A nerdy father of two, a husband of a beautiful and understanding wife, an engineer who loves anime and tinkering with PCs and games, but most of all, loves God.
Amazon is building one of the world’s most powerful artificial intelligence supercomputers in collaboration with Anthropic, an OpenAI rival that is working to push the frontier of what is possible with artificial intelligence. When completed, it will be five times larger than the cluster used to build Anthropic’s current most powerful model. Amazon says it expects the supercomputer, which will feature hundreds of thousands of Amazon’s latest AI training chip, Trainium 2, to be the largest reported AI machine in the world when finished.
Matt Garman, the CEO of Amazon Web Services, revealed the supercomputer plans, dubbed project Rainer, at the company’s Re:Invent conference in Las Vegas today, along with a host of other announcements cementing Amazon’s rising dark-horse status in the world of generative AI.
AI Lab Newsletter by Will Knight
WIRED’s resident AI expert Will Knight takes you to the cutting edge of this fast-changing field and beyond—keeping you informed about where AI and technology are headed. Delivered on Wednesdays.
Garman also announced that Tranium 2 will be made generally available in so-called Trn2 UltraServer clusters specialized for training frontier AI. Many companies already use Amazon’s cloud to build and train custom AI models, often in tandem with GPUs from Nvidia. But Garman said that the new AWS clusters are 30 to 40 percent cheaper than those that feature Nvidia’s GPUs.
Amazon is the world’s biggest cloud computing provider, but until recently, it might have been considered a laggard in generative AI compared to rivals like Microsoft and Google. This year, however, the company has poured $8 billion into Anthropic, and it has quietly pushed out a range of tools through an AWS platform called Bedrock to help companies harness and wrangle generative AI.
At Re:Invent, Amazon also showcased its next-generation training chip, Trainium 3, which it says will offer four times the performance of its current chip. It will be available to customers in late 2025.
“The numbers are pretty astounding” for the next-generation chip, says Patrick Moorhead, CEO and chief analyst at Moore Insight & Strategy. Moorhead says that Trainium 3 appears to have received a significant performance boost from an improvement in the so-called interconnect between chips. Interconnects are critical in developing very large AI models, as they enable the rapid transfer of data between chips, a factor AWS seems to have optimized for in its latest designs.
Nvidia may remain the dominant player in AI training for a while, Moorehead says, but it will face increasing competition in the next few years. Amazon’s innovation “shows that Nvidia is not the only game in town for training,” he says.
Intel heard your screams of anguish, PC gamers. Budget graphics cards that are actually worth your money have all but disappeared this pandemic/crypto/AI-crazed decade, with modern “budget” GPUs going for $300 or more, while simultaneously being nerfed by substandard memory configurations that limit your gaming to 1080p resolution unless you make some serious visual sacrifices.
No more.
Today, Intel announced the $249 Arc B580 graphics card (launching December 13) and $219 Arc B570 (January 16), built using the company’s next-gen “Battlemage” GPU architecture. The Arc B580 not only comes with enough firepower to best Nvidia’s GeForce RTX 4060 in raw frame rates, it has a 12GB memory system target-built for 1440p gaming – something the 8GB RTX 4060 sorely lacks despite costing more.
Intel
As if that wasn’t an appealing enough combination (did I mention this thing is $249?!), Intel is upping the ante with XeSS 2, a newer version of its AI super-resolution technology that adds Nvidia DLSS 3-like frame generation for even more performance, as well as Xe Low Latency (XeLL), a feature that can greatly reduce latency in supported games.
Add it all up and Intel’s Arc B580 seems poised to really, truly shake things up for PC gamers on a budget – something we haven’t seen in years and years. If you’re still rocking an OG GTX 1060, take a serious look at this upgrade. Let’s dig in.
Intel’s debut “Alchemist” Arc GPUs launched in late 2022, rife with all the bugs and issues you’d expect from the first generation of a product as complex as modern graphics cards. Intel diligently ironed those out over the subsequent months, delivering driver updates that supercharged performance and squashed bugs at a torrid pace.
In a briefing with press, Intel Fellow Tom Petersen said a major force during Battlemage’s development was improving software efficiency, to be better able to unleash the full power of Intel’s hardware. But remember, it ran on first-gen hardware, too. Battlemage improves efficiency on that front, using tricks like transforming the vector engines from two slices into a single structure, supporting native SIMD16 instructions, and beefing up the capabilities of the Xe core’s ray tracing and XMX AI instructions to, yes, make everything run smoother and better than before.
Battlemage’s hardware efficiency improvement are illustrated here, showing how a Fortnite frame runs on the new Arc B580 versus last-gen’s (more expensive) A750 — the time to render a frame dropped from 19ms to a silky smooth 13ms
intel
I’ve included a bunch of technical slides above, so nerds can pick through the details. But here’s the upshot: The Arc B580 delivers 70 percent more performance per Xe core than last gen’s Arc A750, and 50 percent more performance per watt, per Intel.
