Scaling LLM Inference Beyond Amdahl`s Limits via Eliminating Non-Scalable Overheads
Fuente:
arXiv
Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911739752218624 |
|---|---|
| author | Zhao, Alan He, Cyril Y. Xu, Wei |
| author_facet | Zhao, Alan He, Cyril Y. Xu, Wei |
| contents | Deployers of online LLM services usually seek to maximize cluster-wide performance given a fixed number of GPUs. Tensor parallelism (TP) is necessary to fit modern models but scales sub-linearly as the TP degree t grows, due to cross-GPU communication and non-scalable runtime work, as predicted by Amdahl's Law. Conversely, increasing t improves memory efficiency and alleviates KV-cache contention and swapping. We identify and validate an empirical optimal TP degree t_e that balances these effects. We present Albireo, a parallel inference system that raises the attainable t_e by shrinking the non-scalable portion via overlap of scheduling and I/O with compute and sequence-parallel sampling, without changing model architectures. Across models and benchmarks, Albireo achieves up to 1.9x higher throughput, 48% lower latency, 28% higher GPU utilization, and 54% lower energy than vLLM; in production it yields up to 2x higher throughput. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2606_01927 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Scaling LLM Inference Beyond Amdahl`s Limits via Eliminating Non-Scalable Overheads Zhao, Alan He, Cyril Y. Xu, Wei Distributed, Parallel, and Cluster Computing Deployers of online LLM services usually seek to maximize cluster-wide performance given a fixed number of GPUs. Tensor parallelism (TP) is necessary to fit modern models but scales sub-linearly as the TP degree t grows, due to cross-GPU communication and non-scalable runtime work, as predicted by Amdahl's Law. Conversely, increasing t improves memory efficiency and alleviates KV-cache contention and swapping. We identify and validate an empirical optimal TP degree t_e that balances these effects. We present Albireo, a parallel inference system that raises the attainable t_e by shrinking the non-scalable portion via overlap of scheduling and I/O with compute and sequence-parallel sampling, without changing model architectures. Across models and benchmarks, Albireo achieves up to 1.9x higher throughput, 48% lower latency, 28% higher GPU utilization, and 54% lower energy than vLLM; in production it yields up to 2x higher throughput. |
| title | Scaling LLM Inference Beyond Amdahl`s Limits via Eliminating Non-Scalable Overheads |
| topic | Distributed, Parallel, and Cluster Computing |
| url | https://arxiv.org/abs/2606.01927 |