高通：走自己的路，让别家拼核去吧

高通 ( Qualcomm Technologies)执行副总裁Murthy Renduchintala开玩笑说，他虽然有个很难发音的姓氏，但他所透露的消息不会出错。 就像身高有六尺多(180~190几公分)的Renduchintala，高通也将自己视为一家顶天立地、能在移动通信领域走自己的路的公司；高通表示将继续开发客制化处理器 (CPU )、绘处理器(GPU)、数字信号处理器(DSP)以及其他核心，并将这些核心以独特的方式整合在该公司的SoC芯片。

Murthy Renduchintala带有浓重的英国口音
K5Wesmc

“我不在乎谁模仿高通，但高通绝不会模仿任何人；”Renduchintala在年度技术研讨会Uplinq的专题演说后接受提问时表示：“ARM A15核心很不错，但对高通来说还不够好，因此我们必须强化它。” 包括三星 (Samsung)与联发科(Mediatek)都声称其最新的8核心SoC，在效能上超越高通的4核心Snapgragon处理器；那两家对手公司的晶片都是使用ARM现成的A15与A7核心，采用ARM的Big.little多核心架构。 “整体的使用者体验比处理器核心的数量重要；”Renduchintala表示：“你可以用那些评测数据来展现你想要的效果，但如果你看广泛的全面性评测资料，Snapdragon还是最好的。 ” 不过Snapdragon在快节奏的芯片市场上似乎发展脚步稍嫌缓慢，其最新成员是在去年底问世，在今年初的国际消费性电子展(CES)亮相，现在才进驻智能手机与平板设备产品。高通应该会在今年底之前发表新一代的Snapdragon。 Renduchintala表示，处理器核心数量只不过是手机处理器SoC芯片的性能指标之一：“我们可以把12或是24个同样的处理器核心整合在一起，其复杂性不会对芯片本身造成影响，但软件才是让众多核心充分利用的关键──能支持8~10个平行任务的即时软件处理程序是非常复杂的，带来的麻烦往往超越其价值。” 高通声称其优势在于将“数量正确”，最优良的CPU/GPU/DSP核心、影像处理功能、蜂窝通信功能等区块结合，将其功耗效率最佳化；出生于英国的Renduchintala将之比喻为拥有最佳球员组合的11人足球队。 Renduchintala不认为ARM的Big.little架构能提升SoC性能、达到功率最佳化：“那只是克服相关挑战的方法之一，我们则有不同的想法；”他指出，ARM搭配大、小核心的多工处理方案，代表不管是哪一个核心都不能把手上的任务做到真正最佳化。 笔者猜想，在高通内部应该会有关于授权其核心以提高营收、扩展开发者生态系统的争议；如其竞争对手Nvidia也表示将开放最新GPU核心Kepler的授权，提供手机处理器应用，但该公司在手机市场面临的竞争压力还多于高通。 本文授权编译自EE Times，版权所有，谢绝转载 本文下一页：4G通信面临标准分歧的挑战

相关阅读：
• 多核处理器攻占三分之二市场，高通拔得头筹
• 下一个十年，4G手机芯片厂商会重演3G悲剧吗？
• 高通Toq想在可穿戴设备市场分一杯羹K5Wesmc

{pagination} 4G通信面临标准分歧的挑战 此外Renduchintala 也提出了打造可全球漫游之4G 手机的困难度；在3G通信时代，只需要支援5个GSM、4个UMTS以及两个CDMA频段，但LTE在全球所使用的频段超过50个，有的还结合2~3个频道以打造宽频连结：“那些排列组合方式很难理解──这是很大的挑战，我们为了解决问题投资不少。”

高通表示他们的Krait核心处理器已经在性能/功耗综合表现上超越了X86
K5Wesmc

针对ARM架构的服务器处理器发展计划，Renduchintala不愿发表意见；但据业界消息高通正在招募工程师人才打造相关产品。他指出：“我们正考虑利用让高通成为智能手机领域领导者的研发资源，进军其他市场领域；但我们一旦要进军某个新领域，就不会做me-too产品，我们一定会做到差异化。” 高通希望其团队追求卓越的策略令人敬佩，不过值得警惕的是，历史上有不少罹患“NIH (not-invented-here)症候群”的领导级科技业者最后遇上麻烦。(编按：NIH意味着“非我所创”，指某些人或企业组织不愿意接受不是出自自己公司、国家的东西/技术) 例如高通的其他两个大竞争对手苹果(Apple)与三星，苹果就是另一个“NIH症候群”的疑似患者，而三星则是以勇于尝试闻名，并对可行的方案坚持到底。 高通与三星之间的关系似乎越来越微妙，后者既是大客户又是竞争对手；而如果高通的幸运之风改变了方向，也许三星愿意尝试各种可行方案，甚至是买卖竞争对手技术的策略可能有一天会派上用场吧？ 编译：Judith Cheng 参考英文原文：Qualcomm Not Big on Big.little，by Rick Merritt

