AMD Radeon 660M (4-CU)

AMD Radeon 660M (4-CU)
AMD Radeon 660M (4-CU) graphics card review

AMD Radeon 660M (4-CU): the same label, but one-third fewer compute units

The Radeon 660M name comes with an unpleasant complication. Most processors with this graphics solution received a version with 6 compute units, but in the Ryzen 3 7335U, AMD kept only 4 CUs, 256 shaders, and a clock speed of up to 1.8 GHz. The name remained unchanged even though performance is noticeably lower. As a result, the Radeon 660M designation in a laptop’s specifications still does not indicate which version is actually installed.

The Radeon 660M (4-CU) is based on the RDNA 2 architecture and belongs to the Rembrandt-R family. Its technological foundation is the same as that of the full Radeon 660M and Radeon 680M, but it has significantly fewer graphics units.

How the 4-CU Radeon 660M differs from the standard version

The difference between the two Radeon 660M variants is not limited to clock speed.

Graphics Example processor CU Shaders GPU clock 3DMark Time Spy Graphics
Radeon 660M (4-CU) Ryzen 3 7335U 4 256 1.8 GHz 1017
Radeon 660M Ryzen 5 7535U 6 384 1.9 GHz 1562

The full Radeon 660M scores approximately 54% higher in Time Spy. That is too large a difference to treat the 4-CU variant as simply another configuration of the same GPU.

The cut-down version’s theoretical performance is approximately 0.92 TFLOPS FP32. DirectX 12 and hardware ray-tracing support are retained, but four CUs do not provide enough performance for games with RT.

Performance compared with other integrated GPUs

In 3DMark Time Spy, the Radeon 660M (4-CU) falls between the weakest Radeon 610M and the full Radeon 660M.

Processor Integrated graphics CU 3DMark Time Spy Graphics Difference
Ryzen 3 7320U Radeon 610M 2 515 -49%
Ryzen 3 7335U Radeon 660M (4-CU) 4 1017 baseline
Ryzen 5 7535U Radeon 660M 6 1562 +54%
Ryzen 5 8540U Radeon 740M 4 1722 +69%
Ryzen 7 7735U Radeon 680M 12 2518 +148%

Compared with the Radeon 610M, the gain is substantial: the Time Spy score nearly doubles. However, the full Radeon 660M is already 54% faster.

The comparison with the Radeon 740M is particularly revealing. It also has four compute units, but its RDNA 3 architecture and clock speed of up to 2.5 GHz allow it to score approximately 1722 points-69% more than the Radeon 660M (4-CU).

The Radeon 680M, with 12 CUs, is already in an entirely different category and outperforms it by approximately 2.5 times.

Gaming tests

Four CUs limit the Radeon 660M much more severely than its name might suggest. 1080p is still viable in older and less demanding games, but demanding titles quickly require lower settings or resolution.

Game Settings Resolution Average FPS
Dota 2 Reborn High 1920x1080 55.9
Dota 2 Reborn Ultra 1920x1080 48.4
GTA V High 1920x1080 35.2
Shadow of the Tomb Raider Medium 1920x1080 25.6
Shadow of the Tomb Raider Low 1280x720 73.2
Cyberpunk 2077 1.6 Low 1920x1080 23.7
Forza Horizon 5 Medium 1920x1080 39
The Witcher 3 High 1920x1080 24.9

The GPU is powerful enough for Dota 2 even at Full HD with high graphics settings. GTA V also remains playable, although 35 FPS on High leaves little headroom.

Shadow of the Tomb Raider clearly demonstrates this iGPU’s limit. At 1080p on medium quality, performance is approximately 26 FPS, while switching to 720p and low settings raises the frame rate above 70 FPS.

Cyberpunk 2077 is more demanding: it delivers approximately 24 FPS even at low settings in 1080p. For more comfortable gameplay, you will need to switch to 720p and use FSR.

Forza Horizon 5 performs better-approximately 39 FPS in 1080p on the Medium preset. Thus, the Radeon 660M (4-CU) is suitable not only for older games, but newer titles require settings to be selected individually for each game.

System memory can cost a third of the performance

The Radeon 660M has no dedicated video memory, so it uses system DDR5 or LPDDR5. For four compute units, memory bandwidth remains an important limitation.

