Builds / RTX 5090 32GB
One RTX 5090 + 128 GB RAM running DeepSeek V4.1 Flash (552B) by streaming experts from NVMe: 5.1 tok/s
Source: github.com ↗More source: huggingface.co ↗📏 How measured: reported by the original source; RigBase re-checked the numbers at that source
- GPU(s)
- RTX 5090 32GB
Amazon · eBay (used) · Newegg - VRAM / unified memory (GB)
- 32
- CPU
- not stated
- Motherboard
- not stated
- RAM
- 125.7 GiB (as reported)
Amazon · eBay (used) · Newegg - PSU
- not stated
- Other
- PCIe 5 NVMe SSD holding the 502 GB GGUF. JigSawPT/llama.cpp branch dsv41-porte: routed experts stream from disk through an 18 GiB VRAM cache and a 72 GiB pinned host tier; engram tables memory-mapped
- Reported total cost
- not stated
- Power
- not stated
Measured speed (generation tokens/sec)
| Model | Quant | Engine | tok/s | Prompt processing (tok/s) | Details |
|---|---|---|---|---|---|
| DeepSeek V4.1 Flash | MXFP4 experts (lossless repack) + Q8_0/BF16 dense + fp8 engram tables | llama.cpp fork dsv41-porte, expert streaming, --reasoning off | 5.1 | not stated | context: 8,192 ctx📏 New content: 4 prompts of mixed content × 3 rounds, TTFT 7.7 s (prompt length not stated). Same prompt again (experts resident): 21.4 tok/s, TTFT 0.26 s. Measured without --moe-stream-direct; 18 GiB VRAM expert cache + 72 GiB pinned host tier |
| DeepSeek V4.1 Flash | MXFP4 experts + Q8_0/BF16 dense + fp8 engram tables | llama.cpp fork dsv41-porte, expert streaming + --moe-stream-direct (O_DIRECT) | 8.9 | not stated | context: 8,192 ctx📏 First pass on the four benchmark prompts; the model card gives 8.9–10.0 tok/s, low end used (3.9–4.9 tok/s without the flag) |
llama.cpp GitHub discussion by JigSawPT, 11 Sep 2026, and the model card of the GGUF (the --moe-stream-direct result was added 23 Sep 2026). Nothing fits in RAM + VRAM; speed depends on the SSD. CPU, board and PSU not stated.
Current GPU street price (estimate, not the builder's cost)
New retail prices in Germany incl. VAT; used cards are often much cheaper. Prices change daily.
- 1× RTX 5090: ≈€7,140 each (≈$8,003), new, approx. price as of 2026-10-24, lowest listing on billiger.de (MSI GeForce RTX 5090 (lowest variant))
New retail prices in Germany incl. VAT; used cards are often much cheaper. Prices change daily.
Source: https://github.com/ggml-org/llama.cpp/discussions/28766
https://huggingface.co/JigSawPT/DeepSeek-V4.1-Flash-GGUF
https://huggingface.co/JigSawPT/DeepSeek-V4.1-Flash-GGUF
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 RTX 5090 run DeepSeek V4.1 Flash?