????3個(gè)月前，英特爾宣布將以167億美元的價(jià)格收購(gòu)Altera公司。對(duì)于這家全球最大的芯片公司來(lái)說(shuō)，這是一筆意義重要的投資，此舉似乎旨在整合半導(dǎo)體行業(yè)。不過最近，對(duì)于英特爾斥巨資收購(gòu)Altera的動(dòng)機(jī)，我們掌握了更多細(xì)節(jié)。

????理由大體可歸結(jié)為英特爾云平臺(tái)團(tuán)隊(duì)副總裁兼總經(jīng)理杰森?韋克斯曼提供的一個(gè)數(shù)據(jù)。韋克斯曼表示，到2020年，英特爾相信將有三分之一的數(shù)據(jù)中心市場(chǎng)使用Altera專攻的芯片技術(shù)。

????Altera生產(chǎn)一種名為現(xiàn)場(chǎng)可編程門陣列（FPGA）的芯片，這種芯片在生產(chǎn)后還可以重新編輯。之前，F(xiàn)PGA只用于特定場(chǎng)合，由于它們面積更大、價(jià)格更高，性能通常也不如通用芯片，所以在普通服務(wù)器和工作站中并不常用。

????在涵蓋數(shù)據(jù)中心的演示中，微軟等客戶正在使用FPGA來(lái)運(yùn)行它們的搜索算法。在關(guān)于加速器的演示中，未來(lái)的機(jī)器學(xué)習(xí)用戶正在目睹FPGA運(yùn)行類神經(jīng)網(wǎng)絡(luò)。總之，英特爾認(rèn)為FPGA會(huì)無(wú)處不在。在網(wǎng)絡(luò)世界，F(xiàn)PGA已經(jīng)在蜂窩基站中亮相，英特爾希望借助FPGA在這一領(lǐng)域獲得更多的市場(chǎng)占有率。

????實(shí)際上，英特爾看待FPGA就像我看待Sriracha辣椒醬一樣。它們無(wú)處不在。

????英特爾的樂觀并不是毫無(wú)道理：這是硬件生態(tài)系統(tǒng)的各個(gè)層面向輕便性和定制化轉(zhuǎn)變的一部分。由于“軟件正在蠶食世界”，對(duì)一個(gè)要花18個(gè)月才能把想法變成某些成品的芯片公司而言，形勢(shì)變得更加困難。至少通過FPGA，你可以提供一款能夠改變的硬件產(chǎn)品，即便實(shí)現(xiàn)這種改變需要不菲的成本。

????一些初創(chuàng)公司正在試圖改變目前的FPGA模塊，讓它變得更加靈活。對(duì)于這些公司來(lái)說(shuō)，這也是一個(gè)機(jī)遇。例如，創(chuàng)業(yè)公司Flex Logic就在試圖制造一種比當(dāng)代芯片更快的FPGA。在軟件方面，BitFusion希望幫助公司在最理想的硬件上運(yùn)行他們的代碼，同時(shí)不需要重現(xiàn)底層芯片的代碼——無(wú)論它是不是FPGA。（財(cái)富中文網(wǎng)）

????譯者：嚴(yán)匡正

????審校：任文科

????Three months ago Intel said it would buy chip maker Altera in a deal valued at $16.7 billion. It was a significant investment for the world’s largest chip company and much was made over the consolidation in the semiconductor industry. On Thursday however, we got a lot more color on why Intel plans to plunk down a considerable hunk of money for Altera.

????It basically boils down to this one statistic offered up by Jason Waxman, the VP & GM of the cloud platforms group at Intel. Waxman said that by 2020 Intel believes a third of the data center market could be using the type of chips that Altera specializes in.

????Altera makes a type of chip called a field programmable gate array, or FPGA, that’s really a fancy way of saying a type of chip that can be reprogrammed after its made. The industry has traditionally used FPGAs in specialty scenarios as opposed to putting them in common servers or workstations because they are larger, cost more, and generally don’t perform as efficiently as general purpose chips.

????In presentations covering the data center, where clients like Microsoft are using FPGAs to run their search algorithms to presentations on accelerators where future machine-learning customers are eyeing FPGAs to run neural networks, Intel sees FPGAs everywhere. In the networking word, FPGAs are already present in cellular base stations, and Intel INTC 1.54% is hoping to gain more market share inside those boxes with, yes, its FPGAs.

????Basically, Intel sees FPGAs like I see Sriracha. They belong everywhere.

????Intel’s optimism isn’t totally unfounded; it’s part of a shift to agility and customization that’s playing out at all levels of the hardware ecosystem. As “software eats the world,” it’s tougher to be a chip company designing and building a product that requires 18 months to go from idea to something physical. At least with an FPGA, you can offer a physical product that can be changed, even if that ability to tweak it comes at a cost.

????There’s an opportunity here too for startups trying to alter the current FPGA model to make it a little bit more flexible. For example, the startup Flex Logic is trying to make an FPGA that is a little bit faster than the current generation of chips. On the software side,BitFusion hopes to help companies run their code on the most optimal hardware for their current job without having to rewrite that code for the chip underneath—be it an FPGA or not.