立即打开
为了这款芯片,AMD赌上了一切

为了这款芯片,AMD赌上了一切

Aaron Pressman 2017-07-03
在过去数十年,曾经的创新能手AMD逐步落伍。现在,CEO苏姿丰要在技术做大胆试验,推动AMD卷土重来。

从四楼办公室的大窗户向外望去,苏姿丰的目光从AMD的奥斯汀园区扫过,然后落在测试新型芯片的实验室大楼上。2016年春天,苏姿丰经常会看着这个大楼,给在此工作的员工发短信、即时消息以及打电话更是家常便饭。当时她正急切地等待着齐柏林飞艇(Zeppelin)的诞生。

齐柏林飞艇是AMD最新微处理器的代号,这款旗舰产品将用于个人电脑和企业级服务器——AMD的未来也有赖于它的成功。作为电机工程博士,苏姿丰2014年成为AMD首席执行官。当时这家芯片制造商生意惨淡,销售额不断滑坡。苏姿丰努力让AMD的产品重新焕发生机,齐柏林飞艇则是她的第一项成果。该产品使用的芯片经过了彻头彻尾的重新设计,有望在运算需求密集的消费者群体中得到青睐,不管是挑剔的游戏玩家,还是运行人工智能和机器学习程序的科技公司。如果这款新产品大获成功,AMD就有可能扭转连年亏损的局面,甚至从英特尔和英伟达等对手的阴影中走出来。

苏姿丰没想到的是,当齐柏林飞艇最终“抵达”奥斯汀时,它一度有可能“坠毁”。

From the wide windows of her fourth-floor office, Lisa Su can look across the Austin campus of Advanced Micro Devices and see the laboratory building where the company’s new chips get tested. In the spring of 2016, Su was looking in that direction quite often, not to mention texting, instant messaging, and calling the staffers who worked there. She was waiting eagerly for a Zeppelin to arrive.

Zeppelin was the code name for AMD’s (amd) newest microprocessor, a flagship chip designed to run in personal computers and corporate servers—and the company’s future was riding on its success. Su, a Ph.D. microprocessor engineer herself, had become CEO in 2014 in the midst of a dismal sales decline for the chipmaker. Zeppelin was the first fruit of her effort to revive AMD’s product line, with redesigned-from-the-ground-up chips that could woo customers with intense computing needs, from finicky video gamers to tech companies running artificial intelligence and machine learning programs. If the new products thrived, AMD stood a chance of reversing years of losses, and even emerging from the shadow of rivals like Intel and Nvidia.

What Su didn’t anticipate was that when the Zeppelin finally got to Austin, it would crash-land.

当AMD的“齐柏林飞艇”芯片项目可能毁于一个设计缺陷时,测试团队进入了苏姿丰所说的“阿波罗13模式”——换句话说就是,决不能失败。摄影:Sarah Lim

测试负责人路易斯·卡斯特罗组建的团队包括80名工程师,他们的任务是检测AMD首款采用齐柏林飞艇架构的芯片,名为Ryzen。然而,2016年4月进行测试的前一天晚上,芯片设计团队主管给卡斯特罗打了电话——设计人员进行电脑模拟时出现了一个缺陷,AMD的第一颗Ryzen芯片有可能“见光死”,甚至都不能让计算机启动起来。如果不能迅速解决这个问题,这个芯片项目或许就得延期几周,甚至几个月。此时苏姿丰正在12800公里外的印度出差,跟他们相隔10个时区,这让情况变得更加复杂。卡斯特罗回忆说:“在你的职业生涯里,你从来没有参与过如此重大的事件。我站起来对自己说,噢,天哪,我该怎么办?”

