AMD Radeon 660M vs AMD Radeon RX 9070 XT Comparison
AMD Radeon 660M
Radeon RX 9070 XT
PERFORMANCE BENCHMARKS
Analysis: AMD Radeon 660M vs AMD Radeon RX 9070 XT
The AMD Radeon 660M and AMD Radeon RX 9070 XT occupy opposite ends of the graphics spectrum, yet both carry the Radeon name. The 660M is a 40 W integrated graphics processor (IGP) built on RDNA 2.0 for mobile devices, while the RX 9070 XT is a 304 W discrete add-in board powered by RDNA 4.0. Benchmark data reveals a decisive performance gulf, but the comparison is more nuanced than raw scores alone, as the 660M's average benchmark score of 13,812 actually edges out the RX 9070 XT's 13,543 in the database's aggregate metric—an artifact of the RX 9070 XT's inclusion of older DirectX 9/10/11 benchmark results that drag down its average.
FAQ
Q: Which GPU is faster in Geekbench OpenCL?
A: The AMD Radeon RX 9070 XT scores 17,428 versus the Radeon 660M's 12,876, a 26.1% advantage for the discrete card.
Q: How large is the gap in Geekbench Vulkan performance?
A: The RX 9070 XT dominates with a score of 65,879, which is 77.6% higher than the 660M's 14,748. This is the largest single-benchmark margin in the comparison.
Q: Do both GPUs support the same modern API features?
A: Yes, both list DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 support, indicating parity in API compatibility despite their generational difference.
Q: What is the transistor density difference between the two chips?
A: The RX 9070 XT's Navi 48 chip packs 151.0 million transistors per mm², while the 660M's Rembrandt chip has a density of 63.0 million per mm²—a more than 2.4x difference favoring the newer 4 nm process.
Q: Which GPU has a higher percentile ranking among all GPUs?
A: Both GPUs sit at the 55th percentile, meaning they are statistically equivalent in this aggregate ranking despite the RX 9070 XT's far higher peak scores.
Q: What is the average benchmark score for each card?
A: The 660M averages 13,812, while the RX 9070 XT averages 13,543. The 660M's average is actually 2% higher, though this reflects the RX 9070 XT's inclusion of legacy DirectX 9/10/11 tests where it scores very low.
The Verdict
The data presents a clear split decision. For users running modern graphics workloads—specifically Vulkan or OpenCL compute—the AMD Radeon RX 9070 XT is the definitive choice. Its Geekbench Vulkan score of 65,879 is 4.5 times higher than the 660M's 14,748, and its OpenCL result of 17,428 beats the 660M's 12,876 by a 26.1% margin. The RX 9070 XT also offers 16 GB of dedicated GDDR6 memory with 644.6 GB/s bandwidth, versus the 660M's system-shared memory, making it the only option for high-resolution textures or memory-intensive compute tasks.
However, the 660M holds its own in aggregate statistics. Its average benchmark score of 13,812 is 2% higher than the RX 9070 XT's 13,543, and it ties the discrete card at the 55th percentile. This is partly because the RX 9070 XT's PassMark DirectX 9 score of 353 and DirectX 10 score of 154 drag down its average, while the 660M has no such legacy results dragging it. The 660M's nearest rivals include the NVIDIA RTX A2000 Mobile (deltaPct -0.1) and AMD Radeon RX 570X (deltaPct -0.4), showing it competes with low-end discrete mobile GPUs. The RX 9070 XT's nearest rivals are far older hardware like the AMD Radeon HD 7770M (deltaPct 0.1) and NVIDIA GeForce GTX 570 (deltaPct 0.2), which reflects its oddly suppressed average score.
The verdict: choose the RX 9070 XT for any modern gaming, ray tracing, or compute workload where raw performance matters. Choose the 660M only if power consumption (40 W TDP vs 304 W), integrated form factor, or the absence of a discrete card requirement is the deciding factor. The RX 9070 XT's launch MSRP is 599 USD.
Head-to-Head Benchmarks
The head-to-head data shows the RX 9070 XT winning both available benchmark comparisons, with zero wins for the 660M. The Geekbench OpenCL test is the closer contest: the RX 9070 XT posts 17,428 against the 660M's 12,876, a 26.1% improvement. This margin is significant but not overwhelming, suggesting that in pure compute throughput the RX 9070 XT's 4,096 shading units and 48.66 TFLOPS FP32 performance translate into a meaningful but not transformative lead over the 660M's 384 shading units and 1,459.2 GFLOPS.
