AMD Radeon 660M vs AMD Radeon R9 370X Comparison

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon R9 370X holds the higher average, with 15,862 points compared to the AMD Radeon 660M’s 13,812 points. This places the R9 370X in the 58th percentile of all GPUs, while the 660M sits in the 55th percentile.

Q: How do these two compare in OpenCL performance?

A: The R9 370X is dramatically ahead in OpenCL, scoring 25,893 versus the 660M’s 12,876. That is a 101.1% advantage, meaning the R9 370X more than doubles the 660M’s OpenCL output.

Q: Which GPU wins in Vulkan performance?

A: The AMD Radeon 660M takes Vulkan by a significant margin. Its score of 14,748 beats the R9 370X’s 9,018, a difference of 38.9% in favor of the 660M.

Q: What are the architecture generations of each GPU?

A: The R9 370X uses GCN 1.0 architecture on a 28 nm process, part of the Pirate Islands (R9 300) generation. The 660M uses RDNA 2.0 architecture on a 6 nm process, part of the Navi II IGP (Rembrandt Mobile) generation.

Q: What is the memory setup for each card?

A: The R9 370X has 2 GB of dedicated GDDR5 memory on a 256-bit bus, delivering 179.2 GB/s bandwidth. The 660M uses system shared memory, with its bandwidth described as system dependent.

Q: What is the power draw difference between the two?

A: The R9 370X has a TDP of 180 W and requires a 450 W suggested power supply with dual 6-pin connectors. The 660M has a TDP of just 40 W and uses no power connectors, as it is an integrated graphics processor.

Architecture Differences

The AMD Radeon R9 370X and the AMD Radeon 660M represent two fundamentally different design philosophies. The R9 370X is a discrete desktop GPU built on the GCN 1.0 architecture, manufactured on a 28 nm process at TSMC. Its chip, codenamed Trinidad, packs 2,800 million transistors into a 212 mm² die, yielding a transistor density of 13.2 million per square millimeter. This is an older, power-hungry design from the Pirate Islands generation, released in August 2015.

The 660M, in contrast, is an integrated graphics processor (IGP) built on RDNA 2.0 architecture. It uses a 6 nm process, also from TSMC, with a chip codenamed Rembrandt. Despite a similar die size of 208 mm², the 660M packs far more transistors: 13,100 million, producing a density of 63.0 million per square millimeter. This nearly fivefold increase in density reflects the newer manufacturing node and the integration of additional components beyond the GPU cores themselves, as Rembrandt is a mobile APU die. The 660M belongs to the Navi II IGP generation, released in January 2022.

Core configurations differ substantially. The R9 370X fields 1,280 shading units, 80 texture mapping units, and 32 raster operation units. The 660M has 384 shading units, 24 TMUs, and 16 ROPs. Notably, the 660M includes 6 ray tracing cores, a feature entirely absent from the R9 370X, which has none. This architectural leap means the 660M supports DirectX 12 Ultimate (12_2), while the R9 370X only reaches DirectX 12 (11_1). Both support OpenGL 4.6, but the 660M offers Vulkan 1.4 versus the R9 370X’s Vulkan 1.2.170.

Clock speeds also tell a story of generational progress. The R9 370X runs at a base of 980 MHz with a boost of 1030 MHz, while the 660M operates at a base of 1500 MHz and boosts to 1900 MHz. The 660M’s higher clocks partially compensate for its fewer shaders, but raw compute output still favors the older card. The R9 370X achieves 2.637 TFLOPS FP32, while the 660M delivers 1,459.2 GFLOPS (approximately 1.46 TFLOPS). The 660M does have a dedicated FP16 rate of 2.918 TFLOPS (2:1), a feature the R9 370X lacks entirely.

Memory architecture is another stark divider. The R9 370X uses dedicated 2 GB GDDR5 on a 256-bit bus, providing 179.2 GB/s of bandwidth. The 660M relies on system shared memory, with bandwidth that is system dependent. This means the R9 370X has a fixed, predictable memory performance profile, while the 660M’s performance scales with the host system’s RAM configuration.

The Verdict

The data points to a clear split depending on workload. For compute-heavy tasks that leverage OpenCL, the AMD Radeon R9 370X is the unequivocal choice. Its OpenCL score of 25,893 more than doubles the 660M’s 12,876, a 101.1% advantage. This is a massive lead in raw compute throughput, driven by the R9 370X’s 1,280 shading units and dedicated 179.2 GB/s memory bandwidth. Users running OpenCL-based applications, such as certain scientific simulations or older compute workloads, should pick the R9 370X without hesitation.

For Vulkan-based workloads, particularly modern gaming or applications that utilize ray tracing, the AMD Radeon 660M is the superior option. Its Vulkan score of 14,748 beats the R9 370X’s 9,018 by 38.9%. The 660M’s support for DirectX 12 Ultimate and its 6 ray tracing cores give it a decisive feature advantage for contemporary graphics APIs. The 660M also operates at a fraction of the power draw, 40 W versus 180 W, making it suitable for portable devices where the R9 370X’s dual-slot, dual 6-pin connector design would be impractical.

The average benchmark scores favor the R9 370X, 15,862 versus 13,812, a 14.8% gap. This places the R9 370X at the 58th percentile and the 660M at the 55th percentile. However, this average masks the divergent strengths. The R9 370X’s nearest rival is the AMD Radeon RX 9060, which scores 16,014, a 1% delta, while the 660M’s nearest rival is the NVIDIA RTX A2000 Mobile, scoring 13,821, a 0.1% delta. These proximity rankings show that each GPU competes in different performance strata.

The verdict is workload-dependent. The R9 370X is for users prioritizing raw OpenCL compute and dedicated memory bandwidth. The 660M is for users prioritizing modern API support, ray tracing, and efficiency. Neither is a general-purpose winner; both are specialists in their respective domains.