Cue Keanu Reeves: Whoa. That’s absolutely bonkers. You almost never see performance leaps that substantial from a single-generation advance anymore!
Intel
That’s at an architectural level; the slide above shows the specific hardware configurations found in the Intel B580 and B570. A couple of things stand out here, first and foremost the memory configuration.
Nvidia and AMD’s current $300 gaming options come with just 8GB of VRAM, tied to a paltry 128-bit bus that all but forces you to play at 1080p resolution. The Arc B580 comes with an ample 12GB of fast GDDR6 memory over a wider 192-bit bus – so yes, this GPU is truly built for 1440p gaming, unlike its rivals. The Arc B570 cuts things down a bit to hit its $219 price tag but the same broad strokes apply.
Also worth noting: Intel’s new GPUs feature a bog standard 8-pin power connector (though third-party models may add a second one to support Battlemage’s ample overclocking chops). No fumbling with fugly 12VHPWR connectors here.
Intel’s homebrew Limited Edition reference GPUs will return for the B580 in a newer, smaller design with blow-through cooling. You’ll also be able to pick up third-party custom cards from the partners shown above, and the B570’s launch in January will be exclusive to custom boards, with no Limited Edition reference planned.
Intel
As part of the launch, Intel is also introducing a redesigned gaming app with advanced overclocking capabilities, including the ability to tweak voltage and frequency offsets.
Intel Arc B580 performance details
Now let’s dig into actual performance, using Intel’s supplied numbers.
Intel says the $249 Arc B580 plays games an average of 25 percent faster than last generation’s higher-tier $279 Arc A770 across a test suite of 40 games. Compared to the competition, Intel says the Arc B580 runs an average of 10 percent faster than Nvidia’s RTX 4060 – though crucially, those numbers were taken at 1440p resolution rather than the 1080p resolution the overly nerfed RTX 4060 works best at.
Intel
Intel also made a point of stressing how the RTX 4060’s limited 8GB of RAM over a 128-bit bus can directly impact performance today. The slide above shows Forza Motorsport running at 1440p resolution. At standard High settings, the RTX 4060 actually holds a performance advantage. As you scale up the stressors, flipping on ray tracing and moving to Ultra settings, the advantage instantly flips, with the B580 taking a clear, substantial lead while the RTX 4060 hits the limits of what’s possible with its memory setup.
Speaking of, Intel says most of the key technologies underlying ray tracing have been improved by 1.5x to 2x in Battlemage compared to the first-gen Arc “Alchemist” offerings. Considering that Intel’s debut Arc cards already went toe-to-toe with Nvidia’s vaunted RTX 40-series ray tracing, there could be a fierce battle brewing in realistic real-time lighting next year – which isn’t something I’d thought I’d say in the $250 segment before even flipping the calendar to 2025. If you’re still rocking a GTX 1060 or 1650 from back in the day, the Arc B580 would be a massive upgrade in both speed and advanced features like ray tracing.
Raw hardware firepower alone is only part of the graphics equation these days, however. Nvidia’s RTX technology forced the power of AI upscaling and frame generation into consideration this decade – and Intel’s new software features are designed to supercharge frame rates and lower latency even further.
Meet XeSS 2 and Xe Low Latency
Intel’s XeSS technology debuted alongside the first-gen Arc cards, serving as an AI upscaling rival to Nvidia’s core DLSS technology. (These render frames at a lower resolution internally, then use AI to supersample the final result, leading to higher performance with little to no loss in visual quality.) But then Nvidia launched DLSS 3, a technology that injects AI-generated “interpolated” frames between every GPU-rendered frame, utterly turbocharging performance in many games and scenarios.
XeSS 2 is Intel’s response to that. While DLSS 3 requires the use of a hardware Optical Flow Accelerator only present in RTX 40-series GPUs, Intel’s XeSS 2 uses AI and Arc’s XMX engines to do the work instead – meaning it’ll also work on previous-gen Arc cards, and the Xe-based integrated graphics found in Intel’s Lunar Lake laptops.
Intel
And as we see with DLSS 3, the performance improvements can be outstanding. Intel says that in its in-house F1 24 tests with the B580, activating XeSS 2 with supersampling and frame generation can improve performance by a whopping 2.8x to 3.9x, depending on the Quality setting used. While the game runs at 48fps at the chosen settings without XeSS 2 enabled, turning on XeSS 2’s Ultra quality lifts that all the way up to 186fps – a literal game changer.
Intel
Support for XeSS 2 is coming to the games shown above, with more to arrive in the coming months. First-gen XeSS hit 150 games to date, so the hope is that XeSS 2 (which uses different APIs for developers to hook into) ramps quickly as well.
