**Editorial Brief**
The question of whether to upgrade from an RTX 5060Ti with 16GB VRAM to either a newer RTX 3080 with 20GB or another RTX 5060Ti is complex and depends on specific use cases. For running inference on large language models like Qwen 3.6 (27B) and Gemma 4 (31B), the key considerations are memory bandwidth, compute capabilities, and cost.
**Takeaways:**
– **Memory Bandwidth**: The RTX 5060Ti with 16GB VRAM may struggle with large language models due to insufficient memory bandwidth. This could limit performance or force more frequent data transfers between GPU and CPU.
– **Compute Capabilities**: The newer RTX 3080 has better compute capabilities, which might be beneficial for certain workloads but is not a strict requirement given the current state of these large language models.
– **Cost vs. Performance**: Both options are around the same price point, so deciding may come down to personal preference and specific needs, especially if one model’s modded status could offer better performance at a similar cost.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
