Starter · FP16
$0.44/hr $0.0001/sec · $321.2/mo at 24/7
Cheapest CUDA — experiments, small models.
VRAM pool
1× NVIDIA T4 16GB
Compute die
NVIDIA T4 16GB — FP16 tensor cores.
GDDR6 memory
16 GB of GDDR6 across the GPU.
PCIe Gen3 interconnect
PCIe Gen3 host links.
Per-second metering
$0.44/hr → 0.0001/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
16 GB of VRAM on this preset. 8B-class models at FP16, LoRA adapters, embeddings, and quick experiments.
Typical loads
student experiments, embeddings, POC builds, small classifiers. Fine on: torch, JAX, raw CUDA, axolotl, vLLM, sglang, ComfyUI.
Billing math
$0.44/hr is $0.0001/sec while running and $0 while parked; park/wake freezes and restores state instead of paying for idle.
Best fit
16 GB of GDDR6 on one PCIe Gen3 card at the catalog's FP16 precision, $0.44/hr. At FP16 weights cost about two bytes per parameter, so 16 GB holds roughly an 8B checkpoint before KV cache, or a 3B model with room to batch (rule of thumb). The single card is one device — there is no multi-GPU fabric to pool — and it is the only row the catalog lists at FP16.
Trade-off
The next rung is l4-1x at $0.87/hr, just under double, for 24 GB instead of 16 GB and the catalog's FP8 mode instead of FP16. At FP8 a 13B checkpoint is about 13 GB of weights, which the L4 holds and the T4 cannot at FP16; the T4 stays cheaper only while your model fits in 16 GB at FP16 and you never need FP8.
$0.44/hr on runaii.cloud, metered per second: $0.0001/sec while the VM runs and $0 while parked. A month of 24/7 uptime would be about $321, 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
16 GB of VRAM. 8B-class models at FP16, LoRA adapters, embeddings, and quick experiments.
No. The catalog lists t4-1x at FP16 — the only preset that is not FP8 or NVFP4. At two bytes per parameter an 8B checkpoint is ~16 GB of weights, the whole card before KV cache (rule of thumb).
$0.44/hr for its single card — the lowest hourly in the catalog, and identical to its per-GPU rate because there is one GPU.
No — one T4 on one PCIe Gen3 link is one 16 GB device. Two workers start at l4-2x: 2× L4 24GB, $1.80/hr.
RTX PRO 6000 slices
from $0.30/hr · 12-48GB
Blackwell slices for light jobs
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
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