Google TPU vs. Nvidia V100

Google as the GPU maker? GPU => Google Processing Unit?

Google 'Cloud TPU' takes machine learning lead from Tesla V100
Mark Tyson on 18 May 2017, 10:01
http://hexus.net/tech/news/industry/105838-google-cloud-tpu-takes-machine-learning-lead-tesla-v100/

"...As CNBC reports, the reveal of the Google Cloud TPU last night is "potentially troubling news for Nvidia, whose graphics processing units (GPUs) have been used by Google for intensive machine learning applications."

In its most recent financial report Nvidia pointed to fast growth in revenues from AI and deep learning, and even cited Google as a notable customer.

Now Google has indicated that it will use its own TPUs more in its own core computing infrastructure. Google is also creating the TensorFlow Research Cloud, a cluster of 1,000 Cloud TPUs that we will make available to top researchers for free...

Last but not least Google is happy to help with software and will bring second-generation TPUs to Google Cloud for the first time as Cloud TPUs on GCE, the Google Compute Engine. It will facilitate the mixing-and-matching of Cloud TPUs with Skylake CPUs, NVIDIA GPUs, and all of the rest of our infrastructure and services to build the best ML system.

So how do Google's new Cloud TPUs perform? Google says each TPU module, as pictured above, can deliver up to 180 teraflops of floating-point performance. That module features 4x Cloud TPU chips (45 teraflops each). These devices are designed to work in larger systems, for example a 64-TPU module pod can apply up to 11.5 petaflops of computation to a single ML (machine learning) training task.

Roughly comparing a Cloud TPU module against the Tesla V100 accelerator, Google wins by providing six times the teraflops FP16 half-precision computation speed, and 50 per cent faster 'Tensor Core' performance. Inference performance of the new Cloud TPU has yet to be shared by Google.

Furthermore, Cloud TPUs "are easy to program via TensorFlow, the most popular open-source machine learning framework," says Google."



Googleflops?

 

If AI will replace software, what will replace Company CEO's?

Jen-Hsun Huang thinks hardware will exterminate code
http://fudzilla.com/news/processors/43697-jen-hsun-huang-thinks-hardware-will-exterminate-code

"Nvida supreme dalek Jen-Hsun Huang believes that software will be exterminated by AI."

Comments...

Steven De Bondt • 10 hours ago
Yes, [lets] all listen to the guy that sells gaming gear to teenagers in a leather jacket. Lets hear him out.

Best comment, ever.

 

Maybe its Vega licensed to google.

 

AI vehicles
AI advertisement
AI designers
AI politicians
AI judges
AI doctors
...

Looks like the general society will be less and less capable of thinking for themselves and be more and more reliant on the tech doing it for them...

I for one like to drive vehicles myself and make my own decisions...and so does brother Jen.

 

AI BIOS Builders / Tuners?

 

Not far away from that, just wait for the next version of Turbo Boost training itself according to each specific chips potential instead of following variables...
Remember my words: It'll be even more boring for the user to take the backseat, as there will be no more room for manual tweaking...silicon lottery will be the only variable.

 

Soooo... it'll all just be "fun and games" moving forward?

Sounds frighteningly horrible

 

A part of me would die if I had an AI driving car. Give me stick shift and a windy road any day.

 

Believe I read that it was asic

 

Maybe ASIC/GPU designed for one thing that is machine learning.

 

A computer algorithm is far better/safer at driving than a Human is (humans in general are pretty bad at it). Furthermore, you are not bogged down by the prospect of having to drive in the first place (which can be/is extremely tiring).
The only reason you enjoy it is because you grew up in a society that made you think it was enjoyable.

Its an utter waste of resources to even 'own' a car since most of the day they just end up parked doing nothing and take up space.
If you are that keen on driving on your own, I think lending out a car when you want it with an AI driver would be a major thing, but if you want to take control of the wheels when you get a hankering for it (out in the country, etc), then by all means, I think it would be ok.

Cities however would likely become off limits to human drivers eventually.

 

Sorry man, you are delusional if you think AI as it is today, or even 50 years from now will safely drive in the "uncontrolled" environment human's are adapted to.

There are only so many patterns that match, the rest are a big "what was that", "oops, too late", "condolences to the bereaved".

It's going to be a vast wasteland of wasted AI parts and code after they are turned on and enough people are injured or die before it's turned off again.

It's already happened a number of times, self or assisted driving is pulled back after AI driving is enabled and fails.

It's a spectacular display of hubris to take technology that's barely crawling along and then release it with full confidence unto an unsuspecting public.

A public that is as bamboozled and enamored of the technology as you are, fooled enough that they put their live's in it's hands - for a grand experiment.

