Efficient ×8 · FP8
$7.2/hr $0.0020/sec · $5,256/mo at 24/7
g2-standard-96 — full-node L4.
VRAM pool
8× NVIDIA L4 24GB
Compute die
NVIDIA L4 24GB — FP8 tensor cores.
GDDR6 memory
192 GB of GDDR6 across the 8 GPUs.
PCIe Gen4 interconnect
PCIe Gen4 host links — 8 GPUs, each on its own link.
Per-second metering
$7.2/hr → 0.0020/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
192 GB of VRAM on this preset. 13-30B dense at FP8 plus LoRA fine-tunes and several concurrent workers.
Typical loads
production chat farms, nightly fine-tunes, concurrent batch jobs. Fine on: torch, JAX, raw CUDA, axolotl, vLLM, sglang, ComfyUI.
Billing math
$7.2/hr is $0.0020/sec while running and $0 while parked; park/wake freezes and restores state instead of paying for idle.
Best fit
192 GB across eight L4 cards — the catalog's full-node L4 row at $7.20/hr, $0.90 per GPU-hour. Eight separate 24 GB FP8 budgets on eight PCIe Gen4 links: eight ~13B checkpoints at once, or one 70B checkpoint split eight ways with ~122 GB left for KV cache (rule of thumb).
Trade-off
The next rung is a different shape, not a bigger bill: g4-1x is one RTX PRO 6000 Blackwell 96GB on PCIe Gen5 at $2.40/hr — less than this preset — but a single card, so a job that needs 96 GB on one device moves there. g4-8x matches the eight-card shape with 8× 96 GB GDDR7 at $19.20/hr.
$7.2/hr on runaii.cloud, metered per second: $0.0020/sec while the VM runs and $0 while parked. A month of 24/7 uptime would be about $5,256, 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
192 GB of VRAM. 13-30B dense at FP8 plus LoRA fine-tunes and several concurrent workers.
192 GB — eight 24 GB L4 cards at $7.20/hr. Each card is a separate device, so a single job sees 192 GB only if it is split in software.
No — $0.90/hr per GPU against $0.87/hr on the single card. The eight-card row costs $0.03/GPU/hr more.
g4-8x: eight RTX PRO 6000 Blackwell cards with 96 GB of GDDR7 each, 768 GB in total, at $19.20/hr — 4× the VRAM per card at 2.67× the hourly.
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
4× NVIDIA L4 24GB
$3.6/hr · FP8
Efficient ×4
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