How much does it cost to rent an H100 per hour?

On the marketplaces an H100 80GB starts around $1.65/GPU-hr on Vast.ai and $2.49 on RunPod on-demand, dropping to roughly $1.10–$1.49 on spot/interruptible. Hyperscalers are far pricier — AWS works out near $12/GPU-hr from a P5 instance. The calculator shows all four side by side for your exact GPU count and hours.

What is the cheapest cloud GPU for training a model?

For small models and hobby fine-tunes a spot RTX 4090 (about $0.22–$0.34/hr on Vast.ai or RunPod) is the cheapest. For serious training the best value per FLOP is usually an A100 80GB or H100 on a marketplace's spot tier. Pick the GPU that fits your model first, then sort by price — the tool highlights the cheapest provider automatically.

How much does it cost to fine-tune an LLM on cloud GPUs?

A QLoRA fine-tune of a 7B–8B model fits one 24GB GPU and a few hours: a spot RTX 4090 for 6 hours is under $3. A larger full fine-tune on 4× A100 80GB for 12 hours is roughly $38 on Vast.ai or $57 on RunPod. Enter your GPU, count, and hours to get the exact figure in USD and LKR.

Is spot or interruptible GPU pricing worth it versus on-demand?

Spot usually saves 30–50% — for example a 24/7 H100 drops from about $1,818 to $1,088 a month on RunPod. The catch is that the provider can reclaim the instance mid-job, so it suits checkpointed training and batch jobs, not a production endpoint that must stay up. The tool shows the exact saving for your job.

Which providers and GPUs does this calculator cover?

Four providers — RunPod, Lambda, Vast.ai, and AWS — across six GPUs: RTX 4090 24GB, L40S 48GB, A100 80GB, H100 80GB, H200 141GB, and B200 180GB. Lambda lists on-demand only (no spot) and does not sell consumer RTX cards; AWS rates are the per-GPU share of its multi-GPU P5/P4de/G6e instances.

Are these GPU prices live and exact?

No. They are illustrative reference rates re-confirmed against each provider's pricing page on the build date and stamped with a last-verified date. GPU rental prices move weekly, and marketplace rates change by the hour. Use the result to budget and compare, then confirm the live rate on the provider before you commit.

How is the LKR figure calculated?

Every USD total is converted at a reference rate of 300 LKR per USD, based on the Central Bank of Sri Lanka indicative rate at the last-verified date. The rupee figure is for planning a budget in local terms; your card or bank will apply its own rate and a foreign-transaction fee on top, so treat it as a guide.

Does the cost include storage and bandwidth?

Storage is optional: enter a persistent-volume size in GB and the tool prorates each provider's per-GB-month list rate to your job's hours. Egress and bandwidth are not modelled — they vary widely by provider and are often free for ingress — so check the provider's data-transfer pricing separately for large datasets.

AI · Cloud GPU pricing

AI GPU Cloud Cost Calculator

Work out what it costs to rent cloud GPUs to train or serve an AI model, and compare the same job across RunPod, Lambda, Vast.ai, and AWS — on-demand and spot — in US dollars and Sri Lankan rupees. Pick a GPU, how many, and how long, and see the cheapest option in seconds.

By Induwara Ashinsana— Executive Director, Ryzera TechnologiesUpdated Jun 5, 2026

Estimate your GPU rental cost

Rates verified · 300 LKR/USD

Popular jobs

GPU model

High-end LLM training & low-latency serving

Number of GPUs

1 – 64

Pricing mode

Usage pattern

Duration

Unit

Storage (GB, optional)

Persistent volume; leave 0 if none

Cheapest job total

$19.80

≈ Rs 5,940 · via Vast.ai

Job duration

12 GPU-hr

12 GPU-hours total (1 × 12 h)

Spot vs on-demand

−$6.60

33.33% off on Vast.ai

Provider comparison

Provider	$/GPU-hr	Job total	Monthly (24/7)
RunPod	$2.49	$29.88	$1,818
Lambda	$2.99	$35.88	$2,183
Vast.aiCheapest	$1.65	$19.80	$1,205
AWS	$12.29	$147.48	$8,972

Breakdown · Vast.ai

Compute · $1.65/GPU-hr × 1 GPU × 12 h	$19.80
Job total	$19.80
In Sri Lankan rupees (300 LKR/USD)	Rs 5,940

Rates are illustrative reference figures — not live quotes — sourced from RunPod, Lambda, Vast.ai, AWS, and last re-confirmed on build day. GPU rental prices move weekly; always check the provider before you commit. Egress and bandwidth vary by provider and are not modelled here.