Welcome to the Alpha Release of Crash Test Dummy's .01a.

Be sure and keep your AI auto-drive updates set for maximum frequency downloads

 

Lot's of assumptions here. Are you taking Social Engineering 101 or something? You have no idea why I enjoy driving, but reaching to deduce it to society telling me it's enjoyable is amateur! Thanks for playing though

 

Actually, I am well aware as to where technology is going and already, self-driving cars are safer in comparison to human drivers.
Most accidents on roads have been committed by human drivers, not AI... because Humans (given the current cultural conditioning) have a tendency to behave in a highly irrational manner and endanger other people's lives.

http://bigthink.com/ideafeed/googles-self-driving-car-is-ridiculously-safe
http://www.huffingtonpost.co.uk/ben-peters/driverless-cars-how-safe-_b_15185596.html
https://www.theguardian.com/sustain...make-our-roads-safer-says-oxbotica-co-founder

Also, technology advances at a faster than exponential pace... whereas most people think in a linear fashion.
The more society becomes technologically advanced, technical and scientific breakthroughs accelerate that much more.
It is an acceleration of acceleration... and AI is already being implemented in R&D which is dozens to thousands of times faster and more accurate.

Think hard as to what exactly is behind the dip in employment rates across the globe in developed (and developing) nations... it is technological automation.
It is becoming easier, faster and much more cost effective to automate even newly 'invented' jobs (btw, the idea we need to invent boringly stupid jobs in the first place to maintain the nonsense of the present system is an absurdity in itself).

I am hardly 'enamored' by technology... I simply acknowledge the premise of how fast things are progressing, and that most other people have a tendency to ridiculously underestimate things.
Also, in case you haven't noticed, Humans tend to use science and technology for virtually everything... and we live in a world where we are intricately dependent on both to survive (Which is rather ironic, considering that most of the population of the globe knows nothing of science or technology in the first place) - which actually makes me wonder just how deeply you studied this particular subject and whether your responses could be biased based on the hypothetical premise you might be a driver who likes driving?

Technology and science also saved quite a lot of lives... yes, they also took away a whole lot of them thanks to wars, but that's mainly due to Human stupidity and lack of education than technology and science themselves.

 

It is rather basic that humans are essentially victims of culture, and you would/should be aware of this had you studied neuroscience and epignetics to a basic level by looking at lectures from Dr's Gabor Mate and Robert Sapolsky.

Where do you think your attractions and interests come from?
They certainly don't materialize out of thin air.
They come from your environment.
You react to environmental stimulus, and given the right conditions, you will start exhibiting various likes/dislikes towards things that were more prominent (btw, most people tend to miss out on how environment impacts behavior because they think they can narrow it down to one or two things in genera, while ignoring there's a huge assortment of factors at play).
You may also be born for a tendency towards certain things, but whether or not those tendencies emerge or not is down to the environment, and again they can change over time thanks to brains plasticity.

But I digress, I took the discussion away from the main topic, and I apologize for that.

 

Let's just stop right there, because you've found the little mentioned secret as to why AI auto-driving will fail - and is so dangerous it needs to be curtailed before it starts.

If all the cars were AI driven cars on roads closed to human drivers, AI auto-drive would work much better, maybe even eventually safe enough to trust with human lives.

Until then, with a mixed road full of human drivers, AI will always fail.

AI can't drive as well as a human driver can to adapt to other human drivers. The unpredictability, irrationality, and stupid random events will catch out the AI auto-driver again and again and again.

It's just not possible to account for all the possibilities, and act on them in the same time window as humans can.

It would take too long to train the AI with all the possibilities for a startup, or even a large company, to invest is a successful project of this kind. Before giving the AI the wheel, data needs to be gathered and integrated on a much larger scale.

It would take a commitment and an investment on the national scale - for every nation - as each locale has their own quirks of driving to account for before the AI will be complete enough to survive.

All the manufacturers sharing the data gathered from monitoring human drivers, and comparing their "answers" and reactions to events to the simultaneously running AI - and do that for a very long time.

Until then it's irresponsible to put human lives at risk as an experiment to gather data for the improvement of AI auto-drive.

 

But isn't that the point?
Current trends are actually pointing into the direction of a driver-less future (or automated driving), when you particularly take into account that the entirety of the economy will also become automated.
Actually, some have gone far to say that 'owning' your own car would be illegal in a relatively short span of time (think in 10 years or so).

Given that we are on the brink of radical technological merges in all areas (mainly thanks to increasing automation and AI), the current trends tend to point into the direction where we will see massive (inevitable even) changes happening rapidly.

 

I've heard the same thing, 40 years ago

It'll come when it comes, not before.

