AI Navigate

The $qs$ Inequality: Quantifying the Double Penalty of Mixture-of-Experts at Inference

arXiv cs.LG / 3/11/2026

Ideas & Deep AnalysisModels & Research

Key Points

  • Mixture-of-Experts (MoE) models offer high training efficiency but suffer a double penalty during inference due to expert routing fragmentation and reduced high-bandwidth memory for KV caches.
  • The authors formalize this structural disadvantage as the $qs$ inequality, which combines sparsity and quality-equivalence factors to predict when MoE underperforms compared to dense models.
  • Evaluation on advanced models like DeepSeek-V3, Qwen3-235B, Grok-1, and Switch-C reveals that reuse fragmentation is a widespread architectural issue, leading dense models to achieve significantly higher throughput at long context lengths.
  • Findings suggest MoE’s training-time FLOP efficiency does not translate directly to inference-time efficiency, especially for long-context scenarios, and indicate distillation into dense models might be a better strategy for inference deployment.
  • The work highlights limitations of MoE for large-scale inference and provides a new framework to assess architectural trade-offs between sparse and dense models.

Computer Science > Machine Learning

arXiv:2603.08960 (cs)
[Submitted on 9 Mar 2026]

Title:The $qs$ Inequality: Quantifying the Double Penalty of Mixture-of-Experts at Inference

View a PDF of the paper titled The $qs$ Inequality: Quantifying the Double Penalty of Mixture-of-Experts at Inference, by Vignesh Adhinarayanan and Nuwan Jayasena
View PDF HTML (experimental)
Abstract:Mixture-of-Experts (MoE) models deliver high quality at low training FLOPs, but this efficiency often vanishes at inference. We identify a double penalty that structurally disadvantages MoE architectures during decoding: first, expert routing fragments microbatches and reduces weight reuse; second, massive resident expert pools reduce high-bandwidth memory (HBM) headroom for the KV cache. This phenomenon, formalized as reuse fragmentation, pushes feed-forward networks (FFNs) into a bandwidth-bound regime, especially at long context lengths.
We introduce the $qs$ inequality, a predictive criterion that identifies when MoE is structurally disadvantaged relative to a quality-matched dense model. This criterion unifies sparsity ($s$), the fraction of parameters activated per token, and the quality-equivalence factor ($q$), the size multiplier required for a dense model to match MoE performance. Our evaluation across frontier models including DeepSeek-V3, Qwen3-235B, Grok-1, and Switch-C demonstrates that this fragmentation is a general architectural phenomenon. For DeepSeek-V3 at 128k context, this results in a 4.5x throughput advantage for a quality-matched dense baseline. Crucially, massive architectures like Switch-C can become infeasible on cluster sizes where a quality-matched dense model remains viable.
Our results suggest that training-time FLOP efficiency is an incomplete proxy for inference-time performance in long-context serving. They also indicate that MoE may be best viewed as a training-time optimization, with distillation into dense models as a possible path toward inference-efficient deployment.
Comments:
Subjects: Machine Learning (cs.LG); Hardware Architecture (cs.AR); Distributed, Parallel, and Cluster Computing (cs.DC); Performance (cs.PF)
MSC classes: 68T07, 68M20
ACM classes: I.2.6; C.4; C.1.2
Cite as: arXiv:2603.08960 [cs.LG]
  (or arXiv:2603.08960v1 [cs.LG] for this version)
  https://doi.org/10.48550/arXiv.2603.08960
Focus to learn more
arXiv-issued DOI via DataCite

Submission history

From: Vignesh Adhinarayanan [view email]
[v1] Mon, 9 Mar 2026 21:48:04 UTC (43 KB)
Full-text links:

Access Paper:

    View a PDF of the paper titled The $qs$ Inequality: Quantifying the Double Penalty of Mixture-of-Experts at Inference, by Vignesh Adhinarayanan and Nuwan Jayasena
  • View PDF
  • HTML (experimental)
  • TeX Source
Current browse context:
cs.LG
< prev   |   next >
Change to browse by:

References & Citations

export BibTeX citation Loading...

BibTeX formatted citation

×
Data provided by:

Bookmark

BibSonomy logo Reddit logo
Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
Links to Code Toggle
Papers with Code (What is Papers with Code?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos

Demos

Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers

Recommenders and Search Tools

Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
IArxiv recommender toggle
IArxiv Recommender (What is IArxiv?)
About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.