Efficient ×4 · FP8
$3.6/hr $0.0010/sec · $2,628/mo at 24/7
g2-standard-48 — multi-worker farm on one VM.
VRAM pool
4× NVIDIA L4 24GB
Compute die
NVIDIA L4 24GB — FP8 tensor cores.
GDDR6 memory
96 GB of GDDR6 across the 4 GPUs.
PCIe Gen4 interconnect
PCIe Gen4 host links — 4 GPUs, each on its own link.
Per-second metering
$3.6/hr → 0.0010/sec. Pause the deployment and the meter hits $0.
Scale-to-zero
Autoscale + region pins (GLOBAL / US / EU / APAC) on the control plane.
Memory budget
96 GB of VRAM on this preset. A 70B-class checkpoint fits (rule of thumb: about one byte per parameter at FP8), or several 13B workers side by side.
Typical loads
single-model 70B serving, team-shared dev capacity. Fine on: torch, JAX, raw CUDA, axolotl, vLLM, sglang, ComfyUI.
Billing math
$3.6/hr is $0.0010/sec while running and $0 while parked; park/wake freezes and restores state instead of paying for idle.
Best fit
96 GB across 4 L4 cards at $3.60/hr, $0.90 per GPU-hour. That is four separate 24 GB FP8 budgets on four PCIe Gen4 links — the catalog's multi-worker farm on one VM: four ~13B checkpoints side by side, or one 70B checkpoint split four ways with about 26 GB left for KV cache (rule of thumb).
Trade-off
l4-8x is $7.20/hr: four more 24 GB cards and 192 GB in total, still $0.90/hr per GPU. The step up adds workers and host links, not per-card VRAM — a checkpoint that needs more than 24 GB on one device never fits on this ladder, whatever the card count.
$3.6/hr on runaii.cloud, metered per second: $0.0010/sec while the VM runs and $0 while parked. A month of 24/7 uptime would be about $2,628, and park/wake means busy-fraction pricing in practice.
Yes. Per-second metering with park/wake: idle freezes the VM state into a signed artifact and stops the meter the same second; the next request wakes it byte-exact
96 GB of VRAM. A 70B-class checkpoint fits (rule of thumb: about one byte per parameter at FP8), or several 13B workers side by side.
Four independent 24 GB FP8 budgets, $3.60/hr in total ($0.90/GPU/hr), each card on its own PCIe Gen4 link.
No — $0.90/hr per GPU against $0.87/hr on the single card. The 4× row is $0.03/GPU/hr more expensive, so you pay for four 24 GB budgets, not a discount.
At FP8 and one byte per parameter, a 70B checkpoint is ~70 GB of weights and leaves about 26 GB of the 96 GB pool for KV cache; four ~13B checkpoints also fit side by side (rule of thumb).
RTX PRO 6000 slices
from $0.30/hr · 12-48GB
Blackwell slices for light jobs
1× NVIDIA T4 16GB
$0.44/hr · FP16
Starter
1× NVIDIA L4 24GB
$0.87/hr · FP8
Efficient
2× NVIDIA L4 24GB
$1.8/hr · FP8
Efficient ×2
8× NVIDIA L4 24GB
$7.2/hr · FP8
Efficient ×8
1× NVIDIA RTX PRO 6000 Blackwell 96GB
$2.4/hr · FP8
Workstation
2× NVIDIA RTX PRO 6000 Blackwell 96GB
$4.8/hr · FP8
Workstation ×2
4× NVIDIA RTX PRO 6000 Blackwell 96GB
$9.6/hr · FP8
Workstation ×4
8× NVIDIA RTX PRO 6000 Blackwell 96GB
$19.2/hr · FP8
Workstation ×8
8× NVIDIA B300 288GB
$96/hr · FP8
Full Precision
8× NVIDIA B200 180GB
$80/hr · NVFP4
Throughput
4× NVIDIA B300 288GB
$48/hr · NVFP4
Minimal
4× NVIDIA B200 180GB
$40/hr · NVFP4
Efficient
16× NVIDIA GB300
$184/hr · FP8
Frontier Max