This is clearly demonstrated by the Lenovo ThinkPad E14 Gen 6 with a Ryzen 3 7335U.

Configuration PassMark 3D Graphics Mark
8 GB DDR5, single-channel, Balanced 2025
24 GB DDR5, dual-channel, Balanced 2609
24 GB DDR5, dual-channel, Performance 2690

After installing a second module, the score increased by approximately 29%. Switching to Performance mode increased the difference from the original configuration to approximately 33%.

In Final Fantasy XIV, the same laptop scored 2051 points with 8 GB of single-channel memory and 3006-3104 points after switching to dual-channel mode.

Therefore, a Ryzen 3 7335U configuration with a single DDR5 module should be considered separately from a laptop with dual-channel memory. For an integrated GPU, the difference is too large to ignore.

Which laptops use the Radeon 660M (4-CU)

The primary processor with this graphics version is the Ryzen 3 7335U. It appears in several production laptops:

  • Lenovo ThinkPad E14 Gen 6;
  • Lenovo ThinkPad E16 Gen 2;
  • Lenovo ThinkBook 16 Gen 7;
  • HP ProBook 465 G11;
  • Acer Aspire 15 A15-41M.

When choosing a laptop, it is important to look specifically at the processor. The Ryzen 3 7335U uses four CUs and has a clock speed of up to 1.8 GHz. The Ryzen 5 7535U and Ryzen 5 7535HS are also called Radeon 660M, but they have six CUs and a clock speed of up to 1.9 GHz.

The identical name hides a difference that exceeds 50% in Time Spy.

What the Radeon 660M (4-CU) is suitable for

It has more than enough performance for Windows, web browsing, office applications, video, and programming. These tasks rarely place a serious load on the GPU.

Gaming capabilities depend on the title. Dota 2 and other lightweight online games run normally at 1080p. GTA V remains playable in Full HD, while Shadow of the Tomb Raider and more demanding games require reduced quality settings.

Cyberpunk 2077 launches, but 23-24 FPS in Full HD on Low clearly demonstrates this GPU’s ceiling. For similar games, it is better to plan from the start on using 720p, low settings, and FSR.

The Radeon 660M (4-CU) is too slow for 3D rendering, modern AAA games, and demanding GPU workloads.

Is a laptop with the Radeon 660M (4-CU) worth buying?

A laptop based on the Ryzen 3 7335U should primarily be viewed as a work machine capable of running undemanding games, rather than as a budget gaming computer.

If a model with a Ryzen 5 7535U is available at a similar price, the difference is substantial. The full Radeon 660M is approximately 54% faster in Time Spy, while the Ryzen 5 also has six CPU cores instead of four.

The Ryzen 5 8540U with the Radeon 740M is even faster: despite having the same four CUs, its graphics are approximately 69% ahead of the Radeon 660M (4-CU).

It is also worth checking the memory configuration separately. Two-channel DDR5 is preferable for the Ryzen 3 7335U. Buying a version with a single module is justified only if you can install the second module yourself.

Conclusion

The Radeon 660M (4-CU) clearly demonstrates why the name of an integrated GPU alone is not enough to assess a laptop. AMD kept the Radeon 660M brand but reduced the number of compute units from six to four and lowered the clock speed to 1.8 GHz.

In the Ryzen 3 7335U, it scores approximately 1017 points in 3DMark Time Spy Graphics. That is roughly twice as fast as the Radeon 610M, but 54% slower than the full Radeon 660M and 69% slower than the Radeon 740M.

In Dota 2, you can expect approximately 56 FPS at 1080p High, while GTA V delivers 35 FPS at 1080p High, and Cyberpunk 2077 reaches only approximately 24 FPS at 1080p Low.

There is one more important factor: memory. Single-channel DDR5 can cost approximately a third of the graphics performance. Therefore, for the Radeon 660M (4-CU), dual-channel memory is not a minor optimization but a mandatory requirement if you expect to use the integrated GPU for gaming.

GPU Comparison

Compare Radeon 660M (4-CU) with other GPUs

Please select GPU from the dropdown list.

