根据Nvidia“丹佛计划”(Project Denver) 即将推出的首款芯片细节，显示出计算机领域在未来两三年将发生什么变化仍有待观察。 2007年，Nvidia悄悄地展开了 Project Denver ，迈向一条成为通用微处理器供货商的漫长艰辛之路。在2011年国际消费电子展(CES)上，Nvidia透露该计划的目标在于开发出瞄准从笔记本计算机到 超级计算机应用的一系列完整SoC。
而在日前举办的2013年度“绘图处理器技术大会”(GTC)上，Nvidia透露基于 Project Denver 的首款商用芯片── Parker 将搭载2014年版的 Maxwell GPU 以及全新的64位客制化ARM CPU，预计2015年上市。
Nvidia对该芯片产品发布的细节有限，但分享了对于业界微处理器发展蓝图的一些愿景，足以让我们好好思考计算机产业的未来发展方向。
Nvidia的下一重大行动就是明年即将推出的 Logan 。这款 Tegra 5 SoC 将在描准智能手机和平板电脑开发的芯片中导入 Kepler GPU 。它将是首款可支持其 Cuda 软件环境的ARM行动 SoC，以实现其通用绘图处理器(GPGPU)的理念。 Cuda 平行运算原本仅用于基于 x86 主机的PC与服务器中。
Nvidia CEO 黄仁勋介绍首款Project Denver芯片── Parker的开发蓝图。
h8pesmc
Nvidia 并展示一款搭载 Tegra 3 ARM 处理器与 Kepler GPU 的 Kaylar 主机板原型，可支持 Cuda 与 OpenGL 4.3 执行。The Linley Group 公司首席分析师Kevin Krewell认为 Kayla 将可用于为 Logan 进行编程。这款 Kayla 主机板原型预计将在今年出样，2014年生产。
本文授权编译自EE Times，版权所有，谢绝转载
本文下一页：迈向统一内存架构
相关阅读：
• 三星一马当先，将在ISSCC展示big.little处理器
• SanDisk与NVIDIA合作推出Tegra 4平板电脑参考设计
• Python支持NVIDIA CUDA: GPU加速计算惠及新一代程序员h8pesmc
{pagination}
迈向统一内存架构
Nvidia 的下一代桌上型产品是 Maxwell GPU 。计划在2014年推出的 Maxwell 将支持可在 GPU 与 ARM CPU 之间共享虚拟内存的能力。 针对封装 GPU 以及ARM或 x86 CPU 的芯片而言，这是简化编程作业的一个关键步骤。有鉴于此，AMD还成立了一个异质系统架构(HSA)基金会，以期建立起开放的标准。HSA成员包括 ARM、Imagination Technologies、微软、高通等公司。
Kevin Krewell与Patrick Moorhead等业界分析师们预期，Nvidia可能会将HSA基金会定义的开放技术应用于 Cuda 中。但这有点像英特尔被迫采用64位指令集的情况一样──由于AMD先定义了可支持微软的64位 x86 架构，而微软也不愿意为英特尔和AMD分别开发两套不同的64位操作系统。
然而，这一次，展开一个全新的架构就像是与AMD、Nvidia和英特尔三方为敌相互竞争一样。据推测，英特尔打算走自己的路，推出可在绘图与 x86 核心间共享内存的SoC。其中还可能包含英特尔现有 GPU 及其外部 Xeon Phi。但确切的计划尚未透露。
h8pesmc
事实上，微软再也不是这个领域中唯一或最大的操作系统供货商了。Google正计划以其 Android 加以取代，期望进一步成为未来客户端计算机的主流。
因此，在迈向共同的架构发展道路上，目前我们已经看到三方微处理器竞争阵营以及两家 OS 竞争供货商(当然如果把苹果 iOS 也算进去的话就有三家了)。未来在前方的发展道路上，还将出现更多曲折。
本文授权编译自EE Times，版权所有，谢绝转载
本文下一页：Parker将选择哪一种种OS？
相关阅读：
• 三星一马当先，将在ISSCC展示big.little处理器
• SanDisk与NVIDIA合作推出Tegra 4平板电脑参考设计
• Python支持NVIDIA CUDA: GPU加速计算惠及新一代程序员h8pesmc
{pagination}
Parker将选择哪一种种OS？
Nvidia开发的 Parker 微处理器将结合其 Maxwell GPU 以及首款64位ARM CPU 。预计在2015年发表的 Parker 将采用Globalfoundries与台积电(TSMC) 20nm FinFET 制程技术制造。
当然， Parker 将成为智能手机和平板电脑的 Tegra 6 SoC。然而，它是否也将应用在笔记本计算机、台式机或服务器领域，目前仍不得而知。
微软专为平板电脑开发了基于ARM的WindowsRT，至今并未取得什么成绩，而且亟需加以大幅改善。微软至少要在平板电脑市场取得成功后，才可能再推向笔记本计算机和台式机。
同时，Google Android 系统在笔记本计算机或台式机上也没听说有什么计划。这一点也不令人意外，因为该公司现在只定位于取得智能手机与平板电脑的主导地位。
