Large Language Models (LLMs) have gained popularity in recent years, driving up the demand for inference. LLM inference is composed of two phases with distinct characteristics: a compute-bound prefill phase followed by a memory-bound decode phase. To efficiently serve LLMs, prior work proposes prefill-decode disaggregation to run each phase on separate hardware. However, existing hardware poorly matches the different requirements of each phase. Current datacenter GPUs and TPUs follow a more-is-better design philosophy that maximizes compute and memory resources, causing memory bandwidth underutilization in the prefill phase and compute underutilization in the decode phase. Such underutilization directly translates into increased serving costs. This paper proposes SPAD (Specialized Prefill and Decode hardware), adopting a less-is-more methodology to design specialized chips tailored to the distinct characteristics of prefill and decode phases. The proposed Prefill Chips have larger systolic arrays and use cost-effective GDDR memory, whereas the proposed Decode Chips retain high memory bandwidth but reduce compute capacity. Compared to modeled H100s, simulations show that the proposed Prefill Chips deliver 8% higher prefill performance on average at 52% lower hardware cost, while the proposed Decode Chips achieve 97% of the decode performance with 28% lower TDP. End-to-end simulations on production traces show that SPAD reduces hardware cost by 19%-41% and TDP by 2%-17% compared to modeled baseline clusters while offering the same performance. Even when models and workloads change, SPAD can reallocate either type of chip to run either phase and still achieve 11%-43% lower hardware costs, demonstrating the longevity of the SPAD design.

In recent years, achieving full autonomy in driving has emerged as a paramount objective for both the industry and academia. Among various perception technologies, Lidar (Light detection and ranging) stands out for its high-precision and high-resolution capabilities based on the principle of light propagation and coupling ranging module and imaging module. Lidar is a sophisticated system that integrates multiple technologies such as optics, mechanics, circuits, and algorithms. Therefore, there are various feasible Lidar schemes to meet the needs of autonomous driving in different scenarios. The ranging performance of Lidar is a key factor that determines the overall performance of autonomous driving systems. As such, it is necessary to conduct a systematic analysis of the ranging performance of different Lidar schemes. In this paper, we present the ranging performance analysis methods corresponding to different optical designs, device selec-tions and measurement mechanisms. By using these methods, we compare the ranging perfor-mance of several typical commercial Lidars. Our findings provide a reference framework for de-signing Lidars with various trade-offs between cost and performance, and offer insights into the advancement towards improving Lidar schemes.

In this era of big data, databases are growing rapidly in terms of the number of records. Fast automatic detection of anomalous records in these massive databases is a challenging task. Traditional distance based anomaly detectors are not applicable in these massive datasets. Recently, a simple but extremely fast anomaly detector using one-dimensional histograms has been introduced. The anomaly score of a data instance is computed as the product of the probability mass of histograms in each dimensions where it falls into. It is shown to produce competitive results compared to many state-of-the-art methods in many datasets. Because it assumes data features are independent of each other, it results in poor detection accuracy when there is correlation between features. To address this issue, we propose to increase the feature size by adding more features based on principal components. Our results show that using the original input features together with principal components improves the detection accuracy of histogram-based anomaly detector significantly without compromising much in terms of run-time.

You can't see the bunny, but the picosecond laser certainly can. In a lab at Stanford, engineers have set up a weird contraption, hiding a toy bunny behind a T-shaped wall. And their complex system of computation and rapidly firing lasers can see around that corner. At least that's the idea behind this technique, which uses the flight paths of the photons in lasers to calculate the shape and position of hidden objects--be they bunnies or passing pedestrians. This system deploys the same very, very precise timing that drives the laser-spewing lidar on a self-driving car.