NVIDIA H200 reports

kv:

fp8_e4m3

User reports DeepSeek V4 Flash with DSpark speculative decoding on an HGX-H200 with 4 GPUs at TP=4 via SGLang. Setup compares the marlin and flashinfer MoE backends. DSpark is faster than EAGLE: 3.2x at bs=1 and +46% throughput at bs=24. The user reports TTFT and accept length benchmarks.

User pretrained a 500M parameter LLM and a 330M image generator from scratch on 8xH200 from modal.com. The total cost was $800, and GGUF weights are included. User mentions pretraining a 1B model next.