Benchmarking Confidential GPU Inference on NVIDIA H100 under Intel TDX
However, the performance cost of enabling confidential execution for GPU-accelerated large language model serving remains workload dependent and operationally important. This paper presents a benchmark study comparing standard non-confidential execution with confidential computing mode on a single NVIDIA H100 80GB GPU hosted in an Intel TDX confidential instance. The evaluation uses two representative language models, Mistral-7B v0.1 and Qwen3-30B-A3B, and measures time to first token, end-to-end request latency, per-request token generation throughput, global token throughput, and closed-loop request throughput under increasing concurrency.
- ▪However, the performance cost of enabling confidential execution for GPU-accelerated large language model serving remains workload dependent and operationally important.
- ▪This paper presents a benchmark study comparing standard non-confidential execution with confidential computing mode on a single NVIDIA H100 80GB GPU hosted in an Intel TDX confidential instance.
- ▪The evaluation uses two representative language models, Mistral-7B v0.1 and Qwen3-30B-A3B, and measures time to first token, end-to-end request latency, per-request token generation throughput, global token throughput, and closed-loop reque
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2607.19353 (cs) [Submitted on 20 May 2026] Title:Benchmarking Confidential GPU Inference on NVIDIA H100 under Intel TDX Authors:Wei Wang, Abdul Hyee Waqas, Burns Smith View a PDF of the paper titled Benchmarking Confidential GPU Inference on NVIDIA H100 under Intel TDX, by Wei Wang and 2 other authors View PDF HTML (experimental) Abstract:Confidential computing is becoming a practical deployment requirement for AI inference workloads that process sensitive inputs or protect proprietary model assets. However, the performance cost of enabling confidential execution for GPU-accelerated large language model serving remains workload dependent and operationally important.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv.org.
Discussion
0 commentsMore from arXiv.org