负责齐柏林飞艇的工程师李·鲁斯克给正在为AMD制造芯片的代工厂打了电话,告诉他们立即停止生产。首席技术官马克·佩珀马斯特出面联系苏姿丰,告诉她这个坏消息。通话很紧急,但两位高管都没有陷入恐慌。苏姿丰的第一反应很果断,那就是测试绝不能推迟。

AMD的测试团队迅速进入了苏姿丰所说的“阿波罗13模式”。四组工程师集思广益,想找到绕过这颗原型芯片中缺陷的解决方案,以便立即开始测试。一回到奥斯汀,苏姿丰直奔实验室给这些人打气,但她也提醒他们,“绝不能失败”。

Louis Castro, who oversees testing, had assembled a team of 80 engineers to check out the first Zeppelin chip, dubbed the Ryzen. But the night before testing was to begin, in April 2016, the head of the chip design team called Castro. A flaw had slipped through the designers’ computer simulations, and the first chip would be dead on arrival, incapable of even booting up a computer. If the problem couldn’t be quickly fixed, the chip might be delayed for weeks or even months. And to complicate matters, Su was 8,000 miles and 10 time zones away on a business trip in India. “You’ve never been part of something as big as this in your career,” Castro recalls. “I sat and thought to myself, Oh, my gosh, what am I going to do?”

Lee Rusk, the engineer in charge of Zeppelin, called the foundry that was making the chip for AMD and told it to stop production immediately. Chief technology officer Mark Papermaster stepped up to call the CEO with the bad news. The conversation was urgent, but neither executive panicked. And Su’s immediate reaction was decisive: Testing couldn’t be delayed.

AMD’s team quickly went into what Su calls “Apollo 13 mode.” Four different teams of engineers brainstormed solutions for getting around the flaw in the prototype chip to start testing immediately. As soon as she got back to Austin, Su headed straight for the lab, encouraging the teams while reminding them that “failure was not an option.”

苏姿丰和路易斯·卡斯特罗(身着RyzenT恤衫)、李·鲁斯克(身着Polo衫)以及另一位工程师在实验室探讨AMD的齐柏林飞艇芯片项目。摄影:Sarah Lim

今天的电脑和手机使用的硅芯片无比复杂。一枚5美分硬币(直径约21毫米)大小的Ryzen芯片中集成了500万个晶体管,分为100层,卡斯特罗的团队发现的缺陷影响着不到万分之一的集成电路。如果它在芯片中的位置更深,在最下面几层,修复所需的时间就可能要了他们的命。但AMD抓住了突破点——实际情况表明,可以花一个月时间在代工厂中纠正这个问题。同时,卡斯特罗的团队想出了绕过这个缺陷进行测试的办法,连这一个月的时间损失都避免了。

AMD迫切需要一场胜利,而且要快,这一点无需累述。10年来,该公司在大多数时间里依赖的策略,包括制造基本但不可或缺的芯片,每一、两年进行一次适度升级,以及通过出低价来赢得竞争,以失败告终。国际数据公司的数据显示,2007-2016年,AMD在PC芯片市场中份额从23%降至10%以下;服务器芯片方面,AMD的市场份额已经不足1%。与此同时,整个PC市场萎缩的速度超过了所有人预期,而取代PC的移动革命基本上跟AMD擦肩而过。AMD已经连续亏损了5年,2015年收入触底时堪堪不到40亿美元,和2011年的高点相比下跌了39%。

应该说,AMD的坏消息是失去了硅谷。在“半导体军备竞赛”中,它和第一名一直相差甚远。但AMD一直在用数不清的方式进行创新——作为芯片制造商,它率先突破了1GHz运算速度障碍,还是首家在一块PC芯片中集成两个处理核心的芯片制造商。在这样的位置上,AMD一直在为压低成本做贡献,并让很多想法畅行无阻,从而惠及无数依赖运算能力的公司。惠普企业CEO、PC和服务器行业老手梅格·惠特曼指出:“在我见过的所有市场中,竞争都是创新的动力。芯片市场有一个广阔的生态系统,这对整个行业来说是件好事。”

The silicon chips at the heart of today’s computers and phones are insanely complex. A single Ryzen chip the size of a nickel has five million transistors, laid out across 100 different layers. The flaw that Castro’s team had found affected fewer than one-hundredth of one percent of the circuits. If it had been located deep in the chip, on one of the lowest layers, repairing it could have been fatally time-consuming. But AMD caught a break: The problem, it turned out, could be corrected at the foundry in a month. And Castro’s team figured out how to get around the flaw for testing, avoiding losing even just that month.

