 |
| Apple M5 Max and M5 Pro-powered MacBook Pros were announced earlier this week. |
It’s been less than a week since Apple quietly unveiled the next generation of its custom silicon, and the internet is already buzzing with the first wave of benchmark results. The new M5 Pro and M5 Max chips have landed in the wild, and early Geekbench scores paint a picture of a classic Apple paradox: absolute dominance in one area, and puzzling stagnation in another.
For professionals and power users eyeing the next upgrade, the numbers coming out of Cupertino’s latest silicon are a mixed bag of jaw-dropping victories and head-scratching trade-offs.
The x86 Killer Returns
If there was any doubt that Apple’s CPU architecture remains the undisputed champion of the consumer processor world, the M5 Max has put it to rest. Early listings show the high-end M5 Max configuration sporting an 18-core CPU, and the results are devastating for the competition.
In raw multi-core and single-core performance, the M5 Max and Pro are outpacing virtually every consumer-grade x86 chip currently on the market. Intel’s latest Raptor Lake updates and AMD’s top-tier Ryzen mobile processors are being left in the dust by the efficiency and raw throughput of Apple’s in-house designs. For video editors, coders, and creatives who live in the export timeline, this generational CPU leap means the M5 MacBook Pros are about to become the fastest mobile workstations money can buy.
GPU Performance: A Reality Check
However, the narrative shifts dramatically when we switch focus to graphics. While the CPU is a heavyweight champion, the GPU performance in these early benchmarks is raising eyebrows.
According to Geekbench’s OpenCL benchmarks, the M5 Max is trading blows with Nvidia’s new RTX 5070 Laptop GPU. At first glance, competing with Nvidia’s latest 50-series mobile card sounds impressive—until you remember that the 5070 is an upper-mid-range part. Meanwhile, the M5 Pro is barely managing to keep pace with a mid-range RTX 5060 Laptop GPU.
Before we declare a GPU crisis, a major caveat is required: OpenCL is essentially a legacy API on macOS. Apple has been pushing developers toward its proprietary Metal framework for years, and OpenCL performance has often lagged as a result. These numbers likely do not represent the true peak of the M5’s graphics architecture, but they do suggest that raw rasterization might not be the massive leap some were hoping for.
The AI Benchmark Surprise: Regressing to Progress?
Beyond CPU and GPU grunt, the most fascinating data point comes from the Geekbench AI test. This benchmark attempts to measure on-device machine learning performance across the CPU, GPU, and the dedicated Neural Engine. While cross-platform comparisons to Windows or Linux machines are difficult due to API differences, we can look at the generational leap from the M4 to the M5 to see just how much "Apple Intelligence" has improved.
The results for the M5 Max and Pro reveal a confusing picture:
M5 Max SoC AI Benchmarks:
- M5 Max Neural Engine: 5,100 (Single) | 44,000 (Half) | 58,000 (Quantized)
- M5 Max CPU: 5,000 (Single) | 8,700 (Half) | 7,000 (Quantized)
- M5 Max GPU: 27,000 (Single) | 43,100 (Half) | 41,000 (Quantized)
M5 Pro SoC AI Benchmarks:
- M5 Pro Neural Engine: 5,000 (Single) | 44,000 (Half) | 57,500 (Quantized)
- M5 Pro CPU: 5,000 (Single) | 8,700 (Half) | 7,011 (Quantized)
- M5 Pro GPU: 20,200 (Single) | 35,000 (Half) | 34,000 (Quantized)
The GPU Lead: As expected, the GPU handles the bulk of the heavy lifting here. The M5 Max GPU commands a roughly 20% lead in single-precision AI workloads over the M4 Max, and the M5 Pro GPU shows a similar 20% uplift over its predecessor.
The CPU Plateau: The CPU scores are virtually unchanged. This aligns with the idea that this generation's CPU is a refinement of the architecture rather than a complete overhaul.
The Neural Engine Anomaly: This is where things get weird. Both the M5 Pro and M5 Max utilize a 16-core Neural Engine—the same core count as the M4 series. While Apple claims the new engine is faster, the Geekbench AI scores tell a different story. In single-precision tests, the Neural Engine scores are actually worse than the M4. However, in half-precision and quantized tests—which are far more common for on-device AI tasks like image generation or large language model processing—the scores have improved by roughly 15% and 10% respectively.
So, is the Neural Engine faster? For the kinds of AI workloads the average user will encounter in macOS Sequoia and future "Apple Intelligence" features, the answer appears to be a cautious "yes." But the regression in single-precision suggests that either the benchmark isn't optimized for the new chip, or Apple has made architectural changes that favor specific types of calculations over others.
The Bigger Picture: Wait for Real-World Tests
It is crucial to remember that these are very early numbers pulled from pre-production devices or early review units. Geekbench, while a useful tool for synthetic comparison, is not the be-all and end-all of performance. It is highly possible that the Geekbench AI test is not yet fully utilizing the new neural accelerators embedded within each GPU core—a feature that could drastically change the final numbers.
Apple is touting up to a 4x faster performance in AI workloads compared to the M2 generation, and while these benchmarks show improvement, they don't yet reflect that kind of quantum leap.
If you are currently in the market for a professional-grade laptop, the base model M5 (found in the entry-level 14" MacBook Pro) offers a fantastic entry point into the ecosystem. You can check the latest pricing on the ** 14" MacBook Pro M5 on Amazon ** to see how the latest generation fits your budget.
Conclusion
The M5 Pro and Max are shaping up to be a study in contrasts. They feature the fastest mobile CPU cores on the planet, solidifying Apple's lead in processor design. Yet, the GPU performance seems to be in a holding pattern, trading blows with mid-to-high-range PC parts rather than pulling away. And the AI story is currently a mess of conflicting data points.
Ultimately, a single benchmark cannot capture the real-world experience of using these chips. The true test of the M5 generation will come from independent reviews that push the silicon through real video exports, 3D rendering, and complex AI model inference. Until then, the numbers suggest that if you need raw CPU power, the M5 Max is the new king. But if you were waiting for a GPU revolution, you might have to wait for the M5 Ultra—or the M6.