When you purchase through links in our articles, we may earn a small commission. Nvidia and its partners are ready to roll out new RTX Spark-powered devices in October, shoving the GPU goliath into the CPU spotlight with a beastly first processor that's augmented by powerful integrated graphics. It's all to chase the allure of local AI. And to set the stage, at IFA Berlin 2026, Nvidia is introducing new software designed to turn your RTX-powered gaming household into a miniature AI factory of its own. Dubbed Nvidia PAIR (short for "Personal AI Router"), the free application takes all the GPUs in your local network and dynamically shoots AI agents and subagents to the best available resources during inference tasks.

A TPU is a chip Google created to deal with cloud-based AI and machine learning tasks in the company's data centers. Known as an AI accelerator, a TPU is optimized to perform millions of mathematical calculations on the fly in AI models, something that NPUs and GPUs do as well. We've been talking a lot about AI and ML, which for many of us will call into play an NPU. As it should, since we've been beaten over the head with them ad nauseum thanks to Apple, the various PC and smartphone makers and the chip manufacturers. If you're unfamiliar, NPUs are the chips you'll find in all your modern smartphones, Macs and PCs.

Nvidia's Vera Rubin platform combines CPUs and GPUs into a single system, reflecting the company's growing ambition to power every layer of AI infrastructure. Nvidia is hyping up its new Vera Rubin chip system this week, revealing new performance benchmarks for the GPU and CPU combo ahead of rival AMD's annual product event in San Francisco on Thursday. During a lengthy technical workshop last week at the company's headquarters in Santa Clara, California, Nvidia executives boasted to a small group of journalists about the chip system's increased power and efficiency capabilities. The biggest takeaway: Nvidia, which has long specialized in making GPUs, is increasingly trying to position itself as a supplier of CPUs that can power AI agents. While GPUs are still the main hardware that companies use to train and run their AI models, the industry's shift toward more complex, agentic systems has increased demand for CPUs, which can orchestrate data flows, networking, and other software tasks.

The scaling law for large language models (LLMs) depicts that the path towards machine intelligence necessitates training at large scale. Thus, companies continuously build large-scale GPU clusters, and launch training jobs that span over thousands of computing nodes. However, LLM pre-training presents unique challenges due to its complex communication patterns, where GPUs exchange data in sparse yet high-volume bursts within specific groups. Inefficient resource scheduling exacerbates bandwidth contention, leading to suboptimal training performance. This paper presents Arnold, a scheduling system summarizing our experience to effectively align LLM communication patterns with data center topology at scale. An in-depth characteristic study is performed to identify the impact of physical network topology to LLM pre-training jobs. Based on the insights, we develop a scheduling algorithm to effectively align communication patterns with the physical network topology in modern data centers. Through simulation experiments, we show the effectiveness of our algorithm in reducing the maximum spread of communication groups by up to 1.67x. In production training, our scheduling system improves the end-to-end performance by 10.6% when training with more than 9600 GPUs, a significant improvement for our training pipeline.

As training billion-scale transformers becomes increasingly common, employing multiple distributed GPUs along with parallel training methods has become a standard practice. However, existing transformer designs suffer from significant communication overhead, especially in Tensor Parallelism (TP), where each block's MHA-MLP connection requires an all-reduce communication. Through our investigation, we show that the MHA-MLP connections can be bypassed for efficiency, while the attention output of the first layer can serve as an alternative signal for the bypassed connection. Motivated by the observations, we propose FAL (First Attentions Last), an efficient transformer architecture that redirects the first MHA output to the MLP inputs of the following layers, eliminating the per-block MHA-MLP connections. This removes the all-reduce communication and enables parallel execution of MHA and MLP on a single GPU. We also introduce FAL+, which adds the normalized first attention output to the MHA outputs of the following layers to augment the MLP input for the model quality. Our evaluation shows that FAL reduces multi-GPU training time by up to 44%, improves single-GPU throughput by up to 1.18, and achieves better perplexity compared to the baseline GPT. FAL+ achieves even lower perplexity without increasing the training time than the baseline. Codes are available at: https://casl-ku.github.io/FAL/

