A lab analysis published October 7, 2026, found smaller Pro-to-Max GPU gains for M5 than for M4 in three graphics tests. In Cinebench 2026, the M5 Max result was about 1.8 times the M5 Pro result, compared with almost 2 times for the 40-core M4 Max over the 20-core M4 Pro. M5 gains were also lower in Cyberpunk 2077 and Solar Bay Extreme.
GPU scaling describes how a benchmark result changes between chip tiers. The analysis also says Apple GPU performance has risen across generations; its findings concern the size of the Pro-to-Max gains in these particular tests.
Three tests show smaller M5 Pro-to-Max gains
| Test | M4 Pro to M4 Max | M5 Pro to M5 Max |
| Cinebench 2026 GPU test | Almost 2×; 40-core M4 Max versus 20-core M4 Pro | About 1.8× |
| Cyberpunk 2077 with ray tracing | 63% increase | About 56% increase |
| Solar Bay Extreme | 97% increase | 78% increase |
The Cinebench result came close to doubling when moving from the 20-core M4 Pro GPU to the 40-core M4 Max GPU. For M5, the reported Max-to-Pro ratio was about 1.8×. These are results from specific benchmark comparisons, not a rule that adding a given number of GPU cores will produce the same gain in every workload.
In Cyberpunk 2077 with ray tracing, the analysis also gave normalized performance-per-GPU-core figures: about 1 frame per second per core for M3, 1.5 for M4, 2.3 for M5 and 1.8 for M5 Max. Those per-core figures are a separate metric from the Pro-to-Max percentage increases in the table.
Results depend on the workload and software
Operating-system versions, benchmark revisions, apps, games and updates to Apple Metal can all affect graphics results over time. Comparisons across chip generations therefore depend on the specific workload and the software versions involved.
The analysis used a small sample and did not cover every MacBook Pro chip-core and memory configuration. Selected Mac Studio and Mac Mini results supplemented the laptop comparisons where desktop results were considered useful proxies; those desktop results are not MacBook Pro measurements.
Graphics benchmarks do not predict AI performance on their own
AI workloads can use both the Neural Engine/NPU and GPU neural accelerators. How those resources are allocated varies by task and system resources, so a graphics benchmark result alone does not predict AI performance.
