Pricing & Instances
Live GPU, CPU pod, and storage rates pulled from the Meshive backend at every docs build — and how instances (pods) are shaped, sized, and limited.
Meshive bills pod runtime in fixed one-minute intervals (60 seconds) — no minimums, no monthly commitments. Rates are reviewed and re-adjusted every two weeks. Rate tables on this page are pulled from the live Meshive catalog when the docs are built; for the full, current catalog (every supported GPU SKU, spot prices, regional variations, and storage SKUs), sign in at console.meshive.ai — the console is always the live source of truth.
GPU pods
| GPU | VRAM | Demand (per GPU · hr) | Spot (per GPU · hr) |
|---|---|---|---|
| A100 PCIe 40GB | 40 GB | $0.8489 | $0.5093 |
| A100 PCIe 80GB | 80 GB | $0.9202 | $0.5521 |
| A100 SXM4 40GB | 40 GB | $0.9759 | $0.5856 |
| A100 SXM4 80GB | 80 GB | $0.9782 | $0.5869 |
| A40 | 48 GB | $0.4687 | $0.2812 |
| B200 | 180 GB | $6.9672 | $4.1803 |
| H100 SXM5 | 80 GB | $3.1105 | $1.8663 |
| H200 | 141 GB | $4.0109 | $2.4065 |
| L4 | 24 GB | $0.3545 | $0.2127 |
| L40S | 48 GB | $0.8496 | $0.5098 |
| RTX 3060 | 12 GB | $0.0652 | $0.0391 |
| RTX 3060 Ti | 8 GB | $0.0753 | $0.0452 |
| RTX 3070 | 8 GB | $0.0889 | $0.0533 |
| RTX 3070 Ti | 8 GB | $0.1009 | $0.0605 |
| RTX 3080 | 10 GB | $0.1203 | $0.0722 |
| RTX 3080 | 12 GB | $0.1203 | $0.0722 |
| RTX 3080 Ti | 12 GB | $0.1288 | $0.0773 |
| RTX 3090 | 24 GB | $0.1714 | $0.1028 |
| RTX 3090 Ti | 24 GB | $0.2320 | $0.1392 |
| RTX 4060 | 8 GB | $0.0659 | $0.0396 |
| RTX 4060 Ti | 16 GB | $0.1154 | $0.0693 |
| RTX 4070 | 12 GB | $0.0962 | $0.0577 |
| RTX 4070 SUPER | 12 GB | $0.0895 | $0.0537 |
| RTX 4070 Ti | 12 GB | $0.1028 | $0.0617 |
| RTX 4070 Ti SUPER | 16 GB | $0.1283 | $0.0770 |
| RTX 4080 | 16 GB | $0.1849 | $0.1109 |
| RTX 4080 SUPER | 16 GB | $0.1895 | $0.1137 |
| RTX 4090 | 24 GB | $0.3839 | $0.2304 |
| RTX 5060 | 8 GB | $0.0816 | $0.0490 |
| RTX 5060 Ti | 16 GB | $0.1098 | $0.0659 |
| RTX 5070 | 12 GB | $0.1223 | $0.0734 |
| RTX 5070 Ti | 16 GB | $0.1712 | $0.1027 |
| RTX 5080 | 16 GB | $0.1995 | $0.1197 |
| RTX 5090 | 32 GB | $0.4299 | $0.2579 |
| RTX 6000 Ada Generation | 48 GB | $0.6823 | $0.4094 |
| RTX A4000 | 16 GB | $0.1121 | $0.0673 |
| RTX A5000 | 24 GB | $0.2466 | $0.1479 |
| RTX A6000 | 48 GB | $0.4263 | $0.2558 |
| RTX PRO 4000 Blackwell | 24 GB | $0.2838 | $0.1703 |
| RTX PRO 4500 | 32 GB | $0.3000 | $0.1800 |
| RTX PRO 4500 Blackwell | 32 GB | $0.3000 | $0.1500 |
| RTX PRO 5000 Blackwell | 48 GB | $0.8716 | $0.5230 |
| RTX PRO 5000 Blackwell | 72 GB | $0.8716 | $0.5230 |
| RTX PRO 5000 Blackwell / RTX PRO 5000 72GB Blackwell | 48 GB | $0.8716 | $0.5230 |
| RTX PRO 5000 Blackwell / RTX PRO 5000 72GB Blackwell | 72 GB | $0.8716 | $0.5230 |
| RTX PRO 6000 Blackwell Workstation Edition | 96 GB | $1.1310 | $0.6786 |
CPU pods — per-vCPU price by CPU generation
| Vendor | Product line | Generation | Demand (per vCPU · hr) | Spot (per vCPU · hr) |
|---|---|---|---|---|
| amd | epyc | Zen 3 | $0.0020 | $0.0012 |
| amd | epyc | Zen 4 | $0.0030 | $0.0018 |
| amd | threadripper | Zen 4 | $0.0020 | $0.0012 |
| amd | threadripper pro | Zen 4 | $0.0020 | $0.0012 |
| amd | ryzen | Zen 4 | $0.0035 | $0.0021 |
| amd | threadripper pro | Zen 5 | $0.0040 | $0.0024 |
| amd | ryzen | Zen 5 | $0.0040 | $0.0024 |
