Skip to content

VRAM reference

DeepSeek-R1-Distill-Qwen-32B on an RTX 4090

No fine-tuning method fits DeepSeek-R1-Distill-Qwen-32B on one RTX 4090 (24 GB) at 2,048 tokens. 4-bit inference: yes, 18 GB of weights, room for 8K tokens of context or 2 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 4090?Longest sequence (batch 1)Largest batch at 2KLargest LoRA rank
QLoRA (4-bit base)26 GBNo———
LoRA (16-bit base)72 GBNo———
Full fine-tuning529 GBNo——n/a
InferenceMemory at 8KOne RTX 4090?Longest contextConcurrent 4K sequences
16-bit (BF16/FP16)68 GBNo——
8-bit35 GBNo——
4-bit21 GBYes8K tokens2
  • QLoRA: no, it needs about 26 GB at 2K tokens, more than the 24 GB available.
  • LoRA: no, it needs about 72 GB at 2K tokens, more than the 24 GB available.
  • Full fine-tuning: no, it needs about 529 GB at 2K tokens, more than the 24 GB available.

Where the memory goes

At 2,048 tokens per sequence, batch 1, LoRA rank 16, gradient checkpointing on
MethodWeightsTraining stateActivationsTotalOn one RTX 409080 GB GPUs needed
QLoRA (4-bit base)19 GB2.1 GB4.5 GB26 GBDoesn't fit1
LoRA (16-bit base)66 GB2.1 GB4.5 GB72 GBDoesn't fit2
Full fine-tuning66 GB459 GB4.5 GB529 GBDoesn't fit8
One sequence, fp16 KV cache
PrecisionWeightsKV cache at 8KTotalAt 128K contextOn one RTX 4090
16-bit (BF16/FP16)66 GB2.1 GB68 GB100 GBDoesn't fit
8-bit33 GB2.1 GB35 GB67 GBDoesn't fit
4-bit18 GB2.1 GB21 GB53 GBFits

The cheapest way to run it

Renting an RTX 4090 costs $0.89 an hour on Runpod Secure Cloud ($0.34 on Runpod Community Cloud). Since no fine-tuning method fits, the table shows the cheapest hardware that does.

Job (default settings)MemoryCheapest rental that fitsPer hour
QLoRA fine-tuning26 GBL40S$1.09
LoRA fine-tuning72 GB2 × L40S$2.18
4-bit inference21 GBRTX 3090$0.50
16-bit inference68 GBA100 80GB$2.19

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 4090 (24 GB)Smallest single GPU
QLoRA (4-bit base)2K tokens × batch 1, rank 1626 GB2 GPUsRTX 5090 (32 GB)
LoRA (16-bit base)2K tokens × batch 1, rank 1672 GB4 GPUsH200 (141 GB)
Full fine-tuning2K tokens × batch 1529 GB25 GPUsNone: needs several
Inference, 16-bit (BF16/FP16)8K-token context × 1 sequence68 GB4 GPUsA100 80GB (80 GB)
Inference, 8-bit8K-token context × 1 sequence35 GB2 GPUsA100 40GB (40 GB)
Inference, 4-bit8K-token context × 1 sequence21 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 4090's dense BF16 peak of 165.2 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 DeepSeek-R1-Distill-Qwen-32B on an RTX 4090?
QLoRA: no, it needs about 26 GB at 2K tokens, more than the 24 GB available. LoRA: no, it needs about 72 GB at 2K tokens, more than the 24 GB available. Full fine-tuning: no, it needs about 529 GB at 2K tokens, more than the 24 GB available.
Can I run DeepSeek-R1-Distill-Qwen-32B on an RTX 4090?
16-bit (BF16/FP16) inference: no, the weights alone are 66 GB. 8-bit inference: no, the weights alone are 33 GB. 4-bit inference: yes, 18 GB of weights, room for 8K tokens of context or 2 concurrent 4K-token sequences.

Sources

  1. [1]Hugging Face. deepseek-ai/DeepSeek-R1-Distill-Qwen-32B config.json
  2. [2]DeepSeek. deepseek-ai/DeepSeek-R1-Distill-Qwen-32B model card
  3. [3]DeepSeek. DeepSeek-R1 (arXiv:2501.12948)
  4. [4]NVIDIA. RTX 4090 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 DeepSeek-R1-Distill-Qwen-32B is worth the GPU time.

Start training free