IT之家1 月 6 日消息，在今日的 CES 2026主题演讲中，AMD 展示了其下一代及首款 2nm EPYC Venice Zen6 CPU 和 Instinct MI455X GPU，专为 Helios AI Racks 设计。 Helios AI 机架最早在 2025 年 AMD 财务分析师日上亮相，该公司承诺将带来“领先的性能数据”，以及在 AI 工作负载方面具有“最佳能效”的性能表现。

当大语言模型（LLM）走向千行百业，推理效率与显存成本的矛盾日益尖锐。 KV Cache （Key-Value Cache）作为提升生成速度的核心技术，却像一个 “吞存巨兽”—— 每增加一个 token，就需要更多显存存储键（Key）和值（Value）向量，最终成为制约模型规模扩张、并发 ...

目前，不同大模型厂商发布的大语言模型在处理超长上下文方面已经有显著突破，最高的已能支持数百万 Token 的输入，例如 MiniMax-M1、Qwen2.5-1M 系列模型，均支持百万Token（1M）级别的超长上下文处理能力。 但是这场有关提升大模型上下文长度的“军备赛”依然不 ...

We've already had a decent look at the patent for AMD's upcoming GPU chiplet technology, but a new patent that was published on April 1 teases a few very interesting new things. AMD's new GPU chiplet ...

A new leak has given us a peek at what could have been Intel’s real flagship GPU from the Battlemage generation. A prototype PCB that surfaced online shows just how ambitious Intel’s graphics division ...

通过堆叠技术，NVIDIA可以将运算核心留在主芯片，而将需要大量面积的SRAM独立成另一层芯片堆叠上去。 台积电的A16制程一大特色是支持背面供电技术，这项技术可以腾出芯片正面的空间，专供垂直信号连接，确保堆叠的LPU能以极低功耗进行高速数据交换。

There's an exciting new graphics card memory technology on the horizon that could see huge gains in one of the most important aspects of GPUs: memory bandwidth. The new GPU SCM with DRAM tech can ...

Previous rumors have suggested that AMD could boost its GPU portfolio by adding a stack of 3D V-Cache, a performance enhancing technique established with its Ryzen CPUs. Now, an engineer with access ...

The AMD RDNA 2 graphics launch event is a little over three weeks away. If anything, the event just became even more important, with the Nvidia CEO's statement about RTX 3000 series graphics card ...

Let's be frank: it was a pleasant surprise when AMD's Radeon RX 6000-family graphics cards were able to provide solid competition for (and even outpace, at times) NVIDIA's GeForce RTX 3000 series. The ...