It’s hard to overstate how badly AMD needed a win, and a quick one to boot. The strategy it relied on over most of the past decade—building basic but essential chips, rolling out modest upgrades every year or two, and underpricing the competition—has broken down. Between 2007 and 2016, AMD’s market share in PC chips sank from 23% to less than 10%, according to IDC; in server chips, it has dropped below 1%. At the same time the overall PC market has shrunk faster than anyone expected, losing ground to a mobile revolution that largely passed the company by. AMD has lost money for five years in a row, and revenue bottomed out at just under $4 billion in 2015, a 39% decline from its 2011 peak.

Bad news for AMD, arguably, is a loss for Silicon Valley. The company has never been anywhere near No. 1 in the semiconductor arms race. But AMD has been an innovator in countless ways—the first chipmaker to break the one-gigahertz speed barrier, the first to put two processing cores in a single chip for PCs. And in that role, it has helped keep costs down and ideas flowing for innumerable businesses that rely on processing power. “Competition drives innovation in every market I’ve ever seen,” says Meg Whitman, CEO of Hewlett Packard Enterprise and now a veteran of the PC and server industries. “A broad ecosystem in the chip market,” she adds, “is a good thing for the industry as a whole.”

 

如今,AMD看来已经不大可能消亡,这在很大程度上要感谢苏姿丰。她的策略以彻底的重新设计为支点,而重新设计可能有助于AMD在类似于超级计算机的处理器市场跃居英特尔和英伟达之上,此类芯片可以同时进行更多的运算,并能迅速读取存储在用户计算机其他部件上的数据。同时,苏姿丰已经开始让AMD摆脱对PC的依赖,并将注意力集中在向三大游戏主机厂商,也就是微软、索尼和任天堂供应芯片的业务上。她还通过向中国合作伙伴提供服务器芯片设计授权提高了AMD的利润。为实现这一切,苏姿丰利用了自己的工程师经验,利用了逾10年来跟IBM建立的关系,还利用了从苹果公司这个迷人的独立王国中挖来的设计人才。

第一批Ryzen芯片今年3月上市,初步反响很强烈。AMD曾承诺,Ryzen的运算速度要比上一代芯片高40%,这些芯片则轻松地超额完成了任务。同时,它们的性能和英特尔的同类芯片相当,价格却不到后者的一半。举例来说,作为高端台式机芯片,Ryzen 7 1800X的售价为499美元,而英特尔酷睿i7-6900K的价格为1089美元。投资者也很兴奋——AMD的股价现已升至12美元左右,而2016年初它还在2美元以下苦苦挣扎。

今年夏天将有更多新产品上市——接下来AMD将推出Epyc服务器芯片,从而对英特尔在这个领域接近垄断的地位发起冲击。随后亮相的将是图形处理芯片,也就是GPU Vega。这些芯片对游戏玩家来说意义重大,但不仅如此,它们还是执行最先进人工智能和机器学习任务的最佳途径。这些任务正是智能助理Siri和Alexa的动力源泉,通用电气等大型企业也通过它们来分析大数据流。实际上,对GPU的需求一直在增长,甚至是在PC市场不断滑坡的情况下。Vega的性能同样引起了外界关注——苹果公司已被说服,将把这款芯片用于即将推出的iMac Pro,也就是今年6月苹果发布的纯黑色时尚一体机。

Today, AMD’s extinction looks less likely—thanks in large part to Su. Her strategy hinges on radical redesigns that could help AMD leapfrog Intel and Nvidia in the market for supercomputer-like processors, chips that do more calculations simultaneously and speed access to data stored on other parts of a user’s computer. At the same time, Su has begun weaning AMD from its dependence on PCs, focusing on deals to supply chips to the three leading video game console makers: Microsoft, Sony, and Nintendo. She has also boosted the bottom line by licensing server chip designs to a Chinese partner. To accomplish it all, Su is drawing on her experience as an engineer, on relationships she built over more than a decade at IBM—and on design talent poached from the glamorous confines of Apple.

