you welcome in long vision—Where we look at the week’s news and strip it of the essentials. let’s work What is the real problem?.
This week: Google’s FLoC offering stops, Meta/Facebook buys RSC – a massive AI supercomputer, and Arm “will IPO” instead of selling to Nvidia.
1. Forget ‘unified group learning’ – long live ‘themes’
First this week: Google acquiesces in the inevitable. Third-party cookies will soon be gone, because people are tired of being tracked. But Google’s FLoC . suggestion was not the answer.
Analysis: FLoC is a failure
Google doesn’t want to invade your privacy. But Google wants to sell well-targeted ads. The main difference between FLoC and Topics is that FLoC containers are created automatically – and therefore are opaque, which can lead to overlap with “protected categories”—eg race or gender. But the Topics API rankings are organized and have names that are descriptive and visible to the user.
Frederick Lardinoy: Google eliminates FLoC, replaces it with themes
Google’s controversial project to replace cookies for interest-based advertising… is dead. In its place, Google today announced a new proposal. … Your browser will learn about your interests as you move across the web.
When you click on a site that supports Topics API for advertising purposes, the browser will share three topics you are interested in… The site can then share this with its advertising partners to decide which ads to show you. … Users will be able to review and remove topics from their lists [or] Turn off themes API completely.
It will remain to be seen if other browser vendors are interested in adding the Themes API. [But] They all quickly turned a cold shoulder toward FLoC. …the plan is to start trying the Topics API at the end of this quarter.
Simple tweaks and rename? “Yes and no,” mananaysiempre quips:
Yes, it is still a matter of building user tracking functionality into user agents. No, not the same approach.
FLoC was grouping users unsupervised, while Topics allocating users into predefined groups. This makes it more transparent and also appears to be designed to overcome the objection of “protected classes”.
But Google has a long way to go to convince garett_spencley:
I’m happy to ditch FLoC, but as an end user I still don’t ask for this nonsense in threads. It doesn’t solve any of my problems. It solves the problems facing the advertising industry.
I would choose to use browsers that do not support this.
2. Machine learning on a Metaverse scale is expensive
Zuckerberg’s new game RSC: An artificial intelligence supercomputer worth $400 million. and that Only the first stage.
Analysis: 3.8 x 1018 Integers per second…a lot
If you rely on DevOps’ Open Compute Project systems, know that Hacker Way’s no longer bee knees. Meta/Facebook is investing huge amounts of money into a configuration dedicated to Ready devices.
Timothy Brickett Morgan: Meta buys its own giant AI supercomputer, rather than builds and unlocks it
If you thought that creating a social network on Facebook required a lot of computing and storage, you’re not going to see anything yet. …to build the Metaverse…it would require an extremely massive amount of supercomputing power.
But the surprise, given Facebook’s over a decade of focus on… unlocking hardware designs through the Open Compute Project, is that… Research Super Computer [RSC] The system is built from…commercially available…servers, storage, and switches. [It] Maybe it’s because…working on a deal with some sellers who want Facebook publicity and are willing to work with the price and parts availability.
[So Meta bought] 760 Nvidia DGX A100 System – Each … a pair of AMD “Rome” processors 64 cores Epyc 7742 … 1 TB of main memory … eight A100 GPU accelerators [and] Eight 200Gb/s HDR InfiniBand controllers — all-flash storage from Pure Storage and… caching servers from Penguin Computing. … at list price, this machine would cost about $400 million … The first stage of the RSC machine is rated at 59 petaflops … (FP64), and 118.6 petaflops FP64 using Tensor Core Matrix motors. …with INT8 processing on Tensor Cores, which is essential for AI inference, the first stage machine is rated at 3.79 exaOPS.
how Much??? It doesn’t sound like news for Kleba:
I used to work at a university where my professor had been working on automatic speech recognition for a long time, but I basically gave up on this kind of research about 10 years ago because he believed that universities simply couldn’t compete budget wise with the big players in the industry. I suppose the same will soon happen for most areas of research related to ML sooner or later.
Indeed, there is a great deal of research innovation in [Natural Language Processing] And [Computer Vision] Comes from the big companies in recent years. Of course there is a discussion to be had about what that means for society as a whole.
But 3.79 exoOPS? This is how the wind blows, according to sungazer:
Give it 15 years or so – that level of computation will be on your desktop. If the desktops are still there.
3. Arm width: no legs
Speaking of Nvidia, as I said before, Arm should not be integrated. It seems less likely. ARM chips are a growing staple in a data center – especially a valuable one “Performance per watt.”
Analysis: ARM M&A DOA – IPO PDQ
The real hornet’s nest of Bloomberg journalists put their names in an anonymous report. The story goes that Nvidia’s proposed purchase of Arm from SoftBank Group is dead. Next stop: IPO.
Ian King, Giles Turner, Peter Elstrom, Dina Bass, David McLaughlin, Ruth David and Dinesh Nair: Nvidia is quietly preparing to give up… try the arm
According to people familiar with the matter…Nvidia…don’t expect the deal to come to an end. … Meanwhile, SoftBank is ramping up preparations for Arm’s initial public offering.
If Nvidia can get past the deal, it would be a massive upheaval. …but it will be an uphill battle. … the world’s largest tech companies rely on Arm’s technology, and fear that they will lose unrestricted access under Nvidia.
Within SoftBank, there are factions. [And] The ordeal created divisions within Nvidia. “We continue to hope that the transaction will be approved,” a SoftBank spokesperson said in an emailed statement.
There is no doubt that post-Brexit Britain wishes to preserve Aram’s independence. Here Simon:
I think Arm has it – of its UK staff – mainly Cambridge – and they’re expanding the campus there. The semiconductor physics group at the university’s Cavendish Laboratory is at the forefront of the world with 18 PhD projects currently.
Guess where groups of top ARM scholars are recruited from? The arm and the university are joined at the hip. That company is not going anywhere.
More broadly, it’s good news, the hatchet believes:
Hopefully, this means that NVIDIA will focus on RISC-V instead. This will be good for everyone
The cues from the story: Enthusiasm should not outweigh appreciation.
You have been reading long vision by Ritchie Jennings. You can contact him at Tweet embed or [email protected].
Photo: Reed Geiger (via Unsplash)