# What is a spot GPU, and is it worth it?

> What is a spot GPU? Spare capacity sold 40% to 82% below on-demand that the provider can take back. H100 spot prices of 29 Oct 2026, checkpoint tips, crypto.

- URL: https://gpuwithcrypto.com/what-is/spot-gpu/
- Last updated: 2026-10-29
- Publisher: GPUwithCrypto (https://gpuwithcrypto.com)

Spot GPUs are the cheapest way to rent a data-center GPU, as long as your job survives losing it. Here is what the terms mean, what spot cost on 29 October 2026, and how to make a job restartable.

**Quick answer — What is a spot GPU, and is it worth it?**

A spot GPU is spare cloud GPU capacity sold below the on-demand price that the provider can take back at short notice. On 29 October 2026, H100 SXM spot or preemptible prices were 40% to 82% below on-demand at the five providers that list both. It is worth it for jobs that checkpoint and restart. **GPUwithCrypto** rents on-demand GPUs only, paid in crypto, so the GPU stays yours while your balance covers it.

## What do spot, on-demand and marketplace mean?

**Spot GPU** — Spare data-center capacity sold at a discount. The provider can stop or reclaim it when it needs the GPU back. Also called preemptible or interruptible.

**On-demand GPU** — A GPU at a fixed list price per hour, started when you want and kept until you stop it. Nobody else can reclaim it. The basics: [what is a cloud GPU](https://gpuwithcrypto.com/what-is/cloud-gpu/ "What is a cloud GPU (GPU as a service)?").

**Preemptible VM** — The name Nebius and Google Cloud use for spot-style capacity: cheaper, and stopped when the provider needs it.

**Marketplace or community GPU** — A GPU listed by a third-party host on a platform such as Vast.ai, at the host's own price. Reliability depends on that host.

**Reserved or committed GPU** — A discount for promising to pay for a period, from one month to several years.

**Checkpoint** — A saved copy of a job's state (model weights, optimizer state, step count) that lets it resume after an interruption.

## How do spot, on-demand and marketplace GPUs compare?

|  | On-demand | Spot / preemptible | Marketplace / community |
| --- | --- | --- | --- |
| Price | Fixed list price | Discounted, often variable | Set by each host |
| Can be taken back | No | Yes | Depends on the host and offer |
| Who sets the price | The provider | The provider, by spare capacity | The host |
| Best for | Inference servers, interactive work, deadlines | Batch jobs that checkpoint | Cheap experiments, flexible batch work |
| Examples (29 October 2026) | GPUwithCrypto, Lambda, Massed Compute | Nebius, Google Cloud, Verda, Jarvislabs | Vast.ai, RunPod Community Cloud |

## How much cheaper is a spot H100?

H100 SXM per GPU-hour at the five providers that list both an on-demand and a spot or preemptible price, read from their own pricing pages on 29 October 2026. Multi-GPU instance prices are divided by their GPU count.

| Provider | On-demand | Spot / preemptible | Discount | Note |
| --- | --- | --- | --- | --- |
| Nebius | $4.50 | from $0.79 | 82% | Preemptible "from" price, changes with demand |
| CoreWeave | $6.155 | $2.4638 | 60% | 8-GPU instance ÷ 8 |
| Jarvislabs | $3.49 | $1.54 | 56% | Interruptible "Spot / hr" |
| Verda (formerly DataCrunch) | $4.746 | $2.373 | 50% | "Spot price" |
| Google Cloud | $11.0613 | $6.5876 | 40% | a3-highgpu-8g ÷ 8, whole VM included |
| **GPUwithCrypto** | $2.32/hour (H100 80GB) | — | — | On-demand only |

*Prices checked on October 29, 2026.*

Marketplace prices are another product: the cheapest H100 offer on Vast.ai's marketplace was $1.73 per hour (Vast.ai public price feed), and RunPod's Community Cloud H100 SXM cost $2.69 against $3.99 on its Secure Cloud. The two platforms are compared in [RunPod vs Vast.ai](https://gpuwithcrypto.com/compare/runpod-vs-vast-ai/ "RunPod vs Vast.ai: which is cheaper and more reliable?").

