> BROMANDER_LABS
HOW MANYUSERS?
The Scanner tells you single-stream speed. Production is concurrent — and what runs out first is almost never the weights. It's KV cache, one per conversation.
All math runs in your browser.
01 — The rig
GPU
How many
02 — Workload
Capacity report
Concurrent users
4
aggregate
143
tok/s · 35.8 each
Limited by KV cache capacity. The pool holds 4 conversations at 8,192 tokens each.
VRAM budget
Weights · 46%Framework · 10%KV pool · 43%
Usable
11.0 GB
KV pool
4.6 GB
KV / user
1.07 GB
Throughput vs concurrency
aggregate (left)per user (right)at 4 users: 143 total · 35.8 each
$ / 1M tokens
$1.340
Tokens / day
12.4M
Rig bandwidth
504 GB/s
Concurrency here is exact — it's pure memory arithmetic. Throughput models the memory roofline only; a real server also hits a compute roofline at high batch, so treat aggregate tok/s as an optimistic ceiling you approach but don't exceed.
─── Your Report Card ───
Live preview
Built by
Bromander Studios — Hire Us