More
Best rig by budgetRent before you buySubmit a buildNewsletterFree datasetAdvertise
Submit build

Builds / 6x RTX 3090 24 GB

6x RTX 3090 on a Ryzen 9 7950X (4 on USB4 eGPU docks): Qwen3.8 Flash-Next at 45.9 tok/s

Source: github.com ↗More source: github.com ↗📏 How measured: reported by the original source; RigBase re-checked the numbers at that source
GPU(s)
6x RTX 3090 24 GB
Amazon · eBay (used) · Newegg
VRAM / unified memory (GB)
144
CPU
AMD Ryzen 9 7950X
Amazon · eBay (used) · Newegg
Motherboard
ASUS ProArt X670E-CREATOR WIFI
Amazon · eBay (used) · Newegg
RAM
192 GB DDR5 (4x48 GB)
Amazon · eBay (used) · Newegg
PSU
not stated
Other
PCIe widths x8/x8/x4/x2/x4/x4, four cards on USB4 eGPU docks; Debian 12; layer split, PLE on CPU, q8_0 KV; lookup speculation off
Reported total cost
not stated
Power
not stated

Measured speed (generation tokens/sec)

ModelQuantEnginetok/sPrompt processing (tok/s)Details
Qwen3.8 Flash-NextUD-Q4_K_XL (~111.3 GB), q8_0 KVllama.cpp-multigpu fork 134489582, layer split45.91,249 @ ~5k-token prompt
est. TTFT @8k ≈ 6.6 s (calculated from measured prefill)		context: 5k prompt (4,992 tokens), 1 slot; 41.9 tok/s at 50k📏 128 greedy generated tokens
Qwen3.8 Flash-NextUD-Q4_K_XL (~111.3 GB), q8_0 KVllama.cpp upstream df03399b8, layer split38.5715 @ ~5k-token prompt
est. TTFT @8k ≈ 11.5 s (calculated from measured prefill)		context: 5k prompt (4,992 tokens), 1 slot; 25.7 tok/s at 50k📏 128 greedy generated tokens

Prompt processing = prefill speed (how fast the prompt is read) as reported by the source, with the prompt length it was measured at; blank = the source does not report it. est. TTFT @8k is calculated from the measured prefill speed (8,192 ÷ prefill tok/s), shown only when prefill was measured on a prompt of 512+ tokens. It is not a measurement and ignores load time and caching. measured TTFT is the source's own time-to-first-token. Why prefill matters for coding assistants.

README of the llama.cpp-multigpu fork (multigpu branch), measured by the author on 5 Oct 2026, comparing the patched fork with upstream llama.cpp on the same machine. The README also has a rented 6x RTX 4090 run (not listed here because it is a cloud machine).

Source: https://github.com/lukolszewski/llama.cpp-multigpu
https://github.com/ggml-org/llama.cpp/discussions/30071

Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.

Rent before you buy: try this model on a cloud GPU for a few dollars first: Runpod (referral link; sign up with Google to get the new-user bonus)

Quick answers: Can 6x RTX 3090 run Qwen3.8 Flash-Next?