The DeepSeek Trial: I rented two Blackwells. One 180K request filled them.
DeepSeek V4 Flash · 2× RTX PRO 6000 Blackwell · TP2 · 192 GB total VRAM
The custom TP2 stack worked and write-heavy throughput looked genuinely promising. Then one 180K request hit the memory ceiling on roughly $25,000 of hardware.