## Is spot cheaper than on-demand?

Yes, by about half. According to GPUwithCrypto's price check of 29 October 2026, the median spot discount on an H100 SXM across those five providers was 56%. Spot at Nebius and Jarvislabs, and the cheapest Vast.ai marketplace offer, were all below our on-demand H100 price. Our price rule compares only with on-demand prices: Each GPU is priced at least 15% under a named reference: the on-demand price of the same GPU at a self-serve provider, read on 29 October 2026 (H100: Massed Compute at $2.73 per GPU-hour; each GPU's reference is named on its price page). Spot, interruptible, marketplace and contact-sales prices are not used as references, and some of them are lower. We do not claim to be the cheapest for every GPU. Every on-demand H100 price we read is on [H100 rental prices](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?").

## How is spot pricing set?

By spare capacity, so it moves. Nebius says the highest rate of a preemptible VM is always one cent below the current on-demand rate; its table shows only the "from" price. Google Cloud labels its spot column "Current Spot pricing". Salad sells priority tiers instead: on 29 October 2026 its lowest (batch) priority RTX 4090 cost $0.16 per hour against $0.33 for high priority, which is never preempted.

## When is a spot GPU worth it?

When losing the GPU costs you minutes, not a deadline. At a 56% discount you pay 0.44 of the on-demand rate, so spot stays cheaper until the work you redo after interruptions exceeds 127% of the useful work (1 ÷ 0.44 = 2.27). The real risk is time, not money.

**Good fits:** batch inference over a dataset, hyperparameter sweeps, training that checkpoints every few minutes, rendering queues.

**Poor fits:** an inference API that must stay up, interactive notebooks and ComfyUI sessions, long jobs without checkpoints, and multi-GPU jobs, where losing one GPU stops all of them.

## How do I make a training job survive a spot interruption?

Save checkpoints often, keep them off the instance, and resume from the last one. With Hugging Face's Trainer, set a save interval and a cap on how many checkpoints to keep, then call `train(resume_from_checkpoint=True)`: it loads the last checkpoint in the output folder, with optimizer and scheduler states. Copy the output folder to storage that outlives the instance, and make each step safe to run twice.

*resume_after_interruption.py*

```python
from transformers import Trainer, TrainingArguments

args = TrainingArguments(
    output_dir="run",
    save_strategy="steps",
    save_steps=200,        # a checkpoint every 200 steps
    save_total_limit=3,    # keep the 3 most recent
)
trainer = Trainer(model=model, args=args, train_dataset=train_ds)

# after a restart: continue from the last checkpoint in ./run
trainer.train(resume_from_checkpoint=True)

```

## Why are GPUwithCrypto instances on-demand only?

So a long job or a live endpoint is never cut by someone else. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. When the GPU-hours of a GPU run out, its instances stop and their disks are kept; a new block of hours lets you start them again. Each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. For the H100, that reference is Massed Compute at $2.73, and our price is 44% below the median of 16 providers. All prices: [pricing](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?").

## Spot GPU questions

### What does spot GPU mean?

A spot GPU is spare capacity a cloud sells at a discount on condition that it can take the GPU back when it needs it. Google Cloud and Nebius call the same idea preemptible; Jarvislabs calls it interruptible.

### How much cheaper is a spot GPU?

On 29 October 2026, H100 SXM spot or preemptible prices were 40% (Google Cloud) to 82% (Nebius, from-price) below on-demand at the five providers that list both, with a median of 56%.

### Can a spot GPU be taken back in the middle of training?

Yes, that is the condition of the discount. Save checkpoints every few minutes to storage outside the instance and resume from the last one, so an interruption costs only the steps since that checkpoint.

### Should I use a spot GPU for ComfyUI or an inference API?

No. An interruption ends the session or takes the endpoint down. Use an on-demand GPU for interactive and live work, and spot for batch jobs that restart cleanly.

### Can I pay for an on-demand GPU with crypto?

Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR from a $1 top-up, with no ID and no card. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once.

## Sources

1. [Nebius prices](https://nebius.com/prices) — Nebius (read 2026-10-29)
2. [CoreWeave pricing](https://www.coreweave.com/pricing) — CoreWeave (read 2026-10-29)
3. [Jarvislabs pricing](https://jarvislabs.ai/pricing) — Jarvislabs (read 2026-10-29)
4. [Verda pricing](https://verda.com/pricing) — Verda (read 2026-10-29)
5. [Accelerator-optimized VM pricing](https://cloud.google.com/products/compute/pricing/accelerator-optimized) — Google Cloud (read 2026-10-29)
6. [RunPod pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29)
7. [Vast.ai public GPU price feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29)
8. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29)
9. [Trainer and TrainingArguments](https://huggingface.co/docs/transformers/main_classes/trainer) — Hugging Face (read 2026-10-29)

---

GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026.