The first Ryzen chips hit the market this March, and early reviews are strong. The chips easily exceeded AMD’s promise of 40% faster processing than the previous generation. And they’re matching the performance of comparable Intel chips at less than half the price; a top-end Ryzen 7 1800X chip for desktop computers, for example, sells for $499, while Intel’s Core i7-6900K is $1,089. Investors are excited too: AMD’s stock, which was barely treading water at under $2 a share in early 2016, now trades at about $12.

This summer will bring more debuts: Up next is a chip for servers called Epyc—taking on Intel’s near monopoly in that category. Later comes Vega, a graphics chip, or GPU. Such chips have not only become important for gamers but also as the best way to run cutting-edge artificial intelligence and machine learning tasks—the kinds that fuel Siri and Alexa and are used by corporate giants like GE to analyze “big data” streams. Indeed, demand for GPUs is growing even as the PC market remains in a slump. The Vega’s performance has been turning heads too: It convinced Apple to use the chip in its upcoming iMac Pro, the sleek, all-black computer it unveiled in June.

AMD企业部门执行副总裁福里斯特·诺罗德手持最近推出的Epyc服务器芯片。英特尔几乎垄断了这个领域,AMD的市场份额不足1%。摄影:Sarah Lim

不过,上述产品尚未经过验证,很难说AMD已经摆脱困境。伯恩斯坦研究公司长期跟踪芯片行业的分析师斯泰西·拉斯刚认为苏姿丰赌对了方向并且理顺了AMD的资产负债表,但她还没有证明AMD能够成功。拉斯刚说:“18个月前AMD的关注点在于它会不会破产,从这个角度来说她干的确实很好。但对于猜测AMD的执行情况,我见过的已经太多了。”苏姿丰的任务就是扫除这样的疑问。

拉斯刚所见证的东西和英特尔紧紧纠缠在一起,作为AMD的对手,英特尔的规模要大得多。两家公司均由来自半导体先驱企业仙童半导体的工程师和高管创立。1968年,罗伯特·诺伊斯、戈登·摩尔和安迪·格鲁夫另起炉灶,组建了英特尔。AMD在一年后成立,创始人是销售员杰里·桑德斯,他自称为来自芝加哥南部的倔小子。AMD的业务在20世纪80年代实现腾飞,主要原因是IBM认为自己的新PC不应完全依赖于英特尔芯片,并将AMD指定为第二家官方供应商。在PC芯片领域,AMD一直是兼容英特尔x86架构的唯一主要可替代产品制造商。但就算在21世纪初的巅峰期,尽管AMD的PC芯片市场份额接近四分之一,它依然排在第二位,而且远远落后于英特尔。

With those products still unproven, however, AMD is hardly out of the woods. Stacy Rasgon, the longtime chip industry analyst at Bernstein Research, believes Su has made the right bets and cleaned up the balance sheet but says she has yet to prove AMD can deliver. “In the context of a company where 18 months ago the concern was, are they going bankrupt or not, she’s doing a really good job,” Rasgon says. “But I have too much history with AMD to bet on their execution.” It’s Su’s mission to vanquish such skepticism.

That history is tightly intertwined with that of AMD’s far bigger rival, Intel. Both companies were founded by engineers and executives from semiconductor pioneer Fairchild. Robert Noyce, Gordon Moore, and Andy Grove struck out in 1968 to form Intel. The AMD group spun out a year later, under salesman Jerry Sanders, a self-proclaimed tough kid from Chicago’s South Side. AMD’s business took off in the 1980s largely because IBM decided it shouldn’t rely only on Intel for chips for its new personal computer and designated AMD as its official second supplier. AMD remains the only major alternative source for PC chips compatible with Intel’s x86 template—but even at its peak in the 2000s, when it sold nearly one out of every four chips found in PCs, it was always a distant second.

