Builds / RTX 5070 Laptop GPU 8 GB
RTX 5070 Laptop 8 GB (ROG Strix G16): Qwen3.8 27B GSQ IQ3_S with 27 of 64 layers on the GPU at 7.0 tok/s, 7.8 with MTP (Kingy)
Last updated:
Source: kingy.ai ↗📏 How measured: reported by the original source; RigBase re-checked the numbers at that source
- GPU(s)
- RTX 5070 Laptop GPU 8 GB
Amazon · eBay (used) · Newegg - VRAM / unified memory (GB)
- 8
- CPU
- Intel Core Ultra 9 290HX
Amazon · eBay (used) · Newegg - Motherboard
- ASUS ROG Strix G16 laptop
Amazon · eBay (used) · Newegg - RAM
- 32 GB
Amazon · eBay (used) · Newegg - PSU
- not stated
- Other
- Windows 11, llama.cpp b12199 (CUDA), Turbo power mode, plugged in; the 12.12 GB file does not fit in 8 GB, so llama.cpp's fit tool put 27 of 64 layers on the GPU (24 for the chat server) and the rest on the CPU
- Reported total cost
- not stated
- Power
- not stated
Measured speed (generation tokens/sec)
| Model | Quant | Engine | tok/s | Prompt processing (tok/s) | Details |
|---|---|---|---|---|---|
| Qwen3.8 27B | GSQ-RCO IQ3_S + MTP head (ISTA-DASLab GGUF, 12.12 GB) | llama.cpp b12199 (CUDA, partial offload) | 7.0 | not stated | context: 32K ctx, one request at a time📏 Kingy test 8 Nov 2026: llama-server, 32K ctx, prompt cache off, temperature 0, seed 42, up to 1,024 reply tokens (all spent thinking); average of 4 prompts (code, explainer, HTML report, question on an 8,217-token document); MTP off; llama-bench tg128 7.4 with -ngl 27 (6.0 with 8K in context); about 19-20 s before the first token on the 8K document |
| Qwen3.8 27B | GSQ-RCO IQ3_S + MTP head (ISTA-DASLab GGUF, 12.12 GB) | llama.cpp b12199 (CUDA, partial offload) | 7.8 | not stated | ⚡ spec-decode (MTP)context: 32K ctx, one request at a time📏 Kingy test 8 Nov 2026: llama-server, 32K ctx, prompt cache off, temperature 0, seed 42, up to 1,024 reply tokens (all spent thinking); average of 4 prompts (code, explainer, HTML report, question on an 8,217-token document); --spec-type draft-mtp --spec-draft-n-max 2, 69% of drafts kept; small gain because most of the model runs on the CPU |
Owner-run test by Kingy (8 Nov 2026). Shows what a 27B model does when it spills out of an 8 GB laptop GPU. Each prompt ran once. The source suggests a smaller model such as Qwen3-8B for everyday use on an 8 GB card.
Source: https://kingy.ai/blog/qwen3-8-27b-gsq-tested/
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)
Speculative-decoding rows on this page are not like-for-like with plain decode. See how much MTP and DFlash really speed things up (measured pairs).