What a Bank Would Ask Before Installing the vCommunity Management Pack
A community Management Pack for VCF Operations that is better engineered than most vendor material, and the four things worth checking before it goes near a regulated estate.
Archive
Latest posts — page 2
Same Stack Underneath, Different Deal About Who Does the Work
Licensing now costs the same wherever you run it, which takes one variable out of every comparison. What separates the four hyperscaler VMware services is who operates the stack, not what is underneath it.
Sovereign Cloud for Regulated Industries: Four Requirements, Not One
Data residency, operational, jurisdictional and technical sovereignty are separate questions. Most estates satisfy two and assume they have satisfied all four.
The Business Case: Building a TCO Model That Survives the Questions
Three numbers decide every infrastructure business case, and most models get two of them badly wrong. Here is how to build one a sceptical CFO cannot dismantle.
What You Already Paid For and Haven't Switched On
A customer asked what database monitoring would cost. Ten minutes in it was clear they already had it. Entitled, not deployed, nobody had looked.
How Much of Your Platform Are You Actually Using?
Fifteen questions, about four minutes. Scores your estate across five areas and tells you what I'd do next in each one. Nothing is sent anywhere.
RAG Is a Distributed Systems Problem
Five stages in the path, and only the last two involve the model. Retrieval is memory bound, generation is GPU bound, and they do not belong on the same hosts.
Which Inference Engine, and Does It Matter?
They implement the same ideas. The difference is what each optimises for and what it costs you to run, which is a less exciting answer than people want.
What Does a Token Actually Cost You?
The arithmetic is easy. The denominator is where models fall apart, and the biggest lever has nothing to do with hardware.
Smaller Numbers, Faster Tokens
Quantisation is usually framed as a quality tradeoff. The more useful framing is bandwidth, because decode re-reads every weight for every token. Tags: architecture, operations
A Quarter of the Card Is Not a Quarter of the Capacity
Passthrough, time-sliced vGPU and MIG do different things to the thing that actually limits you. And fractional slices don't divide capacity the way the spreadsheet says.
Your GPU Is Not Slow, It Is Waiting for Memory
Inference behaves like two workloads with opposite requirements, sharing the same silicon. Once you see that, the strange utilisation numbers start making sense.