Skip to content

VRAM reference

Qwen3 1.7B VRAM requirements

Fine-tuning Qwen3 1.7B takes about 4.7 GB of GPU memory with QLoRA, 7.2 GB with LoRA on 16-bit weights and 35 GB for full fine-tuning, at 2,048 tokens per sequence. Running it takes 5.0 GB in 16-bit or 2.1 GB in 4-bit with a 8K-token context.

By Will Schulz · Architecture and prices checked 15 November 2026 · Prices as of November 2026

Fine-tuning memory

At 2,048 tokens per sequence, batch 1, LoRA rank 16, gradient checkpointing on
MethodWeightsTraining stateActivationsTotalSmallest single GPU80 GB GPUs needed
QLoRA (4-bit base)1.5 GB0.3 GB2.9 GB4.7 GBRTX 3090 (24 GB)1
LoRA (16-bit base)4.1 GB0.3 GB2.9 GB7.2 GBRTX 3090 (24 GB)1
Full fine-tuning4.1 GB28 GB2.9 GB35 GBA100 40GB (40 GB)1

On a 24 GB card such as the RTX 4090: QLoRA: yes, about 4.7 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 512.

Inference memory

One sequence, fp16 KV cache
PrecisionWeightsKV cache at 8KTotalAt 32K contextSmallest single GPU
16-bit (BF16/FP16)4.1 GB0.9 GB5.0 GB7.8 GBRTX 3090 (24 GB)
8-bit2.0 GB0.9 GB3.0 GB5.8 GBRTX 3090 (24 GB)
4-bit1.1 GB0.9 GB2.1 GB4.9 GBRTX 3090 (24 GB)

Which GPUs fit Qwen3 1.7B

One GPU unless a count is shown. Follow a GPU's name for its maximum sequence length, batch size, LoRA rank and rental cost.

GPUQLoRALoRAFull fine-tuningRun, fp16Run, int8Run, int4
RTX 3090 24 GBYesYes2×YesYesYes
RTX 4090 24 GBYesYes2×YesYesYes
RTX 5090 32 GBYesYes2×YesYesYes
L40S 48 GBYesYesYesYesYesYes
A100 40GB 40 GBYesYesYesYesYesYes
A100 80GB 80 GBYesYesYesYesYesYes
H100 SXM 80 GBYesYesYesYesYesYes
H200 141 GBYesYesYesYesYesYes
B200 180 GBYesYesYesYesYesYes

What's specific to Qwen3 1.7B

  • The safetensors hold 2.03B parameters including the tied embeddings, more than the name suggests.
  • Full fine-tuning (about 33 GB before activations) fits on one 48 GB card.
ArchitectureQwen3 1.7B
Parameters2.03B
Layers28
Hidden size2,048
MLP width6,144
Attention heads / KV heads16 / 8 (head dim 128)
Vocabulary151,936 tokens, tied embeddings
Context32K tokens
LicenseApache 2.0
ReleasedApril 2025

Try other settings

Pre-filled for Qwen3 1.7B. Change the sequence length, batch, LoRA rank, context or GPU.

JobMemoryRTX 4090 (24 GB)Smallest single GPU
QLoRA (4-bit base)2K tokens × batch 1, rank 164.7 GBFitsRTX 3090 (24 GB)
LoRA (16-bit base)2K tokens × batch 1, rank 167.2 GBFitsRTX 3090 (24 GB)
Full fine-tuning2K tokens × batch 135 GB2 GPUsA100 40GB (40 GB)
Inference, 16-bit (BF16/FP16)8K-token context × 1 sequence5.0 GBFitsRTX 3090 (24 GB)
Inference, 8-bit8K-token context × 1 sequence3.0 GBFitsRTX 3090 (24 GB)
Inference, 4-bit8K-token context × 1 sequence2.1 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.

Frequently asked questions

Can I fine-tune Qwen3 1.7B on a 24 GB GPU?
On an RTX 3090 or 4090 (24 GB): QLoRA: yes, about 4.7 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 512. LoRA: yes, about 7.2 GB at 2K tokens, batch 1, rank 16; up to 8K-token sequences at batch 1 or batch 5 at 2K, LoRA rank up to 512. Estimates use gradient checkpointing and leave 10% of memory free.
How much VRAM do you need to run Qwen3 1.7B?
About 5.0 GB in 16-bit and 2.1 GB in 4-bit, at 8K tokens of context. The smallest single GPU here for 4-bit is the RTX 3090 (24 GB); for 16-bit, the RTX 3090 (24 GB).
How many GPUs does full fine-tuning Qwen3 1.7B take?
About 35 GB with mixed-precision Adam at 2K tokens, so at least 1 × 80 GB H100 with the training state sharded (FSDP or ZeRO-3). LoRA needs 7.2 GB and QLoRA 4.7 GB.
How long a context can Qwen3 1.7B handle on an 80 GB GPU?
Its KV cache costs 3.8 GB at the full 32K tokens (fp16 cache). On one 80 GB GPU: 32K tokens with 16-bit weights, 32K tokens with 4-bit weights.

Terms used on this page: LoRA, QLoRA, KV cache, gradient checkpointing, quantization. For any model size, use the fine-tuning VRAM calculator; for the method in depth, the Academy lesson Memory first: what hardware can train what.

Sources

  1. [1]Hugging Face. Qwen/Qwen3-1.7B config.json
  2. [2]Alibaba Qwen. Qwen/Qwen3-1.7B model card
  3. [3]Alibaba Qwen. Qwen3 Technical Report (arXiv:2505.09388)
  4. [4]Dettmers et al., 2023. QLoRA: Efficient Finetuning of Quantized LLMs (arXiv:2305.14314)
  5. [5]Hu et al., 2021. LoRA: Low-Rank Adaptation of Large Language Models (arXiv:2106.09685)
  6. [6]Rajbhandari et al., 2019. ZeRO: Memory Optimizations Toward Training Trillion Parameter Models (arXiv:1910.02054)
  7. [7]Korthikanti et al., 2022. Reducing Activation Recomputation in Large Transformer Models (arXiv:2205.05198)
  8. [8]Chen et al., 2016. Training Deep Nets with Sublinear Memory Cost (arXiv:1604.06174)

Architecture read from config.json and parameter count from the Hugging Face safetensors metadata on 15 November 2026.

Find out if you need to fine-tune at all.

Kojen's prompt tuning and model search run for real on your own test cases, on hosted models. Measure the cheaper levers before you rent a GPU.

Start training free