### 🔀 [#128](https://github.com/ikawrakow/ik_llama.cpp/pull/128) - Faster IQ4_XS_R4 on Zen4

| **Author** | `ikawrakow` |
| :--- | :--- |
| **State** | ❌ **Closed** |
| **Created** | 2024-12-08 |
| **Updated** | 2024-12-08 |

---

#### Description

We now get PP-512(LLaMA-3.1-8B) = 254 t/s on a Ryzen-7950X CPU, up from 224 t/s.