Efficient ×2 · FP8
$1.8/hr $0.0005/sec · $1,314/mo at 24/7
g2-standard-24 — two-worker inference.
VRAM pool
2× NVIDIA L4 24GB
Compute die
NVIDIA L4 24GB — FP8 tensor cores.
GDDR6 memory
48 GB of GDDR6 across the 2 GPUs.
PCIe Gen4 interconnect
PCIe Gen4 host links — 2 GPUs, each on its own link.
Per-second metering
$1.8/hr → 0.0005/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
48 GB of VRAM on this preset. 13B-class dense with KV headroom; mid-size LoRA fine-tunes run comfortably.
Typical loads
production chat for 13-30B models, feature prototyping. Fine on: torch, JAX, raw CUDA, axolotl, vLLM, sglang, ComfyUI.
Billing math
$1.8/hr is $0.0005/sec while running and $0 while parked; park/wake freezes and restores state instead of paying for idle.
Best fit
48 GB across 2 L4 cards, each a separate 24 GB device on its own PCIe Gen4 link, $1.80/hr — $0.90 per GPU-hour. The catalog's note for this row is two-worker inference, and the shape matches it: two ~13B checkpoints at about one byte per parameter, or one model split across both cards when a single 24 GB budget is the limit (rule of thumb).
Trade-off
l4-4x is $3.60/hr: 4 cards, 96 GB total, and the catalog calls it a multi-worker farm on one VM. The per-GPU price is flat at $0.90/hr, so the step up buys two more 24 GB budgets and two more PCIe Gen4 links, not a cheaper GPU-hour. No L4 size has NVLink between its cards.
$1.8/hr on runaii.cloud, metered per second: $0.0005/sec while the VM runs and $0 while parked. A month of 24/7 uptime would be about $1,314, 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
48 GB of VRAM. 13B-class dense with KV headroom; mid-size LoRA fine-tunes run comfortably.
No — two separate 24 GB L4 cards, each on its own PCIe Gen4 link, $1.80/hr in total. A model larger than 24 GB has to be split in software across both cards.
$0.90/hr per GPU ($1.80/hr for two), against $0.87/hr on the single-card l4-1x — the 2× row is $0.03/GPU/hr more, not cheaper.
When one 24 GB card or one worker is the bottleneck: two replicas of a ~13B FP8 model, or a job that needs 48 GB split across two PCIe Gen4 cards.
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
4× NVIDIA L4 24GB
$3.6/hr · FP8
Efficient ×4
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