The Geekbench Vulkan test tells a far more dramatic story. The RX 9070 XT scores 65,879, while the 660M manages only 14,748—a 77.6% deficit for the integrated part. This 4.47x ratio likely reflects the RX 9070 XT's dedicated 16 GB GDDR6 memory with 256-bit bus and 644.6 GB/s bandwidth, versus the 660M's system-shared memory with system-dependent bandwidth. Vulkan workloads are often memory-bandwidth sensitive, and the RX 9070 XT's 644.6 GB/s versus the 660M's shared memory explains the massive gap.
Notably, the RX 9070 XT's average benchmark score of 13,543 is actually lower than the 660M's 13,812. This counterintuitive result stems from the RX 9070 XT's PassMark legacy tests: DirectX 9 scores 353, DirectX 10 scores 154, and DirectX 12 scores 78. These low legacy scores pull the average down despite the RX 9070 XT's dominant performance in modern tests. The 660M, with only Geekbench OpenCL and Vulkan results, avoids this penalty entirely.
Specification Differences
The two GPUs differ across nearly every specification field. The RX 9070 XT uses a 4 nm process node versus the 660M's 6 nm node, with the RX 9070 XT's Navi 48 chip containing 53,900 million transistors on a 357 mm² die, compared to the 660M's Rembrandt chip with 13,100 million transistors on 208 mm². Transistor density jumps from 63.0 million per mm² to 151.0 million per mm²—a 2.4x increase.
Clock speeds diverge significantly: the 660M runs at 1500 MHz base and 1900 MHz boost, while the RX 9070 XT runs at 1660 MHz base, 2970 MHz boost, and a 2400 MHz game clock. Memory configurations are entirely different: the 660M uses system-shared memory with system-dependent bandwidth, while the RX 9070 XT has 16 GB of GDDR6 on a 256-bit bus delivering 644.6 GB/s.
Core counts scale dramatically: the RX 9070 XT has 4,096 shading units, 256 TMUs, and 128 ROPs, versus the 660M's 384 shading units, 24 TMUs, and 16 ROPs. Ray tracing cores increase from 6 to 64. Pixel rate jumps from 30.40 GPixel/s to 380.2 GPixel/s, and texture rate from 45.60 GTexel/s to 760.3 GTexel/s. FP32 performance rises from 1,459.2 GFLOPS to 48.66 TFLOPS, while FP16 goes from 2.918 TFLOPS (2:1 ratio) to 48.66 TFLOPS (1:1 ratio).
Power and physical specifications also differ: the 660M has a 40 W TDP, IGP slot width, no power connectors, and PCIe 4.0 x8 interface. The RX 9070 XT has a 304 W TDP, dual-slot width, 2x 8-pin power connectors, a 700 W suggested PSU, and PCIe 5.0 x16 interface. Display outputs differ: the 660M is portable-device dependent, while the RX 9070 XT offers 1x HDMI 2.1b and 3x DisplayPort 2.1a. Production status also differs—the 660M is end-of-life, while the RX 9070 XT is active.
Architecture Differences
The architecture gap is generational: the 660M uses RDNA 2.0 (Navi II IGP, Rembrandt Mobile) while the RX 9070 XT uses RDNA 4.0 (Navi IV, RX 9000 series). The 660M was released on January 3, 2022, with the RX 9070 XT following on March 5, 2025, representing three architecture generations of advancement.
Both use TSMC as foundry, but the process shrinks from 6 nm to 4 nm. The 660M's predecessor is Vega II IGP and its successor is Navi III IGP; the RX 9070 XT's predecessor is Navi III with no successor listed. The 660M's 13,100 million transistors on 208 mm² gives a density of 63.0M per mm², while the RX 9070 XT's 53,900 million transistors on 357 mm² achieves 151.0M per mm²—a more efficient packing of logic and cache.
Ray tracing hardware scales from 6 RT cores on the 660M to 64 on the RX 9070 XT, a 10.7x increase that aligns with the Vulkan performance gap. The FP16 compute ratio also differs: the 660M uses a 2:1 ratio (2.918 TFLOPS), meaning half-rate FP16, while the RX 9070 XT uses a 1:1 ratio (48.66 TFLOPS), meaning full-rate FP16 throughput. This makes the RX 9070 XT significantly more capable for workloads that mix FP16 and FP32 operations, such as certain compute or AI inference tasks.
The RX 9070 XT's memory architecture—dedicated 16 GB GDDR6 with 644.6 GB/s bandwidth—is a fundamental architectural advantage over the 660M's system-shared memory. This affects not just bandwidth but also latency and memory pressure, as the 660M competes with the CPU for system RAM. The 660M's 40 W TDP versus the RX 9070 XT's 304 W TDP also reflects the architectural trade-off between integrated efficiency and discrete performance.