Facebook To Design Its Own Chips For Real-Time Content Filtering
Facebook wants to be able to filter content, including live video streams, in real-time. The company said at a Paris technology conference that it’s going to design its own machine learning (ML) processors to achieve that goal. Facebook had previously designed its own server architecture, motherboards, and its own communication chips for the data center.
High-Performance Content Filtering
Facebook’s chief artificial intelligence scientist Yann LeCun said that the company would like to take down offensive videos, like someone live-streaming a murder or a suicide, as they happen. However, such capability would require “a huge amount of compute power,” as well as consume a lot of power.
At the Viva Technology conference, LeCun noted:
"There’s a huge drive to design chips that are more energy-efficient for that. A large number of companies are working on this, including Facebook.You’ve seen that trend from hardware companies like Intel, Samsung, Nvidia. But now you start seeing people lower in the pipeline of usage having their own needs and working on their own hardware."
Being able to filter content in real-time may not sound like such a bad idea when the company actually takes down violent videos or hate speech, but the same technology could also allow the company to censor certain free speech or other types of non-aggressive content with lightning speed. That means few people would be aware that important content is censored unless the person who posted that content can raise awareness about it through other channels and expose Facebook's actions.
Software Companies Embrace ML Hardware
Google is another example of a company that now builds its own ML hardware that better fits its own requirements. Google surprised many when it built its Tensor Processing Unit (TPU) in 2015 (and announced it in 2016) to run its AlphaGo AI on faster hardware. The chip promised to be a three-generation leap forward with orders of magnitude higher inference performance than Nvidia’s Tesla K80 GPUs.
In 2017, Google launched the even more powerful TPU 2.0, which prioritized training over inference. The company recently announced the TPU 3.0, too, which promises another big increase in performance.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
It’s not just large online services focusing on ML hardware, either. Smartphone makers, for instance, have increasingly adopted ML processors into their devices, promising much better efficiency for certain ML tasks such as improving photo quality, managing battery life, organizing your photos with client-side intelligence, and so on.
This trend doesn’t show any signs of a slowdown, so we should be expecting even more companies adopt ML accelerators in their data centers or consumer products soon.
Silicon Power's 4TB US75 PCIe 4.0 SSD drops to its lowest price this year, just $199
Meta turns to nuclear power for AI training — asking for developer proposals for small modular reactors or larger nuclear solutions
Noctua shares a fan grill design that drops noise levels — 3D print your own 120mm fan grill like the one on the Seasonic x Noctua 1,600W power supply
-
bit_user Something tells me this real-time filtering of all live streams is a stretch goal. Maybe something they can start to think about, after getting a couple generations under their belt.Reply
On the other hand, I do think it's plausible they could deploy real-time filtering based on speech recognition. And that might work well enough to eliminate the majority of certain types of content.
I also think we're going to see some of thees homegrown efforts fizzle, after they build one generation of hardware and look around to see how many of their competitors & potential suppliers have far surpassed their efforts. Google was really ahead of the curve, on this one. Not everybody is a Google. -
therealduckofdeath Designed to help users, will be used to exploit. Facebook is one of those companies having as much connection to reality as the spam abusers they pretend they've got control over now. What was it they said they automatically filtered now? 99 point something percent? I call BS. Just look at any comment section using Facebook and you'll see infinitely more spam today compared to last year. "I make $$$$$ sitting at my computer posting this spam, so a scam company can use my processor power to <insert-your-desired-malice-here>". Facebook is doing as little to counter spam and scams today as they were five years ago. Best scenario they'll keep doing that, worst (most likely) scenario they'll just use these processors to ramp it up.Reply -
mlee 2500 Yeah, Qualcomm went down the path of developing low-power data-center CPU's....Centriq..and recently decided the performance wasn't there with the ARM architecture. Not sure how a software company thinks they can do better.Reply
This smells more like a public relations vapor project. -
bit_user
No, the reason for bailing was financial, from what I've read. Performance actually looked strong, but the speculation I've seen was that their investors felt the financial upside didn't justify the continued expenditures.21007196 said:Yeah, Qualcomm went down the path of developing low-power data-center CPU's....Centriq..and recently decided the performance wasn't there with the ARM architecture.
Talk about a company being a victim of its own success...
Because machine learning hardware is vastly simpler. That's why everybody is coming out of the woodwork with their own new deep learning chip. Seriously, the number of different chips and engines that have been announced is getting ridiculous.21007196 said:Not sure how a software company thinks they can do better.
What would they have to gain by that? The public cares about actual content filtering. Not about how they do it. As long as they aren't filtering what people should, they'll continue to complain.21007196 said:This smells more like a public relations vapor project.
The only advantage I see from this being a bluff is if they're trying to get a price break on hardware from other chip makers. But the market is so crowded with these things that you don't need your own vapor ware to play off against vendors - you can simply play off the vendors against each other. And vendors will be some of the best at detecting a bluff.
To me, the mystery is why they think they can do it better than anyone else, and by a big enough margin to justify the investment. But maybe it's more of a bragging contest with Google (another software company that has some of the fastest deep learning chips known). -
dhayric "Facebook wants to be able to filter content, including live video streams, in real-time"Reply
Dollars to doughnuts says conservative views and content will face the most filtering. -
mlee 2500 21007282 said:
No, the reason for bailing was financial, from what I've read. Performance actually looked strong, but the speculation I've seen was that their investors felt the financial upside didn't justify the continued expenditures.21007196 said:Yeah, Qualcomm went down the path of developing low-power data-center CPU's....Centriq..and recently decided the performance wasn't there with the ARM architecture.