Google已经尝试推出一款精简版 Chrome OS 了，可让浏览器执行在基于云端的笔记本计算机上。这项产品目前已经推出到第二代了，仍未能获得多大的吸引力。
但令人玩味的是，执行 Chrome OS 计划的Google公司资深副总裁Sundar Pichai，同时还负责整并 Chrome OS 和 Android 两个部门。一手打造 Android 的 Andy Rubin 黯然退场，另寻其它机会。
此外，Nvidia公司还表示将在2015年推出 Volta ，这是采用 TSV 技术堆栈内存的下一代 GPU 。该公司并未透露它是否将采用赛灵思公司率先开发的 2.5D 技术──在共同基板的相邻两侧或上下堆栈内存与 GPU ，以实现更远大的目标。
Nvidia CEO 黄仁勋介绍2015年将推出采用TSV技术堆栈内存的下一代 GPU──Volta。
h8pesmc
本文授权编译自EE Times，版权所有，谢绝转载
本文下一页：Nvidia的未来仍将充满变量
相关阅读：
• 三星一马当先，将在ISSCC展示big.little处理器
• SanDisk与NVIDIA合作推出Tegra 4平板电脑参考设计
• Python支持NVIDIA CUDA: GPU加速计算惠及新一代程序员h8pesmc
{pagination}
分析师指出，在2015年以前，包括AMD和其它厂商都可能开发出某种形式的 GPU 和内存堆栈。但我怀疑没人知道究竟在2015年时采用 2.5D 或 3D 是否最适合。但由于 EUV 微影技术延迟导致下一代芯片制程成本越来越高且复杂，这两种技术都将受到更多关注。
因此，我们在GTC只能掌握到一点点有关Nvidia在未来几年的开发计划。其结果是对于未来两三年内客户端计算机领域将出现的变化所知仍十分有限，可预期的是未来仍将充满变量。
h8pesmc
本文授权编译自EE Times，版权所有，谢绝转载
编译：Susan Hong
参考英文原文：Nvidia's road to Denver still in the shadows，by Rick Merritt
相关阅读：
• 三星一马当先，将在ISSCC展示big.little处理器
• SanDisk与NVIDIA合作推出Tegra 4平板电脑参考设计
• Python支持NVIDIA CUDA: GPU加速计算惠及新一代程序员h8pesmc
{pagination}
Nvidia's road to Denver still in the shadows
Rick Merritt
In 2007, Nvidia quietly began work on Project Denver, starting a long hard climb toward being a supplier of general-purpose microprocessors. At CES in 2011 it revealed the goal of that program was to create a soup-to-nuts family of SoCs serving everything from laptops to supercomputers.
At its annual Graphics Technology Conference (GTC) here Tuesday (March 19) Nvidia revealed the first commercial chip it expects to emerge from that program. The chip, Parker, will arrive in 2015 packing a 2014 graphics core called Maxwell and a new 64-bit custom ARM core.
Details are still scarce, but what Nvidia did share shines just a little light on the microprocessor road map for the industry, enough to give me pause about the future of computing.
Nvidia’s next big move is Logan. This Tegra 5 SoC will put a Kepler-class Nvidia graphics processor in a chip aimed at smartphones and tablets. It will be the first ARM-based mobile SOC to support its Cuda software environment for running general purpose graphics (GPGPU) programs. Previously, Cuda worked in PCs and servers that used an x86 host.