 

You boldly made the case that my infinity for driving is based on societies influences. Then preceded to argue the case for AI due to conventional driving as to being 'bogged down' and 'extremely tiring'. The irony wasn't lost on me.

 

Nothing ironic about it.
Every profession, regardless of how much you love it can end up tiring after a certain period of time. You cannot do it ad infinitum.

I also mentioned brain plasticity and how Humans (and their interests) change over time.
Think of it like this... I love doing 3d art... but even I can be put off by it after certain time, or due to various circumstances and get tired of it.
Similarly, you can find a job you seemingly 'love', but it is highly unlikely you will love every second of it... especially because it usually entails working for someone else on a specific schedule, to be creative on demand (good luck with that), and even if you are self-employed, the amount of hurdles you need to go through that have little to do (or are even directly connected) with the work you love doing can bog you down.
You think that I for example enjoy building a mesh from the ground up and then spending tens of hours correcting little mesh errors by rearranging the verticies manually?
I can tell you that rigging a mesh for animation alone is grueling work. I love doing 3d art, but various aspects of it do tire me out... much like it happens in every other field.

Driving is tiring for the simple reason that a lot of the time most drivers complain they cant enjoy the scenery and must pay attention to the road... then there's concentration involved during driving (to avoid crashes, etc.).

So, no, there is no irony here. Just plain facts. Regardless of the profession you may be in or the kind of thing you enjoy, you probably DO get tired of it eventually.
Sex is similar. Try having it 24/7 and see how it goes.
Almost inevitably you will probably take breaks every once in a while at the very least, and at some point will likely reach saturation and won't have it for at least a while.

 

You can't take a break from driving with current autonomous driving AI, you need to be alert and perhaps more involved than normal to catch any judgment errors from the AI and correct them in time.

You are relying too much on the extreme's of human boredom to justify AI driving as a *future* replacement, for 99% of driving it's not an issue.

Until 100% perfect AI driving is available, we can't use it to "take a nap", or relieve boredom. Find another justification

BTW, look out your Windows on your next drive, paying attention to all the interrupts you pay attention to and what interrupts you can ignore - the AI doesn't know which is which...humans do. Try neural networking that

 

I'm not relying on anything. It's the industry that's pushing things into that direction

 

And my longer term observations, since the 1970's, with a big push like now in the early 1980's, similar in scope, shows we are still a long way from ubiquitous autonomous driving.

This article reads just like others I read 40 years ago. They have everything they need, they just need to integrate it all. Except it's all too big, too power hungry, and too expensive.

They don't have everything they need or it would be in production, is the short of it.

Good article:

Connected Cars: The long road to autonomous vehicles
April 3, 2017 - UCSD
http://cwc.ucsd.edu/content/connected-cars-long-road-autonomous-vehicles
March 31, 2017 - San Diego Union
http://www.sandiegouniontribune.com...-fi-connected-automobiles-20170330-story.html

" Work on autonomous vehicles has been going on since the early 1980s, said Takeo Kanade, a computer vision professor and winner of the 2016 Kyoto Prize for Advanced Technology, during a recent event in San Diego."

"...To get there, data from on-board cameras, radar and lidar sensors is being fused together to create a picture of what’s around the car. Lidar, which can cost $60,000 or more, works like radar but uses light instead of radio waves.

At International CES in January, Ford demonstrated a pickle-jar sized laser lidar, mounted near the side-view mirror on a prototype self-driving sedan. (Not long ago, lidar systems were the size of a suitcase.) It pinpointed people and objects around the car — creating a very detailed dynamic view of its surroundings.

But it also required a trunk full of powerful computer servers to process the data — which probably isn’t practical for real life self-driving vehicles."

"“It is going to take time,” said Rebeiz, the UCSD professor. “ But we have the electronics. We have the know-how. It’s just going to take time before we integrate it all together.”"

Sounds good, see you again in 40 *another* years

 

There's actually something new happening:
https://mcity.umich.edu/new-way-test-self-driving-cars-cut-99-9-validation-costs/

As for having everything and not implementing it... easily explained via 'cost efficiency'.
In the industry's eyes, the tech also needs to be low cost and profitable to mass produce.

It's not going to take another 40 years.
Things are accelerating as we speak.
The more automation is integrated into economy, marginal costs come down quicker.

https://www.forbes.com/sites/jeffmc...ake-over-roads-hearts-and-minds/#4b70e4c84a50

 

That's not new, that's been done before, collecting driver reaction information has been around for all those 40 years already passed.

It's a new project repeating old ideas, there will be lots of new projects, research and development all repeating the same mistakes from another angle.

Give it another 40 years, and it will be more ready than it is today.

Until then it's just a fancy plaything to push Electric Cars, and GPU's.