Cheapest Way to Run DeepSeek-V3 (671B MoE) in the Cloud
DeepSeek-V3 introduces revolutionary architectural innovations including Multi-Head Latent Attention (MLA) to compress KV cache by 93% and auxiliary-loss-free load balancing. Self-hosting the full checkpoint requires high-bandwidth multi-GPU clusters, while serverless API providers offer industry-disrupting $0.14/1M token pricing.
Memory & VRAM Requirements Across Quantization Bit-Depths
370 GB (AWQ/GPTQ, runs on 4x H100 80GB or 8x RTX 4090)
680 GB (requires 8x H100 SXM5 80GB or 8x A100 80GB)
1,350 GB (requires 16x H100 80GB cluster)
Cheapest Dedicated Cloud GPUs for DeepSeek-V3 (671B MoE)
Full root SSH container access with CUDA pre-loaded and fast weights downloading.
InfiniBand 3.2 Tbps cluster fabric with ultra-low tensor parallel latency
Pre-configured SGLang / vLLM multi-GPU distributed container
Lowest-cost 8-GPU SXM4 cluster for dedicated private hosting
Cheapest Serverless Token Endpoints for DeepSeek-V3 (671B MoE)
1-Click Docker Launch Command (SGLang)
python3 -m sglang.launch_server \ --model-path deepseek-ai/DeepSeek-V3 \ --tp 8 \ --trust-remote-code \ --port 30000
Dedicated Cloud Pod vs Serverless API: Cost Break-Even for DeepSeek-V3 (671B MoE)
Renting a dedicated instance (e.g., N/A for full 671B (Use DeepSeek-V3 Serverless API) at $1.85/hr) provides 100% reserved GPU memory, full root SSH container control, customizable prompt caches, and zero API rate limits.
Serverless endpoints automatically scale down to zero when no requests are active, completely eliminating idle GPU and stopped volume storage costs.
Production Serving Architecture & Throughput Optimization for DeepSeek-V3 (671B MoE)
Configure --gpu-memory-utilization 0.95 in vLLM to reserve maximum memory pages for non-contiguous KV cache, eliminating memory fragmentation.
Enable --kv-cache-dtype fp8 to halve KV cache footprint, doubling the maximum concurrent context window without degrading reasoning quality.
Batch incoming user prompts at the token level rather than sequence level, sustaining 4x higher throughput during high concurrency.
Get DeepSeek-V3 (671B MoE) Price-Drop & Token Rate Alerts
We monitor DeepSeek-V3 (671B MoE) inference endpoints across DeepInfra, Together AI, Groq, and Cerebras 24/7. Get notified when token rates drop or new serverless endpoints launch.
Frequently Asked Questions: Running DeepSeek-V3 (671B MoE)
What is the cheapest way to access DeepSeek-V3 in production?
Calling DeepInfra's serverless API ($0.14/1M prompt, $0.28/1M completion) is dramatically cheaper than self-hosting unless your application processes more than 400 million tokens per month continuously.
What GPU cluster is needed to self-host DeepSeek-V3 at full speed?
An 8x H100 SXM5 node with NVLink 4 and FP8 support is required to achieve >60 tokens/second per stream. Nebius AI and RunPod provide turnkey 8x H100 pods starting at $19.60–$21.52/hr.