Throughput · NVFP4
$80/hr $0.0222/sec · $58,400/mo at 24/7
Best tokens/$ for large MoE models. Large MoE throughput kings.
VRAM pool
180GB ×8 = 1.4TB
Position
vs A100 8×: ~3× throughput
Compute die
NVIDIA B200 180GB — NVFP4 tensor cores, tuned for large moe throughput kings.
HBM3e memory
1,440 GB of HBM3e across the 8 GPUs.
NVLink 5 interconnect
NVLink 5 across the 8 GPUs — one memory domain.
Per-second metering
$80/hr → 0.0222/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
1,440 GB of VRAM on this preset. Frontier MoE serving at full batch with KV-cache headroom (rule of thumb: about half a byte per parameter at NVFP4).
Typical loads
large MoE serving, high-QPS inference, batch evals at scale. Fine on: torch, JAX, raw CUDA, axolotl, vLLM, sglang, ComfyUI.
Billing math
$80/hr is $0.0222/sec while running and $0 while parked; park/wake freezes and restores state instead of paying for idle.
Best fit
1,440 GB across eight B200 180 GB cards on NVLink 5 at $80/hr, $10 per GPU-hour. The catalog's note calls this the best tokens/$ row for large MoE models; at FP8 weights a 1T-parameter checkpoint is ~1 TB and still leaves ~440 GB of the pool for KV cache (rule of thumb).
Trade-off
b300-8x-fp8 is $96/hr: the same eight-card shape with B300 288 GB cards, 2,304 GB total and the catalog's FP8 mode. That is $16/hr more and ~60% more VRAM, at $12/hr per GPU against $10/hr — precision and capacity for throughput.
$80/hr on runaii.cloud, metered per second: $0.0222/sec while the VM runs and $0 while parked. A month of 24/7 uptime would be about $58,400, 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
1,440 GB of VRAM. Frontier MoE serving at full batch with KV-cache headroom (rule of thumb: about half a byte per parameter at NVFP4).
1,440 GB — eight 180 GB B200 HBM3e cards on NVLink 5 at $80/hr, $10 per GPU-hour.
The catalog lists this row at NVFP4 and labels it the best tokens/$ row for large MoE models; the FP8 counterpart in the catalog is b300-8x-fp8 at $96/hr.
No — b200-8x-fp4 is the only 8-card B200 preset at $80/hr; the 4-card B200 row is $40/hr.
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 RTX PRO 6000 Blackwell 96GB
$19.2/hr · FP8
Workstation ×8
8× NVIDIA B300 288GB
$96/hr · FP8
Full Precision
4× NVIDIA B300 288GB
$48/hr · NVFP4
Minimal
4× NVIDIA B200 180GB
$40/hr · NVFP4
Efficient
16× NVIDIA GB300
$184/hr · FP8
Frontier Max