BlueAlly Sizing PlatformSize on-prem AI infrastructure
Size on-prem AI infrastructure
with formulas, not guesses.
Every number on this site comes from a deterministic formula sourced from NVIDIA, HuggingFace, and Baseten. Claude is in the room as your advisor — it explains, recommends, and warns, but it never calculates. So you can trust the GPU count.
Deterministic
Every VRAM, GPU, storage, and TCO number is computed by typed TypeScript with cited formulas — never by an LLM.
Claude as advisor
Streaming Claude advisor explains the math, flags risks, and recommends optimizations — but cannot return a sizing value.
Excel that calculates
Export to Excel where every cell holds a real formula. Edit a parameter; the workbook recomputes itself.
Sample sizing
Llama 3.1 70B · INT8 · RAG · 50 concurrent users · interactive SLA · N+1 HA
Total VRAM
~140 GB
GPUs (H100)
8
System RAM
512 GB
Storage
~280 GB
Power
~7 kW
3-yr TCO
~$610K
Numbers regenerated live from the calc engine. The wizard will show you the exact values for your configuration.