AMD Radeon R9 M375X vs NVIDIA GeForce GTX 660 Comparison

NVIDIA GeForce GTX 660 and AMD Radeon R9 M375X are two end-of-life graphics cards with a three-year gap between their release dates, the GTX 660 launching in September 2012 and the R9 M375X in May 2015. The benchmark data positions them close in overall standing, with the GTX 660 at the 45th percentile of all GPUs and the R9 M375X at the 43rd percentile, yet their actual performance scores reveal a decisive separation. The GTX 660 achieves an average benchmark score of 9022, while the R9 M375X trails at 8325, a difference that places the NVIDIA card roughly 8% ahead on aggregate. This gap is consistent across the two shared tests, with the GTX 660 winning both head-to-head comparisons by wide margins. For a builder choosing between these legacy parts, the data points clearly toward the GTX 660 for raw compute performance, though the R9 M375X’s architecture offers its own trade-offs.

Head-to-Head Benchmarks

The head-to-head data covers two compute workloads: Geekbench OpenCL and Geekbench Vulkan. In OpenCL, the NVIDIA GeForce GTX 660 scores 11347, while the AMD Radeon R9 M375X scores 8273. That is a delta of 37.2% in favor of the GTX 660, a substantial lead that reflects the NVIDIA card’s higher shading unit count and memory bandwidth. The Vulkan test shows a similar outcome: the GTX 660 posts 11415 against the R9 M375X’s 8377, a 36.3% advantage. Both results are unambiguous wins for the GTX 660, giving it a 2-0 record in the head-to-head set.

Breaking down why the GTX 660 leads, the specification sheet shows it carries 960 shading units versus 640 on the R9 M375X, and its texture units number 80 against 40. The pixel rate also favors NVIDIA at 20.64 GPixel/s versus 16.24 GPixel/s, and the texture rate is nearly double at 82.56 GTexel/s compared to 40.60 GTexel/s. Memory bandwidth is another major divider: the GTX 660’s 144.2 GB/s is exactly double the R9 M375X’s 72.00 GB/s, which explains why the NVIDIA card sustains higher throughput in compute-bound tasks. The GTX 660’s FP32 output is listed as 1.981 TFLOPS, while the R9 M375X delivers 1,299.2 GFLOPS, a 52% raw compute advantage for the older card.

Context from nearest rivals reinforces the gap. The GTX 660’s average score of 9022 sits within 0.2% of the NVIDIA TITAN V CEO Edition (9037) and 0.4% below the GeForce GTX 560 (9058), while it beats the AMD Radeon 550X (8918) by 1.2% and the Radeon Pro WX 5100 (8863) by 1.8%. The R9 M375X’s average of 8325 is 0.7% above the Quadro K1200 (8265) but 1.2% below the GeForce GTX 675MX (8427) and 1.6% below the GeForce MX330 (8458). In short, the GTX 660 competes in a higher performance tier than the R9 M375X, despite being released earlier.

The Verdict

The data’s verdict is straightforward: the NVIDIA GeForce GTX 660 is the stronger card for anyone prioritizing compute performance in OpenCL or Vulkan workloads. Its wins of 37.2% and 36.3% in the two shared benchmarks are not marginal; they represent a generation-level gap in throughput. The GTX 660 also has a higher average benchmark score (9022 vs 8325), a higher percentile ranking (45th vs 43rd), and a superior specification sheet across every major compute metric, including shading units, texture units, ROPs, memory bandwidth, and FP32 output. If all you care about is benchmark scores, the choice is the GTX 660, no contest.

However, the R9 M375X is not without reason for existence. It is a mobile-oriented part with no listed TDP, power connectors, or slot width, and its 123 mm² die is smaller and denser (12.2M transistors per mm²) than the GTX 660’s 221 mm² die (11.5M per mm²). The R9 M375X also supports DirectX 12 (11_1) versus the GTX 660’s DirectX 12 (11_0), which could matter for specific API feature checks. But the data shows no benchmark where the R9 M375X wins, and its 1,500 million transistors against the GTX 660’s 2,540 million indicates a less capable chip overall. For a builder with a desk and a power supply, the GTX 660 is the practical pick. For a constrained laptop or small-form-factor system where the R9 M375X’s lower power profile (unlisted, but implied by its mobile design) matters, the AMD card is the safer fit — but it will cost you performance.

Architecture Differences

The architectural split is clear: the GTX 660 uses NVIDIA’s Kepler architecture (chip GK106), while the R9 M375X uses AMD’s GCN 1.0 (chip Tropo). Both are built on TSMC’s 28 nm process, and both are end-of-life, but that is where the similarities end. The GTX 660’s die is 221 mm² and houses 2,540 million transistors, while the R9 M375X’s die is 123 mm² with 1,500 million transistors. This makes the GTX 660 a larger, more complex chip, while the R9 M375X is a smaller, denser design — 12.2M transistors per mm² versus 11.5M.

