广告
阿波罗登月回顾、AI基准测试、究竟什么是车?
00:00
00:00

BRIAN SANTO: I'm Brian Santo, EETimes editor in chief, and you're listening to EETimes on Air. This is your briefing for the week ending July 19th.

BRIAN SANTO:我是EETimes总编辑Brian Santo,你正在的是EETimes全球联播。以下是截至719日的一周新闻要点。

This week…

本周…

Artificial intelligence is a vastly complex market. There’s a fierce competition among hardware vendors to be the best platform for AI applications. But first, you have to know what it means to be “the best.” This week, analyst Karl Freund from Moor Insights talks to us about the latest AI benchmarks.

人工智能是一个非常复杂的市场。硬件供应商之间的竞争异常激烈,竞相角逐最好的AI应用平台。但首先,你必须知道“最好”是指什么。本周,Moor Insights分析师Karl Freund向我们介绍了最新的AI基准测试。

Researchers are re-imagining what – fundamentally – a vehicle is. One company just put everything other than the chassis – literally everything – entirely inside the wheels. We’ll explore that and other proposals.

研究人员在重新想象:从根本上说车是什么。有一家公司将所有东西都放在车轮内,除底盘外。我们将探讨这个想法以及其他方案。

And it’s the 50th anniversary of the Apollo 11 space mission – the first time people set foot on the moon. This week we look back – and also look forward to going back to the moon. 

接下来是阿波罗11号太空飞行50周年,纪念人类第一次踏上月球。本周我们不但回顾过去,也展望未来,期待着重返月球。

GEORGE LEOPOLD "The dust is a major problem. It sticks to everything. It's highly abrasive material. It gets into everything. I think on the first lunar landing it became a big problem, so you have to find a way to manage that before we can go back there."

GEORGE LEOPOLD “灰尘是一个主要问题,它会附着到任何东西上。它是高度磨蚀的材料,会进入任何东西内。我认为在第一次登月时它就是一个大问题,所以在我们能够返回月球之前必须找到一种办法来控制它。“

BRIAN SANTO: We’ll get back the moon in a moment.

BRIAN SANTO:我们稍后就会返回月球(话题)

First up: Artificial intelligence is going to be the next big – no, ginormous – market for computer chips. There’s fierce competition among IC vendors to design platforms that can host various AI applications. Naturally, each wants to prove their chips are the best, and for that – they need benchmarks. There’s a new set called M-L PERF. EE Times Editor Rick Merritt talks to analyst Karl Freund, who wrote the story for us about the new benchmarks.

首先,人工智能将成为计算机芯片的下一个大市场,不是“大”,而是“巨大”。 IC供应商们正在激烈竞争,都在设计可以支持各种AI应用的平台。当然,每家公司都想证明他们的芯片是最好的,为此就需要基准测试。有一个新的基准测试称为M-L PERFEE Times编辑Rick Merritt与分析师Karl Freund就此进行了交谈,后者为我们撰写了有关这个新基准测试的文章。

以下内容是英文原稿,Enjoy!

RICK MERRITT: Thanks for making some time today, and thanks last week for covering for us the new results that came out of M-L PERF. So what did we learn about kind of who's in the lead of this AI race from those results?

KARL FREUND: I think what we learned from the benchmarks is that both Google and Nvidia have just some incredible hardware and have made significant investments in software. A lot of the performance improvements actually came from the software innovations in tuning and really focusing on the AI algorithms, and so they're both able to achieve some very, very impressive speed ups with their hardware GPUs in the case of Nvidia and the TPU in the case of Google.

RICK MERRITT: Very good. Now, we both know that there's dozens of companies that are coming out with AI training accelerators. A lot of them are startups. Why are none of them yet putting out any results?

KARL FREUND: That's a great question, Rick. You know, I think the anticipation of the 40-some startups around the world that are all building hardware to accelerate artificial intelligence, expectations are set very high by the companies and by the industry. And those expectations are difficult to meet. So there's really not a lot of hardware available today from these startups. And what is available is primarily for inference, the actually processing of training neural networks. Not the development of the neural networks themselves, a process called training.

In the realm of training, there's really one two players: Nvidia and Google. Intel hopefully will have a competitive product by the end of this year. And the training market is right now the largest market. But in the long term, the inference market is probably going to be the bigger market opportunity.