相关阅读：
• 多核处理器攻占三分之二市场，高通拔得头筹
• 下一个十年，4G手机芯片厂商会重演3G悲剧吗？
• 高通Toq想在可穿戴设备市场分一杯羹K5Wesmc

{pagination} Qualcomm Not Big on Big.little Rick Merritt SAN DIEGO, Calif. — Murthy Renduchintala jokes about how hard it is to pronounce his surname, but there's no mistaking the message of the executive vice president of Qualcomm Technologies Inc. Like six-foot-something Renduchintala, Qualcomm sees itself as a tall, capable presence that goes its own way in mobile technology. It expects to continue developing custom CPU, graphics, DSP, and other cores and combining them in unique ways in its SoCs. "I don't care who copies Qualcomm, but Qualcomm will never copy anyone," Renduchintala said in a press Q&A at the Uplinq event here. "The ARM A15 core is good, but it's not good enough for Qualcomm, so we have to enhance it." Murthy Renduchintala spoke with press after his Uplinq keynote. Murthy Renduchintala spoke with press after his Uplinq keynote. Both Samsung and Mediatek claim their latest eight-core SoCs beat Qualcomm's quad-core Snapdragon in benchmarks. The two rivals both use ARM's off-the-shelf A15 and A7 cores combined using ARM's Big.little multicore architecture. "What's important is the aggregate user experience rather than superficially the number of CPUs," said Renduchintala. "You can use benchmarks to show whatever you want, but if you take a broad panorama of benchmarks, Snapdragon is still the best technology out there." But Snapdragon is getting a bit long in the tooth in the fast-paced silicon world. Its latest members were announced late last year, demonstrated at CES in January, and are now emerging in smartphones and tablets. Qualcomm will probably announce follow-ons before the end of the year. Meanwhile Renduchintala downplays CPU core count as a metric for the quality of a mobile SoC. "We could integrate 12 or 24 cores by tiling identical CPUs together. The complexity is not in laying down the silicon but in the software to make use of that many cores -- real-time software processing with 8 to 10 parallel tasks is pretty complicated and more trouble than it's worth." Qualcomm claims it gains advantage by combining "the right number" of best-of-breed CPU, GPU, DSP, imaging processing, cellular, and other cores optimized for performance per watt. Renduchintala, raised in England, compares it to an 11-man soccer team with the right mix of players. He dismisses ARM's Big.little approach of ganging performance and power optimized cores into an SoC. "It's one way of addressing the challenge, but we have a different idea" he said, adding that the ARM approach of multitasking between big and little cores implies that neither is well optimized for the task at hand. Of LTE & NIH syndrome I suspect there's debate inside Qualcomm about licensing its cores as a way to gain revenue and expand its ecosystem of developers. Rival Nvidia announced it will license Kepler, its latest GPU core for use in mobile SoCs, but it is under more competitive pressures than Qualcomm in mobile. Qualcomm claims its Krait cores beat the Intel x86 in performance/watt. (Click here to enlarge.) Qualcomm claims its Krait cores beat the Intel x86 in performance/watt. (Click here to enlarge.) Separately, Renduchintala called attention to the difficulties building a 4G phone with global roaming. In 3G you needed to support five GSM, four UMTS, and two CDMA bands. But LTE is used in more than 50 bands globally, some combining two or three channels to create a broadband link. "The permutations are beyond comprehension -- that's a big challenge, and we are investing heavily in solving it." Renduchintala declined to comment on any plans for an ARM server SoC, despite Qualcomm's ads advertising for engineers to join a team to build them. "We're looking to leverage the R&D we put into being leaders in smartphones to be in other markets," he said. But as for server SoCs, "if we were to enter into that space it would not be a me-too product; it will have to be differentiated." I applaud Qualcomm's strategy that expects excellence of its people. However, I'd also caution there's a long history of not-invented-here (NIH) syndrome leading tech companies into trouble. Qualcomm's two biggest rivals are Apple, another candidate for NIH syndrome, and Samsung, a giant known for trying everything and sticking with what works. Qualcomm has an increasingly interesting relationship with Samsung, which is both a top customer and competitor. The Korean giant's willingness to try whatever works -- even buying or selling competitor technology -- might come in handy if the winds of fortune ever change direction down in balmy San Diego.