How it works

The calculator is pure arithmetic over published per-GPU-hour rates. It does not call any API or run a model — every figure comes from a rate table re-confirmed against the four providers' pricing pages on the last-verified date below. Each provider lists a rate in US dollars per GPU per hour, separately for on-demand and spot/interruptible capacity.

1. Normalise the duration

Your duration is converted to hours: hours = value × { hours: 1, days: 24, months: 730 }. The 730-hour month is the standard cloud “month-hour” convention (24 × 365.25 ÷ 12) used by AWS and most providers for monthly billing.

2. Compute cost per provider

For the selected GPU, count, and pricing mode, each provider's compute cost is rate × number of GPUs × hours. Optional storage adds GB × (per-GB-month ÷ 730) × hours, prorating each provider's persistent-volume list price to the job length. The two sum to the job total.

3. Always-on monthly projection

The monthly column projects a 24/7 deployment as rate × number of GPUs × 730 plus monthly storage — the number you would pay to keep an inference endpoint live for a full month.

4. Spot saving and the cheapest pick

The tool finds the lowest-cost provider that offers both on-demand and spot for your GPU and reports the saving in dollars and percent — spot typically cuts 30–50% in exchange for the instance being reclaimable mid-job. The cheapest available provider overall is highlighted, and the headline total is converted to rupees at the reference rate of 300 LKR per USD.

5. Cross-check

Marketplaces such as RunPod and Vast.ai bill per second, not per hour. As an internal check the same compute cost is recomputed by converting the hourly rate to a per-second rate (rate ÷ 3600) × GPUs × (hours × 3600); the per-hour and per-second methods agree to the cent, the same idea as checking a tax figure against the regulator's own formula.

Worked examples

Fine-tune on 4× A100 80GB for 12 hours (one-off, on-demand)

Hours: 12 × 1 = 12 (unit = hours)
RunPod: $1.19 × 4 GPU × 12 h = $57.12
Lambda: $2.49 × 4 × 12 = $119.52
Vast.ai: $0.80 × 4 × 12 = $38.40 ← cheapest
AWS: $5.12 × 4 × 12 = $245.76
Cheapest = Vast.ai $38.40 (≈ Rs 11,520). RunPod still beats Lambda by $62.40.

Always-on inference, 1× H100 80GB for a month — spot vs on-demand

Hours: 1 × 730 = 730 (unit = months)
RunPod on-demand: $2.49 × 1 × 730 = $1,817.70 / month
RunPod spot: $1.49 × 1 × 730 = $1,087.70 / month
Spot saving: 1,817.70 − 1,087.70 = $730.00 / month (40.2%)
Cheapest monthly across providers: Vast.ai on-demand $1.65 × 730 = $1,204.50

Edge case — a fractional half-hour run on 1× A100 80GB

Hours: 0.5 × 1 = 0.5
RunPod: $1.19 × 1 × 0.5 = $0.595 → shown as $0.60
No silent zero or NaN: durations below an hour still bill, because marketplaces meter per second.
Try 0 GPUs or a negative duration and the tool returns a specific error instead of a misleading $0.

Frequently asked questions

Sources & references

Per-GPU-hour rates and the USD/LKR reference were last re-confirmed against the sources above on 2026-06-05. GPU rental prices move frequently — these are illustrative reference figures for budgeting and comparison, not live quotes. Confirm the live rate on the provider before you commit.

Related tools

LiveAI

Fine-Tuning Cost Calculator

Estimate the one-time training cost and ongoing monthly inference cost of fine-tuning an LLM (GPT-4o, GPT-4o mini, GPT-3.5 Turbo, Together AI Llama) from your token count, epochs, and usage — in USD and LKR.

Open tool

LiveAI

Transcription Cost Calc

Estimate what it costs to transcribe audio with the major speech-to-text APIs — OpenAI Whisper, Deepgram Nova, AssemblyAI Best and Nano, Google Cloud Speech-to-Text, and Azure AI Speech. Enter a duration and get a side-by-side per-provider comparison in USD and LKR, with each vendor's billing-rounding rule, free tier, effective cost per hour, and a monthly-to-annual projection.

Open tool

LiveAI

LLM VRAM Calculator

Estimate the GPU VRAM needed to run or fine-tune any open LLM (Llama 3, Mistral, Qwen, Gemma, DeepSeek) at a given precision, context, and batch size — and check whether it fits your GPU. Formulas cited, runs in your browser.

Open tool

Rate this tool

Be the first to rate

Comments & feedback

Spotted a bug or want an improvement? Tell us — our team reviews every comment, and good ideas get built. Comments are public and anonymous.

Found a bug, a stale rate, or want another GPU or provider added?

Email me at [email protected] — most fixes ship within 24 hours.