RICK MERRITT: All right. Good enough. So what's the milestone here, Karl?

KARL FREUND: Well actually, I think what's next is going to be the inference benchmarks. They have not been published yet by any of the players, and I think that is an opportunity for the startups and large companies, Qualcomm and Intel, to be able to demonstrate their prowess in inference processing.

I would also note that the cost of the training is still quite high. The systems that ran these benchmarks are massive supercomputers. I estimate the Nvidia supercomputer that ran this benchmark is probably close to about a $40 million US list price. So training is still very expensive, costs have got to come down, and the second thing is that we hopefully will see some inference benchmarks that we can use to assess the startups' potential in penetrating that market.

RICK MERRITT: All right. Karl Freund, analyst for AI, with more insights and strategy. Thanks so much for being on EETimes On Air.

BRIAN SANTO: Innovation in the automotive industry is hardly confined to electric vehicles or autonomous vehicles. EE Times has come across several wild new proposals for how to build a car in the future. International editor Junko Yoshida talked with our London correspondent Nitin Dahad about it.

JUNKO YOSHIDA: Nitin, you recently reported on an Israeli startup who has developed a modular EV platform you said. Presumably all the components previously found under the hood are now in the wheel, as you wrote. What does that entail? Can you explain?

NITIN DAHAD: Yes, Junko. I had spoken to Daniel Barel, co-founder of the Israeli startup REE, which has just come out of stealth mode with a wheel design that integrates the motor, steering, suspension, drive train, sensing, electronics, rim and tire all in the wheel. That's quite a mouthful. But they do claim to have put all of that into the wheel.

JUNKO YOSHIDA: Wow. All right, so that's a lot. So take me from the top. What sort of advantages will such a platform bring to car OEMs and Tier Ones?

NITIN DAHAD: It’s quite interesting, actually. I just attended an intelligent mobility conference in London, where Benny Daniel, VP with Frost & Sullivan, said the existing E&E architecture in cars is really left wanting. And he implied that conventional platforms are outdated. He said the direction of the industry is actually a skateboard platform, one which is clearly adopted by Tesla. He said that by 2025, thirteen OEM brands will launch battery/electric vehicles on a skateboard platform.

JUNKO YOSHIDA: Just for our listeners, can you explain what “skateboard platform” means?

NITIN DAHAD: Actually even I found it very difficult to understand when I first read it in the REE press release when they sent it to me. But I think it's essentially what it says. It's a skateboard. And this was sort of invented by GM back in 2002. What you've got you is, the platform is on what looks like a skateboard. And then what REE have done is actually put everything for the car into the wheel.

JUNKO YOSHIDA: So, you said that Tesla is using a skateboard platform. What they're doing is different from what REE is pitching here, right? Can you tell me the difference?

NITIN DAHAD: The difference is the skateboard platform that GM advocated is basically saying, let's make it a clean platform where everything can be built on the platform and everything above can be a space. But I think a lot of companies are using some for of that platform, and Tesla is obviously using that. But if you look at the Tesla, they do use a lot of the space quite effectively above that platform. And what REE is actually doing is integrating everything into the wheel, so that the whole auto platform is contained within the skateboard, and then that opens up the flexibility for OEMs to use that single platform to create everything on top, from robotaxis to cars to trucks.

JUNKO YOSHIDA: Oh, I see. So that's essentially making it as a “universal platform,” as you wrote.

NITIN DAHAD: Correct.

JUNKO YOSHIDA: All right. So as good as it sounds, though, I'm a bit of a skeptic here because I wonder... the OEMs, traditional OEMs, may resist this idea because they have invested so much for the current automotive design. We're talking about changing the architecture in terms of the wiring and cabling they used to do to this new skateboard. Plus, the OEMs already have a production, well-oiled production lines, that are designed for the mass production of the conventional cars. So is REE, this startup, are they ready to respond to the potential resistance from OEMs?

NITIN DAHAD: Well, yes, you would have obviously a lot of resistance. There's a lot of existing manufacturing infrastructure. But if you think about it, if it's eliminating-- I think one of the aspects you said was about all the cabling and infrastructure-- if the wheel is really getting rid of all of that, then they'll actually probably welcome that. But I'll actually say, any new disruptive technology, there's going to be resistance, but you just need a small handful of early adopters to see what's possible and get it out there in the market.

