AMD Radeon 890M vs NVIDIA GeForce GTX 660 Comparison

The AMD Radeon 890M and NVIDIA GeForce GTX 660 represent two distinct eras of GPU design, separated by a decade of architectural evolution. The 890M is a modern integrated processor built on TSMC's 4 nm node, while the GTX 660 is a legacy discrete card from 2012 built on 28 nm. Benchmark data shows the 890M leads decisively in every head-to-head comparison, but the GTX 660 still holds relevance in specific legacy scenarios.

Head-to-Head Benchmarks

The most striking disparity appears in Geekbench Vulkan results. The AMD Radeon 890M scores 40,808 points against the GTX 660's 11,415, a delta of 257.5% in favor of the integrated part. This is not a marginal improvement; the 890M more than triples the older card's output. Vulkan's low-overhead nature favors the RDNA 3.5 architecture's modern command processing, while Kepler's 2012-era driver stack struggles to extract comparable efficiency from the same API.

Geekbench OpenCL tells a similar story, though with slightly less extreme separation. The 890M posts 37,254 points versus 11,347 for the GTX 660, a 228.3% advantage. OpenCL compute workloads benefit heavily from the 890M's 1,024 shading units and 5.939 TFLOPS of FP32 throughput, compared to the GTX 660's 960 shading units and 1.981 TFLOPS. The raw compute gap is nearly 3x, and the benchmark reflects that almost exactly.

In terms of average benchmark scores, the 890M's 9,210 average places it just 0.1% behind the AMD Radeon Vega 8 (9,221) and 0.7% behind the NVIDIA GeForce GTX 960 (9,273). The GTX 660's 9,022 average sits 0.2% behind the NVIDIA TITAN V CEO Edition (9,037) and 0.4% behind the GeForce GTX 560 (9,058). Both cards occupy the 45th percentile among all GPUs, indicating they are statistical peers in overall performance distribution, yet the head-to-head tests reveal the 890M's dominance in modern API workloads.

The GTX 660 does not win a single head-to-head benchmark in this comparison. Its only recorded scores come from Geekbench OpenCL, Vulkan, and Metal (4,305), none of which appear in the direct comparison set. The 890M wins both available head-to-head tests, giving it a 2-0 record. The delta percentages are so lopsided that any discussion of "close competition" would be misleading.

Where Each One Wins

The AMD Radeon 890M wins in every scenario where modern API support matters. Its DirectX 12 Ultimate (12_2) and Vulkan 1.4 support enable features the GTX 660 cannot access, which translates directly to higher scores in contemporary benchmark suites. The 890M's 16 ray tracing cores provide hardware acceleration that the GTX 660 completely lacks, making it the clear choice for any workload that leverages RT effects. Its 5.939 TFLOPS FP32 and FP16 (1:1) performance also gives it massive compute headroom for GPGPU tasks like OpenCL compute, where it outperforms the GTX 660 by over 200%.

The NVIDIA GeForce GTX 660 wins in exactly one category: legacy software compatibility. Its DirectX 12 (11_0) support means it runs older DirectX 11 titles with mature, battle-tested drivers. The card's 2 GB of dedicated GDDR5 memory on a 192-bit bus with 144.2 GB/s bandwidth provides predictable, local memory performance that does not depend on system RAM. For users running pre-2015 games that never received Vulkan or DX12 patches, the GTX 660's 980 MHz base and 1,032 MHz boost clocks offer stable, consistent frame pacing that the 890M's system-shared memory cannot guarantee.

The 890M's memory configuration is "System Dependent," meaning its bandwidth varies with the host laptop's RAM. This creates a scenario where the GTX 660's 144.2 GB/s fixed bandwidth can outperform the 890M in memory-bound legacy workloads, even though the 890M's 92.80 GPixel/s pixel rate and 185.6 GTexel/s texture rate dwarf the GTX 660's 20.64 GPixel/s and 82.56 GTexel/s respectively. The GTX 660 also wins on physical interface: its PCIe 3.0 x16 connection provides more dedicated lanes than the 890M's PCIe 4.0 x8.

The Verdict

The data unequivocally favors the AMD Radeon 890M for any modern use case. Its 257.5% Vulkan advantage and 228.3% OpenCL advantage are not incremental gains; they represent generational leaps in compute efficiency and API utilization. The 890M's 4 nm process node, 34,000 million transistors, and RDNA 3.5 architecture deliver performance that the GTX 660's 28 nm Kepler design cannot approach. For gaming, content creation, or compute workloads built after 2017, the 890M is the only rational choice.

