TLDR Tech / DevOps
How we built the most performant DeepSeek V3.2, MiniMax-M2.5 and Qwen 3.5 397B on DigitalOcean NVIDIA HGX™ B300 GPU Droplets
DigitalOcean announced general availability of DeepSeek V3.2, MiniMax-M2.5, and Qwen 3.5 397B on its Serverless Inference platform, which achieved the fastest output speeds among all tested providers, with DeepSeek V3.2 delivering 230 tokens per second and sub-1-second time to first token for 10,000 input tokens. The performance was achieved through NVIDIA's HGX B300 GPUs with 288GB memory, NVFP4 quantization for 1.8x smaller memory footprint, and custom optimizations to the vLLM serving framework in collaboration with Inferact.
Read the full story