If it takes off and it can demonstrate significant cost savings and shorter time to deliver that return on investment, then there might be gradual adoption. And I think that happens-- I mean, I remember (I don't want to keep on) when I was involved with a startup, we were disruptive, but eventually it takes off.

JUNKO YOSHIDA: So cut to the chase. When will the new platform be ready?

NITIN DAHAD: The CEO told us they will present early prototypes to partners during 2019, but he wasn't really more specific than that. They're still deciding on what market to address first, because obviously if it's a universal framework, it can go for anything. As we said, from trucks to robotaxis to cars. But I don't think cars will be their first market. As you said, there's going to be resistance. So it will be probably the low-speed vehicles and commercial vehicles which are the low-hanging fruit. And that’s what he said they'll probably target first.

JUNKO YOSHIDA: All right, very good. Thank you so much, Nitin.

NITIN DAHAD: Thank you.

BRIAN SANTO: This Saturday, July 20th, will be the 50th anniversary of the first step taken on the moon. We’ve got a set of stories on the website to commemorate the event, including a series we did to mark the 40th anniversary, along with several brand new articles, including a few by George Leopold, who is not only a long-time EE Times editor, but is also an authority on the NASA space program. George has authored a book on one of America’s great astronauts, Gus Grissom, who participated in the Apollo, Gemini, and Mercury programs. Welcome back to the podcast, George.

GEORGE LEOPOLD: Hey, Brian, good to be back with you.

BRIAN SANTO: I have a really clear memory from 50 years ago. My mom and dad getting me out of bed and dragging me downstairs to watch Neil Armstrong walk on the moon. As sleepy as I was, it was incredibly exciting. And I think every kid alive in the 1960s considered becoming an astronaut. How about you? What are your memories from back then?

GEORGE LEOPOLD: Well, like a lot of folks my age, my generation, we sort of remember that night as magical. It was... the night of July 20th was like Christmas Eve. Two guys coming... fuzzy figures coming down that ladder. It was unforgettable. I guess even at a young age, you sort of sensed that this was something historic, something that perhaps for the first time in a long time sort of pulled the whole planet together. And we did it; it wasn't just the Americans, but we all did it. So it was a very magical thing, and a bit ephemeral but nevertheless I think it was significant. And maybe those images alone made that $180 billion in 2019 dollars worth the effort.

BRIAN SANTO: Ten years ago, EE Times ran a special report, looking back on Apollo 11 on the 40th anniversary. Now we’ve re-posted that on the website by the way – on EETimes.com. You contributed to that report, and you were already pretty prepared for it, right?

GEORGE LEOPOLD: Yeah, 10 years ago we did a special report, the 40th anniversary of Apollo 11. One of the people we interviewed, Allen Bean, said the reason everybody showed up for the 40th anniversary was that he checked the actuarial tables and realized that a lot of these guys wouldn't be here for the 50th. And unfortunately, he was right. There are only a few of the people who walked on the moon or saw the whole circle of the Earth alive today. So, luckily, too, the Apollo 11 astronauts-- the crew of the first moon landing-- are here. Mike Collins and Buzz Aldrin

We're profiling the great Mike Collins in EETimes this week. He's one of my favorites. His conclusion was that, What did this all mean? It was about leaving. It was about leaving the Earth for the first time and looking back and seeing the whole Earth and realizing what we have down here. So I think it changed everybody's perspective. And of course the famous "Earthrise" photograph and so forth. It changed our whole view of where we fit in the universe. So again, that alone probably made the whole thing worth it.

BRIAN SANTO: So what’s changed in the 10 years since then?

GEORGE LEOPOLD: Well, I guess, you know, in the last 10 years since we took a look at this, the 40th anniversary to the 50th anniversary of Apollo 11, we've seen the whole commercial space industry grow up. If we're going to get back to the moon, it's probably going to be an industry-NASA effort, with the industry guys supplying most of the hardware and NASA sort of looking over their shoulder and making sure the software works and getting everything right. Because everything has got to be right. This is a dangerous, high-risk business. Space is unforgiving.

At some point, Elon Musk or Jeff Bezos or one of these guys is going to have an accident. Someone's going to get killed, and then we'll find out how resilient their business model is. But that's just the nature of space travel.