Training Transformer models on long sequences in a distributed setting poses significant challenges in terms of efficiency and scalability. Current methods are either constrained by the number of attention heads or excessive communication overheads. To address this problem, we propose StarTrail, a multi-dimensional concentric distributed training system for long sequences, fostering an efficient communication paradigm and providing additional tuning flexibility for communication arrangements. Specifically, StarTrail introduces an extra parallel dimension and divides the peer-to-peer communication into sub-rings to substantially reduce communication volume and avoid bandwidth bottlenecks. Through comprehensive experiments across diverse hardware environments and on both Natural Language Processing (NLP) and Computer Vision (CV) tasks, we demonstrate that our approach significantly surpasses state-of-the-art methods that support Long sequence lengths, achieving performance improvements of up to 77.12% on GPT-style models and up to 114.33% on DiT (Diffusion Transformer) models without affecting the computation results.

As the adoption of Generative AI in real-world services grow explosively, energy has emerged as a critical bottleneck resource. However, energy remains a metric that is often overlooked, under-explored, or poorly understood in the context of building ML systems. We present the ML.ENERGY Benchmark, a benchmark suite and tool for measuring inference energy consumption under realistic service environments, and the corresponding ML.ENERGY Leaderboard, which have served as a valuable resource for those hoping to understand and optimize the energy consumption of their generative AI services. In this paper, we explain four key design principles for benchmarking ML energy we have acquired over time, and then describe how they are implemented in the ML.ENERGY Benchmark. We then highlight results from the early 2025 iteration of the benchmark, including energy measurements of 40 widely used model architectures across 6 different tasks, case studies of how ML design choices impact energy consumption, and how automated optimization recommendations can lead to significant (sometimes more than 40%) energy savings without changing what is being computed by the model. The ML.ENERGY Benchmark is open-source and can be easily extended to various customized models and application scenarios.

The computational sparsity of Mixture-of-Experts (MoE) models enables sublinear growth in compute cost as model size increases, thus offering a scalable path to training massive neural networks. However, existing implementations suffer from low GPU utilization, significant latency overhead, and a fundamental inability to leverage task locality, primarily due to CPU-managed scheduling, host-initiated communication, and frequent kernel launches. To overcome these limitations, we develop FlashMoE, a fully GPU-resident MoE operator that fuses expert computation and inter-GPU communication into a single persistent GPU kernel. FlashMoE enables fine-grained pipelining of dispatch, compute, and combine phases, eliminating launch overheads and reducing idle gaps. Unlike existing work, FlashMoE obviates bulk-synchronous collectives for one-sided, device-initiated, inter-GPU (R)DMA transfers, thus unlocking payload efficiency, where we eliminate bloated or redundant network payloads in sparsely activated layers. When evaluated on an 8-H100 GPU node with MoE models having up to 128 experts and 16K token sequences, FlashMoE achieves up to 9 higher GPU utilization, 6 lower latency, 5.7 higher throughput, and 4 better overlap efficiency compared to state-of-the-art baselines--despite using FP32 while baselines use FP16. FlashMoE shows that principled GPU kernel-hardware co-design is key to unlocking the performance ceiling of large-scale distributed ML.

Large language models (LLMs) power many modern applications, but serving them at scale remains costly and resource-intensive. Current server-centric systems overlook consumer-grade GPUs at the edge. We introduce SpecEdge, an edgeassisted inference framework that splits LLM workloads between edge and server GPUs using a speculative decoding scheme, exchanging only token outputs over the network. SpecEdge employs proactive edge drafting to overlap edge token creation with server verification and pipeline-aware scheduling that interleaves multiple user requests to increase server-side throughput. Experiments show SpecEdge enhances overall cost efficiency by 1.91 through achieving 2.22 server throughput, and reduces inter token latency by 11.24% compared to a server-only baseline, introducing a scalable, cost-effective paradigm for LLM serving.

Large language models (LLMs) power many modern applications, but serving them at scale remains costly and resource-intensive. Current server-centric systems overlook consumer-grade GPUs at the edge. We introduce SpecEdge, an edge-assisted inference framework that splits LLM workloads between edge and server GPUs using a speculative decoding scheme, exchanging only token outputs over the network. SpecEdge employs proactive edge drafting to overlap edge token creation with server verification and pipeline-aware scheduling that interleaves multiple user requests to increase server-side throughput. Experiments show SpecEdge enhances overall cost efficiency by 1.91