Key Takeaways
1. The M5 Pro and Max chips show significant enhancements in CPU performance, with the 18-core CPU outperforming most consumer-grade x86 chips.
2. GPU performance is lacking, with the M5 Max competing closely with the RTX 5070 and the M5 Pro struggling against the RTX 5060 due to OpenCL optimization issues on macOS.
3. The M5 Max GPU shows a 20% improvement in single-precision AI tasks over the M4 Max, while the M5 Pro shows a similar 20% improvement over the M4 Pro.
4. CPU scores remain mostly unchanged, and the 16-core Neural Engine has decreased in single-precision scores, though improvements are noted in half-precision and quantized tests.
5. Current benchmarks may not fully reflect the M5 generation’s AI capabilities, and true performance will be better evaluated through independent reviews with practical workloads.
Apple’s M5 Pro and Max were introduced just a few days ago, and initial benchmarks have started to emerge. These chips show clear enhancements in GPU and CPU performance compared to their predecessors, with the 18-core CPU in the M5 Max and Pro outperforming nearly all consumer-grade x86 chips available today.
GPU Performance Insights
However, the GPU performance is somewhat lacking, as the M5 Max competes closely with the RTX 5070 Laptop GPU in Geekbench OpenCL, while the M5 Pro struggles to keep pace with a mid-range RTX 5060 Laptop GPU in the same benchmark. It’s important to note that OpenCL is no longer optimized for macOS, which likely contributes to these less-than-stellar results.
AI Performance Comparison
Beyond just CPU and GPU numbers, Geekbench also evaluates on-device AI performance. Due to the differences in APIs between macOS and other platforms, cross-platform comparisons might not be entirely accurate. Nonetheless, we can assess the M5 Pro and Max against their earlier versions to understand how Apple has advanced in this area.
When we look at the Geekbench AI test results for the M5 Max SoC, the findings are as follows:
In comparison to the M4 Max, the M5 Max GPU appears to have about a 20% advantage in single-precision tasks. Similarly, the M5 Pro GPU shows an improvement of around 20% over the M4 Pro.
Neural Engine Performance
Curiously, the CPU scores have remained largely unchanged, and the 16-core neural engine’s scores have actually decreased in single-precision, which seems illogical. However, results in half-precision and quantized tests have improved by approximately 15% and 10%, respectively.
It’s also worth noting that the earlier models also had a 16-core Neural Engine. While Apple asserts that the new Neural Engine is quicker, they haven’t specified by how much. The base model M5 (14″ MacBook Pro currently priced at $1,450 on Amazon), alongside the M5 Pro and M5 Max SoCs, all come equipped with the same 16-core Neural Engine.
These initial figures should be taken with caution, as it’s possible that Geekbench AI hasn’t fully leveraged the neural accelerators in the M5 generation’s GPU cores yet. Apple claims that performance in AI tasks can be up to 4x faster, but current benchmarks paint a different picture.
Given that a single benchmark doesn’t accurately reflect real-world performance, the true AI capabilities of the M5 Pro and M5 Max SoCs will only be determined through independent reviews using practical workloads.
