Blog post #51

---

This wasn’t a shopping trip. We have a couple of workstations at home, and I spent an evening with Claude tuning the one I work at — no new purchases, just getting what’s already on the desk to actually perform. A few principles kept repeating, and they’re worth writing down.

---

The cable is the bottleneck

I checked my main monitor’s refresh rate and Windows said 59.95 Hz. On a 165 Hz panel. I’d been working on a 60 Hz desktop without realizing it.

First reflex: bad driver. Wrong. The culprit was an old HDMI cable — the 1.4-era kind that caps 1440p around 75 Hz. The monitor was never the limit, and neither was the GPU. It was a cable from a drawer. The fix is a DisplayPort cable, but the lesson is broader: before blaming the expensive component, check the cheap one in the middle. I’d have happily believed it was “just a 60 Hz panel” if I hadn’t looked.

Sharpness over Hz for text

For staring at code all day, resolution beats refresh rate. The screens I use are around 109 ppi; a 27” 4K panel is ~163 ppi — that’s the difference you actually feel reading text, far more than extra hertz. If I optimize further, it’s pixels, not frames.

Measure before you buy

The cleanest answer of the night wasn’t “buy RAM.” It was: find out if you’re short first. The question isn’t whether 32GB beats 16GB in the abstract — it’s whether my workload hits the ceiling. So the test is concrete: open Task Manager → Performance → Memory under a genuinely heavy session and watch whether memory spills to disk. If it thrashes, more RAM is a real upgrade. If you sit at 70%, it does nothing and the money goes elsewhere. At today’s inflated DDR4 prices, that’s not a hunch you pay for.

Knowing when the cutoff matters

This is the one I want to remember. When I asked what screen I’d buy today, Claude named specific models — then I pushed: did you look that up, or is that training data? It was training data, and we searched. The recommendation changed.

That’s the useful instinct: a model’s knowledge cutoff matters intensely for some questions and not at all for others. Current product models, today’s prices, what’s in stock — all stale the moment the cutoff passes, so verify them. But reading the specs off my own machine, the physics of why an old cable caps refresh rate, how to read Task Manager — none of that expires. Same evening, same assistant, and the right move flipped from “trust it” to “go check” depending purely on whether the answer lived in fast-moving reality or in things that don’t change. Knowing which kind of question you’re asking is half the skill of working with these tools.

Reflections

I sat down to optimize a desk and came away with three settings fixed for free and a shopping list ordered by real impact instead of vibes. Every decision came down to the same move: don’t guess, look. Look at the refresh dropdown. Look at Task Manager under load. Look up whether the model you’re about to recommend still exists — and notice when you don’t need to.

— Stefan