Skip to main content
Vantaige

Hardware Sizing & Buying Guide

The Ultimate GPU Guide for Local LLMs

Forget raw gaming benchmarks. Learn why memory bandwidth dictates your tokens-per-second, which VRAM tier you need for each model class, and how Apple Silicon compares to Nvidia.

1. The Golden Rule of AI Hardware: Bandwidth Over Compute

In 3D gaming and video rendering, raw compute performance (TFLOPS) and CUDA core count are the primary metrics. In large language model inference, the bottleneck is almost 100% memory bandwidth.

During token generation, the GPU cores must read every single parameter of the model from memory once for every single token generated. The mathematical speed limit is defined by a simple formula:

Theoretical Max Speed (Tokens/Sec) = Memory Bandwidth (GB/s) ÷ Model Weight Size (GB)

Consider a 40 GB model (e.g. 70B at Q4_K_M):

System DDR5 RAM (~60 GB/s)~1.5 tokens/sec

PCIe bus + slow RAM bandwidth makes generation virtually unusable.

M3 Max Mac (~300 GB/s)~7.5 tokens/sec

Very usable for reading and coding; completely silent and power-efficient.

RTX 4090 (1,008 GB/s)~25.2 tokens/sec

Ultra-fast interactive chatting, but requires dual cards to hold 40GB.

2. The 4 VRAM Capacity Tiers: What Actually Runs?

Entry Level8 GiB VRAM
8GB Tier (RTX 3060 Laptop / 4060 8GB)

What Runs: 3B models at FP16, or 7B/8B models at Q4_K_M with short context (up to 4k–8k tokens).

What Fails: 14B models, 32B models, long 32k context, and concurrent browser tasks.

Best For: Students, hobbyists, and experimenting with small lightweight SLMs.
The Sweet Spot12–16 GiB VRAM
12GB–16GB Tier (RTX 3060 12GB / 4060 Ti 16GB)

What Runs: 8B models at full unquantized FP16, 14B models at Q5_K_M with 32k context, or small MoEs.

Hardware Highlights: The RTX 3060 12GB remains the best budget AI card on the market ($280). The RTX 4060 Ti 16GB is the cheapest 16GB entry point.

Best For: Daily coding assistants, local RAG pipelines, and standard productivity.
Enthusiast Standard24 GiB VRAM
24GB Tier (RTX 3090 / 4090)

What Runs: 27B to 32B models at Q4/Q5, dense 14B at FP16, and massive 64k+ context windows with zero compromise.

Pro Tip: A used RTX 3090 24GB (~$700) delivers 90% of the AI value of a $1,700 RTX 4090 because both have identical 24GB capacity.

Best For: Serious developers, local fine-tuning (LoRA/QLoRA), and 32B coding models.
70B Paradise48–128 GiB Memory
48GB–96GB Tier (Dual 3090s / Mac Studio)

What Runs: Full Llama 3.3 70B and Qwen 2.5 72B at Q4_K_M or Q8_0 completely in memory at snappy interactive speeds.

Build Choices: Dual RTX 3090s in a PC workstation ($1,500 total build) OR an Apple Mac Studio M2/M3 Max/Ultra (64GB to 192GB).

Best For: Enterprise teams, power users wanting GPT-4-class reasoning completely offline.

3. Apple Silicon Mac vs. Nvidia PC: Which Should You Buy?

Apple Silicon (Unified Memory)
  • Massive VRAM Capacity: Up to 128GB or 192GB unified RAM on a single desktop.
  • Near Silent & Low Power: Draws 80–120W under full load compared to 850W+ for dual-GPU PCs.
  • Slower Bandwidth: 300 to 800 GB/s means lower peak tokens/second than high-end Nvidia cards.
  • Ecosystem: Native MLX and llama.cpp Metal backends run great; vLLM/CUDA tools are unavailable.
Nvidia Workstation (CUDA Ecosystem)
  • Maximum Generation Speed: 1,008 GB/s bandwidth on RTX 4090 produces blazing fast 30–60 tok/s.
  • Industry Standard: 100% compatibility with CUDA, PyTorch, vLLM, TensorRT-LLM, FlashAttention-2.
  • VRAM Wall: Hard 24GB ceiling per consumer card. Running 70B requires complex multi-GPU builds.
  • Power & Heat: Requires 1000W+ power supplies, high-airflow cases, and dedicated cooling.

