# High performance

Powered by the NVIDIA Blackwell architecture, the NVIDIA? GeForce RTX? 5090 features 32 GB of ultra-fast GDDR7 memory, offering unprecedented AI performance for gamers and creators. It supports full ray tracing and minimizes latency for a superior gaming experience, capable of handling advanced models and the most challenging creative workloads.

The new iPad mini is an ultra-portable device powered by the robust A17 Pro chip and supports the Apple Pencil Pro, delivering outstanding performance and versatility. It features an 8.3-inch Liquid Retina display with all-day battery life and comes pre-installed with the latest iPadOS 18 system. This new device not only offers exceptional performance but also boasts a stunning design, available in colors including blue, purple, starlight, and space gray. Starting at $499, the iPad mini offers 128GB of storage—double that of the previous generation—delivering excellent value for users.

The Jamba 1.5 Open Model Family is the latest series of AI models launched by AI21, built on the SSM-Transformer architecture. It is capable of processing ultra-long texts with high speed and quality, outperforming other similar products on the market. These models are designed specifically for enterprise applications, taking into account resource efficiency, quality, speed, and the ability to solve critical tasks.

Mooncake is a leading large language model (LLM) service offered by Moonshot AI, serving as the service platform for Kimi. It utilizes a decoupled architecture centered around KVCache, achieving decoupling caching by separating prefill and decoding clusters and leveraging underutilized CPU, DRAM, and SSD resources within GPU clusters. At the heart of Mooncake lies its KVCache-centered scheduler, which balances maximizing overall effective throughput while ensuring compliance with latency-related service level objectives (SLOs). Different from traditional research, Mooncake addresses high-load scenarios by implementing a prediction-based early rejection strategy. Experiments demonstrate that Mooncake excels in long-context scenarios, achieving up to a 525% throughput increase compared to baseline methods in certain simulated environments while adhering to SLOs. Under real-world workloads, Mooncake's innovative architecture enables Kimi to handle over 75% of requests.

AQChatServer is a high-speed and convenient anonymous online real-time chat room powered by AI. Built on Netty and protobuf protocols for high performance, it is comparable to game backend development. The entire process does not require HTTP protocol, and supports the sending and receiving of text, images, files, audio, and video.

Jamba is an open-weights language model based on the hybrid SSM-Transformer architecture, delivering top-tier quality and performance. It combines the strengths of Transformer and SSM architectures, achieving outstanding results in inference benchmarks while providing a 3x throughput increase in long-context scenarios. Jamba is currently the only model of this scale that can support a 140,000-character context on a single GPU, offering exceptional cost-effectiveness. As a foundational model, Jamba is designed for developers to fine-tune, train, and build customized solutions.