Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Validation of GPU Computation in Decentralized, Trustless Networks

Eric Boniardi, Stanley Bishop, Alison Haire

TL;DR

Verifying GPU computations in decentralized, trustless networks is challenging due to non-determinism across GPU architectures. The paper shows exact recomputation is inadequate and that TEEs and Fully Homomorphic Encryption have practical drawbacks, then develops probabilistic verification frameworks by adapting model fingerprinting, semantic similarity, and GPU profiling. It introduces two semantic-verification schemes—Binary Reference with trusted nodes and Ternary Consensus—that enable trustless validation through offline threshold optimization and multi-node consensus. An empirical study on the Chatbot Arena dataset demonstrates about 76% accuracy with a threshold $t^* \approx 0.5$, supporting the viability of semantic-verification approaches for GPU workloads in untrusted networks. The work provides a foundation for computational integrity in GPU-accelerated decentralized systems and outlines future directions for profiling-based verification and broader deployment across heterogeneous hardware.

Abstract

Verifying computational processes in decentralized networks poses a fundamental challenge, particularly for Graphics Processing Unit (GPU) computations. Our investigation reveals significant limitations in existing approaches: exact recomputation fails due to computational non-determinism across GPU nodes, Trusted Execution Environments (TEEs) require specialized hardware, and Fully Homomorphic Encryption (FHE) faces prohibitive computational costs. To address these challenges, we explore three verification methodologies adapted from adjacent technical domains: model fingerprinting techniques, semantic similarity analysis, and GPU profiling. Through systematic exploration of these approaches, we develop novel probabilistic verification frameworks, including a binary reference model with trusted node verification and a ternary consensus framework that eliminates trust requirements. These methodologies establish a foundation for ensuring computational integrity across untrusted networks while addressing the inherent challenges of non-deterministic execution in GPU-accelerated workloads.

Validation of GPU Computation in Decentralized, Trustless Networks

TL;DR

Verifying GPU computations in decentralized, trustless networks is challenging due to non-determinism across GPU architectures. The paper shows exact recomputation is inadequate and that TEEs and Fully Homomorphic Encryption have practical drawbacks, then develops probabilistic verification frameworks by adapting model fingerprinting, semantic similarity, and GPU profiling. It introduces two semantic-verification schemes—Binary Reference with trusted nodes and Ternary Consensus—that enable trustless validation through offline threshold optimization and multi-node consensus. An empirical study on the Chatbot Arena dataset demonstrates about 76% accuracy with a threshold , supporting the viability of semantic-verification approaches for GPU workloads in untrusted networks. The work provides a foundation for computational integrity in GPU-accelerated decentralized systems and outlines future directions for profiling-based verification and broader deployment across heterogeneous hardware.

Abstract

Verifying computational processes in decentralized networks poses a fundamental challenge, particularly for Graphics Processing Unit (GPU) computations. Our investigation reveals significant limitations in existing approaches: exact recomputation fails due to computational non-determinism across GPU nodes, Trusted Execution Environments (TEEs) require specialized hardware, and Fully Homomorphic Encryption (FHE) faces prohibitive computational costs. To address these challenges, we explore three verification methodologies adapted from adjacent technical domains: model fingerprinting techniques, semantic similarity analysis, and GPU profiling. Through systematic exploration of these approaches, we develop novel probabilistic verification frameworks, including a binary reference model with trusted node verification and a ternary consensus framework that eliminates trust requirements. These methodologies establish a foundation for ensuring computational integrity across untrusted networks while addressing the inherent challenges of non-deterministic execution in GPU-accelerated workloads.
Paper Structure (21 sections, 6 figures)

This paper contains 21 sections, 6 figures.

Table of Contents

  1. Introduction
  2. Background and Literature Review
  3. Verifying Non-Deterministic GPU Computations in Distributed Networks
  4. Cryptographic Verification Methods
  5. Fully Homomorphic Encryption (FHE)
  6. Trusted Execution Environments (TEEs)
  7. Methodological Approaches from Adjacent Technical Domains for Verification Applications
  8. Fingerprinting AI Models
  9. Semantic Similarity
  10. GPU Performance Profiling
  11. Implementation of Cross-Domain Techniques for Computational Verification
  12. From Fingerprints to Proofs: Verifying AI Inference
  13. Semantics as Security: Verifying AI Through Meaning
  14. GPU Fingerprints: Profiling as Proof of Computation
  15. Data
  16. ...and 6 more sections

Figures (6)

  • Figure 1: Model Fingerprinting Process
  • Figure 2: Semantic Similarity of Input Embeddings
  • Figure 3: GPU Profiling
  • Figure 4: Fingerprinting Preliminary Results
  • Figure 5: Semantic Preliminary Results
  • ...and 1 more figures