GPU Comparison

AMD Radeon Pro 455 and AMD Radeon 660M are two very different mobile graphics solutions separated by nearly six years of architectural evolution. The Pro 455 is a discrete MXM module from the GCN 4.0 era, while the 660M is a modern RDNA 2.0 integrated graphics processor. The benchmark data shows a clear overall winner: the Radeon 660M takes both shared test categories, boasting an average benchmark score of 13812 against the Pro 455’s 12831, placing it in the 55th percentile of all GPUs versus the Pro 455’s 52nd percentile. This page breaks down where each part excels, what separates them internally, and how their head-to-head results translate into real-world usage.

Where Each One Wins

The Radeon 660M wins every benchmark category shared between the two, making it the outright performance leader in this comparison. However, the Pro 455 retains a distinct advantage in form factor and compatibility: it is a discrete MXM module with 2 GB of dedicated GDDR5 memory, whereas the 660M is an IGP relying on system shared memory with bandwidth described as "System Dependent." For users with older Mac Pro systems that accept MXM modules, the Pro 455 offers a plug-and-play upgrade path that the 660M, being soldered into a mobile APU, cannot match.

In raw compute workloads, the 660M’s lead is substantial. Its Geekbench OpenCL score of 12876 outperforms the Pro 455’s 10336 by 19.7%, and its Vulkan score of 14748 beats the Pro 455’s 12240 by 17%. These deltas indicate that the 660M is not just marginally faster but significantly ahead in both general-purpose compute and modern graphics API workloads. The 660M also brings hardware ray tracing to the table with 6 dedicated RT cores, a feature entirely absent from the Pro 455, which relies on GCN 4.0’s older shader-based approach.

The Pro 455’s wins are narrower and more situational. Its 128-bit memory bus and 81.28 GB/s dedicated bandwidth give it predictable memory performance, unlike the 660M’s shared memory setup that varies with system RAM. For legacy DirectX 12 (12_0) applications, the Pro 455 supports that feature level, while the 660M goes further with DirectX 12 Ultimate (12_2). In terms of raw pixel throughput, the 660M actually leads with 30.40 GPixel/s versus the Pro 455’s 13.68 GPixel/s, but the Pro 455’s dedicated memory may still offer lower latency in certain memory-bound scenarios. Ultimately, the data favors the 660M in every measurable benchmark, leaving the Pro 455 as the choice only for specific hardware compatibility needs.

Architecture Differences

The two GPUs come from completely different design philosophies and manufacturing eras. The Radeon Pro 455 uses the Baffin chip built on GlobalFoundries’ 14 nm process, packing 3,000 million transistors into a 123 mm² die with a transistor density of 24.4M per mm². In contrast, the Radeon 660M uses the Rembrandt chip fabricated by TSMC on a 6 nm node, containing 13,100 million transistors across a 208 mm² die, achieving a density of 63.0M per mm², more than 2.5 times the transistor density of the Pro 455.

The architectural leap from GCN 4.0 to RDNA 2.0 is stark. The Pro 455 features 768 shading units, 48 TMUs, and 16 ROPs, while the 660M has only 384 shading units and 24 TMUs but the same 16 ROPs. Despite having half the shader count, the 660M produces higher FP32 throughput at 1,459.2 GFLOPS versus the Pro 455’s 1,313.3 GFLOPS, thanks to its higher clock speeds, the 660M runs between 1500 MHz base and 1900 MHz boost, while the Pro 455’s clock data is incomplete. The 660M also supports FP16 at a 2:1 ratio with 2.918 TFLOPS, whereas the Pro 455 offers FP16 at a 1:1 ratio, capping it at the same 1,313.3 GFLOPS as FP32.

Memory architecture diverges completely. The Pro 455 has 2 GB of GDDR5 on a 128-bit bus with 81.28 GB/s bandwidth and a memory clock of 1270 MHz (5.1 Gbps effective). The 660M uses system shared memory with no dedicated VRAM, making its bandwidth "System Dependent." This means the Pro 455 offers consistent memory performance, while the 660M’s memory speed hinges on the host system’s RAM configuration. The 660M compensates with PCIe 4.0 x8 connectivity versus the Pro 455’s PCIe 3.0 x8, potentially reducing data transfer bottlenecks.

Feature support also differs significantly. The 660M supports DirectX 12 Ultimate (12_2), Vulkan 1.4, and includes 6 ray tracing cores, while the Pro 455 is limited to DirectX 12 (12_0), Vulkan 1.3, and has no RT cores. Both support OpenGL 4.6. The 660M’s power envelope is slightly higher at 40 W TDP versus the Pro 455’s 35 W TDP, but the Pro 455 requires an MXM module slot while the 660M is an IGP integrated directly into the processor. The Pro 455 was released in late 2016 and is end-of-life, while the 660M launched in early 2022 and is also end-of-life, but represents a much more modern design generation.