4. Building a Dual-GPU Workstation for 70B Models

If your goal is to run dense 70B models at full speed on a PC, a dual-card configuration is the most cost-effective path:

1. Motherboard PCIe Lanes

Ensure your motherboard supports x8 / x8 bifurcation on CPU lanes, with at least 3 to 4 slot spacing between cards for cooling.

2. Power Supply (PSU)

Budget at least a 1,200W or 1,500W Platinum/Titanium PSU to handle the transient power spikes of dual RTX 3090s or 4090s.

3. Inference Engine

Use vLLM with Tensor Parallelism (`--tensor-parallel-size 2`) or llama.cpp multi-GPU to shard weights across both cards.

Frequently Asked Questions

Because in local LLMs, VRAM capacity is king. A 24GB RTX 3090 can fit 32B models at Q4 or dense 14B models at FP16 with a 32k context window completely on-device. The 16GB card cannot fit 32B models at full speed and is forced to offload layers to slow system RAM. Furthermore, the RTX 3090 features 936 GB/s memory bandwidth across a 384-bit bus, outperforming the RTX 4080's 717 GB/s.

8GB is enough to learn and run 7B/8B models at Q4_K_M with short context (up to 4k–8k tokens). However, it is tight for modern coding agents, large document analysis, or running a desktop display simultaneously. If buying new, 12GB to 16GB is the recommended minimum floor.

Yes! Apple Silicon's unified memory architecture allows the GPU cores to access the entire RAM pool without PCIe bottlenecks. A 64GB Mac Studio can run Llama 3.3 70B at Q4_K_M comfortably (~45 GB total) at 15–22 tokens/second. A 128GB or 192GB Mac Studio can run 70B at full FP16 or run large MoE models that would otherwise require $10,000+ in PC hardware.

When a model fits 100% in VRAM, data moves at 500 to 1,000 GB/s. When layers are offloaded to system RAM, data must travel over the PCIe slot (PCIe 4.0 x16 tops out at ~31.5 GB/s, and DDR5 RAM runs at 60–90 GB/s). Because generation is memory-bandwidth bound, offloading even a few layers often drops throughput from 45 tok/s to 3–6 tok/s.

For 70B models, yes! A single RTX 4090 has 24GB of VRAM and cannot fit a quantized 70B model without severe offload. Dual RTX 3090s give you 48GB of pooled VRAM (via Tensor Parallelism in vLLM or layer splitting in llama.cpp), allowing you to run 70B Q4_K_M completely in VRAM at 25+ tokens/second for roughly the same total price as one 4090.

Test your exact GPU against the model catalog

Select your GPU in our VRAM Calculator to see which models fit, or compare cloud rental prices.

Open VRAM Calculator

Knowledge & Deep Dives

VRAM & Local AI Knowledge Hub

Everything you need to know about sizing hardware, understanding model architecture, and running local LLMs without out-of-memory crashes.

50+ Terms8 min reference
Master the technical terminology: GGUF vs. Safetensors, K-quants, KV cache, GQA, MoE total vs. active parameters, and Apple unified memory.
Interactive search & filter
Practical hardware takeaways
Alphabetical index
Explore guide
Interactive Widget6 min read
Why context length explodes memory usage, the exact KV cache formula, MHA vs. GQA vs. MLA (DeepSeek), and how to calculate cache in seconds.
Interactive KV sizing widget
MHA vs GQA vs MLA math
Quantized cache (FP8 / Q4)
Explore guide
Config Decoder7 min read
Inspect Hugging Face repositories like an ML engineer: decode config.json, identify true context windows, and avoid the MoE parameter trap.
Interactive config.json inspector
MoE active vs total trap
GGUF naming decoded
Explore guide
License Matrix5 min read
The legal and practical differences between true OSI Open Source AI and open-weight models like Llama 3, DeepSeek, Qwen 2.5, and Gemma 2.
Full license comparison matrix
Commercial usage caps
Synthetic distillation rules
Explore guide