Basic

Label Name
AMD
Platform
Integrated
Launch Date
January 2023
Former Codename
Rembrandt R
GPU Lithography
6 nm
Model Name
AMD Radeon 660M
Generation
Radeon 600M
Boost Clock
1800 MHz
Bus Interface
Integrated
RT Cores
4
Compute Units
4
Foundry
TSMC
Process Size
6 nm
Architecture
RDNA 2

Memory Specifications

Memory Size
Up to 2 GB Shared
Memory Type
DDR5 / LPDDR5
Memory Bus
?
The memory bus width refers to the number of bits of data that the video memory can transfer within a single clock cycle. The larger the bus width, the greater the amount of data that can be transmitted instantaneously, making it one of the crucial parameters of video memory. The memory bandwidth is calculated as: Memory Bandwidth = Memory Frequency x Memory Bus Width / 8. Therefore, when the memory frequencies are similar, the memory bus width will determine the size of the memory bandwidth.
128-bit
Bandwidth
?
Memory bandwidth refers to the data transfer rate between the graphics chip and the video memory. It is measured in bytes per second, and the formula to calculate it is: memory bandwidth = working frequency × memory bus width / 8 bits.
Up to 76.8 GB/s DDR5-4800 / 102.4 GB/s LPDDR5-6400

Display and Media

AMD FreeSync
Yes
AV1 Encode/Decode
Decode only
DisplayPort Extensions
Adaptive-Sync, HBR3, HDR Metadata
H.264 Hardware Encode/Decode
Encode / Decode
H.265 HEVC Hardware Encode/Decode
Encode / Decode, 8-bit / 10-bit
HDCP Version
2.2
HDMI Version
2.1
Intel Quick Sync Video
No
Max Resolution DP
3840x2160 @ 60Hz
Max Video Decode Bandwidth
1080p60 MPEG2, 1080p60 VC1, 1080p240 8bpc VP9, 1080p240 10bpc VP9, 1080p240 H.264, 1080p240 8bpc H.265, 1080p240 10bpc H.265, 1080p240 JPEG 8bpc
Max Video Encode Bandwidth
1080p205 8bpc H.264, 1440p115 8bpc H.264, 2160p51 8bpc H.264, 1080p205 8bpc H.265, 1440p115 8bpc H.265, 2160p51 8bpc H.265
Number of Displays Supported
4
Outputs
HDMI 2.1 / DisplayPort 2.0 / USB Type-C DisplayPort Alt Mode
USB Type-C DisplayPort Alternate Mode
Yes
Wireless Display
Miracast

Theoretical Performance

FP16 (half)
?
An important metric for measuring GPU performance is floating-point computing capability. Half-precision floating-point numbers (16-bit) are used for applications like machine learning, where lower precision is acceptable. Single-precision floating-point numbers (32-bit) are used for common multimedia and graphics processing tasks, while double-precision floating-point numbers (64-bit) are required for scientific computing that demands a wide numeric range and high accuracy.
1.8432 TFLOPS
FP32 (float)
?
An important metric for measuring GPU performance is floating-point computing capability. Single-precision floating-point numbers (32-bit) are used for common multimedia and graphics processing tasks, while double-precision floating-point numbers (64-bit) are required for scientific computing that demands a wide numeric range and high accuracy. Half-precision floating-point numbers (16-bit) are used for applications like machine learning, where lower precision is acceptable.
0.92 TFLOPS

AI Features

Intel Deep Learning Boost on GPU
No

Miscellaneous

Shading Units
?
The most fundamental processing unit is the Streaming Processor (SP), where specific instructions and tasks are executed. GPUs perform parallel computing, which means multiple SPs work simultaneously to process tasks.
256
OpenCL Version
2.2
OpenGL
4.6
CUDA
No
DirectX
DirectX 12

Benchmarks

FP32 (float)
Score
0.92 TFLOPS
3DMark Time Spy
Score
1017
Vulkan
Score
11226
OpenCL
Score
9903

Compared to Other GPU

FP32 (float) / TFLOPS
1.007 +9.5%
0.988 +7.4%
0.28 -69.6%
3DMark Time Spy
3817 +275.3%
1864 +83.3%
1205 +18.5%
Vulkan
59828 +432.9%
34688 +209%
OpenCL
A2
35144 +254.9%
20836 +110.4%
12393 +25.1%
884 -91.1%