LLM VRAM Calculator

Calculate precise GPU memory requirements for running Large Language Models locally. Select your hardware, configure parameters, get instant results.

Model Configuration

bpw = effective bits per weight including GGUF block metadata overhead

4,096
5124K8K32K64K128K256K

Select Your CPU (for CPU-only inference)

Select Your GPU

RTX 5090
32 GB
RTX 5080
16 GB
RTX 5070 Ti
16 GB
RTX 5070
12 GB
RTX 4090
24 GB
RTX 4080 S
16 GB
RTX 4080
16 GB
4070 Ti S
16 GB
RTX 4070 Ti
12 GB
4070 Super
12 GB
RTX 4070
12 GB
4060 Ti 16
16 GB
4060 Ti
8 GB
RTX 4060
8 GB
3090 Ti
24 GB
RTX 3090
24 GB
3080 Ti
12 GB
RTX 3080 12
12 GB
RTX 3080
10 GB
3070 Ti
8 GB
RTX 3070
8 GB
3060 Ti
8 GB
RTX 3060
12 GB
RTX 6000 Ada
48 GB
RTX A6000
48 GB
RTX A5000
24 GB
A100
80 GB
A100 40
40 GB
H100
80 GB
H200
141 GB
B200
192 GB
1 GPU(s)

Tensor parallelism across identical GPUs. Effective VRAM scales linearly.

Estimated VRAM

4.75 GB
for 7B @ Q4 · 4K context
0% of 24 GB (RTX 4090)
Fits comfortably
Model Weights 3.50 GB
KV Cache (Context) 0.29 GB
Activation Buffer 0.50 GB
CUDA / Driver Overhead ~0.50 GB
Safety Margin 0.46 GB
Tokens/sec est.
~45 t/s
Batch Size est.
~8

Recommendation

Your selected GPU can handle this configuration

How It Works

1. Model Weights

(Params × effective_bpw) ÷ 8
Uses effective bits-per-weight including GGUF K-quant block metadata overhead (e.g. Q4_K_M = 4.83 bpw, not 4.0).

2. KV Cache

2 × layers × kv_dim × ctx × 2B
Uses actual KV head count for GQA models. DeepSeek MLA uses compressed 576-dim latent vectors.

3. Total VRAM

Weights + KV + Activations + CUDA + 5%
Includes CUDA/driver overhead (~0.5-1.5 GB). Training adds gradients + AdamW optimizer states (8B/param).

Popular Configurations

Frequently Asked Questions

Complete Guide to LLM VRAM Requirements

What is VRAM and Why Does It Matter for LLMs?

VRAM (Video Random Access Memory) is the dedicated memory on your graphics card used to store data that the GPU needs to process quickly. When running Large Language Models (LLMs) locally, VRAM is the most critical resource because the entire model must fit into this memory for fast inference. Unlike system RAM, VRAM offers significantly higher bandwidth (up to 1 TB/s on modern GPUs), which is essential for the massive matrix multiplications that LLMs perform.

If a model doesn't fit in VRAM, the system will fall back to using system RAM, which is 10-50x slower. This results in extremely slow token generation (often less than 1 token per second) or complete failure to load the model.

What is LLM Quantization?

Quantization reduces the precision of model weights to save memory. GGUF K-quant formats add block metadata (scale/min values per 256-weight block), so the effective bits-per-weight (bpw) is always higher than the nominal bit depth. AWQ and GPTQ use group-based quantization with ~3% overhead from scale factors.

Format Effective bpw Quality 7B Weights
FP16 / BF1616.00 bpwOriginal quality~14.0 GB
Q8_0 (GGUF)8.50 bpwNear-lossless~7.4 GB
Q6_K (GGUF)6.57 bpwExcellent~5.7 GB
Q5_K_M (GGUF)5.67 bpwVery good~5.0 GB
Q4_K_M (Recommended)4.83 bpwGood~4.2 GB
AWQ / GPTQ (4-bit)4.00 bpwGood~3.5 GB
Q3_K_M (GGUF)3.89 bpwAcceptable~3.4 GB
IQ2_XS (GGUF)2.05 bpwLow / constrained HW~1.8 GB

Q4 quantization (4-bit) is the sweet spot for most users, offering a good balance between model quality and memory usage. Modern quantization methods like GGUF (used by llama.cpp) and AWQ (used by vLLM) preserve model quality remarkably well even at 4-bit precision.

Understanding Context Window and KV Cache

The context window determines how many tokens the model can "remember" at once. A larger context window allows the model to process longer documents and maintain coherence over extended conversations. However, the KV cache (Key-Value cache) that stores attention states grows linearly with context length.

For a 7B model with 32 layers, each additional 1,000 tokens of context requires approximately 0.06 GB of additional VRAM. At 128K context, this alone can add 7-8 GB of VRAM usage on top of the model weights.

GPU Recommendations for LLM Inference

Budget (8-12 GB VRAM)

RTX 3060 12GB, RTX 4060, RTX 4070. Can run 7B models at Q4 with moderate context. Good for experimentation and small-scale use.

Mid-Range (16 GB VRAM)

RTX 4070 Ti Super, RTX 4080. Handles 7B-13B models comfortably at Q4 with large context windows. Ideal for most developers.

High-End (24 GB VRAM)

RTX 3090, RTX 4090. Can run 30B-34B models at Q4. The best consumer option for serious LLM work. RTX 3090 offers the best value.

Enterprise (48+ GB VRAM)

RTX 6000 Ada, A100, H100. Required for 70B+ models or training. Consider cloud instances (AWS, GCP, Lambda Labs) for cost efficiency.

Tips to Reduce VRAM Usage

Popular LLM Models and Their VRAM Requirements

Here's a quick reference for popular models at Q4 quantization with 4K context:

Model Parameters VRAM (Q4) Min GPU
Llama-3.2 3B3B~2.3 GBRTX 3060
Phi-3 Mini3.8B~2.8 GBRTX 3060
Mistral 7B7B~5.1 GBRTX 3060
Llama-3 8B8B~5.6 GBRTX 3060
Qwen2.5-32B32B~20 GBRTX 3090/4090
Llama-3 70B70B~46 GB2x RTX 4090
DeepSeek V3/R1671B (MoE)~370 GB8x H100 / Cloud