伊隆馬斯克的「巨像」（Colossus）僅用19天建成
使用10萬塊英偉達GPU，打破AI紀錄
伊隆馬斯克的xAI公司僅用19天就建構了全球最強大的AI超級電腦之一－「巨像」（Colossus），這堪稱一項令人驚嘆的工程壯舉。它使用了10萬塊英偉達H200 GPU，其建造速度甚至讓英偉達的執行長都感到震驚，他稱其速度「超人」。這種規模的項目通常需要數年時間。 xAI僅用122天就完成了整個項目，並且已經啟動並運行了相關模型。
位於孟菲斯的「巨像」（Colossus）旨在支持xAI的Grok模型，並推動科學研究的進步。這台超級電腦代表了馬斯克在人工智慧領域策略的轉變。曾經直言不諱的質疑者馬斯克，如今似乎正全力以赴加速人工智慧的發展。
「巨像」的未來遠不止於此。 xAI 計劃將其 GPU 數量翻倍至 20 萬個，這將進一步突破 AI 訓練的極限。這項擴展將使 xAI 在規模、速度和運算能力方面領先於競爭對手。
Colossus 不僅僅是另一個技術里程碑。它代表 AI 基礎設施的轉折點——在這個轉折點上，時間線縮短，目標不斷擴大，探索的步伐也加快。
您認為 AI 硬體的如此快速進步是突破還是警告訊號？
關注“Secrets of the Past”，了解更多塑造未來的前沿故事
#elonmusk #xai #nvidia #supercomputer #grokai #ai #artificialintelligence #futureofai #aitraining
Elon Musk’s Colossus built in 19 days
A record-breaking AI leap using 100000 Nvidia GPUs
In a stunning engineering feat, Elon Musk’s xAI has constructed one of the most powerful AI supercomputers in the world—Colossus—in just 19 days. Using 100000 Nvidia H200 GPUs, the build time shocked even Nvidia’s CEO, who called the speed “superhuman.” Projects of this scale typically require years. xAI completed the full project in only 122 days and already has models up and running.
Located in Memphis, Colossus is designed to support xAI’s Grok model and fuel advancements in scientific research. The supercomputer represents a shift in Musk’s approach to artificial intelligence. Once a vocal skeptic, Musk now seems all-in on accelerating AI development.
The future of Colossus doesn’t stop here. xAI is planning to double its GPU count to 200000, which could push the limits of AI training even further. This expansion could place xAI ahead of competitors in scale, speed, and computing power.
Colossus is more than just another tech milestone. It represents a turning point in AI infrastructure—one where timelines shrink, ambitions grow, and the pace of discovery quickens.
Do you think such rapid progress in AI hardware is a breakthrough or a warning sign?
Follow Secrets of the Past for more cutting-edge stories shaping the future
Go to Article from Barron's»
This isn’t the first Iranian research organization to release specifications for a high performance computer system. In 2007, the Iranian High Performance Computing Research Center said it had assembled a Linux-based supercomputer using 216 Opteron processing cores. That center said it was the fastest system in Iran to date.
這不是伊朗研究機構首度發布高效能電腦系統的規格。2007年，「伊朗高效能電腦運算研究中心」稱，它利用了216個Opteron處理核心，組裝出一套Linux作業系統為基礎的超級電腦。該中心宣稱，這是伊朗歷來最快的（電腦）系統。
to date：片語，迄今為止。
Up to now, until the present time, as in To date we've received no word from them. [First half of 1900s]
try-on
1. Test the fit or look of a garment by putting it on, as in Do you want to try on this dress? This expression is also put astry on for size , which is sometimes used figuratively, as in The teacher wanted to try the new method on for size before agreeing to use it. [Late 1600s] 2. Test the effectiveness or acceptability of something, as in The actors decided to try on the new play out of town. [Late 1800s] Also see try out.
沒有留言:
張貼留言