Skip to content

Cost reference

What it costs to fine-tune Gemma 3 27B

The cheapest setup we compared is LoRA on 2 × L40S: about $0.23 per million training tokens, so $2.29 for 10 million and $22.94 for 100 million. Full fine-tuning costs about $4.06 per 10 million tokens on 3 × B200.

By Will Schulz · Checked 15 November 2026 · Prices as of November 2026

GPU-hours and cost by GPU and method

Per epoch over the token count shown; multiply by your number of epochs. GPU-hours are the same however many GPUs the job is split over.

GPU · methodGPUs for memoryTokens/s per GPUGPU-hours per 1M1.0M tokens10M tokens100M tokens
H100 SXM · LoRA$4.39/hr13,6050.08$0.34$3.38$33.82
H100 SXM · QLoRA$4.39/hr12,8120.10$0.43$4.34$43.36
H100 SXM · Full fine-tuning$4.39/hr72,4040.12$0.51$5.07$50.74
B200 · LoRA$7.99/hr18,2020.03$0.27$2.71$27.06
B200 · QLoRA$7.99/hr16,3980.04$0.35$3.47$34.69
B200 · Full fine-tuning$7.99/hr35,4680.05$0.41$4.06$40.59
L40S · LoRA$1.09/hr21,3200.21$0.23$2.29$22.94
L40S · QLoRA$1.09/hr11,0290.27$0.29$2.94$29.42
L40S · Full fine-tuning$1.09/hr11 (over 8)8800.32———
RTX 5090 · LoRA$1.39/hrDoesn't fit7640.36———
RTX 5090 · QLoRA$1.39/hr15960.47$0.65$6.48$64.82
RTX 5090 · Full fine-tuning$1.39/hrDoesn't fit5090.55———

Compared with fine-tuning APIs

OptionPer 1M tokens10M tokens100M tokens
Rented GPU: LoRA on L40S (our estimate)$0.23$2.29$22.94
Together AIGemma 3 27B not listed——
OpenAI gpt-4.1-mini fine-tuning (closed model; closed to new users)$5.00$50.00$500

API prices are per training token processed (dataset tokens × epochs), seen on 15 November 2026. Together AI charges a minimum per job. OpenAI's pricing page says fine-tuning is closed to new users. Rented GPU estimates leave out setup, failed runs and idle time, which an API absorbs.

Your run

Pre-filled for Gemma 3 27B: 10M tokens, 3 epochs, LoRA on an H100.

Estimated cost

$10.15

H100 SXM-hours

2.3

Time on 1 GPU

2.3 hours

Tokens per second per GPU

3,605

30M tokens processed. Fits on one H100 SXM at 2,048-token sequences. GPU-hours assume perfect scaling across GPUs.

Assumptions

  • Compute per token: 6 FLOPs per active parameter for full fine-tuning (forward plus backward) and about 4 for LoRA and QLoRA, which skip the frozen weights' gradients. Gemma 3 27B uses 27.432B parameters per token.
  • Utilisation: 40% of each GPU's dense BF16 peak. torchtune measured 3,083 tokens a second for Llama 3.1 8B LoRA on one RTX 4090, about 60% on the same accounting, so well-tuned code can beat these figures.
  • QLoRA runs at 78% of LoRA's speed, the ratio torchtune measured (2,413 vs 3,083 tokens a second), because the 4-bit weights are dequantised on the fly.
  • Memory decides the GPU count, from our Gemma 3 27B VRAM estimates at 2,048-token sequences. Consumer cards are counted as single GPUs only.
  • Prices: on-demand Runpod Secure Cloud rates (page dated 12 November 2026). Community Cloud and spot capacity cost less.

Frequently asked questions

How much does it cost to fine-tune Gemma 3 27B?
About $0.23 per million training tokens at the cheapest setting we compared (LoRA on 2 × L40S, $1.09/hour per GPU), so about $2.29 for 10M tokens and $22.94 for 100M. Budget for several runs: the first rarely wins.
How long does it take to fine-tune Gemma 3 27B on an H100?
LoRA processes about 3,605 tokens a second on one H100 at 40% utilisation, so 10M tokens take about 46 minutes. Full fine-tuning needs 7 H100s for memory and about 1.5× the compute.
Is it cheaper to fine-tune Gemma 3 27B through an API?
Together AI's November 2026 price list doesn't include Gemma 3 27B, so the choice is rented GPUs or a hosted model. Rented GPU time costs about $2.29 per 10M tokens by our estimate, plus setup and idle time.
Should I use LoRA or full fine-tuning for Gemma 3 27B?
Start with LoRA or QLoRA. Full fine-tuning Gemma 3 27B needs about 445 GB, so 7 × 80 GB GPUs with sharded optimiser state, and about 1.5× the compute per token. LoRA usually gets close for format, style and task adaptation; move to full fine-tuning only when evaluation shows LoRA has plateaued.

Sources

  1. [1]Hugging Face. google/gemma-3-27b-it config.json
  2. [2]Google DeepMind. google/gemma-3-27b-it model card
  3. [3]Google DeepMind. Gemma 3 Technical Report (arXiv:2503.19786)
  4. [4]Hu et al., 2021. LoRA: Low-Rank Adaptation of Large Language Models (arXiv:2106.09685)
  5. [5]Dettmers et al., 2023. QLoRA: Efficient Finetuning of Quantized LLMs (arXiv:2305.14314)
  6. [6]Kaplan et al., 2020. Scaling Laws for Neural Language Models (arXiv:2001.08361)
  7. [7]PyTorch. torchtune: memory and throughput benchmarks (Llama 3.1 8B on RTX 4090)
  8. [8]Together AI. Fine-tuning pricing (seen 15 November 2026)
  9. [9]OpenAI. API pricing: fine-tuning (seen 15 November 2026)
  10. [10]Runpod. GPU pricing (updated 12 November 2026)

Try the cheaper levers first.

Prompt tuning and model search in Kojen cost API calls, not GPU-hours, and often close the gap before any fine-tuning.

Start training free