苏姿丰在AMD奥斯汀总部的实验室中。“我正在进行自己的一系列战斗,而且我觉得很过瘾。” 摄影:Sarah Lim

在桑德斯及其继任者赫克特·迪·热苏斯·鲁伊斯领导下,AMD在20世纪90年代以及21世纪初进入全盛时期,推出了K6等速度很快的创新型PC芯片,并用这些产品证明AMD不仅仅是英特尔的复制品。2006年,AMD的股价达到逾42美元的高点。但就在这一年,鲁伊斯做出了斥资54亿美元收购图形芯片厂商ATI的重大决定。ATI的技术未能带来AMD所期待的推动力。此外,本次收购让AMD在债务重担以及合并冲销的影响下步履维艰,这些年来持续亏损。2008-2011年,AMD更换了四任CEO。

这就是苏姿丰接手的烂摊子。她生于台湾,2岁时随家人迁往纽约。父母告诉苏姿丰,她可以成为钢琴师、医生或者工程师。第三种选择引起了幼年苏姿丰的共鸣,那时她经常把弟弟的电动玩具车拆开,然后试着再装回去。她上了著名的布朗克斯科技高中,随后进入麻省理工。在这里,苏姿丰获得了电机工程学士、硕士和博士学位,对微处理器的兴趣也首次在她心里扎了根。

AMD reached its prime in the 1990s and 2000s under Sanders and his successor, Hector de Jesus Ruiz, introducing fast and innovative PC chips like the K6 that proved the company was more than an Intel clone. Its stock reached a high of over $42 a share in 2006. But Ruiz made the fateful decision that year to buy graphics-chip maker ATI for $5.4 billion. ATI’s technology never gave AMD the boost it hoped for. What’s more, the deal led to years in the red for the company as it juggled a heavy debt load and merger write-offs. From 2008 to 2011, AMD went through four CEOs.

That was the mess inherited by Su. Born in Taiwan, she moved to New York City with her family at age 2. Her parents told Lisa she could be a concert pianist, a doctor, or an engineer. The third choice resonated with a kid who regularly took apart her younger brother’s electric cars and tried to put them back together. She attended the prestigious Bronx High School of Science, then MIT—where her interest in microprocessors first took root—for an undergraduate degree, master’s, and Ph.D. in electrical engineering.

获得博士学位的苏姿丰在麻省理工的毕业典礼上,1994年。图片提供:AMD

在德州仪器工作了很短一段时间后,苏姿丰进入了IBM。在IBM的逾10年光阴里,她一直致力于追求价格更低、速度更快的芯片,而且遇到了至关重要的导师尼古拉斯·多诺弗里奥。后者是IBM的传奇人物。从大型机到普通PC,多诺弗里奥什么都做过,他安排苏姿丰担任CEO特别技术助理。当时的CEO是来自美国运通的郭士纳,而苏姿丰的工作是让郭士纳了解最新的重大技术进展,并且确保技术培训上的欠缺不会妨碍郭士纳进行决策。苏姿丰回忆说:“我的收获是这份工作让我看到了大公司CEO如何思考问题。” 郭士纳的强项在于简化可选方案,从而把注意力集中在新技术究竟怎样帮助消费者的问题上。

苏姿丰受到了启发,产生了担任管理工作的想法。同时,随着她不断取得成功,苏姿丰感到她在IBM做不到这一点(她强调,女性身份一直都不是障碍。如果说有什么的话,那就是她觉得自己很幸运,总是能遇到不受性别问题困扰的上司)。2007年,Freescale Semiconductor邀请苏姿丰担任首席技术官,她也抓住了这个机会。Freescale Semiconductor是摩托罗拉剥离出的业务,曾为阿波罗登月项目生产芯片,当时需要进行全民调整。苏姿丰搬到了奥斯汀,最终成为Freescale Semiconductor价值10亿美元的网络芯片业务负责人,并帮助公司在2011年上市。

After a brief stint at Texas Instruments, Su went to IBM, where she spent more than a decade focused on the race for cheaper, faster chips. She also met a crucial mentor in Nicholas Donofrio, an IBM legend who had worked on everything from mainframes to the original PC. Donofrio arranged for Su’s appointment as special technical assistant to Lou Gerstner, the then CEO who left American Express to run Big Blue. Su’s job was to keep Gerstner abreast of major technology developments and ensure his lack of technical training didn’t hinder his decision-¬making. “What I got as a benefit from it was watching how the CEO of a major corporation thinks,” Su recalls now. ¬Gerstner’s strength lay in simplifying the available options, homing in on how new technologies could actually help customers.

