Model Guides
Step-by-step setup guides for specific AI models on dedicated GPU servers. From LLM deployment to vision model hosting and speech model hosting, each guide includes configuration, optimisation tips, and GPU recommendations.
RTX 4090 24GB for Gemma 2 9B: Sliding-Window Attention, 155 t/s and Conversation-Strong Output
Gemma 2 9B-it on the RTX 4090 24GB - FP8 weights at 9.5GB, 155 tokens/sec, with full notes on sliding-window attention, soft-cap logits and the FlashAttention version pinning that matters.
RTX 4090 24GB for Hermes 3 8B: Agent Backend, Tool Calling and Concurrency
Hermes 3 8B on the RTX 4090 24GB - 195 t/s FP8, 1,140 t/s aggregate, 99.1% tool-call adherence, full 128k…
RTX 4090 24GB for Mistral 7B v0.3: Throughput and Sizing
Production deployment guide for Mistral 7B v0.3 on the RTX 4090 24GB: VRAM math, throughput by batch and quant, sliding…
RTX 4090 24GB for Mistral Nemo 12B: 128k Context at FP8 with deep VRAM math
Deep deployment guide for Mistral Nemo 12B on the RTX 4090 24GB - GQA-optimised KV math, full 128k context at…
RTX 4090 24GB for Mistral Small 3 (24B): The Best Mid-Size Dense Model on One Card
Mistral Small 3 24B fits a single RTX 4090 24GB as AWQ INT4 with 32k context, 85 t/s decode, and…
RTX 4090 24GB for Llama 3.1 70B INT4: Best Single-Card 70B Setup
Exhaustive deployment guide for Llama 3.1 70B AWQ INT4 on a single RTX 4090 24GB: line-by-line memory math, max-model-len calculation,…
RTX 4090 24GB for Llama 3.1 8B: Deployment Guide
Production-grade deployment guide for Llama 3.1 8B on the RTX 4090 24GB: VRAM math broken down line by line, throughput…
RTX 4090 24GB for Llama 3.2 Vision 11B: OCR, Multi-Image Reasoning and Latency
Llama 3.2 Vision 11B on the RTX 4090 24GB - 13.5GB FP8 footprint, 220ms image encode, 115 t/s decode, multi-image…
RTX 4090 24GB for Mixtral 8x7B: The Original Open MoE on a Single Card
Mixtral 8x7B AWQ on the RTX 4090 24GB - 14GB MoE weights, 12.9B active per token, 85 t/s decode, 32k…
RTX 4090 24GB for Multimodal LLMs
Vision-language models on the RTX 4090 24GB - Llama 3.2 Vision 11B, Qwen 2.5-VL 7B, LLaVA, image+video understanding throughput, OCR,…
Explore GPU Hosting Solutions
From the blog to your next deployment — pick the right platform for your workload.
Dedicated GPU Hosting
Bare-metal servers with a dedicated GPU, NVMe, full root access, and 1Gbps networking from our UK datacenter.
Browse GPU ServersOpen Source LLM Hosting
Deploy LLaMA, Mistral, DeepSeek, and more on dedicated hardware with no per-token API fees.
Explore LLM HostingVision Model Hosting
Deploy YOLO, PaddleOCR, Stable Diffusion, and other vision models on GPU-accelerated servers.
Explore Vision HostingSpeech Model Hosting
Deploy Whisper, Coqui, Bark, and other speech models with low-latency inference.
Explore Speech HostingMultimodal Model Hosting
Vision-language models, audio-language models — deploy multimodal AI on dedicated GPUs.
Explore MultimodalTokens/sec Benchmarks
Real-world tokens per second data across every GPU we offer, tested on popular LLMs.
View BenchmarksReady to deploy your AI workload?
Dedicated GPU servers from our UK datacenter. NVMe storage, 1Gbps networking, full root access.
65 Victory Avenue
PE7 1XU, England
GPUAIR © All Rights Reserved.