Click on image to enlarge.
Nvidia chief executive Jen-Hsun Huang shows the road to Parker, the first Project Denver chip.
To some extent the part plays catch up with its competitors that already provide similar features using OpenCL, said Kevin Krewell, a principal analyst with The Linley Group (Mountain View, Calif.).
Nvidia demonstrated it already has Cuda running on a new board called Kayla that sports a Kepler graphics chip and Tegra 3 ARM processor. Krewell called Kayla “a science experiment and a software development platform” to seed programming work for Logan which will sample this year and be in production in 2014.
The race to unified memory
The next big step is Maxwell, Nvidia’s next big graphics core. It will arrive in 2014 sporting a capability to share virtual memory between the graphics and ARM host processor.
This is a key step in simplifying the job of programming chips that pack graphics and host ARM or x86 processors. It is so important, AMD launched an alliance called the Heterogeneous Systems Architecture (HSA) group to set open standards for how to do it. HSA includes ARM, Imagination Technologies, Microsoft, Qualcomm and many others.
Analysts such as Krewell and Patrick Moorhead expect Nvidia will probably borrow the open techniques the HSA group defines and plug them into Cuda, using different marketing names. That’s pretty much what Intel did with its 64-bit instructions because AMD defined the 64-bit x86 architecture first working with Microsoft and the Windows giant refused to support two different flavors of the technology in its operating systems.
But this time, the move to a new architecture is a three-way game with AMD, Nvidia and Intel all competing. Presumably Intel is taking its own route to SoCs that share a common pool of cache coherent memory between graphics and x86 cores It likely will embrace both its current on-chip graphics cores and its external Xeon Phi. But it has yet to reveal its exact plans.
There’s another new wrinkle this time. Microsoft is not the only or even the biggest player in operating systems anymore. Google is set to displace it with Android as the client OS observers expect to run in the majority of tomorrow’s client computers.
So we have three competing microprocessor camps and two competing OS providers (three if you count Apple which will do its own thing) all racing toward one common architecture. There could be a lot of twists ahead in this road map.
Parker arrives amid OS uncertainty
For its part, Nvidia will combine its Maxwell graphics core with its first 64-bit ARM core to create Parker in 2015. It will be made in a FinFET process which by then should be readily available as a sort of second-generation 20-nm node from Globalfoundries and TSMC.
Parker will certainly power smartphones and tablets as a Tegra 6. What’s less clear is whether it will have a life in notebooks, desktops or servers.
Microsoft’s WindowsRT for ARM—aimed at tablets--has failed to gain significant traction and badly needs a makeover. Until it gains traction in tablets, Microsoft is not likely to try to drive it into notebooks and desktops. And we are still waiting to hear anything about Windows Server for ARM.
Meanwhile, Google has been mum on any plans for Android in notebooks or desktops. No surprise really--it is only now well positioned for a dominant role in smartphones and tablets.
Google has experimented with a slimmer Chrome OS layer that essentially puts a browser on a cloud-based laptop client. The effort, now in a second generation, has failed to gain much traction.
But, interestingly, the Google exec (Sundar Pichai) who ran the Chrome OS initiative was put in charge of a merged Chrome OS and Android group. Android founder Andy Rubin left the team to explore other opportunities, as the old euphemism goes.
One more wrinkle: Nvidia said in 2015 it will also roll out Volta, a next generation graphics processor using stacked memory with through silicon vias. It did not say if it will use a 2.5-D technology pioneered by Xilinx that lays memory and graphics cores next to each other on a common substrate or actually stack them on top of each other, a more ambitious goal.
Click on image to enlarge.
Huang discloses the 2015 graphics chip Volta chip with stacked memory.
Analysts say AMD and others are likely to do some form of graphics and memory stack by 2015. I suspect no one knows yet whether the 2.5- or 3-D techniques will be the best fit come 2015. But both these techniques are getting a lot of attention as the next steps in silicon process technology become increasingly costly and complex due to the delays in extreme ultraviolet lithography.
So at GTC we learned just a little bit of what Nvidia plans to do over the next couple years. The result is we now see how little we know about what anyone will do in the client computing space for the next two or three years. But we can see it is going to be something of a wild ride.
责编：Quentin
扫码分享到好友
