VRAM Calculator for AI Models
How much VRAM do you need?
This free VRAM calculator shows exactly how much GPU memory a model needs. Pick a model and quantization level, and see if it fits your hardware in seconds.
Pick a model
Search by name, or scroll the list. Custom param count at the bottom of the panel.
QuantizationQuantization compresses model weights to use less memory. Lower quant = less VRAM but slightly lower quality. Q4_K_M is the most popular choice for models 14B+.
Recommended for models 14B and above. Best balance of size and quality.
Context lengthHow many tokens the model can see at once. Larger context = more VRAM for KV cache. 32K is the sweet spot for most coding tasks. 8K is too short for repo-level work.
Pick a model or enter a parameter count to see the VRAM estimate.
Frequently asked questions
How much VRAM do I need to run a local LLM?+
As a rule of thumb, budget about 0.5 to 0.6 GB of VRAM per billion parameters at Q4_K_M quantization, then add extra for KV cache and system overhead. A 7B model needs roughly 5-6 GB, a 27B model needs roughly 17-18 GB, and longer context windows push the number higher. Use the calculator above for an exact estimate instead of the rule of thumb.
What does this VRAM calculator do?+
It estimates the VRAM a local LLM needs based on three inputs: model size, quantization level, and context length. The result breaks down model weights, KV cache, and overhead separately, then checks the total against common consumer and workstation GPUs so you can see if a model fits your hardware.
Is this VRAM calculator free?+
Yes. It runs entirely in your browser, with no sign-up and no account required. It is one of several free tools built for anyone evaluating local AI models, whether or not they use Bodega One Code.
Does quantization change how much VRAM a model needs?+
Yes, significantly. Lower-bit quantization uses less VRAM at a small quality cost. Q4_K_M is the most common choice for models 14B and above. For models under 8B, Q4 quantization causes a noticeable accuracy drop on coding tasks, so Q6_K or Q8_0 is a better trade-off if you have the VRAM to spare.
Can I calculate VRAM for a model that isn't on the list?+
Yes. Scroll to the bottom of the model dropdown and enter a custom parameter count in billions. The calculator applies the same quantization and KV cache math to your custom figure.
Every model on this page runs in Bodega One Code.
Pick a model, connect a provider, start coding. No config files. Free for personal use.
Download FreeEstimates based on published model architectures and quantization specifications. Actual usage may vary by 1-2 GB depending on runtime, driver version, and system configuration. KV cache calculated at FP16. VRAM figures do not include memory used by other applications. Last verified March 2026.