The GPU rig
One 96GB Blackwell + two 3090s for KV cache: the tinkerer's flagship at half the price.
The path the 'three 170hx' commenters were pointing at, one generation newer. A 96GB RTX PRO 6000 runs the 100B-class models with room left, and stacked cheap 3090s act as extra KV-cache memory so long contexts do not strangle the big card. Think of it as a pickup truck: louder than the family car, hauls anything, and the upgrade path is a PCIe slot, not a new vehicle. You are assembling: two cards, a case that fits both, and a power plan.
You have built a PC before and enjoyed it. You want CUDA the day a model drops, you quantize models yourself, and you treat thermal design as part of the job. If your main use is image generation, video work, or fast batch processing, the NVIDIA stack is the whole point.
last check: Sep 29, 2026
The parts
What it runs
The 400B+ MoE class at usable quality: GLM 5.3 full is 371GB at 4-bit. Kimi K3 at 2.8T is 700GB at 2-bit, which no desktop holds.
Plan 1,100 to 1,300W at the wall under full load: a dedicated circuit, not a power strip. Three GPUs at load sound like a hair dryer in the next room. Dust filters are a real purchase.
Three boxes, 384GB of pooled memory, CUDA-native. The models that need a rack are the models that do everything, and this is the smallest room they fit in. Batch workloads, agent fleets, a family of services all answering from your own hardware.
See the $20k tier →