Skip to content

VRAM reference

Llama 3.1 70B on a B200

Yes, you can fine-tune Llama 3.1 70B on one B200 (180 GB) with QLoRA or LoRA. 4-bit inference: yes, 40 GB of weights, room for 128K tokens of context or 91 concurrent 4K-token sequences.

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

Verdict by method

Fine-tuningMemory at 2K tokensOne B200?Longest sequence (batch 1)Largest batch at 2KLargest LoRA rank
QLoRA (4-bit base)48 GBYes32K tokens21512
LoRA (16-bit base)150 GBYes4K tokens364
Full fine-tuning1,134 GBNo——n/a
InferenceMemory at 8KOne B200?Longest contextConcurrent 4K sequences
16-bit (BF16/FP16)144 GBYes32K tokens15
8-bit73 GBYes128K tokens68
4-bit42 GBYes128K tokens91
  • QLoRA: yes, about 48 GB at 2K tokens, batch 1, rank 16; up to 32K-token sequences at batch 1 or batch 21 at 2K, LoRA rank up to 512.
  • LoRA: yes, about 150 GB at 2K tokens, batch 1, rank 16; up to 4K-token sequences at batch 1 or batch 3 at 2K, LoRA rank up to 64.
  • Full fine-tuning: no, it needs about 1,134 GB at 2K tokens, more than the 180 GB available.

Where the memory goes

At 2,048 tokens per sequence, batch 1, LoRA rank 16, gradient checkpointing on
MethodWeightsTraining stateActivationsTotalOn one B20080 GB GPUs needed
QLoRA (4-bit base)40 GB3.3 GB5.6 GB48 GBFits1
LoRA (16-bit base)141 GB3.3 GB5.6 GB150 GBFits3
Full fine-tuning141 GB988 GB5.6 GB1,134 GBDoesn't fit16
One sequence, fp16 KV cache
PrecisionWeightsKV cache at 8KTotalAt 128K contextOn one B200
16-bit (BF16/FP16)141 GB2.7 GB144 GB184 GBFits
8-bit71 GB2.7 GB73 GB114 GBFits
4-bit40 GB2.7 GB42 GB83 GBFits

The cheapest way to run it

Renting a B200 costs $7.99 an hour on Runpod Secure Cloud ($5.98 on Runpod Community Cloud). LoRA on it processes about 3,189 tokens a second by our estimate, so 10 million training tokens take 52 minutes and cost about $6.96.

Job (default settings)MemoryCheapest rental that fitsPer hour
QLoRA fine-tuning48 GBA100 80GB$2.19
LoRA fine-tuning150 GB4 × L40S$4.36
4-bit inference42 GBL40S$1.09
16-bit inference144 GB2 × A100 80GB$4.38

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

JobMemoryB200 (180 GB)Smallest single GPU
QLoRA (4-bit base)2K tokens × batch 1, rank 1648 GBFitsA100 80GB (80 GB)
LoRA (16-bit base)2K tokens × batch 1, rank 16150 GBFitsB200 (180 GB)
Full fine-tuning2K tokens × batch 11,134 GB8 GPUsNone: needs several
Inference, 16-bit (BF16/FP16)8K-token context × 1 sequence144 GBFitsB200 (180 GB)
Inference, 8-bit8K-token context × 1 sequence73 GBFitsH200 (141 GB)
Inference, 4-bit8K-token context × 1 sequence42 GBFitsL40S (48 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 B200's dense BF16 peak of 2,250 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 Llama 3.1 70B on a B200?
QLoRA: yes, about 48 GB at 2K tokens, batch 1, rank 16; up to 32K-token sequences at batch 1 or batch 21 at 2K, LoRA rank up to 512. LoRA: yes, about 150 GB at 2K tokens, batch 1, rank 16; up to 4K-token sequences at batch 1 or batch 3 at 2K, LoRA rank up to 64. Full fine-tuning: no, it needs about 1,134 GB at 2K tokens, more than the 180 GB available.
Can I run Llama 3.1 70B on a B200?
16-bit (BF16/FP16) inference: yes, 141 GB of weights, room for 32K tokens of context or 15 concurrent 4K-token sequences. 8-bit inference: yes, 71 GB of weights, room for 128K tokens of context or 68 concurrent 4K-token sequences. 4-bit inference: yes, 40 GB of weights, room for 128K tokens of context or 91 concurrent 4K-token sequences.
What does fine-tuning Llama 3.1 70B on a B200 cost?
With LoRA at our planning rate of 40% utilisation, about 52 minutes for 10 million training tokens, around $6.96 at $7.99 an hour (Runpod Secure Cloud, November 2026).

Sources

  1. [1]Hugging Face. meta-llama/Llama-3.1-70B-Instruct config.json
  2. [2]Meta. meta-llama/Llama-3.1-70B-Instruct model card
  3. [3]Meta. The Llama 3 Herd of Models (arXiv:2407.21783)
  4. [4]NVIDIA. B200 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 Llama 3.1 70B is worth the GPU time.

Start training free