Which LLM can my PC run?

It depends on your GPU's VRAM and memory bandwidth. As a rule of thumb, an 8 GB card runs 7–8B models at Q4, 12 GB runs up to ~14B, 16–24 GB runs 24–32B, and 48 GB+ (or Apple Silicon unified memory) runs 70B models. The Local AI Advisor detects your exact hardware and lists every model that fits, how fast it runs, and the best quantization to use.

How much VRAM do I need to run an LLM?

A rough rule: model VRAM ≈ parameters (in billions) × bits-per-weight ÷ 8, plus the KV cache for your context and ~1 GB of overhead. An 8B model at Q4_K_M needs about 4.8 GB of weights, so ~6–7 GB total — comfortable on an 8 GB card. A 70B model at Q4 needs roughly 40 GB. The advisor computes the exact footprint for any model, quant and context length.

What GPU do I need to run a 70B model locally?

A 70B model at Q4_K_M needs roughly 40 GB of memory, so a 48 GB card (like an RTX A6000), two 24 GB GPUs, or Apple Silicon with 64 GB+ of unified memory can run it fully in memory. On a 24 GB card it only runs by offloading part of the model to system RAM, which drops speed sharply — a 24–32B model is the sweet spot for a single 24 GB GPU.

How fast will an LLM run on my hardware (tokens per second)?

Single-stream text generation is memory-bandwidth bound: each token reads the active weights once, so tokens/sec ≈ usable memory bandwidth ÷ active-weight bytes. For example an RTX 4090 (~1008 GB/s) runs an 8B model at Q4 around 120+ tokens/sec — far faster than reading speed. The advisor estimates realistic tokens/sec for every GPU, model and quantization.

Do I need a GPU to run local AI?

No — you can run smaller models on the CPU alone, but it is much slower because system RAM bandwidth is far lower than GPU VRAM bandwidth. A dedicated GPU, or Apple Silicon with its fast unified memory, is strongly recommended for anything above ~3B parameters if you want a usable, interactive speed.

Which quantization should I use?

Q4_K_M is the default sweet spot — about 97.5% of full quality at ~4.8 bits per weight. Use Q5_K_M or Q6_K for near-lossless output if the model still fits in memory, and only drop to Q3, IQ3 or Q2 when you need to squeeze a larger model into limited VRAM.

Is Ollama or LM Studio faster for local LLMs?

Throughput is nearly identical — both run llama.cpp / GGUF under the hood, so speed is set by your hardware and quantization, not the app. Choose Ollama for CLI and automation, LM Studio for a friendly GUI (and its MLX backend on Apple Silicon), or vLLM / TabbyAPI when you need to serve many concurrent requests.

Can my laptop run local LLMs?

Yes. A laptop with a 6–8 GB dedicated GPU runs 7–8B models comfortably, and Apple Silicon MacBooks with 16 GB+ of unified memory run 8–14B models well (larger models with more memory). The advisor detects laptop GPUs and integrated graphics, and estimates the slower CPU-class speed when there is no dedicated GPU.

Why do Mixture-of-Experts models like Qwen3-30B-A3B feel so fast?

MoE models only activate a fraction of their parameters per token — Qwen3-30B-A3B activates about 3B of its 30B parameters. Because generation speed depends on active-weight bytes, it decodes roughly as fast as a 3B model while answering with 30B-scale quality, which is why it feels much quicker than a dense 32B model.

What AI models can the RTX 6000 Ada run?

The RTX 6000 Ada pairs 48GB of VRAM with 960 GB/s of bandwidth — and bandwidth is what sets local LLM speed. Here's exactly what it can run, how fast, and at which quantization. Every number below is computed, not guessed.

At a glance

VRAM

48GB

Bandwidth

960GB/s

Runnable models

15/ 15

Largest model

70BIQ4_XS

Top picks for this card

Best for coding

Qwen2.5-Coder 32B

IQ4_XS · 43 tok/s · 19 GB

Best for reasoning

DeepSeek-R1-Distill 32B

IQ4_XS · 43 tok/s · 19 GB

Best for chat

Qwen3 32B

IQ4_XS · 43 tok/s · 19 GB

Every model on the RTX 6000 Ada

Model	Fit	Best quant	Memory	Speed
Llama 3.2 3B 3.2B · Llama	Perfect	FP16 / BF16	7.8 GB	116.6 tok/s
Qwen3 4B 4B · Qwen	Perfect	FP16 / BF16	9.4 GB	93.2 tok/s
Mistral 7B v0.3 7.2B · Mistral	Perfect	FP16 / BF16	16 GB	52.7 tok/s
Qwen2.5-Coder 7B 7B · Qwen	Perfect	FP16 / BF16	15 GB	54.4 tok/s
Llama 3.1 8B 8B · Llama	Perfect	FP16 / BF16	18 GB	47.6 tok/s
Qwen3 8B 8B · Qwen	Perfect	FP16 / BF16	18 GB	47.5 tok/s
Gemma 3 12B 12B · Gemma	Perfect	Q8_0	15 GB	57.7 tok/s
Phi-4 14B 14B · Phi	Perfect	Q8_0	17 GB	50.2 tok/s
Qwen3 14B 14B · Qwen	Perfect	Q8_0	17 GB	50.2 tok/s
DeepSeek-R1-Distill 14B 14B · DeepSeek	Perfect	Q8_0	17 GB	49.9 tok/s
Mistral Small 3 24B 24B · Mistral	Perfect	Q5_K_M	19 GB	43.6 tok/s
Gemma 3 27B 27B · Gemma	Perfect	Q4_K_M	19 GB	44.7 tok/s
Qwen3 30B-A3B (MoE) 30.5B · Qwen	Perfect	Q8_0	34 GB	217.9 tok/s
Qwen3 32B 32B · Qwen	Perfect	IQ4_XS	19 GB	43 tok/s
Llama 3.3 70B 70B · Llama	Recommended	IQ4_XS	41 GB	19.8 tok/s

Fit and speed assume a 8K context on 48GB VRAM with 32GB system RAM for overflow. Tokens/sec is single-stream decode. Click any model for its full hardware requirements.

Frequently asked

How many local LLMs can the RTX 6000 Ada run?

15 of the 15 models we test fit on the RTX 6000 Ada's 48GB of VRAM at a sensible quantization. The largest is Llama 3.3 70B (70B) at IQ4_XS, generating about 19.8 tokens/sec.

How fast is the RTX 6000 Ada for local AI?

Single-stream generation is memory-bandwidth bound, and the RTX 6000 Ada has 960 GB/s. On an 8B model at Q4 that translates to roughly 47.5 tokens/sec — comfortably faster than reading speed.

What should I run the RTX 6000 Ada with?

Ollama or LM Studio (CUDA) for one-click setup, or vLLM/TabbyAPI for serving.

Can the RTX 6000 Ada run 70B models?

Yes — a 70B model runs, but it fits comfortably in memory.