AMD Radeon HD 8690M vs AMD Radeon R7 M350 Comparison

The Verdict

The benchmark data positions the AMD Radeon R7 M350 as the stronger mobile GPU of the two, winning the single head-to-head benchmark with a 13.9% margin in Geekbench OpenCL. Its average benchmark score of 6327 sits well above the HD 8690M's 6137, and it also holds a commanding lead in raw compute specifications. The HD 8690M, however, is not obsolete; it remains competitive in specific legacy scenarios and its memory configuration offers a distinct advantage in bandwidth that the R7 M350 cannot match. For users prioritizing raw compute throughput and modern API support, the R7 M350 is the clear choice. For those running older titles or tasks that are highly sensitive to memory bandwidth, the HD 8690M's GDDR5 memory could provide a tangible benefit, despite its lower overall compute scores. Both GPUs are end-of-life, so the decision is purely about matching the existing hardware's strengths to the workload.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon R7 M350 has an average benchmark score of 6327, compared to the AMD Radeon HD 8690M's 6137. This places the R7 M350 in a percentile rank of 36 among all GPUs, while the HD 8690M also sits at the 36th percentile, indicating they are in the same general performance tier despite the score difference.

Q: How do the two compare in the Geekbench OpenCL test?

A: In the Geekbench OpenCL test, the R7 M350 scores 6991, while the HD 8690M scores 6137. This results in a 13.9% delta, with the R7 M350 winning the head-to-head comparison.

Q: What are the key architectural differences between the two GPUs?

A: The R7 M350 uses the Meso chip based on GCN 3.0 architecture, while the HD 8690M uses the Sun chip based on the older GCN 1.0 architecture. The R7 M350 is fabricated with 1,550 million transistors on a 125 mm² die, whereas the HD 8690M has 690 million transistors on a 56 mm² die. The R7 M350 also supports DirectX 12 (12_0), while the HD 8690M only supports DirectX 12 (11_1).

Q: Is there a difference in memory bandwidth?

A: Yes, there is a significant difference. The HD 8690M has a memory bandwidth of 32.00 GB/s using GDDR5 memory, while the R7 M350 has a bandwidth of 16.00 GB/s using DDR3 memory. Both use a 64-bit memory bus.

Q: Which GPU has more shading units?

A: The AMD Radeon R7 M350 has 384 shading units, which is 64 more than the HD 8690M's 320 shading units. The R7 M350 also has 24 texture mapping units (TMUs) compared to 20 on the HD 8690M, while both have 8 ROPs.

Q: What is the difference in their release dates?

A: The HD 8690M was released on 2013-02-28, while the R7 M350 came later on 2015-05-04. The HD 8690M's predecessor is "London" and its successor is "Gem System", which is the generation of the R7 M350.

Architecture Differences

The two GPUs represent distinct generations of AMD's graphics technology. The R7 M350 is built on the GCN 3.0 architecture using the Meso chip, while the HD 8690M is based on the older GCN 1.0 architecture with the Sun chip. This generational leap is reflected in several key areas. The R7 M350 is a substantially larger chip, packing 1,550 million transistors onto a 125 mm² die, compared to the HD 8690M's 690 million transistors on a 56 mm² die. Interestingly, the transistor density is nearly identical, at 12.4M / mm² for the R7 M350 and 12.3M / mm² for the HD 8690M, indicating that the size increase is primarily for additional compute resources rather than a denser design.

The architectural shift also brought changes in API support. The R7 M350 supports DirectX 12 (12_0), while the HD 8690M is limited to DirectX 12 (11_1). Both support OpenGL 4.6 and Vulkan 1.2.170. This means newer games and applications that leverage DirectX 12 features will require the R7 M350 for full compatibility. The R7 M350 also supports FP16 compute at a 1:1 ratio with FP32, offering 779.5 GFLOPS for both, whereas the HD 8690M's FP16 capability is not specified in the data. This suggests the R7 M350 has a more modern compute pipeline capable of handling half-precision workloads, which are increasingly used in machine learning and certain graphics effects.

The HD 8690M belongs to the "Solar System" generation (HD 8600M), with its predecessor being "London". The R7 M350 is part of the "Gem System" (R7 M300) generation, and its successor is "Polaris Mobile". This places the two cards at different points in AMD's mobile GPU roadmap, with the R7 M350 representing a more recent design iteration.

