AMD Radeon Pro 450 vs NVIDIA GeForce GTX 660 Comparison

The Verdict

The benchmark data reveals a clear split: the AMD Radeon Pro 450 dominates in Metal workloads, while the NVIDIA GeForce GTX 660 wins in OpenCL and Vulkan. The Radeon Pro 450 scores 12,893 in Geekbench Metal, a 199.5% advantage over the GTX 660’s 4,305. That is a decisive margin, reflecting the Pro 450’s newer GCN 4.0 architecture and its optimization for Apple’s Metal API. However, the GTX 660 counters with 11,347 in OpenCL (20.7% ahead) and 11,457 in Vulkan (7.9% ahead). The average benchmark score favors the Pro 450 at 10,804 versus 9,022, a 19.8% overall lead. The Pro 450 also sits in the 49th percentile of all GPUs, while the GTX 660 sits in the 45th.

For users working within a Metal-centric ecosystem, particularly macOS environments where the Radeon Pro 450 is typically deployed, the data clearly favors AMD. The Pro 450’s nearest rivals include the NVIDIA Quadro K2200 (0.4% behind), the GeForce MX350 (0.7% ahead), and the GeForce GTX 560 Ti (1.1% behind), placing it in a competitive mid-range bracket. The GTX 660, by contrast, ranks near the NVIDIA TITAN V CEO Edition (0.2% behind) and the GeForce GTX 560 (0.4% ahead), showing it competes with older high-end cards despite being a generation older. The GTX 660’s launch MSRP was 229 USD, stated once per database convention.

The verdict from the data: choose the AMD Radeon Pro 450 for Metal-based compute and modern API compatibility, and choose the NVIDIA GeForce GTX 660 for raw OpenCL throughput and legacy Vulkan performance. The GTX 660 offers higher shading unit count (960 versus 640) and more texture mapping units (80 versus 40), but the Pro 450’s newer process node and architectural efficiency narrow the gaps in real-world workloads. The GTX 660 wins two of three head-to-head benchmarks, but the Pro 450’s single dominant win in Metal outweighs those in overall average score.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon Pro 450 has an average benchmark score of 10,804, while the NVIDIA GeForce GTX 660 averages 9,022. This places the Pro 450 in the 49th percentile of all GPUs, compared to the GTX 660’s 45th percentile.

Q: How do the two cards compare in Geekbench Metal performance?

A: The AMD Radeon Pro 450 scores 12,893 in Geekbench Metal, which is 199.5% higher than the GTX 660’s 4,305. This is the largest performance gap between the two across all tested workloads.

Q: Does the NVIDIA card win any benchmarks?

A: Yes, the NVIDIA GeForce GTX 660 wins in Geekbench OpenCL with a score of 11,347 versus 9,002 (a 20.7% lead) and in Geekbench Vulkan with 11,457 versus 10,518 (a 7.9% lead).

Q: What is the transistor density difference between the two?

A: The AMD Radeon Pro 450, built on a 14 nm process, has a transistor density of 24.4 million per square millimeter. The NVIDIA GeForce GTX 660, built on a 28 nm process, has a density of 11.5 million per square millimeter.

Q: Which card offers better memory bandwidth?

A: The NVIDIA GeForce GTX 660 provides 144.2 GB/s of memory bandwidth over a 192-bit bus, while the AMD Radeon Pro 450 provides 81.28 GB/s over a 128-bit bus.

Q: What is the TDP difference between the two cards?

A: The AMD Radeon Pro 450 has a TDP of 35 W, while the NVIDIA GeForce GTX 660 has a TDP of 140 W. This makes the Pro 450 substantially more power-efficient on paper.

Architecture Differences

The AMD Radeon Pro 450 uses the Baffin chip based on GCN 4.0 architecture, fabricated by GlobalFoundries on a 14 nm process. It packs 3,000 million transistors into a 123 mm² die, yielding a transistor density of 24.4 million per square millimeter. This modern node allows for high efficiency, reflected in its 35 W TDP. The Radeon Pro 450 is part of the Radeon Pro Mac (400 Series) generation, indicating its target market of professional Mac systems.

The NVIDIA GeForce GTX 660 uses the GK106 chip based on Kepler architecture, manufactured by TSMC on a 28 nm process. It contains 2,540 million transistors on a 221 mm² die, giving a transistor density of 11.5 million per square millimeter. The older, larger process node contributes to a significantly higher 140 W TDP. The GTX 660 belongs to the GeForce 600 generation, a consumer-focused lineup from 2012.

Architecturally, GCN 4.0 supports DirectX 12 (12_0), OpenGL 4.6, and Vulkan 1.3. Kepler supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. The Pro 450’s newer architecture also provides FP16 support at a 1:1 ratio with FP32 (1,024.0 GFLOPS each), while the GTX 660 has no listed FP16 capability. The GTX 660’s FP32 throughput is 1.981 TFLOPS, nearly double the Pro 450’s 1,024.0 GFLOPS, but the Pro 450’s architectural efficiency and API support make it more future-proof in the recorded data.