The NVIDIA GeForce GTX 660 should only be selected by users with specific legacy requirements. Its 2 GB GDDR5 frame buffer is fixed and predictable, its 140 W TDP is manageable in a dual-slot form factor, and its 241 mm length fits standard desktop cases. The card's launch MSRP was 229 USD, though its end-of-life production status means it now exists only in used markets. For someone running a Windows 7-era build with DirectX 11-only games, the GTX 660 remains functional. However, its 1.981 TFLOPS FP32 and 82.56 GTexel/s texture rate are less than one-third of the 890M's capabilities.

The 890M's 15 W TDP is particularly notable against the GTX 660's 140 W. That 125 W difference means the integrated AMD part delivers superior performance at a fraction of the power draw. The 890M requires no power connectors and occupies an IGP slot, while the GTX 660 needs a 6-pin connector and a 300 W suggested PSU. For mobile or small-form-factor systems, the 890M is categorically superior.

FAQ

Q: How much faster is the AMD Radeon 890M in Vulkan benchmarks?

A: The 890M scores 40,808 in Geekbench Vulkan, which is 257.5% higher than the GTX 660's 11,415.

Q: Does the GTX 660 have any ray tracing capability?

A: No. The GTX 660 has no RT cores, while the 890M includes 16 dedicated ray tracing cores within its RDNA 3.5 architecture.

Q: What is the memory bandwidth difference?

A: The GTX 660 has a fixed 144.2 GB/s from its 192-bit GDDR5 bus. The 890M's bandwidth is "System Dependent," meaning it varies based on the host system's shared memory configuration.

Q: Which card has higher FP32 compute performance?

A: The 890M delivers 5.939 TFLOPS, which is 3x higher than the GTX 660's 1.981 TFLOPS.

Q: Are these cards in the same performance percentile?

A: Yes, both occupy the 45th percentile among all GPUs, though the 890M has a higher average benchmark score of 9,210 versus 9,022 for the GTX 660.

Q: What is the process node difference?

A: The 890M uses TSMC's 4 nm process with 34,000 million transistors on a 233 mm² die. The GTX 660 uses TSMC's 28 nm process with 2,540 million transistors on a 221 mm² die.

Architecture Differences

The AMD Radeon 890M is built on the Strix Point chip using RDNA 3.5 architecture, representing the Navi III IGP generation for mobile platforms. It uses a 4 nm process from TSMC with 34,000 million transistors packed into a 233 mm² die, yielding a transistor density of 145.9M per mm². The GPU has 1,024 shading units, 64 texture mapping units, 32 ROPs, and 16 ray tracing cores. Its base clock is 400 MHz with a boost of 2,900 MHz. The 890M supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, making it fully compliant with modern API standards. It operates at a 15 W TDP with no power connectors, using PCIe 4.0 x8 interface and system-shared memory.

The NVIDIA GeForce GTX 660 uses the GK106 chip with Kepler architecture, part of the GeForce 600 generation. It is built on TSMC's 28 nm process with 2,540 million transistors on a 221 mm² die, giving a transistor density of just 11.5M per mm². The GPU has 960 shading units, 80 TMUs, and 24 ROPs, with no ray tracing or tensor cores. Its base clock is 980 MHz with a boost of 1,032 MHz, and memory runs at 1,502 MHz (6 Gbps effective). The card features 2 GB of GDDR5 memory on a 192-bit bus with 144.2 GB/s bandwidth. It supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. The GTX 660 draws 140 W, requires a 6-pin power connector and 300 W PSU, and uses a PCIe 3.0 x16 interface. It measures 241 mm in length, uses a dual-slot cooler, and provides 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2 outputs. The card is end-of-life, having been released in September 2012 with a launch MSRP of 229 USD.

The architectural gap is vast: RDNA 3.5's 4 nm node enables 13.4x more transistors per mm² than Kepler's 28 nm node. The 890M's 16 RT cores and FP16 (1:1) support have no equivalents in the GTX 660, which lacks both hardware ray tracing and native FP16 throughput. The 890M's system-shared memory model contrasts sharply with the GTX 660's dedicated 2 GB GDDR5, creating different performance characteristics depending on workload memory patterns. The 890M's IGP form factor and 15 W TDP make it suitable for thin-and-light laptops, while the GTX 660's 140 W dual-slot design requires a desktop chassis with adequate airflow and power delivery.