The MacBook Pro That Questions Every Dollar You Spent

I've watched product launches become detached from reality since the mid-1990s. Back then, vendors promised megahertz would save us all. Today it's cores and bandwidth. The numbers get bigger, the promises stay familiar, and someone still has to test whether any of it matters.

Alex Ziskind did that testing with Apple's new M5 MacBook lineup, and his findings present an uncomfortable question for anyone who bought the expensive model: What exactly did you pay for?

The Spec Sheet Doesn't Tell the Story

On paper, the progression looks sensible. The base M5 has 10 CPU cores and one fan. The M5 Pro doubles to 18 cores with two fans. The M5 Max matches those cores but quadruples the GPU cores to 40 and doubles memory bandwidth to a theoretical 614 GB per second.

Apple's website lists these numbers cleanly. What it doesn't show is how rarely those differences surface in actual work.

Ziskind ran the standard browser benchmarks first—Speedometer and JetStream, the tests that measure everyday web responsiveness. The M5 scored 61.4. The M5 Max scored 61.7. "Everyday browser responsiveness, web app snappiness, you're not going to feel a big difference between these machines," he noted.

That pattern held across multiple real-world scenarios. A TypeScript monorepo with 253 packages and over 5,000 files? The base M5 took 48 seconds for a cold build. The M5 Max took 42 seconds. In batch mode, all three machines were essentially identical at 13 seconds.

"If you're a web dev working with TypeScript and Node, the base M5 is really fine," Ziskind said. "And that's an uncomfortable thought when you've spent three times as much on the Max."

Where the Money Goes (and Doesn't)

The testing reveals a clear dividing line. For single-threaded work or anything that doesn't sustain maximum load, the base M5 keeps pace. The moment you ask all cores to work hard for more than a few minutes, the architecture differences emerge.

A Python Mandelbrot computation using all available cores showed the split clearly. The M5 took 29.4 seconds. The M5 Pro finished in 12.2 seconds. The M5 Max? 11.7 seconds—half a second faster than the Pro.

That half-second gap repeated across sustained workloads. An Xcode benchmark with 70+ CocoaPods: M5 Pro took 89 seconds, M5 Max took 85 seconds. A full WebKit build from source: M5 Pro finished in 16.1 minutes, M5 Max in 15.3 minutes.

The base M5 fell dramatically behind in these tests, but not because its cores are slower. Thermal constraints explain the gap. With only one fan and no high power mode option, the M5 throttles under sustained load. Ziskind ran the Xcode benchmark after a long build when the machine was already hot. Performance dropped 60% compared to running it cold. The M5 Pro lost 37% under the same conditions. The M5 Max barely moved.

"The M5 doesn't just have fewer cores," Ziskind observed. "It can't sustain what it has."

The WebKit build showed the most dramatic performance gap in the entire test suite. The base M5 took 56.5 minutes—almost an hour versus 15 minutes for the Pro and Max. That's 3.7 times slower. For developers who regularly compile large projects, the base M5 simply isn't in the same category.

The AI Exception

For months, the M5 Pro and M5 Max looked nearly identical in Ziskind's testing. Then he moved to local AI model inference, and the story changed completely.

Local language model performance depends on two factors: memory bandwidth and GPU compute. Ziskind measured both directly before running any models. The M5 achieved 115 GB/s in practical memory bandwidth. The M5 Pro hit 251 GB/s. The M5 Max reached 350 GB/s. Apple's spec sheets promise higher theoretical numbers, but these measured ratios—roughly 1x, 2.1x, and 3x—proved predictive.

GPU compute showed even clearer scaling. Using the MLX framework optimized for Apple Silicon, the M5 delivered 8.9 teraflops, the M5 Pro 19.7 teraflops, and the M5 Max 42.4 teraflops. That's nearly perfect scaling with the 10, 20, and 40 GPU core counts.

Running an 8.2 GB language model confirmed the predictions. The M5 generated 25 tokens per second. The M5 Pro hit 51. The M5 Max reached 87.

"This is what bandwidth buys you," Ziskind said. "After spending all this time watching the Pro keeping up, this is the first test where I don't feel completely stupid for buying the Max."

But the real differentiation came with larger models. A 15 GB model at 4-bit quantization ran on the M5 Pro at 17.6 tokens per second and on the M5 Max at 25.1 tokens per second. On the base M5 with 16 GB of total RAM? It crashed. Not slow—broken.

A 65 GB mixture-of-experts model couldn't even load on the M5 Pro, which maxes out at 64 GB of RAM. The M5 Max with 128 GB ran it at 23 tokens per second. "This is the why you buy the Max demo," Ziskind noted. "Not because it's a little faster, because entire categories of models simply don't exist on any other machine."

The Storage Speed Asterisk

SSD performance added another layer of complexity. Sequential read speeds showed dramatic differences thanks to PCIe Gen 5 on the Pro and Max models. The M5 Pro and Max achieved around 13,000-13,500 MB/s for reads. The base M5, still on PCIe Gen 4, managed about half that.

But copy the entire WebKit source tree—half a million tiny files, 43 GB total—and the gap shrinks to 18% between the M5 and M5 Max. For internal operations involving many small files, all three machines handle metadata at roughly the same speed.

External storage tells yet another story. The M5 on Thunderbolt 4 maxed out around 1.4 GB/s writes and 3.9 GB/s reads to an external SSD. The M5 Pro and Max on Thunderbolt 5 hit 4-9.6 GB/s. But when copying those same half-million small files, the M5 and M5 Pro were essentially identical at 67-69 MB/s. The bottleneck wasn't bus bandwidth—it was IOPS and file system metadata.

Only the M5 Max, with its different storage controller, broke through at 336 MB/s—five times faster than the others. "When somebody says my SSD benchmark is at 7 gigs a second, ask them what they're copying," Ziskind suggested.

What the Testing Actually Shows

The pattern is clear enough. For most software development work, the M5 Pro and M5 Max perform nearly identically. The Pro costs significantly less. For web development with TypeScript, even the base M5 proves adequate despite costing a third as much as the Max.

The base M5 falls behind dramatically only under sustained multi-core loads—large compilations, video encoding, anything that pushes all cores hard for minutes at a time. Thermal constraints, not core speed, create that gap.

The M5 Max justifies its premium in exactly one scenario: running large AI models locally. Not because it's faster at everything, but because it's the only machine that can hold certain models in memory at all.

"The M5 Pro is the real developer sweet spot," Ziskind concluded. "Professional iOS and macOS developers, full-stack developers with Docker-heavy workflows, and serious AI enthusiasts who want to run 70 billion parameter models."

That assessment rings true. The question is whether buyers will believe it. Apple has spent decades training customers to associate higher numbers with better experiences. Sometimes that's accurate. Sometimes you're paying for capability you'll never access.

The uncomfortable part isn't that the base M5 struggles with certain tasks. It's that the M5 Pro handles nearly everything the Max does while costing substantially less. The Max's advantages are real but narrow. Knowing which side of that line your work falls on determines whether you made a smart purchase or an expensive mistake.

—Bob Reynolds is Senior Technology Correspondent at Buzzrag