RUUZEN
AI

Gaming laptop tested as a local AI workstation for programming

An enthusiast evaluated the capabilities of the MAIBENBEN X17F with RTX 5060 for running large language models and solving complex tasks in C++ and Rust.

17 Nov, 15:22Read in Russian

The gaming laptop MAIBENBEN X17F, equipped with an AMD Ryzen 9 8940HX processor and an NVIDIA GeForce RTX 5060 graphics card, was tested as a local AI workstation. The experiment demonstrated that even a device with 8 GB of video memory can run large language models without connecting to cloud services.

For the tests, Qwen3.8-Flash-Next Coder, GPT-OSS 20B, Qwen3-Coder-30B-A3B, and DeepSeek-Coder-V2-Lite were used. The models were evaluated using two engines—Strata and Ollama. The test bench included tasks such as creating an audio synthesizer in C++20, developing a distributed storage system in Rust, and debugging an existing software project.

One of the most notable results was the performance of Qwen3.8-Flash-Next Coder. Despite its size of approximately 58.4 GB, the model delivered an average of 41 tokens per second on short queries and 39.6 tokens per second during prolonged generation. In certain code correction rounds, the speed reached 75–83 tokens per second.

To work with such models, both the GPU’s video memory and the laptop’s RAM were used simultaneously. The experiment’s author also noted the need to configure the Windows swap file and enable the device’s performance mode. However, the testing revealed issues with long contexts, inference engine performance, and Windows 11’s system protection mechanisms.

The experiment showed that running large AI models locally is possible even on a mid-range gaming laptop, though performance and stability depend on software settings and available memory.