Memory configurations differ significantly. The GTX 660 uses a 192-bit bus with a 144.2 GB/s bandwidth, whereas the R9 M375X uses a 128-bit bus with 72.00 GB/s. Both have 2 GB of GDDR5, but the GTX 660’s memory clock is 1502 MHz (6 Gbps effective) against the R9 M375X’s 1125 MHz (4.5 Gbps effective). The GTX 660 also has more execution resources: 960 shading units, 80 TMUs, and 24 ROPs, versus 640 shading units, 40 TMUs, and 16 ROPs on the R9 M375X. Clock speeds are closer — the GTX 660 runs at 980 MHz base and 1032 MHz boost, while the R9 M375X runs at 925 MHz base and 1015 MHz boost — but the NVIDIA card’s higher resource count and memory bandwidth create the performance gap.

Feature support shows minor divergences. Both support OpenGL 4.6 and Vulkan, but the GTX 660’s Vulkan version is 1.2.175 versus 1.2.170 on the R9 M375X. DirectX support is the reverse: the AMD card lists DirectX 12 (11_1), while the NVIDIA card lists DirectX 12 (11_0). Neither card has ray tracing or tensor cores, as those fields are null. The GTX 660 is a dual-slot card with a 241 mm length, a 1x 6-pin power connector, a 300 W suggested PSU, and a 140 W TDP, while the R9 M375X has no listed TDP, slot width, power connectors, or dimensions — consistent with its mobile-oriented positioning.

FAQ

Q: Which card scores higher in Geekbench OpenCL?

A: The NVIDIA GeForce GTX 660 scores 11347, which is 37.2% higher than the AMD Radeon R9 M375X’s 8273.

Q: Does the R9 M375X win any benchmark in the head-to-head data?

A: No. The head-to-head set includes only Geekbench OpenCL and Vulkan, and the GTX 660 wins both, giving it a 2-0 record.

Q: How do the two cards compare on average benchmark score?

A: The GTX 660 has an average score of 9022, placing it at the 45th percentile of all GPUs, while the R9 M375X averages 8325 at the 43rd percentile, a gap of roughly 8% in favor of the NVIDIA card.

Q: What are the memory bandwidth figures for each card?

A: The GTX 660 has a 192-bit bus with 144.2 GB/s bandwidth, while the R9 M375X has a 128-bit bus with 72.00 GB/s, exactly half the bandwidth.

Q: Which card has more shading units?

A: The GTX 660 has 960 shading units, compared to 640 on the R9 M375X, a 50% increase for the NVIDIA card.

Q: Are both cards built on the same process node?

A: Yes, both use TSMC’s 28 nm process, but the GTX 660’s die is 221 mm² with 2,540 million transistors, while the R9 M375X’s die is 123 mm² with 1,500 million transistors.

Where Each One Wins

The GTX 660 wins in every measured compute workload. In OpenCL, it beats the R9 M375X by 37.2%, and in Vulkan, it leads by 36.3%. Its higher pixel rate (20.64 GPixel/s vs 16.24 GPixel/s), texture rate (82.56 GTexel/s vs 40.60 GTexel/s), and FP32 output (1.981 TFLOPS vs 1,299.2 GFLOPS) make it the superior choice for any GPU-accelerated task that relies on raw parallelism, such as rendering, scientific compute, or heavy multimedia processing. The GTX 660 also has a wider memory bus and double the bandwidth, which helps in workloads that stream large datasets.

The R9 M375X has no benchmark wins, but its strengths lie in its physical and power profile. With no listed TDP, power connectors, or slot width, it is clearly designed for mobile or low-power systems where the GTX 660’s 140 W TDP and dual-slot cooler would be impractical. Its smaller die (123 mm² vs 221 mm²) and lower transistor count also suggest lower manufacturing cost and potentially lower power draw, though the data does not quantify that. For a desktop builder with a 300 W PSU and a full-length slot, the GTX 660 is the obvious winner. For a laptop or a compact chassis where space and thermals are the primary constraints, the R9 M375X is the only one of the two that fits the bill, but you will accept a 36-37% performance penalty in compute benchmarks.

Specification Differences

The cards differ on nearly every specification where data exists. Process node is the same (28 nm, TSMC), but die size differs: GTX 660 is 221 mm², R9 M375X is 123 mm². Transistor counts are 2,540 million versus 1,500 million, with transistor density at 11.5M/mm² for NVIDIA and 12.2M/mm² for AMD. Shading units are 960 vs 640, TMUs are 80 vs 40, and ROPs are 24 vs 16. Base clocks are 980 MHz vs 925 MHz, boost clocks are 1032 MHz vs 1015 MHz, and memory clocks are 1502 MHz (6 Gbps effective) vs 1125 MHz (4.5 Gbps effective). Memory bus width is 192-bit vs 128-bit, bandwidth is 144.2 GB/s vs 72.00 GB/s, and memory size is 2 GB GDDR5 on both, but the GTX 660’s faster memory makes the difference. Pixel rate is 20.64 GPixel/s vs 16.24 GPixel/s, texture rate is 82.56 GTexel/s vs 40.60 GTexel/s, and FP32 is 1.981 TFLOPS vs 1,299.2 GFLOPS. The GTX 660 has a 140 W TDP, dual-slot width, a 1x 6-pin power connector, a 300 W suggested PSU, a 241 mm length, and display outputs of 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2, while the R9 M375X has none of these fields listed. API support shows the GTX 660 with DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175, versus DirectX 12 (11_1), OpenGL 4.6, and Vulkan 1.2.170 on the AMD card. Release dates are September 2012 for the GTX 660 and May 2015 for the R9 M375X, with the GTX 660 having a launch MSRP of 229 USD.