Su aspired to run a company, and as her star rose she felt she couldn’t do that at IBM. (Being a woman, she stresses, was never an obstacle: If anything, she feels lucky to have always had bosses who were free of gender hang-ups.) In 2007, Freescale Semiconductor, a Motorola spinoff that made chips for the Apollo moon missions and was in need of an overhaul, offered Su the role of chief technology officer, and she seized the opportunity. She relocated to Austin, where she eventually ran Freescale’s $1 billion networking chip division and helped the company go public in 2011.

苏姿丰和时任IBM CEO的郭士纳(2000年前后)。图片提供:AMD

当时多诺弗里奥已经从IBM退休,并且进入了AMD董事会,帮助AMD筹划扭亏战略。一次,到奥斯汀参加董事会会议的多诺弗里奥请苏姿丰在豪华的Barton Creek度假村吃晚饭。两人都担心对方因为苏姿丰离开IBM而生气。为缓解紧张气氛,多诺弗里奥点了一瓶非常贵的思福山坡精选赤霞珠。红酒入杯时,双方显然都不再心怀不满。多诺弗里奥记得苏姿丰离开IBM的原因,因而转换了话题。AMD的问题根深蒂固,但也有天才工程师和独家知识产权。多诺弗里奥回忆道,当时自己说:“这个时机太适合你了。”“你说的太对了。”苏姿丰爽快地答应下来,并在2012年进入了AMD。2014年,她成了CEO。

那时AMD已经组建了设计梦之队,可以重塑它的芯片。多诺弗里奥请来了如今的AMD首席技术官马克·佩珀马斯特,后者是IBM老臣,史蒂夫·乔布斯曾邀请他帮苹果开发用于iPhone的系列芯片。反过来,佩珀马斯特也帮AMD引进了其他明星设计师,其中最著名的当属2013年加盟的拉加·库德里。他当时在苹果任职,被外界普遍视为图形芯片预言家。库德里彻底调整了苹果的产品线,以便对接高分辨率的Retina显示屏。

By then, Donofrio had retired from IBM and joined AMD’s board to help craft a turnaround strategy. In Austin for a board meeting, Donofrio invited Su to dinner at the tony Barton Creek Resort. Each worried that the other was angry over Su’s departure from IBM. To break the tension, Donofrio ordered a very expensive California cabernet, Shafer Hillside Select, and as the wine flowed it became obvious that neither harbored a grudge. Remembering why Su had left IBM, Donofrio flipped the script. AMD had deeply rooted problems but also had incredible engineering talent and unique intellectual property. “It’s so ripe for you,” Donofrio recalls saying. “It’s so right.” Su jumped at the bait and joined the company in 2012. By 2014, she was CEO.

By then AMD had put together a design dream team that could reinvent its chips. Donofrio recruited Mark Papermaster, now the CTO, an IBM veteran whom Steve Jobs had wooed to help Apple develop its own line of chips for the iPhone. Papermaster in turn helped the company bring in other rock-star designers—most notably Raja Koduri in 2013. Widely respected as a graphics-chip visionary, Koduri was at Apple at the time, overhauling its product line to handle high-resolution Retina screens.

首席技术官马克·佩珀马斯特,IBM老臣,史蒂夫·乔布斯曾邀请他帮苹果开发用于iPhone的系列芯片。 摄影:Sarah Lim

在库德里的家人和朋友看来,从如日中天的苹果跳到陷入困境的AMD简直是疯了。妻子觉得他遭遇了中年危机。但库德里渐渐相信,其他平台或许会取代移动平台,成为芯片制造创新的核心。连续几个小时一直盯着iPhone手机屏幕后,库德里突然顿悟。他回忆说:“人们最终会希望把这个东西一直带在身边,而不仅仅是放在口袋里。大家会希望随时接触到这里的信息。”实现这一点的也许是虚拟现实,也许是人工智能型数字助理,亦或是库德里无法预见的某种混合体。但它一定需要更多的高性能运算,进而提升对新型芯片的需求。为重振旗鼓而不顾一切的AMD正好成为他按照清晰方案设计此类芯片的场所。库德里说:“如果在工作时好像自己一无所有一样,那你就会做出一些相当有趣的东西。”

