今日主播周明陽：

當(dāng)?shù)貢r(shí)間周三，超威半導(dǎo)體公司（AMD）舉行“Advancing AI”發(fā)布會(huì)，發(fā)布了備受期待的MI300系列芯片產(chǎn)品，包括Instinct MI300X加速器和Instinct MI300A加速處理單元，兩款新產(chǎn)品可用于訓(xùn)練和運(yùn)行大型語言模型，比前代產(chǎn)品擁有更好的內(nèi)存容量和更高的能效。

MI300X擁有超過1500億個(gè)晶體管，內(nèi)存是英偉達(dá)H100芯片的2.4倍，據(jù)稱在訓(xùn)練大型語言模型方面與H100芯片相當(dāng)，但在推理方面表現(xiàn)更好。MI300A則能提供更高性能的計(jì)算、更快的模型訓(xùn)練以及30倍的能效提升，其內(nèi)存容量是H100芯片的1.6倍。Instinct MI300X加速器從發(fā)布會(huì)之日起開始發(fā)貨，Instinct MI300A加速器也已經(jīng)進(jìn)入量產(chǎn)階段。

目前，多家企業(yè)宣布將采用AMD的新型人工智能芯片。微軟宣布將MI300X放入其Azure虛擬機(jī)中；META宣布將在其數(shù)據(jù)中心部署MI300處理器；甲骨文也稱將在云服務(wù)中使用AMD的新款芯片。AMD首席執(zhí)行官蘇姿豐（Lisa Su）表示，預(yù)計(jì)新產(chǎn)品將使其潛在市場(chǎng)總額增長(zhǎng)至450億美元。

那么，AMD的芯片將給AI行業(yè)帶來哪些新變化，友商們又將如何應(yīng)對(duì)？值得期待。

更多資訊請(qǐng)看今日雙語“外企頭條”↓

AMD 推出新型人工智能芯片

AMD Launches New AI Chip

據(jù)外媒報(bào)道，當(dāng)?shù)貢r(shí)間12月6日，超威半導(dǎo)體（AMD）股價(jià)收漲9.9%，報(bào)每股128.37美元，市值達(dá)2073億美元。

According to foreign media reports, on December 6, AMD shares rose 9.9% to close at $128.37, with a market value of $207.3 billion.

報(bào)道稱，這是該股自5月份以來表現(xiàn)最好的一天，也是自6月份以來的最高收盤價(jià)。

It was the stock’s best day since May and the highest close since June.

在當(dāng)天舉行的“Advancing AI”發(fā)布會(huì)上，AMD表示，其新型高端芯片將于明年開始大量出貨。

At the "Advancing AI" launch event, AMD said it’s new high-end chip starts shipping in significant quantities next year.

該公司還表示，未來四年，人工智能芯片的總市場(chǎng)規(guī)模可能會(huì)攀升至4000億美元，是該公司此前預(yù)期的兩倍。

The company also said that the company believes the total market for AI chips could climb to $400 billion over the next four years, twice as high as the company previously believed.

有分析指出，發(fā)布會(huì)凸顯出AMD在利用迅速擴(kuò)張的人工智能市場(chǎng)方面仍然處于非常有利的地位，該公司最終可能占據(jù)整個(gè)人工智能芯片市場(chǎng)的10%左右。

Some analysis points out that AMD’s event highlighted how AMD remains extremely well positioned to take advantage of the rapidly expanding AI TAM, and AMD could end up with about 10% of the total AI chip market.

谷歌公司發(fā)布人工智能模型“雙子座”

Google announces Gemini AI mode

谷歌公司12月6日宣布推出該公司所謂“規(guī)模最大、功能最強(qiáng)”的人工智能模型“雙子座”。

On December 6, Google announced the launch of Gemini, the company's "largest and most powerful" artificial intelligence model.

美國(guó)銀行分析師表示，谷歌母公司Alphabet今年一直面臨著外界對(duì)谷歌人工智能能力的擔(dān)憂，因此“品牌良好”、具有競(jìng)爭(zhēng)力的模式可能對(duì)其消費(fèi)者搜索活動(dòng)和云企業(yè)銷售有好處。

Analysts at Bank of America said that Alphabet, parent company of Google has been under pressure from concerns over Google’s AI capabilities this year, so a “well-branded,” competitive model could have upsides for its consumer search activity and Cloud enterprise sales.

還有分析指出，“雖然現(xiàn)在還為時(shí)過早，但‘雙子座’的推出代表了谷歌的重大創(chuàng)新，因?yàn)槲覀冞M(jìn)入了生成式人工智能商業(yè)化和廣泛分布的第二年。”

“While it remains early, the Gemini launch represents significant innovation for Google as we enter year 2 of commercialized and widely distributed availability of Generative AI,” another analyst noted.

谷歌高管表示，Gemini的性能優(yōu)于OpenAI的GPT-3.5聊天機(jī)器人，但該公司沒有透露它與OpenAI的最新型號(hào)GPT-4 Turbo的比較。盡管如此，Gemini仍表明人工智能有進(jìn)一步變現(xiàn)的機(jī)會(huì)。

Google executives said Gemini outperforms OpenAI’s GPT-3.5 chatbot, but the company didn’t share how it compares with OpenAI’s latest model GPT-4 Turbo. Still, Gemini shows there’s an opportunity to further monetize AI.