| amd | threadripper | Zen 5 | $0.0040 | $0.0024 |
| intel | xeon | 4th Gen | $0.0020 | $0.0012 |
RAM — per-GB price by DDR generation
| DDR | Demand (per GB · hr) | Spot (per GB · hr) |
|---|---|---|
| DDR0 | $0.0030 | $0.0015 |
| DDR3 | $0.0030 | $0.0015 |
| DDR4 | $0.0050 | $0.0025 |
| DDR5 | $0.0080 | $0.0040 |
Storage
| Storage type | Disk type | Price (per GB · month) |
|---|---|---|
| hostPath | NVMe | $0.0600 |
| nfs | NVMe | $0.0600 |
Numbers are pulled from the production API at every docs build. The console at console.meshive.ai is always the live source of truth.
Instances
Section titled “Instances”A Meshive instance is a pod: a container with dedicated GPU, vCPU, RAM, and storage carved out of a host machine. There are no fixed instance SKUs — you compose the shape, and the console bounds every slider by what the live fleet can actually serve (which is why the options you see can differ from day to day).
GPU pods
Section titled “GPU pods”-
Pick a GPU model and count. The model list and the count options come from the fleet in real time — if the largest host with your GPU has 4 of them, the slider tops out at 4.
-
Size the companions. vCPU and RAM are set per GPU, pre-filled with recommended values and bounded by the host’s hardware. The platform-wide rule of thumb, per GPU:
vCPU RAM Minimum ⌈VRAM / 4⌉ VRAM (GB) Maximum 2× minimum 2× minimum Example — 1× 32 GB GPU: 8–16 vCPU and 32–64 GB RAM. Hosts must reserve these companions GPU-first, so a GPU pod never finds the CPU/RAM around its GPU already sold off (how that works).
-
Add storage. Local volumes start at 10 GB and go up to whatever the host has free; storage is billed per GB·month at the rates in the table above.
You pay the GPU rate × count plus the vCPU/RAM/storage rates — each line item is visible in the create flow before you confirm.
CPU pods
Section titled “CPU pods”CPU pods rent vCPU and RAM without a GPU — including spare CPU capacity on GPU hosts. You choose the vCPU count and RAM, with RAM selectable from 0.5 to 8 GB per core (matching the usual compute / balanced / memory shapes — 2, 4, and 8 GB per core). The rate depends on the silicon: per-vCPU pricing is tiered by CPU vendor/generation, and per-GB RAM pricing by DDR generation — exactly the tiers in the tables above.
Demand vs. spot
Section titled “Demand vs. spot”Every compute SKU has two rates:
| Demand | Spot | |
|---|---|---|
| Price | Standard rate | Discounted |
| Availability | Runs until you stop it | May be reclaimed and reallocated to a different node |
| Local storage | ✓ | ✗ — local volumes would be lost on reallocation, so spot pods must use NFS storage instead |
Use spot for interruption-tolerant work (training with checkpoints, batch processing); keep services and stateful workloads on demand.
Limits
Section titled “Limits”- Shape ceilings are fleet-driven, not fixed — GPU count, vCPU, and RAM max out at the largest matching host currently online. The GPU table above is the live lineup.
- Local volumes: 10 GB minimum per volume, host capacity as the ceiling, demand pods only.
- Hardware is pinned: a pod stays on the machine it was placed on — to move to a different GPU, save your image/data and recreate (how placement is chosen).
- Billing granularity is one minute, on a running-pod basis.