今年的 CES 真可谓是八仙过海，黄仁勋、苏姿丰、陈力武等"经典面孔"齐亮相； 不过台上谈的已不只限于显卡、算力和制程，还在于 AI 接下来要被带 去哪里。 在 AMD 的专场演讲中，苏妈甩出一个大胆判断： "未来五年内，将有 50 亿人每天使用 AI，超过世界人口的一半。" ——什么概念？就是这个增长速度将远超互联网早期阶段，自 ChatGPT 在 2022 年底发布以来，AI 活跃用户已从 100 万暴涨至 10 亿 +。 值得一提的是，这场演讲还请来了"AI 教母"李飞飞。 李飞飞并不是来站台新品的，她和苏妈主要探讨空间智能和世界模型，这也是她已耕深 20 余年的领域。 对于云端，基于下一代 MI455 GPU 的 Helios 机架级平台成为全场焦点：单机架集成 72 块 AI GPU，算力高达 2.9 ExaFLOPS，可通过成千上万个机架拼接 成超大训练集群，直指千亿参数大模型的核心战场。 谈到云端算力的未来，苏姿丰毫不掩饰 AMD 的野心： "全球人工智能运行在云端，而云端运行在 AMD 平台上。" 另外，她还指出，下一代 Instinct 数据中心 AI 加速器平台 MI500 系列，将在 2 ...

作者 | 允毅、木子 今年的 CES 真可谓是八仙过海，黄仁勋、苏姿丰、陈力武等"经典面孔"齐亮相； 不过台上谈的已不只限于显卡、算力和制 程，还在于 AI 接下来要被带去哪里。 在 AMD 的专场演讲中，苏妈甩出一个大胆判断： "未来五年内，将有 50 亿人每天使用 AI，超过世界人口的一半。" ——什么概念？就是这个增长速度将远超互联网早期阶段，自 ChatGPT 在 2022 年底发布以来，AI 活跃用户已从 100 万暴涨 至 10 亿 +。 值得一提的是，这场演讲还请来了"AI 教母"李飞飞。 李飞飞并不是来站台新品的，她和苏妈主要探讨空间智能和世界模型，这也是她已耕深 20 余年的领域。 谈到云端算力的未来，苏姿丰毫不掩饰 AMD 的野心： "全球人工智能运行在云端，而云端运行在 AMD 平台上。" 另外，她还指出，下一代 Instinct 数据中心 AI 加速器平台 MI500 系列，将在 2027 年推出并全面转向 2nm 工艺，并放出狠 话：希望借此在四年内 AI 芯片性能提升 1000 倍（远超摩尔定律啊...）。 与此同时，AMD 还在推动把 AI 从云端下放到本地，而他们的一个很核心的 ...

Core Viewpoint - The RTX 5070 has been criticized as the "worst 70 series ever" due to its performance not meeting expectations set during its launch, where it was claimed to rival the RTX 4090 at a significantly lower price of $549 [1][2][41]. Performance Analysis - The RTX 5070's performance does not match that of the RTX 4090, and in many cases, it underperforms compared to the RTX 4080 and 4070 Ti [4][18]. - In practical tests, the RTX 5070 is only 15% faster than the two-year-old RTX 4070 and just 3% faster than the 4070 Super, indicating a lack of significant improvement [19]. - At 1440p resolution, the RTX 4090 outperforms the RTX 5070 by an average of 63%, with the latter struggling to run certain games due to its 12GB memory limitation [18][20]. - The RTX 5070's performance at 4K resolution is also disappointing, with frame rates dropping to 31 FPS in demanding titles, which is significantly lower than the RTX 4090 [20][23]. Technical Specifications - The RTX 5070 features 6,144 CUDA cores and 12GB of GDDR7 memory, which is above the 8GB standard for high-resolution gaming [7][8]. - Its power consumption is rated at 250W, which is 30W higher than the 4070 Super, raising concerns about its efficiency compared to other models in the series [10][11]. - The cooling design of the RTX 5070 has been criticized for being less effective, resulting in higher temperatures than other tested graphics cards [13]. Market Competition - The upcoming AMD Radeon RX 9070 series, priced at $550, is expected to compete directly with the RTX 5070, potentially offering better performance and more memory [26][27]. - The competition from AMD could pose a significant threat to the RTX 5070's market position, as consumers may prefer a more powerful alternative at a similar price point [25]. Product Issues - Recent reports indicate that the RTX 5080 and 5090 have been experiencing serious issues, including black screen crashes and hardware recognition failures [30][31]. - There are also concerns regarding the RTX 5090 having fewer ROPs than advertised, which could negatively impact gaming performance [32][34]. - NVIDIA has acknowledged these issues but has stated that they affect a small percentage of units, suggesting that the problems have been addressed in production [36][37].