Builds / RX 7900 XTX 24GB
RX 7900 XTX (ROCm 7.2.1): KRAFTON Raon-1 mini preview IQ4_XS at 64.6 tok/s, 75.5 with opt-in fused kernels + backend sampler reuse
Last updated:
Source: github.com ↗📏 How measured: reported by the original source; RigBase re-checked the numbers at that source
- GPU(s)
- RX 7900 XTX 24GB
Amazon · eBay (used) · Newegg - VRAM / unified memory (GB)
- 24
- CPU
- not stated
- Motherboard
- not stated
- RAM
- not stated
- PSU
- not stated
- Other
- ROCm 7.2.1; patched KRAFTON llama.cpp fork from an open pull request with opt-in Raon kernel flags (not defaults); ctx 8192, 1 slot, -ngl 99, batch/ubatch 2048/512, 12 threads
- Reported total cost
- not stated
- Power
- not stated
Measured speed (generation tokens/sec)
| Model | Quant | Engine | tok/s | Prompt processing (tok/s) | Details |
|---|---|---|---|---|---|
| Raon-1 mini Preview | IQ4_XS GGUF | llama.cpp KRAFTON fork + PR #1 kernel profile (ROCm 7.2.1) | 64.57 | not stated | context: 8K ctx, Korean prompts, thinking off📏 PR #1 to krafton-ai/llama.cpp (25 Oct 2026): weighted decode tok/s over 4 Korean prompts x 2 repeats, greedy, prompt cache off, CPU-side sampling in the patched binary; 63.00 with thinking on |
| Raon-1 mini Preview | IQ4_XS GGUF | llama.cpp KRAFTON fork + PR #1 kernel profile (ROCm 7.2.1) | 75.46 | not stated | context: 8K ctx, Korean prompts, thinking off📏 Same PR: --backend-sampling with LLAMA_SERVER_BACKEND_SAMPLER_REUSE=1; 71.73 with thinking on; identical outputs to reuse off on 26 matched requests |
Contributor's measurements in an open pull request to KRAFTON's llama.cpp fork (25 Oct 2026). Every row uses the patched binary with opt-in fused ROCm kernels enabled, so these are not stock-fork speeds. Weighted TPS is server-side decode time only (sum of generated tokens / generation time), not end-to-end.
Source: https://github.com/krafton-ai/llama.cpp/pull/1
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 Radeon RX 7900 XTX run Raon-1 mini Preview?