这种“一无所有”的精神正在产生回报。去年AMD的收入同比上升7%,达到42亿美元。到今年底,图形芯片业务有望发挥更大作用。2015年,苏姿丰把AMD的所有图形芯片业务整合成一家名为Radeon Technology的新子公司,并任命库德里为负责人。从那时至今,Radeon Technology的员工人数增加了60%,成为AMD最大的团队。苏姿丰说,AMD的PC芯片“已经占据了主舞台。现在我们要说,图形业务同样是一等公民”。

随着使用新型Vega芯片的产品今夏开售,这项策略很快就会迎来真正的考验。CEO黄仁勋(也是在台湾出生的美国人,也是电机工程师)领衔的英伟达是图形芯片领域的重量级选手,特别是在软件方面,该公司特有的CUDA平台是大数据分析的主导性工具。虽然分析师所说的“GPU运算”市场很小,但它正在快速成长。伯恩斯坦分析师拉斯刚估算,去年这个市场的销售额不到5亿美元,到2020年将达到90亿美元。AMD正通过打造开源软件平台来追赶CUDA。但就连AMD高层都承认,他们的起步时间已经远远落后。

与此同时,英特尔绝不会退出对PC和服务器市场的争夺。COE科再奇把重心放在数据中心客户上。今年5月,英特尔发布了酷睿i9,也就是性能更强大的台式机系列产品。该公司PC芯片业务新任负责人格雷格·布莱恩特表示:“近一段时间我们在这个领域确实一直在加速。”

梅格·惠特曼认为AMD能和更高级别的选手较量。我们注意到,《财富》500强(2015年AMD曾因收入未达标而落选)中的女性CEO寥寥无几,惠特曼和苏姿丰是其中之二,而且此前苏姿丰曾请惠特曼传授担任CEO的建议。惠特曼认为惠普企业的服务器业务将成为AMD Epyc芯片的主要受益者(和买家)。她指出:“为什么苏姿丰能在别人失败的地方取得成功?她让公司把注意力集中在打造出色的产品上。从头至尾,就是这样。”

苏姿丰定期到世界各地出差时都会讲这个故事。今年6月一个阳光明媚的日子,她再次来到波士顿地区,和一些潜在客户会面。在她获得博士学位差不多25年后,苏姿丰的母校麻省理工邀请她给大约500名博士毕业生讲话。苹果CEO蒂姆·库克(毕业于奥本大学和杜克大学商学院)会在第二天的全校毕业典礼上发言,苏姿丰则是博士生毕业典礼上的演讲嘉宾。

Jumping from Apple, at the height of its dominance, to struggling AMD seemed crazy to Koduri’s friends and family. His wife thought he was having a midlife crisis. But Koduri had come to believe that other platforms might displace mobile as the locus for innovation in chipmaking. Staring at the display of an iPhone for hours on end, he had an epiphany. “Man eventually wants to carry this thing around with him all the time, not just in his pocket,” Koduri recalls. “You want access to this information all the time.” That might be attained through virtual reality or digital assistants fueled by artificial intelligence, or some mix Koduri couldn’t foresee. But it was bound to boost demand for high-performance computing—and new kinds of chips. And AMD’s desperation for renewal made it the place where he could design such chips with a clean slate. “If you work like you have nothing to lose, you can do some pretty interesting things,” he says.

The “nothing to lose” ethos is beginning to pay dividends. Last year AMD’s revenue rose 7% over the previous year, to $4.2 billion. By the end of this year, the graphics business could be pulling more weight. In 2015, Su put all AMD’s graphics-chip work under Koduri in a new unit called the Radeon Technology Group. Radeon’s headcount has risen 60% since then, to 3,200, making it the largest team in the company. AMD’s PC chips “had taken the limelight,” Su says. “Now we’re saying that graphics is also a first-class citizen.”

