The creator tested the MacBook Pro M5 Max against the M4 Max for local AI tasks, finding that the M5 Max offers a significant performance boost—typically between 2x to 3x faster in prompt processing—though it falls short of the 4x improvement claimed by Apple. These results highlight the variability in real-world AI workloads and emphasize the importance of hands-on benchmarking over theoretical expectations.
In this video, the creator visits Apple HQ Australia to test and compare the performance of the MacBook Pro M5 Max against the M4 Max, specifically focusing on local AI tasks. Initially, many believed the M5 Max would be only about 20% faster, while the creator predicted it could be two and a half times faster. Previous tests with the M5 Pro showed it was roughly twice as fast as the M4 Max in prompt processing, even outperforming the M3 Ultra in some cases, which led to a revised expectation that the M5 Max might be four times faster than the M4 Max.
The creator conducted live benchmarking tests using different AI models, including the popular Quinn 3.5 and Gemma 4 26B. For Quinn 3.5, the M5 Max completed prompt processing in 15.65 seconds, while the M4 Max took 45.98 seconds, indicating about a 3x speed improvement. However, token generation was faster on the M4 Max in some cases. For the Gemma model, the M5 Max showed a 2x improvement in prompt processing speed compared to the M4 Max, but interestingly, the M4 Max was faster in token generation for this model.
These mixed results led to some confusion, as the performance gains varied between 2x and 3x rather than the expected 4x improvement touted by Apple. The creator acknowledged that the M5 Max model tested was the 36GB version with a 32-core GPU, which is less powerful than the top-tier 40-core GPU model. This difference likely accounts for some of the discrepancy, suggesting that the fastest M5 Max could be around 2.4x to 3.6x faster than the M4 Max, but still not quite reaching the 4x mark.
Despite not hitting the 4x speed increase, the creator emphasized that the M5 Max still represents a significant performance boost over the M4 Max, especially in prompt processing tasks for local AI. The video also highlights the importance of real-world testing and benchmarking, as theoretical or advertised improvements may not always fully translate in practical use cases. The creator invites viewers with access to both models to share their own benchmark results to help clarify the performance differences further.
In conclusion, the video provides a detailed and transparent comparison of the M5 Max and M4 Max MacBook Pros for AI workloads, debunking some exaggerated claims while confirming substantial improvements. The creator’s hands-on approach and willingness to revise predictions based on actual data offer valuable insights for users considering an upgrade. Overall, the M5 Max is faster, but the exact performance gain depends on the specific model and workload, with improvements ranging from 2x to just under 4x.