Specification Differences

The specification sheet reveals a clear progression in compute hardware from the HD 8690M to the R7 M350. The R7 M350 has 384 shading units, 24 TMUs, and 8 ROPs, while the HD 8690M has 320 shading units, 20 TMUs, and 8 ROPs. This 20% increase in shading units and TMUs translates directly into higher compute throughput. The R7 M350's FP32 performance is 779.5 GFLOPS, compared to the HD 8690M's 624.0 GFLOPS, a difference of roughly 25%. Pixel rate is also higher on the R7 M350 at 8.120 GPixel/s versus 7.800 GPixel/s, and texture rate is 24.36 GTexel/s versus 19.50 GTexel/s.

Clock speeds are close but slightly favor the R7 M350. The R7 M350 has a base clock of 1000 MHz and a boost clock of 1015 MHz, while the HD 8690M has a base clock of 925 MHz and a boost of 975 MHz. Memory, however, is a major point of divergence. The R7 M350 comes with 4 GB of DDR3 memory on a 64-bit bus, running at 1000 MHz with an effective speed of 2 Gbps, resulting in a bandwidth of 16.00 GB/s. The HD 8690M, in contrast, has 2 GB of GDDR5 memory on the same 64-bit bus, running at 1000 MHz with an effective speed of 4 Gbps, yielding a bandwidth of 32.00 GB/s. This means the HD 8690M has double the memory bandwidth of the R7 M350, despite having half the memory capacity. Both GPUs use a PCIe 3.0 x8 bus interface.

Head-to-Head Benchmarks

The only direct benchmark comparison available is the Geekbench OpenCL test, which the R7 M350 wins decisively. The R7 M350 scores 6991, while the HD 8690M scores 6137. This is a 13.9% advantage for the R7 M350. This margin is consistent with the raw compute differences, as the R7 M350 offers more shading units, higher clocks, and greater FP32 throughput. The R7 M350's victory in this test aligns with its position among its nearest rivals, where it sits just 0.7% behind the NVIDIA Quadro K620 (which scores 6282) and 0.9% behind the NVIDIA GeForce RTX 5070 Ti SUPER (which scores 6270). The R7 M350's average score of 6327 is also nearly identical to the AMD Radeon Pro WX 4100's 6330, showing it performs in a similar class to professional workstation GPUs.

The HD 8690M, despite its lower score, is not far off the pace. Its 6137 score places it only 0.3% ahead of the Intel Iris Pro Graphics 6200 (6117) and 1% ahead of both the NVIDIA RTX A400 (6078) and the NVIDIA GeForce MX230 (6077). Interestingly, the HD 8690M's score is 1.2% lower than the AMD Radeon HD 6950's 6210, showing that the older desktop card still holds a slight edge. The data shows that while the R7 M350 has a clear lead in this benchmark, the HD 8690M remains competitive within its own performance tier, especially when considering its memory bandwidth advantage could help in specific workloads not captured by this compute-centric test.

Where Each One Wins

The R7 M350 wins in scenarios that leverage raw compute power and modern API features. Its 13.9% lead in Geekbench OpenCL indicates superior performance in general-purpose GPU computing tasks, such as video encoding, physics simulations, and any workload that utilizes the shading units heavily. The higher FP32 throughput of 779.5 GFLOPS (versus 624.0 GFLOPS) makes it better suited for compute-intensive applications. Its support for DirectX 12 (12_0) means it can run newer games that require this API level, whereas the HD 8690M's DirectX 12 (11_1) support may be a limiting factor. The R7 M350 also has double the memory capacity at 4 GB, which is beneficial for modern games and applications that require larger frame buffers for high-resolution textures. The R7 M350 also offers FP16 support, which the HD 8690M lacks, making it a better choice for emerging workloads that use half-precision arithmetic.

The HD 8690M wins in tasks that are heavily dependent on memory bandwidth. Its 32.00 GB/s of bandwidth is exactly double the R7 M350's 16.00 GB/s. In gaming, this can translate to better performance in scenarios where the GPU needs to stream large amounts of texture data quickly, particularly at lower resolutions where the compute units are not the bottleneck. Older games, which were designed for the GCN 1.0 architecture and may not fully utilize the newer features of GCN 3.0, could run at comparable or even better frame rates on the HD 8690M due to its faster memory. The HD 8690M's smaller die size (56 mm² vs 125 mm²) and lower transistor count (690 million vs 1,550 million) also suggest it may be more power-efficient for the same workload, although TDP data is not available. For users with systems that have limited cooling or power headroom, the HD 8690M could be the more practical choice, despite its lower compute scores.