来源:http://bit.ly/2DyEddv
Tesla Raises the Bar for Self-Driving Carmakers
In unveiling the specs of his new self-driving car computer at this week’s
Tesla Autonomy Day investor event, Elon Musk made several things very clear
to the world.
Tesla 在本周面向投资人的"Tesla Autonomy Day" 上发布了其新自驾车电脑的
规格。同时,Elon Musk 也在活动上向世界传递出了几条非常清楚的讯息。
First, Tesla is raising the bar for all other carmakers.
首先,Tesla 正在提高所有其他汽车制造商的门槛。
Second, Tesla’s self-driving cars will be powered by a computer based on two
of its new AI chips, each equipped with a CPU, GPU, and deep-learning
accelerators. The computer delivers 144 trillion operations per second
(TOPS), enabling it to collect data from a range of surround cameras, radars
and ultrasonics and power deep neural network algorithms.
其次,Tesla 的自动驾驶车将由一台基于两颗全新AI 芯片的电脑所驱动,每颗
芯片都配备了CPU、GPU 以及深度学习加速器。该电脑运算能力每秒可达到144
TOPS,这使其能够从一系列的环绕镜头、雷达和超音波雷达中收集数据,同时驱
动深度神经网络算法。
Third, Tesla is working on a next-generation chip, which says 144 TOPS isn’t
enough.
第三,Tesla 正在研发下一代芯片,这表示144 TOPS 的运算能力是不够的。
At NVIDIA, we have long believed in the vision Tesla reiterated: self-driving
cars require computers with extraordinary capabilities.
NVIDIA 长期以来一直相信Tesla 不断重申的愿景:自动驾驶车辆需要性能卓越
的电脑。
Which is exactly why we designed and built the NVIDIA Xavier SoC several
years ago. The Xavier processor features a programmable CPU, GPU and deep
learning accelerators, delivering 30 TOPs. We built a computer called DRIVE
AGX Pegasus based on a two chip solution, pairing Xavier with a powerful GPU
to deliver 160 TOPS, and then put two sets of them on the computer, to
deliver a total of 320 TOPS.
这也正是几年前我们设计并制造NVIDIA Xavier SoC 芯片的初衷。NVIDIA Xavier
处理器具有可编程的CPU、GPU 以及深度学习加速器,可提供30 TOPS 的运算能
力。我们还打造了一个名为DRIVE AGX Pegasus 的电脑。DRIVE AGX Pegasus基
于双芯片解决方案,一个Xavier 搭配一个强大的独立GPU,可以提供160 TOPS的
运算能力,将两组这样的芯片装载到一台电脑中,可实现总共320 TOPS 的运算
能力。
And as we announced a year ago, we’re not sitting still. Our next-generation
processor Orin is coming.
如同一年前所宣布的那样,我们未曾止步。新一代处理器–Orin 即将到来。
That’s why NVIDIA is the standard Musk compares Tesla to—we’re the only
other company framing this problem in terms of trillions of operations per
second, or TOPS.
这也是为什么Musk 将NVIDIA作为与Tesla 比较的参考基准 — 我们是唯一一间
除Tesla 之外提供每秒数兆次级运算(或称TOPS)作为解决方案的公司。
But while we agree with him on the big picture—that this is a challenge that
can only be tackled with supercomputer-class systems—there are a few
inaccuracies in Tesla’s Autonomy Day presentation that we need to correct.
尽管我们同意他所描述的自动驾驶发展蓝图,即我们面临着一个只能用超级电脑
级的系统来应对挑战的未来,但我们依然需要对一些Tesla Autonomy Day 中不
准确的介绍进行更正。
It’s not useful to compare the performance of Tesla’s two-chip Full Self
Driving computer against NVIDIA’s single-chip driver assistance system. Tesla
’s two-chip FSD computer at 144 TOPs would compare against the NVIDIA DRIVE
AGX Pegasus computer which runs at 320 TOPS for AI perception, localization
and path planning.
拿Tesla 的双芯片全自动驾驶电脑与NVIDIA 的单芯片驾驶辅助系统进行性能对
比是没有意义的。运算能力达到144 TOPS 的Tesla 的双芯片FSD 电脑应与能在
AI 感知、定位以及路径规划方面提供320 TOPS 的NVIDIA DRIVE AGX Pegasus
进行比较。
Additionally, while Xavier delivers 30 TOPS of processing, Tesla erroneously
stated that it delivers 21 TOPS. Moreover, a system with a single Xavier
processor is designed for assisted driving AutoPilot features, not full
self-driving. Self-driving, as Tesla asserts, requires a good deal more
compute.
