Does it fit
Northwind wants the claims model on their own two 80 GiB cards; the sheet says 70 GB of weights. Work out whether it fits with a KV cache at their concurrency, in the units the card uses, and set the admission limit that keeps it fitting.
- intermediate
- about 25 min
- 4 objectives
- par 8 steps
- Northwind Insurance
This mission is part of premium, with every challenge and incident in the lab. The brief below is the whole problem; the box, the hints and the debrief are what you unlock.
Every answer, concept and course, every paper breakdown, all published video lessons, hands-on FDE Lab missions, Premium PDF guides and companion files, the full practice-test bank and work-sample downloads. Referral Premium excludes guide PDFs and their companion files.
6 months · One payment · No auto-renewal
Study alongside free video lessons.
- Customer
- Northwind Insurance (synthetic customer; the model and hardware figures on this box are illustrative and made up for the exercise)
- Situation
- Northwind's platform team has two 80 GiB accelerators and a vendor sheet for a 35-billion-parameter model at 16-bit weights. They want 32 concurrent claims conversations at an 8k context. Their question is one sentence: does it fit, and if not, what do we cut.
- Your role
- You are the FDE sizing the deployment. You have the vendor sheet, the sizing tool on this box, and the serving config (max concurrency) that the tool checks against.
- Goal
- Convert the weights to the unit the card reports, compute the KV cache at the requested concurrency, decide whether it fits, and set an admission limit that does.
- 1.State the weights in GiB (you commit to an answer)
- 2.State the KV cache at the requested concurrency (you commit to an answer)
- 3.Say whether the requested deployment fits (you commit to an answer)
- 4.Set an admission limit that fits with headroom