BRIAN SANTO: If we might be going back to the moon, I have to bring this up, because it’s new information to me. I always knew that without an atmosphere for protection, radiation would be a serious problem. And recently we discovered on the plus side, there might be ice to mine on the moon, a source of oxygen to breathe and hydrogen for fuel. But what I just learned is that moondust is incredibly vicious, harsh stuff. It’s way tougher to go back to the moon than I ever realized. Way many more challenges. Is this any more involved an effort than, say, going to Mars or asteroid mining?

GEORGE LEOPOLD: Yeah, if we go back to the moon, half the problem is finding a way to get back. Right now we don't have a way to get back, but they're talking a lot about it. But we'll see if any of that pans out. But one of the big problems, as you point out, is yeah, the dust is a major problem. It sticks to everything, it's highly abrasive material, it gets into everything. I think on the first lunar landing they were getting moondust in the connections between their suits and their gloves, and it became a big problem. So you have to find a way to manage that before we can go back there and do whatever we're going to do on the moon in terms of getting at the water that you could use as fuel to go to Mars or out to an asteroid, something like that.

I asked the writer Homer Hickham, who maybe some of the listeners knows from "October Sky" about the asteroid mining. He was of course a miner's son in West Virginia, and he said he looked at the board of directors of some of these companies, startups, that are proposing asteroid mining, and he said they didn't have any miners on their board of directors. So they need to focus in on the details here to get it right.

BRIAN SANTO: Okay, last question: Are you doing anything to celebrate on July 20th? 

GEORGE LEOPOLD: Well, on July 20th I will be at Purdue University, the alma mater of the subject of my biography, Gus Grissom, as well of course as the alma mater of the first man on the moon, Neil Armstrong. The engineering building at Purdue University is named for Neil Armstrong. There's a statue of his out in front. They're obviously, and rightly so, quite proud of him, and also the last man on the moon, Gene Cernan, is also a moonwalker and a Purdue graduate. So I think Purdue has more astronauts than any other... has produced more astronauts than any other university. So they're quite proud of their tradition. And as they say at Purdue, "boiler up."

BRIAN SANTO: Okay. That was George Leopold, EE Times editor and author of the book, “Calculated Risk: The Supersonic Life and Times of Gus Grissom.” It’s a great read. Available wherever fine books are sold.

So have you had your fill of nostalgia yet? The answer is no, you have not.

Another milestone in space exploration field happened 44 years ago this week. On July 17th, 1975, Apollo 18 and Soyuz 19 docked together, the first time two vessels from two different countries met in space.

Also on July 17th, this time way back in 1850, astronomers at the Harvard Observatory snapped the first photograph of a star. It was taken through the Great Refractor, one of the two most powerful telescopes on Earth at the time; it had a 20-foot-long mahogany veneer tube with a lens 15 inches in diameter. The photo – technically a daguerreotype – was of the star Vega, the second brightest star visible in the northern hemisphere.

On July 18, 1968, Robert Noyce, Andy Grove and Gordon Moore incorporated a company. They smooshed together the words "integrated" and "electronics" to create the company’s name.

I interviewed Bob Noyce in 1988 when he was named the chairman of Sematech. Sematech is consortium of US chip vendors and US fab equipment suppliers dedicated to researching and developing new manufacturing technologies. It was then only a year old. It was a response to the growing sophistication and success of rival Japanese companies. At the time, there was a lot of consternation about that. I asked Noyce about the threat of foreign competition. His answer was: you do everything you can to remain competitive yourself, and if everyone helps to make the pie bigger, what's the problem with competition?

I wonder how many people would agree with that today.

Well, that’s your Weekly Briefing for the week ending July 19th.

This podcast is Produced by AspenCore Studio. It was Engineered by Taylor Marvin and Greg McRae at Coupe Studios. The Segment Producer was Kaitie Huss.

The transcript of this podcast can be found on EETimes.com, complete with links to the articles we refer to. Be sure to join us next week for your July 26th Weekly Briefing on EE Times On Air. I’m Brian Santo.

感谢收听本期推送,全球联播 (EE|Times On Air) 现已同期在喜马拉雅以及蜻蜓FM上线,欢迎订阅收听!
广告