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.

GeForce RTX 4090 vs GeForce RTX 5090 for local AI

Which is the better card for running LLMs locally? The verdict comes down to memory bandwidth and VRAM — here's the head-to-head with real, computed tokens/sec, not marketing numbers.

Verdict

For local LLMs the GeForce RTX 5090 is the stronger pick — about 78% faster on typical models thanks to its 1792 GB/s of memory bandwidth, and its 32GB of VRAM runs a 123B model.

Head to head

	GeForce RTX 4090	GeForce RTX 5090
VRAM	24 GB	32 GB
Memory bandwidth	1008 GB/s	1792 GB/s
FP16 compute	83 TFLOPS	105 TFLOPS
Largest LLM	123B	123B
Launch price	$1599	$1999
Released	2022	2025

Real LLM speed (tokens/sec, Q4_K_M)

Model	GeForce RTX 4090	GeForce RTX 5090	Diff
Qwen3 8B	152.5	271.1	+78%
Qwen3 14B	89.1	158.5	+78%
Qwen3 32B	40.1	71.2	+78%

Single-stream decode at 8K context, Q4_K_M, computed from each card’s real memory bandwidth. A model that spills to system RAM on the smaller card shows a larger gap.

Frequently asked

Is the GeForce RTX 5090 faster than the GeForce RTX 4090 for LLMs?

On average across typical models the GeForce RTX 5090 is about 78% faster. Local generation is memory-bandwidth bound, and the GeForce RTX 5090 has the higher bandwidth (1792 GB/s), which is what usually decides it.

Is upgrading from the GeForce RTX 4090 to the GeForce RTX 5090 worth it for AI?

On a 14B model at Q4 the GeForce RTX 5090 is about 78% faster (89.1 → 158.5 tok/s). It also has 8GB more VRAM, letting you run larger models.

Which has more VRAM, the GeForce RTX 4090 or GeForce RTX 5090?

The GeForce RTX 5090 has more — 32GB vs 24GB. More VRAM means larger models fit fully in memory instead of spilling to slow system RAM.