The Radeon Pro 450 uses a PCIe 3.0 x8 interface and an MXM Module slot width, designed for portability and integration into mobile or compact systems. The GTX 660 uses a PCIe 3.0 x16 interface and a dual-slot form factor, with a length of 241 mm (9.5 inches). Display outputs also differ: the Pro 450 is "Portable Device Dependent," while the GTX 660 offers 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2.

Specification Differences

Both cards share 2 GB of GDDR5 memory, but the bus widths differ: the Radeon Pro 450 uses a 128-bit bus, while the GTX 660 uses a 192-bit bus. This results in a bandwidth advantage for NVIDIA: 144.2 GB/s versus 81.28 GB/s. Memory clocks also differ, with the Pro 450 running at 1270 MHz (5.1 Gbps effective) and the GTX 660 at 1502 MHz (6 Gbps effective).

The shading unit counts differ substantially: the GTX 660 has 960 shading units, while the Pro 450 has 640. Texture mapping units also favor NVIDIA: 80 versus 40. Raster operation units favor NVIDIA as well: 24 versus 16. Pixel rate is 20.64 GPixel/s for the GTX 660 versus 12.80 GPixel/s for the Pro 450, and texture rate is 82.56 GTexel/s versus 32.00 GTexel/s.

The GTX 660 has explicit clock speeds, with a base clock of 980 MHz and a boost clock of 1032 MHz. The Pro 450 has no listed base or boost clock in the database. The GTX 660 also requires a 1x 6-pin power connector and a 300 W suggested PSU, while the Pro 450 has no listed power connector or PSU requirement. Release dates differ by over four years: the GTX 660 launched on 2012-09-05, while the Pro 450 launched on 2016-10-29. Both are end-of-life, but the GTX 660 has recorded predecessor (GeForce 500) and successor (GeForce 700) lines, while the Pro 450 has none.

Head-to-Head Benchmarks

The most striking result is in Geekbench Metal. The AMD Radeon Pro 450 scores 12,893 versus the GTX 660’s 4,305, a 199.5% delta. This means the Pro 450 delivers roughly triple the Metal performance. For any workload leveraging Apple’s Metal API, the Pro 450 is the clear choice. This single result anchors the Pro 450’s higher average benchmark score.

In Geekbench OpenCL, the NVIDIA GeForce GTX 660 takes the lead with 11,347 against the Pro 450’s 9,002, a 20.7% difference. The GTX 660’s higher FP32 throughput (1.981 TFLOPS versus 1,024.0 GFLOPS) and wider memory bus contribute to this result. OpenCL workloads that scale with raw compute throughput will favor the GTX 660.

In Geekbench Vulkan, the GTX 660 again wins with 11,457 versus 10,518, a 7.9% margin. This is a closer contest than Metal or OpenCL, indicating that the Pro 450’s Vulkan driver and architecture are competitive despite lower raw specs. The GTX 660’s Vulkan support is version 1.2.175, while the Pro 450 supports Vulkan 1.3, but the older card still edges out the newer one in this test.

The win count is 1 for the AMD Radeon Pro 450 and 2 for the NVIDIA GeForce GTX 660. However, the magnitude of the Pro 450’s Metal victory (199.5%) far exceeds the GTX 660’s wins in OpenCL (20.7%) and Vulkan (7.9%). This asymmetry explains why the Pro 450’s average score (10,804) remains higher despite losing two of three tests.

Where Each One Wins

The AMD Radeon Pro 450 is the winner in Metal applications. Its 12,893 Geekbench Metal score, 199.5% above the GTX 660, makes it the superior choice for macOS environments, professional creative suites, and any software that relies on Metal for GPU acceleration. The Pro 450’s lower TDP (35 W versus 140 W) also makes it suitable for compact, power-constrained systems like all-in-one Macs or portable workstations. Its newer GCN 4.0 architecture with Vulkan 1.3 support suggests better long-term driver compatibility in the recorded data. The Pro 450’s nearest rivals, such as the Quadro K2200 and GTX 560 Ti, are all within 1.1% of its average score, indicating it sits in a stable mid-range performance tier.

The NVIDIA GeForce GTX 660 wins in OpenCL and Vulkan. Its OpenCL score of 11,347 (20.7% ahead) and Vulkan score of 11,457 (7.9% ahead) make it the better choice for cross-platform compute workloads that favor these APIs. The GTX 660’s higher memory bandwidth (144.2 GB/s), more shading units (960), and higher pixel and texture rates give it an edge in bandwidth-sensitive tasks. Its nearest rivals include the TITAN V CEO Edition and GTX 560, with the GTX 660 slightly ahead of both, showing it remains competitive against much newer and older hardware alike. The GTX 660 also offers a boost clock of 1032 MHz, which may help in sustained workloads.

The data does not support a single universal winner. Instead, the choice depends on the target API and use case. For Metal-centric workflows, the Radeon Pro 450 is the only rational selection. For OpenCL or Vulkan compute, the GTX 660 is the stronger performer. The Pro 450’s average score advantage (19.8%) and higher percentile ranking (49 versus 45) suggest it is the better all-rounder in the database’s aggregate metrics, but only by a modest margin. The GTX 660’s two benchmark wins demonstrate that raw architectural age does not preclude competitive performance in specific workloads.