RTX 3090 vs RTX 4090 for Local LLM Inference: 2025 Comparison

The RTX 3090 and 4090 both offer 24GB VRAM — the threshold where most 13B-34B models fit entirely in GPU memory. The 4090 is significantly faster but costs $1700-2000 new. A used 3090 can be had for $600-800. Is the 4090's extra speed worth 2-3x the cost for local LLM users? The answer is nuanced.

For LLM inference specifically: The 3090 and 4090 have similar memory bandwidth (936 GB/s vs 1,008 GB/s) — the factor that actually limits inference speed. The 4090's compute advantage matters far less than you'd expect. See detailed benchmarks at LLMHardware.io.

Benchmark Comparison

On Llama 3.1 13B Q4_K_M: RTX 3090 achieves ~55-60 tokens/second, RTX 4090 achieves ~70-75 tokens/second — about 25% faster. On a 34B Q4_K_M (fits in 24GB): 3090 gets ~22 t/s vs 4090's ~28 t/s. For inference, the 4090 is faster but not dramatically so. For fine-tuning with QLoRA, the gap widens significantly because the 4090's faster compute matters more.

The Value Case for RTX 3090

At $650-800 used, the RTX 3090 is the best price-per-VRAM card available. You get 24GB for less than many 12GB cards cost new. The only caution: 3090s run very hot (330W TDP) and older cards in the used market may have worn thermal pads. Factor in a repaste job ($50-80 at a shop) for long-term reliability. The 3090 Ti offers slightly more bandwidth and compute for ~$100 more.

When the 4090 Is Worth It

Buy the 4090 if: you fine-tune models regularly, you run high-concurrency serving (the faster compute helps with batching), or you want the fastest single-GPU inference available. The 4090's lower power consumption (450W vs 350W rated, but 4090 throttles more efficiently under load) and better thermals also make it preferable for always-on server use. For occasional local chat use, the 3090 is the rational choice.

Comments

Post a Comment

Popular posts from this blog

AMD RX 9070 XT for Local LLMs: ROCm Performance Guide (2026)

The AMD RX 9070 XT launched in early 2026 at $599 with 16GB GDDR6 and 648 GB/s memory bandwidth. For local LLM users who want a mid-range GPU without NVIDIA prices, this changes the calculus significantly. Here is what it actually delivers. RX 9070 XT Specs vs Competition GPU VRAM Bandwidth Price LLM Score RX 9070 XT 16 GB 648 GB/s $599 ~60 tok/s RTX 4070 Ti Super 16 GB 672 GB/s $799 ~62 tok/s RTX 4080 Super 16 GB 736 GB/s $999 ~68 tok/s RTX 4090 24 GB 1,008 GB/s ~$1,100 ~90 tok/s At $599 with 16GB VRAM, the RX 9070 XT is the best value GPU for local LLMs in the mid-range tier in 2026. See the full comparison at LLMHardware.io's AMD GPU guide . ROCm Setup on Linux (2026) # Ubuntu 22.04/24.04 sudo apt install rocm-hip-libraries hip-base # or use the AMD installer: wget https://repo.radeon.com/amdgpu-install/6.3/ubuntu/jammy/amdgpu-install_6.3.60300-1_all.deb sudo dpkg -i amdgpu-install*.deb sudo amdgpu-install --usecase=rocm,lrt,openclsdk # Verify ROCm sees your GPU r...
Read more

llama.cpp flash attention: how to enable and benchmark

Image
Flash attention computes attention with far less memory traffic. The math is identical; the memory movement is not. In llama.cpp it matters most in two situations: long context windows and tight video memory (VRAM). With flash attention enabled, the key-value cache shrinks substantially. A 32k-token context window becomes realistic on a 16GB card, and the prompt-processing phase speeds up, so long documents get ingested faster. The catch is compatibility: not every model and quantization combination supports it, and on some setups the speedup is modest. If you run short chats on a strong GPU, you may barely notice. Where it shines is retrieval-augmented generation (RAG) and long-document work on mid-range cards. Enable it, compare your tokens per second, and keep it if your workload benefits. For the common case of pasting a long PDF and asking questions, it is close to free performance. See how context length changes VRAM needs .
Read more

SillyTavern with Local LLMs: Complete Setup for AI Companions

Image
SillyTavern is the most feature-rich frontend for character-based AI conversations. What began as an AI roleplay tool has evolved into a powerful platform with persona management, memory systems, extensive prompt engineering tools, and connections to every major local inference backend. Here's how to set it up with local models. Quick connect: SillyTavern connects to Ollama, KoboldCpp, llama.cpp server, and TabbyAPI (ExLlamaV2) out of the box. Any of the hardware setups at LLMHardware.io provides a solid backend for SillyTavern. Installation Install via git: git clone https://github.com/SillyTavern/SillyTavern && cd SillyTavern && npm install then node server.js . Access at localhost:8000. For persistent installation, use the included start.sh script. SillyTavern uses Node.js and runs on any platform. The companion app (STMP) enables mobile access to your SillyTavern instance via Tailscale or local network. Character Cards and the TavernAI Format Chara...
Read more