Workstation ×8 · FP8
$19.2/hr $0.0053/sec · $14,016/mo at 24/7
g4-standard-384 — full-node Blackwell.
VRAM pool
8× NVIDIA RTX PRO 6000 Blackwell 96GB
Compute die
NVIDIA RTX PRO 6000 Blackwell 96GB — FP8 tensor cores.
GDDR7 memory
768 GB of GDDR7 across the 8 GPUs.
PCIe Gen5 interconnect
PCIe Gen5 host links — 8 GPUs, each on its own link.
Per-second metering
$19.2/hr → 0.0053/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
768 GB of VRAM on this preset. 70B+ dense checkpoints fit with room for KV cache (rule of thumb: about one byte per parameter at FP8).
Typical loads
multi-tenant inference, parallel eval harnesses, 70B+ dense serving. Fine on: torch, JAX, raw CUDA, axolotl, vLLM, sglang, ComfyUI.
Billing math
$19.2/hr is $0.0053/sec while running and $0 while parked; park/wake freezes and restores state instead of paying for idle.
Best fit
768 GB across eight RTX PRO 6000 Blackwell cards at $19.20/hr, $2.40 per GPU-hour — the catalog's full-node RTX PRO 6000 row. Eight separate 96 GB GDDR7 budgets on PCIe Gen5, or one ~700B FP8 checkpoint split eight ways with ~68 GB left for KV cache (rule of thumb).
Trade-off
The next rung is b200-4x-fp4 at $40/hr — about double — and it is a different architecture, not more of the same: four B200 180 GB cards, 720 GB of HBM3e on NVLink 5 at NVFP4. That is slightly less total VRAM than 768 GB but a shared GPU fabric and the catalog's 4-bit mode; for independent workers, g4-8x gives eight cards and more VRAM for half the hourly.
$19.2/hr on runaii.cloud, metered per second: $0.0053/sec while the VM runs and $0 while parked. A month of 24/7 uptime would be about $14,016, 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
768 GB of VRAM. 70B+ dense checkpoints fit with room for KV cache (rule of thumb: about one byte per parameter at FP8).
768 GB — eight 96 GB GDDR7 cards at $19.20/hr, $2.40 per GPU-hour.
For HBM3e and NVLink 5: b200-4x-fp4 is four 180 GB B200 cards on a shared fabric at $40/hr, against eight independent 96 GB GDDR7 cards at $19.20/hr.
A ~700B FP8 checkpoint split eight ways with ~68 GB left for KV cache, or eight independent 70B checkpoints — one per card (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
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 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