Unverified50% confidenceFactExact time
32-Tile AMX将矩阵寄存器资源扩展至8-Tile的四倍,单核可同时驻留的矩阵数据量从约8KB提升至约32KB
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/4/2026
First Seen
Valid until: 11/2/2026
Sources
RosaicLabs与Atom RTL揭示x86生态隐秘变局
Related Claims
UnverifiedAMX的核心组件包括默认8个二维Tile寄存器(每个最大可达1KB,即16行×64字节)、TILECFG指令,以及TDPBF16PS、TDPBSSD等矩阵乘法指令77% similarUnverifiedAMD64将通用寄存器从8个扩展到16个,寄存器宽度从32位扩展到64位,引入REX前缀,并保持向后兼容x8668% similarUnverified开发者提供五种量化格式:INT8 ConvRot(4.5GB)、FP8 E4M3(4.2GB)、bf16(8.3GB)、NVFP4(2.9GB)、MXFP8(4.7GB)67% similarUnverified使用IQ4_XS配合-fa 1 -ctk q8_0 -ctv q8_0对KV缓存进行q8_0量化后,可加载完整32k上下文,常驻内存约16.6GB67% similarUnverifiedTaking the Q4_0 format as an example, every 32 FP32 weights are packed into one quantization block containing one scale factor and 32 4-bit integers, averaging approximately 4.5 bits per parameter, reducing memory usage to approximately 1/8 of the original.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/681692API
curl https://kongchang.com/api/v1/knowledge/claims/681692MCP
get_claim(id=681692)