Generating a token means reading the model's active weights once, so a CPU server's speed ceiling is its memory bandwidth divided by the model's size. The arithmetic, the catch in prompt processing, why two sockets are not twice as fast, and the workloads a big-memory EPYC box genuinely suits.
A handful of caching edges in front of one origin covers most of what a commercial CDN does for a European audience, at a fraction of the per-terabyte price. The architecture, the nginx configuration that does most of the work, how users reach the nearest edge, and where the do-it-yourself version loses.
Most VPS plan pages lead with a price and a vCPU count and leave out the numbers that decide how the machine behaves. The nine to find before you pay, what a good answer looks like, the red flags, and a worked example of matching a workload to a plan.
One rented machine, a free hypervisor, and as many virtual machines as the hardware will hold. How to budget RAM, lay out ZFS, get networking right on a data-centre switch port — the part that differs from a homelab — and keep the backups off the box.
A VPS plan lists vCPUs the way a server lists cores, but a vCPU is a scheduling promise, not a piece of silicon. What dedicated and shared vCPUs actually guarantee, the one number that exposes contention, and a ten-minute test you can run on any plan.
What you can rent is not this year's Xeon against this year's EPYC but an eight-year spread of generations at prices seven times apart. Per-core speed, memory bandwidth, PCIe lanes and price per core — and which of those differences your workload can actually feel.
Most discrepancies on a new machine are a mis-seated DIMM, a drive pulled from another server, or a NIC that negotiated the wrong speed — all trivial to fix on day one and expensive in month four. An hour of acceptance testing, with the pass conditions.
Both protocols get video from an encoder into your origin; neither is what viewers receive. Judged on that one job, they differ sharply under packet loss — and the SRT latency setting decides whether it helps at all. A comparison, with the pipeline around it.
Capacity is the easy half of a backup. What decides whether it is worth having is how long the restore takes, whether anyone verified the archive, and whether an attacker holding your production credentials can also delete the backups.
Kernel settings are changeable; the distance between your server and your users is physics. Work out the latency floor from the map, understand why interconnection decides how close you get to it, and test from where the users actually are before committing.
Copying data is solved; sequencing is what fails. Lower the TTL two days out, sync in two passes so the window moves only a delta, replicate rather than copy databases, and rehearse the whole cutover once a week early. Plus the six things that always break.
No one can quote a player count without knowing the title, but the shape of the limit is knowable. The tick deadline binds first, memory decides instances per box, and bandwidth binds the fleet rather than the instance. Each has arithmetic you can run on your own numbers.