Specification Differences

The two GPUs differ across nearly every specification category. The process node is a major differentiator: the R9 370X uses 28 nm, while the 660M uses 6 nm. Transistor counts are 2,800 million for the R9 370X versus 13,100 million for the 660M, despite similar die sizes of 212 mm² and 208 mm² respectively. Transistor density is 13.2M per mm² for the R9 370X and 63.0M per mm² for the 660M.

Clock speeds favor the 660M, with a base of 1500 MHz and boost of 1900 MHz, compared to the R9 370X’s 980 MHz base and 1030 MHz boost. Memory configurations are entirely different: the R9 370X has 2 GB GDDR5 on a 256-bit bus with 179.2 GB/s bandwidth, while the 660M uses system shared memory with system-dependent bandwidth.

Shader resources heavily favor the R9 370X: 1,280 shading units versus 384, 80 TMUs versus 24, and 32 ROPs versus 16. However, the 660M uniquely offers 6 ray tracing cores. Pixel rates are close, 32.96 GPixel/s for the R9 370X versus 30.40 GPixel/s for the 660M, but texture rates diverge sharply, 82.40 GTexel/s versus 45.60 GTexel/s. FP32 performance is 2.637 TFLOPS for the R9 370X versus 1,459.2 GFLOPS for the 660M, while FP16 is only available on the 660M at 2.918 TFLOPS.

Power requirements are starkly different: the R9 370X has a 180 W TDP, dual-slot width, and needs dual 6-pin connectors plus a 450 W suggested PSU, while the 660M has a 40 W TDP, is an IGP with no power connectors, and has no suggested PSU. The R9 370X uses PCIe 3.0 x16, while the 660M uses PCIe 4.0 x8. Display outputs are 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2 on the R9 370X, versus portable device dependent outputs on the 660M. API support differs: DirectX 12 (11_1) versus DirectX 12 Ultimate (12_2), and Vulkan 1.2.170 versus 1.4. The R9 370X has physical dimensions of 221 mm length and 111 mm height, while the 660M has none. The R9 370X had a launch MSRP of 199 USD.

Head-to-Head Benchmarks

The two recorded benchmarks show a perfect split, with each GPU winning one test. The first test, Geekbench OpenCL, is a decisive victory for the AMD Radeon R9 370X. Its score of 25,893 utterly eclipses the 660M’s 12,876. The delta is 101.1%, meaning the R9 370X delivers more than double the OpenCL performance. This is the single largest margin in any comparison between these two cards. The R9 370X’s 1,280 shading units and dedicated GDDR5 memory with 179.2 GB/s bandwidth clearly dominate in this compute-oriented API. For reference, the R9 370X’s OpenCL score alone pushes its average benchmark to 15,862, while the 660M’s average is dragged down by this result.

The second test, Geekbench Vulkan, flips the script entirely. The AMD Radeon 660M scores 14,748, while the R9 370X manages only 9,018. The delta here is 38.9% in favor of the 660M. This is a substantial margin, though not as large as the R9 370X’s OpenCL lead. The 660M’s RDNA 2.0 architecture, with its 6 ray tracing cores and modern Vulkan 1.4 support, clearly excels in this API. The R9 370X’s older GCN 1.0 architecture and Vulkan 1.2.170 support hold it back significantly.

These two results illustrate the generational shift between the cards. The R9 370X was designed for an era where OpenCL was a primary compute interface, and its massive shader count and memory bandwidth pay dividends there. The 660M was designed for a modern era where Vulkan and DirectX 12 Ultimate dominate, and its architectural efficiencies shine through. The win count is even at one apiece, but the magnitudes differ: the R9 370X’s OpenCL win is more than double, while the 660M’s Vulkan win is less than half again. This asymmetry means the R9 370X has the larger single victory, but the 660M’s win is in the API that matters more for contemporary gaming.

Where Each One Wins

The AMD Radeon R9 370X wins in scenarios that demand raw compute throughput in OpenCL. Its 101.1% lead over the 660M in this API makes it the clear choice for any OpenCL-based application, whether that is legacy compute workloads, certain scientific software, or older games that used OpenCL for acceleration. The R9 370X also wins on memory bandwidth, with 179.2 GB/s of dedicated bandwidth versus the 660M’s system-dependent shared memory. This makes it preferable for tasks that are bandwidth-sensitive, where a fixed, high-bandwidth pool of GDDR5 is advantageous. Its higher pixel rate of 32.96 GPixel/s and texture rate of 82.40 GTexel/s further cement its lead in traditional rasterization throughput. The R9 370X also holds the overall average score advantage, 15,862 versus 13,812, and the higher percentile ranking at 58th versus 55th.

The AMD Radeon 660M wins in modern graphics API scenarios. Its Vulkan score of 14,748 is 38.9% ahead of the R9 370X, making it the preferred option for Vulkan-based games and applications. The 660M’s 6 ray tracing cores and DirectX 12 Ultimate support give it a feature set that the R9 370X simply cannot match, making it the choice for ray-traced workloads and games that leverage DirectX 12 Ultimate features. Its FP16 capability of 2.918 TFLOPS (2:1) is a unique advantage for workloads that can utilize half-precision math. The 660M also wins decisively on power efficiency, with a 40 W TDP versus 180 W, and its IGP form factor requires no power connectors or extra slot space. This makes it the only viable option for portable devices, thin-and-light laptops, or any system where power draw and physical space are constraints. Its PCIe 4.0 x8 interface is also more modern than the R9 370X’s PCIe 3.0 x16, though the practical impact depends on the host system. Ultimately, the 660M is the pick for modern gaming and efficiency, while the R9 370X is the pick for legacy compute and bandwidth-heavy workloads.