AI Infra Wiki
搜索
Search
暗色模式
亮色模式
探索
标签: power
此标签下有9条笔记。
2026年10月07日
Beyond LLM Serving: Characterizing Vision-Language-Action Workloads for Embodied AI System Design
inference
latency
benchmark
hardware
gpu
memory-bandwidth
power
agentic-ai
architecture
2026年10月06日
Quantitative Architecture Fundamentals
architecture
cpu
benchmark
power
wse
2026年10月06日
Constable: Safely Eliminating Load Instruction Execution
architecture
cpu
load-elimination
isca
power
ooo
2026年10月06日
Constable Load Elimination
architecture
cpu
load-elimination
ooo
power
llm
isca
2026年10月06日
Divide and Conquer: Scalable Performance and Energy in MCM GPUs
architecture
gpu
chiplet
noc
interconnect
topology
mesh
routing
packaging
power
benchmark
2026年9月28日
HeteroReason: FPGA–GPU 解耦投机推理加速
architecture
accelerator
gpu
inference
decode
latency
throughput
speculative-decoding
disaggregated-inference
llm
serving
power
2026年9月08日
Huawei’s τ Chip Was Supposed to Melt?(LogicFolding / Hybrid Bonding)
3d
hybrid-bonding
power
huawei
interconnect
packaging
architecture
cpu
chiplet
2026年9月03日
LEAP: LLM Inference on IMC-NoC with Balanced Dataflow and Fine-Grained Parallelism
noc
interconnect
mesh
accelerator
memory
llm
inference
kv-cache
serving
throughput
latency
architecture
pipeline
power
batching
prefill
decode
disaggregated-inference
dataflow
2026年9月02日
CHIPSMORE: Compute-in-Interconnect and -Memory Chiplets for Multi-Mode Multi-Request LLM Inference
chiplet
noc
interconnect
mesh
accelerator
sram
memory
llm
inference
kv-cache
serving
throughput
latency
packaging
architecture
pipeline
power
batching