Infrastructure

GPU Compute

BRICQS provisions dedicated GPU instances on Azure for every model deployment. Compute is allocated per-deployment, billed by the second, and automatically scaled to zero when idle.

Available GPU tiers

GPU T4NVIDIA T4

16 GB VRAM · 4 vCPU · 28 GB RAM

Good for: LLaMA 3 8B, Mistral 7B, Phi-3 Mini

$0.526/hr

GPU A10NVIDIA A10

24 GB VRAM · 6 vCPU · 55 GB RAM

Good for: LLaMA 3 70B, Mixtral 8x7B, CodeLlama 34B

$1.10/hr

GPU A100NVIDIA A100

80 GB VRAM · 12 vCPU · 220 GB RAM

Good for: LLaMA 3 405B, Llama 2 70B (quantized 4-bit)

$3.20/hr

Note: CPU-only deployments (no GPU) are available for lightweight inference or non-GPU workloads at standard BRICQS compute pricing.

Auto-scaling

Every deployment configures a replica range. BRICQS scales within that range based on concurrent request load:

bashbricqs deploy my-api \
  --model meta-llama/Llama-3-8B-Instruct \
  --min-replicas 0 \
  --max-replicas 3

min_replicas: 0Scale to zero when idle. First request after idle incurs a cold-start (~15–45s for GPU model pull).min_replicas: 1Always-on. No cold start. Billed continuously even when idle.max_replicasHard cap on parallel instances. Requests beyond capacity are queued or rejected.

Compute regions

West US 2 • Live

All GPU tiers available · T4 / A10 / A100

East US — Coming soon

Phase 4 multi-region — planned

GPU hours and billing

Compute is billed from the moment a replica starts to when it stops. Scale-to-zero deployments accumulate zero cost when idle. Your 30-day GPU hour total and projected spend are visible in the Dashboard and in the Deployments summary strip.

bash# Check GPU usage for your account
curl https://api.bricqsai.com/v1/usage/summary \
  -H "Authorization: Bearer <api-key>"

Set a monthly GPU budget cap per project to prevent unexpected overage:

bricqs projects budget set my-project --gpu-hours 100 --action warn

--action warn surfaces a Deployment Intelligence alert at 80% of the cap. Use --action block to hard-stop new deployments once the cap is reached.