此外,Xavier 拥有30 TOPS 的运算能力,但Tesla 错误地宣称它只拥有21 TOPS
的处理性能。而且,使用单个Xavier 处理器的系统是针对辅助驾驶AutoPilot
的特点设计的,而不是为了全自动驾驶而设计。正如Tesla 所说,自动驾驶需要
更大量的计算。
Tesla, however, has the most important issue fully right: Self-driving cars—
which are key to new levels of safety, efficiency, and convenience—are the
future of the industry. And they require massive amounts of computing
performance.
然而,Tesla 在最重要问题上的看法是完全正确的:自动驾驶车辆是提高安全、
效率和便利性的关键所在,也是整个行业的未来。而做到这些需要大量的运算能
力。
Indeed Tesla sees this approach as so important to the industry’s future
that it’s building its future around it. This is the way forward. Every
other automaker will need to deliver this level of performance.
事实上,Tesla 认为强大的运算能力对自动驾驶行业的发展未来十分重要,这也
成为了Tesla 未来发展的核心。这是发展向前的方向。每个汽车制造商都应该提
供这种水平的性能。
There are only two places where you can get that AI computing horsepower:
NVIDIA and Tesla.
如果你想获得强大的AI 计算性能,只有两种选择:NVIDIA和Tesla。
And only one of these is an open platform that’s available for the industry
to build on.
但两者之中只有一个能够提供可供整个行业使用的开放平台。
=============================================================================
这篇文章是Tesla 在发布其FSD 电脑后没几个小时Nvidia 就公布的文章,在捧
Tesla 的同时也不忘纠错,最后顺便还强调一下只有他们有在卖跟Tesla 同级的
自动驾驶电脑(不跟我们买你就自己开发吧 XD)。
从这篇文章大概就能看出大约需要怎么样的能力才足够供给全自动驾驶车辆来做
运算。不过Nvidia 也还是省去了自己的缺点不说。首先是他们有320 TOPS 运算
能力的Pegasus 还在开发当中,可能至少还要1~2 年才能提供成品。其次就是功
耗太高,TDP 500W。而Tesla 在设计之初就将功耗考虑进去,目标是要做到小于
100W,最后的成果是72W。这就跟当初Google 在发展深度学习时放弃以GPU 为主,
自行研发TPU 一样,能耗是一个很重要的关键,有兴趣可以参考以下的文章:
仅需1/5成本:TPU是如何超越GPU,成为深度学习首选处理器的
http://bit.ly/2UWKF8L
Google 研发了一块芯片,省下建资料中心的钱还推动机器学习的发展
http://bit.ly/2DAAXyd
同时Elon 也提到,全自动驾驶每英里的电耗是250W,若以平均时速30 km/h 来
计算,那大约是每小时要耗费约4.69度电,而若不用72W 的Tesla FSD 电脑改用
500W 的Pegasus,等于每小时要再额外花费约0.43度电,这相当于整体功耗多了
8.67%,关于功耗这问题就是Nvidia 没有提的,他们还没解决。
另外成本也是一个重点,Tesla 在设计之初就决定了要兼顾性能、功耗、与成本,
最后做出的成果也是性能比起以前Nvidia 提供的Drive PX 2 要好上一个数量级,
同时成本只有Drive PX 2 的80%。
╭────────────┬────┬────┬───╮
│ │芯片数量│运算性能│功耗 │
├────────────┼────┼────┼───┤
│Tesla FSD Computer │ 2 │144 TOPS│ 72 W│
│Drive PX 2(Tesla HW 2.5)│ 3 │ 12 TOPS│ 60 W│
│Drive PX Xavier │ 1 │ 30 TOPS│ 30 W│
│Drive PX Pegasus │ 4 │320 TOPS│ 500 W│
╰────────────┴────┴────┴───╯
Tesla FSD Computer,两颗自研的SoC 芯片,功耗72W:
https://i.imgur.com/60khZxi.jpg
Drive PX 2,两颗Parker SoC 芯片 + 两颗辅助Pascal GPU,功耗250W:
https://i.imgur.com/aaZj5db.jpg
基于Drive PX 2 所订制的Tesla HW 2.5,两颗SoC 芯片 + 一颗辅助GPU,功耗60W:
https://i.imgur.com/4L5wnkN.png
Drive PX Xavier,单颗SoC 芯片,功耗30W:
https://i.imgur.com/oOhaAw0.jpg
Drive PX Pegasus,两颗Xavier SoC 芯片 + 两颗GPU,功耗500W:
https://i.imgur.com/sinVMOh.jpg
https://i.imgur.com/OXuzhcU.jpg