Skip to content

VRAM reference

Phi-4 on an RTX 5090

Yes, you can fine-tune Phi-4 on one RTX 5090 (32 GB) with QLoRA. 4-bit inference: yes, 8.2 GB of weights, room for 16K tokens of context or 24 concurrent 4K-token sequences.

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

Verdict by method

Fine-tuningMemory at 2K tokensOne RTX 5090?Longest sequence (batch 1)Largest batch at 2KLargest LoRA rank
QLoRA (4-bit base)13 GBYes8K tokens6128
LoRA (16-bit base)33 GBNo———
Full fine-tuning238 GBNo——n/a
InferenceMemory at 8KOne RTX 5090?Longest contextConcurrent 4K sequences
16-bit (BF16/FP16)31 GBNo——
8-bit16 GBYes16K tokens16
4-bit9.9 GBYes16K tokens24
  • QLoRA: yes, about 13 GB at 2K tokens, batch 1, rank 16; up to 8K-token sequences at batch 1 or batch 6 at 2K, LoRA rank up to 128.
  • LoRA: no, it needs about 33 GB at 2K tokens, more than the 32 GB available.
  • Full fine-tuning: no, it needs about 238 GB at 2K tokens, more than the 32 GB available.

Where the memory goes

At 2,048 tokens per sequence, batch 1, LoRA rank 16, gradient checkpointing on
MethodWeightsTraining stateActivationsTotalOn one RTX 509080 GB GPUs needed
QLoRA (4-bit base)9.1 GB1.0 GB3.0 GB13 GBFits1
LoRA (16-bit base)29 GB1.0 GB3.0 GB33 GBDoesn't fit1
Full fine-tuning29 GB205 GB3.0 GB238 GBDoesn't fit4
One sequence, fp16 KV cache
PrecisionWeightsKV cache at 8KTotalAt 16K contextOn one RTX 5090
16-bit (BF16/FP16)29 GB1.7 GB31 GB33 GBDoesn't fit
8-bit15 GB1.7 GB16 GB18 GBFits
4-bit8.2 GB1.7 GB9.9 GB12 GBFits

The cheapest way to run it

Renting an RTX 5090 costs $1.39 an hour on Runpod Secure Cloud ($0.69 on Runpod Community Cloud). QLoRA on it processes about 1,115 tokens a second by our estimate, so 10 million training tokens take 2.5 hours and cost about $3.46.

Job (default settings)MemoryCheapest rental that fitsPer hour
QLoRA fine-tuning13 GBRTX 3090$0.50
LoRA fine-tuning33 GBL40S$1.09
4-bit inference9.9 GBRTX 3090$0.50
16-bit inference31 GBL40S$1.09

On-demand rates seen on 15 November 2026: Runpod Secure Cloud, Lambda for the A100 40GB. Consumer cards count as single GPUs only; data-centre GPUs up to eight per node.

Try other settings

JobMemoryRTX 5090 (32 GB)Smallest single GPU
QLoRA (4-bit base)2K tokens × batch 1, rank 1613 GBFitsRTX 3090 (24 GB)
LoRA (16-bit base)2K tokens × batch 1, rank 1633 GB2 GPUsA100 40GB (40 GB)
Full fine-tuning2K tokens × batch 1238 GB9 GPUsNone: needs several
Inference, 16-bit (BF16/FP16)8K-token context × 1 sequence31 GB2 GPUsA100 40GB (40 GB)
Inference, 8-bit8K-token context × 1 sequence16 GBFitsRTX 3090 (24 GB)
Inference, 4-bit8K-token context × 1 sequence9.9 GBFitsRTX 3090 (24 GB)

A job fits when the estimate is within 90% of the card's memory. Training assumes gradient checkpointing and FlashAttention; multi-GPU counts assume the memory splits evenly (FSDP, ZeRO-3 or tensor parallelism).

How we estimate it

Every figure comes from the model's published architecture (layers, hidden size, attention and KV heads, vocabulary) rather than a rule of thumb, using the same functions as the calculator above. Sizes are in GB (10⁹ bytes).

  • Full fine-tuning: 16 bytes per parameter for mixed-precision Adam (bf16 weights and gradients, fp32 master weights and two moments), as in the ZeRO paper.
  • LoRA: frozen bf16 weights (2 bytes per parameter) plus 16 bytes per adapter parameter, rank 16 on every linear layer. For mixture-of-experts models, adapters go on attention only and the experts stay frozen.
  • QLoRA: NF4 weights with double quantisation (4.127 bits per parameter); embeddings and the output head stay in bf16, as bitsandbytes leaves them.
  • Activations: gradient checkpointing and FlashAttention. Each layer's bf16 input is kept, one layer is recomputed at a time, and the output logits take 8 bytes per vocabulary entry per token (fp32 values and gradients). Chunked cross-entropy, as in Unsloth or Liger kernels, cuts that last term.
  • Inference: 2, 1 or about 0.56 bytes per weight (4-bit with group-wise scales), plus an fp16 KV cache of 2 × layers × KV heads × head dim × 2 bytes per token; sliding-window layers only cache their window.
  • Fits means the estimate is within 90% of the card's listed memory; the rest covers the CUDA context, kernels and fragmentation. Multi-GPU counts assume memory splits evenly.

Speed estimates use RTX 5090's dense BF16 peak of 209.5 TFLOPS, about 4 FLOPs per active parameter per token for LoRA and 40% utilisation; see fine-tuning and training costs for the method.

Frequently asked questions

Can I fine-tune Phi-4 on an RTX 5090?
QLoRA: yes, about 13 GB at 2K tokens, batch 1, rank 16; up to 8K-token sequences at batch 1 or batch 6 at 2K, LoRA rank up to 128. LoRA: no, it needs about 33 GB at 2K tokens, more than the 32 GB available. Full fine-tuning: no, it needs about 238 GB at 2K tokens, more than the 32 GB available.
Can I run Phi-4 on an RTX 5090?
16-bit (BF16/FP16) inference: no, the weights alone are 29 GB. 8-bit inference: yes, 15 GB of weights, room for 16K tokens of context or 16 concurrent 4K-token sequences. 4-bit inference: yes, 8.2 GB of weights, room for 16K tokens of context or 24 concurrent 4K-token sequences.
What does fine-tuning Phi-4 on an RTX 5090 cost?
With QLoRA at our planning rate of 40% utilisation, about 2.5 hours for 10 million training tokens, around $3.46 at $1.39 an hour (Runpod Secure Cloud, November 2026).

Sources

  1. [1]Hugging Face. microsoft/phi-4 config.json
  2. [2]Microsoft. microsoft/phi-4 model card
  3. [3]Microsoft. Phi-4 Technical Report (arXiv:2412.08905)
  4. [4]NVIDIA. RTX 5090 specifications
  5. [5]Runpod. GPU pricing (updated 12 November 2026)
  6. [6]Lambda. GPU cloud pricing
  7. [7]Dettmers et al., 2023. QLoRA: Efficient Finetuning of Quantized LLMs (arXiv:2305.14314)
  8. [8]Hu et al., 2021. LoRA: Low-Rank Adaptation of Large Language Models (arXiv:2106.09685)
  9. [9]Rajbhandari et al., 2019. ZeRO: Memory Optimizations Toward Training Trillion Parameter Models (arXiv:1910.02054)
  10. [10]Korthikanti et al., 2022. Reducing Activation Recomputation in Large Transformer Models (arXiv:2205.05198)

Measure before you rent the GPU.

Prompt tuning and model search in Kojen run on hosted models against your own test cases, so you know whether fine-tuning Phi-4 is worth the GPU time.

Start training free