LM Studio vs Ollama benchmarked on Linux with an RTX 3090
Start here
A Docker-based test on a single RTX 3090 compared LM Studio and Ollama running the same Qwen3.5 9B and 27B GGUF files, verified by SHA-256. Ollama was about 10% faster on the 9B model and 5% faster on the 27B for single-request generation, while LM Studio reached 90.2 tokens/s on four concurrent 27B requests versus Ollama's 33.1 tokens/s. The comparison comes with caveats. LM Studio's app is closed source, and its Docker image, lmstudio/llmster-preview, is marked a technical preview described as having "CPU-only support on x86 systems"; the tester says GPU use therefore requires unsupported steps and that LM Studio does not support this setup. Since July 2025, LM Studio has been free for home and internal business use under its terms.
Generated from reports · updated 2 h ago
Report timeline
Follow the reports to see the event from different sides.
- Hacker News · AILM Studio vs. Ollama on Linux: Same Qwen3.5 GGUF Files Benchmarked on an RTX 3090
A Docker-based test on one RTX 3090 compared LM Studio and Ollama running the same Qwen3.5 9B and 27B GGUF files, verified by SHA-256. Ollama was about 10% faster on the 9B model and 5% faster on the 27B for single-request generation, while LM Studio hit 90.2 tokens/s on four concurrent 27B requests versus Ollama's 33.1 tokens/s. LM Studio's official Docker image is CPU-only, so GPU use requires unsupported steps.
Heat for this event
Not enough continuous observations yet to draw a trend.