i7-12700K + RTX 5060 Ti 16GB + RTX 3060 12GB: Qwen3.8 27B from 21 to 51 tok/s (split, MTP, single-GPU)
| GPU(s) | RTX 5060 Ti 16GB + RTX 3060 12GB Amazon · eBay (used) · Newegg |
|---|---|
| VRAM / unified memory (GB) | 28 |
| CPU | Intel Core i7-12700K Amazon · eBay (used) · Newegg |
| Motherboard | |
| RAM | 64 GB Amazon · eBay (used) · Newegg |
| PSU | |
| Other | Mismatched pair: 5060 Ti on PCIe 5.0, 3060 negotiating PCIe 3.0 x4; llama.cpp layer split -ts 5/3 (enabling P2P changed nothing) |
| Reported total cost | |
| Power |
Measured speed (generation tokens/sec)
📏 How measured: reported by the original source; RigBase re-checked the numbers at that source
| Model | Quant | Engine | tok/s | Details |
|---|---|---|---|---|
| Qwen3.8 27B | Q4_K_M, layer split 5/3 across both GPUs | llama.cpp (May 2026 build) | 21.2 | |
| Qwen3.8 27B | Q4_K_M, split + MTP draft (n_max=2) | llama.cpp master (Oct 2026) | 37.6 | ⚡ spec-decode (MTP) |
| Qwen3.8 27B | UD-IQ4_XS on the 5060 Ti alone, 8K ctx, q8 KV | llama.cpp master (Oct 2026) | 28.0 | |
| Qwen3.8 27B | UD-IQ4_XS on the 5060 Ti + MTP draft on the 3060 (n_max=2) | llama.cpp master (Oct 2026) | 51.4 | ⚡ spec-decode (MTP) |
JESVS blog, 6 Oct 2026. Prefill 1,144 tok/s at the 5/3 split; row split mode dropped prefill to 228 tok/s. On code-edit prompts MTP reached 46.9 tok/s (split) and 58.1 tok/s (single-GPU sprint config, 8K context). A 35B-A3B MoE distill on the same pair went 111.8 -> 118.6 tok/s on llama.cpp master (not listed: exact model not named).
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 5060 Ti 16GB: ≈€739 each (≈$828), new, approx. price as of 2026-10-16, lowest listing on billiger.de (Asus Dual GeForce RTX 5060 Ti OC 16 GB GDDR7)
- 1× RTX 3060: ≈€429 each (≈$481), new, approx. price as of 2026-10-16, lowest listing on billiger.de (Asus GeForce RTX 3060 V2 Dual 12 GB GDDR6)
New retail prices in Germany incl. VAT; used cards are often much cheaper. Prices change daily.
Source: https://jesvs.com/posts/llama-cpp-dual-gpu-optimization/
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)