Soon will come the real test of the strategy, as products with the new Vega chips go on sale this summer. Nvidia, under CEO Jensen Huang (another ¬Taiwanese-born American trained as an electrical engineer), is a powerful player in graphics, particularly on the software side, where its proprietary CUDA platform dominates as a tool for big-data analysis. Although the market for what analysts call “GPU compute” is small, it’s growing fast. From under $500 million in sales last year, it will hit close to $9 billion by 2020, Bernstein’s Rasgon estimates. AMD is building an open-source software platform to catch up to CUDA. But by its executives’ own admission, it’s starting far behind.

Intel, meanwhile, is far from folding its tent in the competition for PCs and servers. CEO Brian Krzanich is focusing attention on data center customers, and in May Intel unveiled a higher-performance desktop line dubbed the Core i9. “We really stepped on the accelerator in this space a while back,” says Greg Bryant, the new head of Intel’s PC chip unit.

Meg Whitman, for one, thinks AMD will punch above its weight. Notably, Whitman and Su are among the few women who have reached the ranks of CEO at a Fortune 500 company (though AMD’s recent revenue woes bumped it off that list in 2015), and Su consulted Whitman early on for advice about being a CEO. Whitman thinks Hewlett Packard Enterprise’s core server business will be a big beneficiary (and buyer) of AMD’s Epyc chip. “Why did Lisa succeed where others failed?” she asks. “She has focused that company on building great product. Beginning, middle, end of story.”

Su regularly travels the world to pitch that story. And on a sunny day this June she was back in the Boston area, mingling with some likely future customers. Almost 25 years after she earned her own doctorate, Su’s alma mater, MIT, invited her to speak to the 500 or so graduates receiving a Ph.D. Apple CEO Tim Cook (an Auburn University and Duke business school grad) would address the school’s full graduation the next day, but it was Su who spoke at the event where the doctoral recipients got their ceremonial hoods.

苏姿丰出席麻省理工博士生毕业典礼,2017年6月。她在发言中展现了自己富有竞争性的一面,苏姿丰说:“你们要向我保证,你们会努力工作,从而确保今后会有许多哈佛MBA给麻省理工的博士生打工。” 摄影:Aaron Pressman

苏姿丰在自己写的发言稿(很多公司高管都会找人代劳)中向毕业生们提出了挑战,要求他们有远大梦想,把握自己的命运,并且改变这个世界。她还显露出了自己富有竞争性的一面。她在最后说:“你们要向我保证,你们会努力工作,从而确保今后会有许多哈佛MBA给麻省理工的博士生打工。”这句话赢得了听众由衷的喝彩。随后,她的名气得到了印证——毕业生们请她合影,教授们则上前和她讨论芯片设计以及摩尔定律。

智能手机前,苏姿丰认真地摆着姿势,并和大家亲切交谈,随后躲进了附近一家狭小的中餐馆,这是她学生时代最喜欢的就餐场所之一。她坐在丈夫身边,点了塑料菜单上几样最辣的菜。放松下来后,苏姿丰总结了她的发言主旨:“我正在进行自己的一系列战斗,而且我觉得很过瘾。你们每个人都可以选择自己想要的战斗,而且都可以获胜。”(财富中文网)

本文将刊登在2017年7月1日出版的《财富》上,题为《拿出崭新芯片,赌上一切》(Betting It All, With Brand-New Chips)。

译者:Charlie

审稿:夏林

In a speech she wrote herself (a task many execs delegate), Su challenged the graduates to dream big, make their own luck, and change the world. Her competitive streak also came out. “Promise me that you will work hard at ensuring that there are lots of Harvard MBAs working for MIT Ph.D.s in the future,” she concluded to hearty applause. Afterward her celebrity status was confirmed as graduates asked her to pose for selfies and professors approached to discuss chip design and Moore’s law.

Su dutifully posed for the smartphones and chatted amiably before retreating to a nearby hole-in-the-wall Chinese restaurant, a favorite from her student days. Sitting next to her husband and ordering some of the spiciest dishes from the plastic menus, a relaxed Su summed up what she was trying to get across in her speech. “I’m fighting my set of wars, and I’m having a great time,” Su said. “Each of you can pick a war that you want to fight, and you can win it.”

A version of this article appears in the July 1, 2017 issue of Fortune with the headline "Betting It All, With Brand-New Chips."

热读文章
热门视频
扫描二维码下载财富APP