Machine learning force fields (MLFFs) offer high accuracy at reduced computational cost, but their repeated evaluation at every molecular dynamics (MD) timestep leads to prohibitive runtime, limiting simulations to short physical timescales. In this work, we accelerate Orb v2, a Graph Neural Network–based MLFF, by targeting its dominant Attention Interaction Network (AIN), which accounts for approximately 97% of inference time. We identify on-chip memory constraints and a latency-memory trade-off in block-based computation as the key challenges to efficient acceleration. To address these challenges, we propose AccelOrb, a memory-aware FPGA acceleration of AIN based on a single streaming kernel that employs a non-uniform tiling strategy to balance latency and memory footprint, and selectively maps parameters beyond BRAM. Implemented on a resource-constrained AMD Alveo U50 FPGA, our approach achieves 9.19× and 8.18× speedup over CPU and GPU baselines while operating within tight power and resource budgets.

HTTP’s intrinsic request-and-response traffic pattern makes most web servers often idle, leaving a potential to accelerate page loads. We present the notion of HTTP steady connections, which fully utilizes the server’s available network bandwidth during a page load using the promising HTTP/3 server push, transforming the intermittent workload of loading a page into a more steady one. To construct a proper server push policy to achieve this, we separate the structure of a page, which is a relatively static factor, from the page load environments including client and network characteristics, which are generally dynamic and unknown to servers. We formulate a deadline-based sequencing problem using a page load model with dependency graphs and design a feedback-based reprioritization mechanism within HTTP server push to reactively match client progress robustly. Experiments with a prototype and a wide range of real-world pages show that HTTP steady connections significantly improve web page loads compared with state-of-the-art accelerators, even under packet losses and without any prior knowledge of network environments.