Old Mac Pro shines with new Vulkan support for LLMs
2016's 'Trash Can' Mac Pro now runs advanced AI models effectively.
Get AI news that actually matters
One email a day. Zero fluff. Join 10,000+ professionals.
The 2016 Mac Pro, affectionately known as the 'Trash Can', has proven its resilience in the evolving landscape of AI technology. Originally costing nearly £10,000, this machine was primarily used as a Kubernetes single-node development platform. However, recent updates to Vulkan support have unlocked its potential, allowing its D700 GPUs to run advanced models like Qwen 3.5 and Qwen 2.5 coder. With 64GB of RAM and 24 logical cores, it can now handle significant workloads, outputting 11 t/s and 22 t/s respectively at 70k context.
The standout revelation is the performance of Qwen 3.5, which significantly outperforms Claude Sonnet 4.6 in planning tasks, particularly in complex C# .NET applications. This unexpected capability highlights the Mac Pro's longevity and adaptability in AI applications, challenging the notion that only the latest hardware can effectively run large language models (LLMs). As professionals explore the potential of older hardware, the Mac Pro serves as a reminder that powerful computing can come from unexpected sources, encouraging further experimentation in the AI community.
- 2016 Mac Pro, originally £10,000, now runs advanced AI models with Vulkan support.
- D700 GPUs output 11 t/s and 22 t/s for Qwen 3.5 and Qwen 2.5 coder.
- Qwen 3.5 outperforms Claude Sonnet 4.6 in complex C# planning tasks.
Why It Matters
Demonstrates that older hardware can still deliver substantial AI performance.