Talk about a company being a victim of its own success...
Because machine learning hardware is vastly simpler. That's why everybody is coming out of the woodwork with their own new deep learning chip. Seriously, the number of different chips and engines that have been announced is getting ridiculous.21007196 said:Not sure how a software company thinks they can do better.
What would they have to gain by that? The public cares about actual content filtering. Not about how they do it. As long as they aren't filtering what people should, they'll continue to complain.21007196 said:This smells more like a public relations vapor project.
The only advantage I see from this being a bluff is if they're trying to get a price break on hardware from other chip makers. But the market is so crowded with these things that you don't need your own vapor ware to play off against vendors - you can simply play off the vendors against each other. And vendors will be some of the best at detecting a bluff.
To me, the mystery is why they think they can do it better than anyone else, and by a big enough margin to justify the investment. But maybe it's more of a bragging contest with Google (another software company that has some of the fastest deep learning chips known).
Nearly everything a company does is based on the Financial returns (or at least should be). The market and value proposition of Centriq was tied to it's performance and cost savings (in this case, power and cooling overhead). To say Centriq had "strong" performance running datacenter loads and applications is a stretch, and I say that as one of the few people who got their hands on the reference design models. There's also problems with simply getting OS's and Apps not only compiled, but *optimized* for a new platform. It's an uphill battle. No, the performance was NOT sufficient even when considering the power and cooling savings. If it were, then the Value Proposition to customers would have made it Financially Viable for Qualcomm.
But yeah, it's just one more in a long line of products Qualcomm successfully engineered and developed, only to see the market never materialize (Globalstar, MediaFlo, Centriq....I'd even say Mirasol, but it had technical difficulties being delivered, unlike the others).
As far as what does Facebook gain with it as a PR stunt? That's easy and obvious. You might have noticed them getting beat up by privacy issues lately, and Zuckerberg grasping for positive news when he gets grilled by Congress and the EU. Announcing efforts like this, even if entirely academic, makes them look like they're serious about addressing the problem. As you said, the market is already quickly getting saturated by companies with more semiconductor experience. -
bit_user
Thanks for sharing. It's always good to hear first-hand experiences.21010775 said:To say Centriq had "strong" performance running datacenter loads and applications is a stretch, and I say that as one of the few people who got their hands on the reference design models. There's also problems with simply getting OS's and Apps not only compiled, but *optimized* for a new platform. It's an uphill battle. No, the performance was NOT sufficient even when considering the power and cooling savings.
What I had read wasn't that it couldn't be profitable, but that it wouldn't generate enough profit, thereby watering down Qualcomm's overall profitability. I don't know if that's true, but it's not inconceivable that even a successful product might still be regarded as disappointment by Wall St.21010775 said:If it were, then the Value Proposition to customers would have made it Financially Viable for Qualcomm.
You can read a lot of such musings over here:
https://www.anandtech.com/comments/12755/qualcomms-server-lead-anand-chandrasekher-leaves-company
-
stdragon Maybe instead they can just reverse-engineer the positron brain inside Zucks head. If only he had an emotion chip......Reply -
mlee 2500 Thanks for the article link...it's interesting to read Qualcomm, whose mobile phone margins have for several years now suffered from commoditization and increased competition from lower cost shops like Mediatek, decide that what I assumed to be a diversification play would not be profitable enough. Perhaps they looked at all their other new-market projects (Automotive, VR, ARM laptops) and decided that Centriq was the one to be sacrificed for their limited development resources and desire to be more focused.Reply
In either event, I feel for the hundreds of engineers and other folks who for years must have poured allot of blood, sweat, and tears into the Centriq product to make it functionally viable, only to see it shelved.
21011589 said:
Thanks for sharing. It's always good to hear first-hand experiences.21010775 said:To say Centriq had "strong" performance running datacenter loads and applications is a stretch, and I say that as one of the few people who got their hands on the reference design models. There's also problems with simply getting OS's and Apps not only compiled, but *optimized* for a new platform. It's an uphill battle. No, the performance was NOT sufficient even when considering the power and cooling savings.
What I had read wasn't that it couldn't be profitable, but that it wouldn't generate enough profit, thereby watering down Qualcomm's overall profitability. I don't know if that's true, but it's not inconceivable that even a successful product might still be regarded as disappointment by Wall St.21010775 said:If it were, then the Value Proposition to customers would have made it Financially Viable for Qualcomm.
You can read a lot of such musings over here:
https://www.anandtech.com/comments/12755/qualcomms-server-lead-anand-chandrasekher-leaves-company
-
bit_user
Worse yet, if this was only to appease some short/medium-term investors. Qualcomm has been on the receiving end of activist investors' litigation, before. Basically, accusing them of spending too much on R&D, I guess instead of paying dividends. Maybe that's why Qualcomm's later Kyro cores are just tweaked versions of ARM's existing cores.21016366 said:In either event, I feel for the hundreds of engineers and other folks who for years must have poured allot of blood, sweat, and tears into the Centriq product to make it functionally viable, only to see it shelved.
Anyway, it could always get sold. Someone like Apple or Amazon could even buy it, if they wanted their own cloud platform.