This MacBook Pro Makes Me Feel Stupid

Testing the M5 Max vs. Pro Chips

Initial Impressions

• The speaker compares the M5 Max to previous models, expressing discomfort with performance expectations.

• They highlight a personal experience of testing both chips under similar workloads, leading to surprising results.

Specifications Overview

• Key specifications include 10 CPU cores for the M5 and 18 for the Pro and Max, affecting performance significantly.

• Memory bandwidth increases with each model: 153 GB/s (M5), 307 GB/s (Pro), and 614 GB/s (Max). These are theoretical numbers that will be tested later.

SSD Performance Insights

Real-world SSD Testing

• The speaker emphasizes real-world testing over synthetic benchmarks, noting significant differences in SSD speeds across models due to PCIe Gen 5 adoption in Pro and Max.

• Higher capacity drives yield better performance due to more NAND chips for parallel access; examples include read speeds of up to 13,500 MB/s on the M5 Max.

File Operations Comparison

• Copying large files shows substantial speed differences: M5 at 2.1 GB/s vs. M5 Max at 10.4 GB/s for sequential copies confirms earlier benchmarks on PCIe Gen 5's impact.

• For many small file operations, all three machines perform similarly due to internal SSD controller limitations rather than bus bandwidth constraints.

Thunderbolt Performance Analysis

External Drive Speed Tests

• When copying files to external drives via Thunderbolt connections, significant disparities emerge: M5 on Thunderbolt 4 achieves lower speeds compared to Pro and Max on Thunderbolt 5.

• The bottleneck shifts from bus bandwidth to IOPS when handling numerous small files; the M5 Max excels here with a write speed of 336 MB/s compared to others' much slower rates.

AI Workflow Management Tools

Introduction of Verden Tool

• Verden is introduced as an AI tool that streamlines coding workflows by managing multiple AI models simultaneously for better output consolidation and planning efficiency.

Benchmarking Developer Workflows

Browser Benchmarks

• Classic browser benchmarks like Speedometer show minimal differences between models due to their similar single-core performance capabilities; everyday tasks feel equally responsive across devices.

Multi-Core Performance Testing

• In multi-core tasks such as Python computations, the additional cores in Pro and Max lead to significantly faster processing times compared to the base M5 model which lags behind substantially in this area.

Xcode Build Times Comparison

Compilation Speed Results

• Xcode benchmark tests reveal that while the base M5 is nearly twice as slow as its counterparts during builds, it performs poorly against older models like the three-year-old M2 Max in specific scenarios like CocoaPods builds.

Large Project Builds

WebKit Release Build Test

• A full clean WebKit release build showcases dramatic time differences: M5 takes almost an hour while Pro and Max complete it within approximately 15 minutes—highlighting thermal constraints impacting sustained workloads on lower-tier models.

TypeScript Mono Repo Compilation

TypeScript Compilation Results

• In non-batched cold builds using a large mono repo test case, there’s only a slight difference between base M5 and higher-end models indicating that web development tasks may not require top-tier hardware unless working with very large projects or complex setups.

Local LLM Performance Metrics

Memory Bandwidth Impact

• Local LLM performance is heavily influenced by memory bandwidth; measured values show significant scaling advantages for higher-end models during token generation tasks confirming earlier theoretical predictions about memory specs.

Thermal Throttling Observations

Inference vs Compilation

• During inference tasks involving AI completions, all machines maintain consistent performance without throttling issues unlike compilation tasks where thermal limits affect lower-tier devices significantly.

Conclusion on Model Selection

Recommendations Based on Use Cases

• For general productivity including browsing or light coding work, base M5 suffices but serious developers should consider investing in either Pro or Max based on their specific workload needs especially if they handle larger compilations or advanced AI modeling tasks regularly.