Head-to-Head Benchmarks

The head-to-head data is unambiguous: the Radeon 660M wins both shared benchmark tests by wide margins. In Geekbench OpenCL, the 660M scores 12876 against the Pro 455’s 10336, a delta of 19.7% in favor of the newer chip. This is the larger of the two gaps and reflects the 660M’s superior FP32 throughput and architectural efficiency. For compute-heavy tasks like OpenCL-accelerated rendering or physics simulations, the 660M delivers nearly 20% more performance, which is a substantial generational improvement.

In Geekbench Vulkan, the 660M again takes the lead with 14748 versus 12240, a 17% advantage. Vulkan is a modern low-level API that benefits from RDNA 2.0’s improved command processor and shader efficiency. The Pro 455’s GCN 4.0 architecture, while competent for its time, cannot match the newer design’s ability to extract performance from Vulkan’s parallel workload distribution. The 660M’s Vulkan score is actually higher than its OpenCL score, suggesting particularly strong driver optimization for that API, while the Pro 455 shows the opposite trend, its OpenCL score (10336) is lower than its Vulkan score (12240), but both trail the 660M.

Interestingly, the Pro 455 has a Geekbench Metal score of 15916 that the 660M does not have a comparable result for in this data set. This means on Apple platforms where Metal is the primary API, the Pro 455 may still hold its own, given that its Metal score exceeds its Vulkan and OpenCL results significantly. However, without a Metal benchmark for the 660M, a direct comparison in that API is impossible from the available data. The average benchmark scores reinforce the 660M’s overall edge: 13812 average versus 12831 for the Pro 455, a difference of about 7.6% across all recorded tests.

FAQ

Q: Which GPU has higher raw compute performance?

A: The AMD Radeon 660M leads in FP32 with 1,459.2 GFLOPS versus the Pro 455’s 1,313.3 GFLOPS. In FP16, the 660M offers 2.918 TFLOPS (2:1 ratio), while the Pro 455 is capped at 1,313.3 GFLOPS (1:1 ratio).

Q: Does either GPU support hardware ray tracing?

A: Only the AMD Radeon 660M includes 6 dedicated ray tracing cores. The Radeon Pro 455 has no RT cores and relies on traditional shader-based rendering.

Q: How do their memory configurations differ?

A: The Pro 455 has 2 GB of dedicated GDDR5 memory on a 128-bit bus with 81.28 GB/s bandwidth. The 660M uses system shared memory with bandwidth described as "System Dependent," meaning it varies based on the host system’s RAM.

Q: Which GPU is better for modern DirectX games?

A: The Radeon 660M supports DirectX 12 Ultimate (12_2), while the Pro 455 only supports DirectX 12 (12_0). This gives the 660M access to more advanced rendering features in compatible titles.

Q: What are the power requirements for each?

A: The Pro 455 has a 35 W TDP and requires an MXM module slot with no power connectors. The 660M has a 40 W TDP and is an IGP, drawing power through the host processor.

Q: Which GPU has better Vulkan performance?

A: The Radeon 660M scores 14748 in Geekbench Vulkan, which is 17% higher than the Pro 455’s 12240. The 660M also supports Vulkan 1.4 versus the Pro 455’s Vulkan 1.3.

The Verdict

The data clearly favors the AMD Radeon 660M as the superior graphics processor in nearly every measurable way. It wins both head-to-head benchmarks, OpenCL by 19.7% and Vulkan by 17%, and carries a higher average benchmark score of 13812 versus 12831. Its architecture is more modern, with RDNA 2.0 features like ray tracing, DirectX 12 Ultimate support, and significantly higher transistor density on a 6 nm process. The 660M also offers better peak throughput in both FP32 and FP16, despite having half the shading units of the Pro 455, thanks to higher clock speeds and architectural efficiency.

The Radeon Pro 455’s only clear advantages are its dedicated 2 GB GDDR5 memory and MXM module form factor. For users with legacy systems that accept MXM modules, the Pro 455 provides a discrete GPU upgrade path that the 660M cannot offer, since the 660M is an integrated processor component. Its Metal benchmark score of 15916 also suggests strong performance on Apple platforms, though no direct comparison with the 660M is available in that API.

For most workloads, including modern gaming, compute tasks, and Vulkan-based applications, the Radeon 660M is the unequivocal choice. Its performance lead is substantial and consistent across all shared benchmarks. The Pro 455 remains relevant only for specific hardware compatibility scenarios where a dedicated MXM GPU is required, or where Apple’s Metal API is the primary workload and the Pro 455’s 15916 Metal score proves advantageous. Otherwise, the 660M’s newer architecture, higher compute throughput, and superior API support make it the better-performing GPU in this comparison.