趋近智
The AMD Ryzen AI 9 HX 370 (32GB system, 24GB GPU-accessible) provides 24 GB of memory with 90 GB/s bandwidth, supporting local inference via ROCm and Vulkan acceleration.
总显存
24 GB
可用显存上限
24 GB
显存带宽
89.6 GB/s
最大推理模型级别
32B Models (Q4)
热设计功耗
54W
随上下文增长模拟的显存需求。当曲线超过参考线时将导致显存不足(OOM)。
量化精度:
不同量化精度下可运行的最大模型参数级别。
上下文长度:
4位量化
32B Models
8位精度
14B Models
16位全精度
8B Models
本地模型训练与适配器微调支持能力。
微调上下文:
QLoRA
32B Models
LoRA
14B Models
全参数微调
3B Models
Guidance for optimal precision and operational ceilings on AMD Ryzen AI 9 HX 370 (32GB system, 24GB GPU-accessible).
Inference Sweet Spot
Optimized for 14B models at uncompressed or Q8 precision, and 32B models at Q4 with up to 32k context.
Bandwidth & Speed Profile
With 89.6 GB/s aggregate bandwidth, batch size 1 inference operates in a memory-bandwidth bound regime. Generates approximately 20 tok/s on an 8B Q4 model and 2 tok/s on a 70B Q4 model.
APX AI
在线