← Dev Docs

HOMO-1 Avatar Model — GPU Cost Analysis

Realtime streaming & video generation with text or audio inputs

RealtimeVideo GenText InputAudio Input

GPU Comparison: A100 vs L4

SpecA100 (avatar-a100)L4 (avatar-l4)
VRAM40 GB (or 80 GB)24 GB
On-demand~$3.67/hr~$0.84/hr
Spot~$1.80/hr~$0.34/hr
FlashHead Lite FPS~96~70-90
Max concurrent sessions31-2

L4 is ~77% cheaper on spot but supports fewer concurrent sessions and slightly lower FPS.

Per-GPU Cost (1 Instance)

Provider/sec/min/hour
GCP On-demand$0.00102$0.0612$3.67
GCP Spot (current)$0.00050$0.0300$1.80
RunPod Community$0.00029$0.0173$1.04
RunPod Serverless$0.00032$0.0192$1.15
Modal$0.00030$0.0182$1.09

1 Session / GPU

No multi-session — cost per session equals full GPU cost.

Provider/sec/min/hour/month
GCP On-demand$0.00102$0.0612$3.67$2,679
GCP Spot$0.00050$0.0300$1.80$1,314
RunPod Community$0.00029$0.0173$1.04$759
RunPod Serverless$0.00032$0.0192$1.15pay-per-use
Modal$0.00030$0.0182$1.09pay-per-use

3 Sessions / 1 GPU (multi-session)

Same GPU cost split across 3 concurrent users — cheapest per-user option.

Provider/sec/session/min/session/hour/session
GCP On-demand$0.00034$0.0204$1.22
GCP Spot$0.00017$0.0100$0.60
RunPod Community$0.00010$0.0058$0.35

3 Sessions / 3 GPUs (horizontal scaling)

3x the cost — each user gets a dedicated GPU.

Provider/sec (total)/min (total)/hour (total)
GCP On-demand$0.00306$0.1836$11.01
GCP Spot$0.00150$0.0900$5.40
RunPod Community$0.00087$0.0519$3.12
RunPod Serverless$0.00096$0.0576$3.45
Modal$0.00091$0.0546$3.27

Takeaway

  • ✓ Cheapest per-user: 3 sessions on 1 GPU with RunPod Community at $0.35/hr/session
  • ✓ Best always-on dedicated: GCP Spot at $1.80/hr ($1,314/mo)
  • ✓ Best serverless (pay-per-use): RunPod Serverless or Modal at ~$1.10/hr, zero idle cost
  • ✓ Multi-session (3:1) cuts per-user cost by ~67% vs dedicated GPUs