Builds / RTX 5090 Laptop GPU 24 GB
RTX 5090 Laptop GPU (24 GB) + 64 GB RAM: Qwen3.8 Flash-Next at 24.4 tok/s on llama.cpp, 85.6 with the Strata engine
Last updated:
- GPU(s)
- RTX 5090 Laptop GPU 24 GB
Amazon · eBay (used) · Newegg - VRAM / unified memory (GB)
- 24
- CPU
- Intel Core Ultra 9 290HX Plus (24 threads)
Amazon · eBay (used) · Newegg - Motherboard
- not stated
- RAM
- 64 GB DDR5-5600, 2 channels
Amazon · eBay (used) · Newegg - PSU
- not stated
- Other
- Laptop; the card cannot hold the model, so experts sit partly in system RAM. One request at a time, 256 tokens out, greedy
- Reported total cost
- not stated
- Power
- Board power (not wall), 4K prompt: 61.1 W (llama.cpp) / 126.3 W (Strata); floating 135-150 W Dynamic Boost limit
Measured speed (generation tokens/sec)
| Model | Quant | Engine | tok/s | Prompt processing (tok/s) | Details |
|---|---|---|---|---|---|
| Qwen3.8 Flash-Next | IQ3_S (ISTA-DASLab GGUF), experts of 34/48 layers in system RAM, no draft head | llama.cpp (3cf03257), 8 threads, 512 micro-batch | 24.4 | measured TTFT: 19.1 s @ 4,038-token prompt | context: 4,038-token prompt📏 Median of three runs, 4K prompt, as printed in the source table; 20.7 tok/s at a 32K prompt🔌 61.1 W (board) · 0.40 tok/s per W |
| Qwen3.8 Flash-Next | IQ3_S + model draft head, 8,120 of 24,576 experts on the card | Strata engine 0.1.25 (draft head) | 85.6 | measured TTFT: 1.9 s @ 4,038-token prompt | ⚡ spec-decode (MTP)context: 4,038-token prompt📏 Median of three runs, 4K prompt; drafts from the model's own draft head (speculative). 87.1 tok/s at 32K. Run shared the laptop with another job🔌 126.3 W (board) · 0.68 tok/s per W |
| Qwen3.8 Flash-Next | IQ3_S, experts of 34/48 layers in system RAM, no draft head | llama.cpp (3cf03257), 8 threads, 512 micro-batch | 20.7 | measured TTFT: 126 s @ 32K-token prompt | context: 32,038-token prompt📏 Median of three runs at a 32K prompt🔌 68.8 W (board) · 0.30 tok/s per W |
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.
strata→signal research, published 12 Oct 2026; laptop runs 2-3 Oct 2026 (UTC), back to back. Median of three runs. llama.cpp ran near its defaults (8 threads, 512-token micro-batch, flash attention, 8-bit KV), not tuned; with a 4,096-token micro-batch its 4K wait dropped to 7.6 s. Strata is a third-party engine built on llama.cpp. The same article's desktop cards are on a separate RigBase build page.
- 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.
Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.
Copy into a README or blog. Links back to this build. See badge docs (CC BY 4.0 attribution). The SVG does not invent speeds — numbers stay on the build page.
Markdown
[](https://getrigbase.com/builds/rtx-5090-laptop-64gb-qwen38-flash-next-strata/)
HTML
<a href="https://getrigbase.com/builds/rtx-5090-laptop-64gb-qwen38-flash-next-strata/" rel="noopener"><img src="https://getrigbase.com/badge/rtx-5090-laptop-64gb-qwen38-flash-next-strata.svg" alt="Measured on RigBase" height="20"/></a>
Preview: 
· How to use
Quick answers: Can RTX 5090 run Qwen3.8 Flash-Next?