0.7%. This positions the R9 370X as a mid-pack modern competitor, while the GTX 660 Ti hovers near the bottom of the same performance tier.

Architecture Differences

These GPUs come from different architectural philosophies despite sharing the same 28 nm TSMC process node. The AMD Radeon R9 370X uses the Trinidad chip built on GCN 1.0, part of the Pirate Islands (R9 300) generation. The NVIDIA GeForce GTX 660 Ti uses the GK104 die with Kepler architecture, from the GeForce 600 family. Both are end-of-life products, but they launched three years apart: the GTX 660 Ti arrived in August 2012, the R9 370X in August 2015.

Transistor counts and die sizes tell a surprising story. The GTX 660 Ti packs 3,540 million transistors onto a 294 mm² die, while the R9 370X uses 2,800 million transistors on a much smaller 212 mm² die. That gives the AMD chip a higher transistor density — 13.2 million per mm² versus 12.0 million for NVIDIA. Despite the smaller transistor budget, the R9 370X still manages competitive performance.

Core configurations differ significantly. The R9 370X has 1,280 shading units, 80 texture mapping units, and 32 ROPs. The GTX 660 Ti counters with 1,344 shading units, 112 TMUs, but only 24 ROPs. The NVIDIA card's extra TMUs help explain its higher texture rate of 109.8 GTexel/s versus 82.40 GTexel/s for AMD. However, the R9 370X's higher ROP count contributes to its 32.96 GPixel/s pixel rate, ahead of the GTX 660 Ti's 27.44 GPixel/s.

Clock speeds are close but favor AMD slightly. The R9 370X runs at a 980 MHz base and 1030 MHz boost, while the GTX 660 Ti operates at 915 MHz base and 980 MHz boost. Memory clocks are similar, with the NVIDIA card at 1502 MHz (6 Gbps effective) and the AMD at 1400 MHz (5.6 Gbps effective). Both use 2 GB of GDDR5, but the R9 370X has a 256-bit memory bus versus the GTX 660 Ti's 192-bit bus. That yields 179.2 GB/s of bandwidth for AMD against 144.2 GB/s for NVIDIA.

The FP32 compute figures are nearly identical — 2.637 TFLOPS for the R9 370X and 2.634 TFLOPS for the GTX 660 Ti — yet the benchmark results diverge wildly depending on the API. This suggests the architectures handle workloads differently rather than one being universally faster.

Both cards draw power through dual 6-pin connectors and recommend a 450 W PSU. The R9 370X has a 180 W TDP, while the GTX 660 Ti is more efficient at 150 W. Physical dimensions differ too: the AMD card is 221 mm long, the NVIDIA card 241 mm. Both are dual-slot designs. Display outputs are identical: 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2.

API support shows a notable divergence. The R9 370X supports DirectX 12 (11_1), while the GTX 660 Ti only reaches DirectX 12 (11_0). Both support OpenGL 4.6. Vulkan support is close, with the NVIDIA card at version 1.2.175 and AMD at 1.2.170. The gap in Vulkan performance is therefore not explained by API version differences.

The Verdict

The benchmark data supports a clear but nuanced recommendation. The AMD Radeon R9 370X is the better overall card, with a 5.3% higher average benchmark score and a superior OpenCL result by 71.3%. This makes it the choice for compute-oriented workloads and general-purpose performance. The R9 370X also offers more memory bandwidth (179.2 GB/s versus 144.2 GB/s) and a higher pixel rate, which can matter in certain rendering scenarios.

However, the NVIDIA GeForce GTX 660 Ti is not without its arguments. Its Vulkan score of 15,012 crushes the R9 370X's 9,018 by 39.9%, making it the stronger option for Vulkan-based games and applications. The GTX 660 Ti also draws 30 W less power (150 W versus 180 W), runs cooler per watt, and has a higher texture rate of 109.8 GTexel/s.

The choice depends on the workload. Data-driven buyers who prioritize compute performance and modern API overhead should select the R9 370X. Gamers targeting Vulkan titles specifically may find the GTX 660 Ti delivers better frame pacing in those scenarios. The R9 370X's launch MSRP was 199 USD, while the GTX 660 Ti launched at 299 USD — a difference reflected in the AMD card's stronger value proposition, though both are now end-of-life products.

FAQ

Q: Which card has the higher average benchmark score?

A: The AMD Radeon R9 370X, with an average benchmark score of 15,862 compared to the NVIDIA GeForce GTX 660 Ti's 15,063 — a 5.3% difference.

Q: How large is the OpenCL performance gap?

A: The R9 370X scores 25,893 in Geekbench OpenCL, while the GTX 660 Ti scores 15,113. That represents a 71.3% advantage for AMD.

Q: Does the GTX 660 Ti beat the R9 370X in any benchmark?

A: Yes. In Geekbench Vulkan, the GTX 660 Ti scores 15,012 against the R9 370X's 9,018, a 39.9% win for NVIDIA.

Q: Which card has more memory bandwidth?

A: The R9 370X, with 179.2 GB/s over a 256-bit bus. The GTX 660 Ti offers 144.2 GB/s over a 192-bit bus.

Q: What are the TDP ratings for these GPUs?

A: The R9 370X has a 180 W TDP, while the GTX 660 Ti is rated at 150 W. Both recommend a 450 W PSU.

Q: How do these cards compare to modern GPUs in the nearest-rival data?

A: The R9 370X sits within 1.7% of the RTX 3060 Ti's average score, while the GTX 660 Ti trails the RX 7600 by 0.7%. The GTX 660 Ti's closest rival is the Radeon Pro 560X, just 0.1% behind.

Where Each One Wins

AMD Radeon R9 370X wins in:

• Compute workloads: The 71.3% OpenCL lead over the GTX 660 Ti is the single largest margin in the head-to-head data.
• Overall performance: A 5.3% higher average benchmark score and a one-point percentile advantage place it ahead in general-purpose tasks.
• Pixel throughput: 32.96 GPixel/s versus 27.44 GPixel/s, aided by 32 ROPs against 24.
• Memory bandwidth: 179.2 GB/s over a 256-bit bus offers a 24.3% advantage over the GTX 660 Ti's 144.2 GB/s.
• DirectX 12 feature level: Support for 11_1 versus 11_0 on the NVIDIA card.

NVIDIA GeForce GTX 660 Ti wins in:

• Vulkan performance: A 39.9% lead in Geekbench Vulkan suggests better low-level API optimization on Kepler.
• Texture throughput: 109.8 GTexel/s versus 82.40 GTexel/s, thanks to 112 TMUs against 80.
• Power efficiency: 150 W TDP versus 180 W, a 16.7% reduction in power draw.
• Transistor count: 3,540 million versus 2,800 million, offering more raw hardware resources.
• Shading units and TMUs: 1,344 shaders and 112 TMUs outnumber the R9 370X's 1,280 shaders and 80 TMUs.

The data ultimately shows two cards optimized for different eras of graphics workloads. The R9 370X leverages its wider memory bus and modern architecture for compute-heavy tasks, while the GTX 660 Ti's Kepler design excels in texture-bound scenarios and Vulkan API efficiency. Buyers should weigh the specific applications they intend to run, since neither card dominates universally.