Injecting AI frames between tradition frames has a side effect though – it increases latency, or the reaction between your mouse click and the action occurring onscreen, because the interpolated AI frames can’t respond to your commands. Enter Intel’s Xe Low Latency feature.
XeLL essentially cuts out a bunch of the ‘middleman’ rendering and logic queues that happen behind the scenes in a frame, letting your GPU render a frame much, much faster than typical. (Nvidia’s awesome Reflex technology works similarly.) Activating it drastically lowers latency. You can tangibly feel the improvement in games that don’t have frame generation active, but by enabling it alongside XeSS 2, it claws back the latency created by frame generation.
You can witness the improvements possible in the slide below, which shows the performance of an F1 24 frame with a variety of XeSS features (supersampling, frame gen, XeLL) active. It really illustrates the need for a latency-reduction feature alongside frame generation.
Intel
Latency reduction is so critical to frame gen “feeling right” that Intel requires developers to include XeLL as part of the wider XeSS 2 package, following in Nvidia’s footsteps. As with DLSS 3 and Reflex, you may see the options presented separately in some games, while others will silently enable them together – it’s up to the developer.
Battlemage brings the heat?
Always take vendor numbers with a big punch of salt. We’ve seen vendor benchmark controversies over the years, including this year. Corporate marketing exists to sell stuff to you first and foremost. Hashtag: Wait for benchmarks et cetera et cetera.
All that said, while Battlemage doesn’t push for the bleeding edge of performance, I’m wildly excited by what I see on paper here. Budget GPUs have been an absolute quagmire ever since the pandemic, with none of the current Nvidia or AMD offerings being very compelling. They feel like rip-offs.
Intel
Intel’s Arc B580 and B570 feel like genuine value offerings, finally giving gamers without deep pockets an enticing 1440p option that’s actually affordable – something we haven’t seen this decade despite 1440p gaming becoming the new norm. Delivering better-than-4060 performance and 12GB of VRAM for $250 is downright killer if Intel hits all its promises, especially paired with what looks to be a substantial increase to Arc’s already-good ray tracing performance. And with XeSS 2 and XeLL, Intel is keeping pace with Nvidia’s advanced features – assuming developers embrace it as wholeheartedly as they’ve done with first-gen XeSS.
Add it all up and I’m excited for a truly mainstream GPU for the first time in a long time. The proof is in the pudding (again, wait for independent benchmarks!) but Intel seems to be brewing up something spicy indeed with Battlemage and the Arc A580.
Like everywhere else on the internet, LinkedIn is awash in AI-generated content. It’s a perfect fit. As first reported by Wired, a new study has found that more than half of the posts on LinkedIn were constructed using some form of generative AI. Anyone who has spent any amount of time on LinkedIn won’t be shocked.
Wired had exclusive access to a study performed by AI detection startup Originality AI. According to the publication, Originality scanned 8,795 public English LinkedIn posts that are more than 100 words long and published from January 2018 to October 2024. Of those, 54 percent were likely AI-generated. According to the study, there was a huge spike in 2023 when OpenAI released ChatGPT but it’s leveled off.
LinkedIn is a social media site aimed at helping people get a job and build a professional network. Interactions on the site have long felt like an unnecessary corporate meeting or sterile job interview. The site has been steeped in corporate culture and stilted corporate speech—that kind of dittoing aggressively bland talk that’s drained of all color and joy. It’s the kind of writing LLMs are perfect at replicating.
In the corporate world, it’s best to talk in buzzwords and jargon. LinkedIn even has a tool built in for premium subscribers that lets them cut out third-party sites like ChatGPT. After entering a minimum of 20 words into a post, subscribers can click a button and use AI to repackage their corporate content for the world.
In a world where pictures of shrimp Jesus are offending us on Facebook and grotesque Musk-as-chad pictures flood X, AI has found its perfect home on LinkedIn. But not all are happy. “Some people engaged positively, appreciating the clarity and structure of the posts. Others were skeptical or critical, often focusing on the fact that AI was involved rather than the content itself,” Entrepreneur Zack Fosdyck told Wired. “I find it fascinating how polarizing this technology can be, especially since tools like calculators or spellcheck, which are also forms of assistance, are widely accepted.”
The difference is that calculators and spellcheck do not serve to substitute and replace basic human interaction. Context matters too. It’s impossible for an AI-generated LinkedIn post to offend me. But if I caught a friend using Google’s new systems to generate a personal response to a text message? I’d be pissed.
Yesterday, Lance Eliot—a “world-renowned AI scientist” who once appeared on 60 Minutes—published an op-ed on Forbes that advocated for using ChatGPT to make Thanksgiving peaceful. Why bother engaging with your family when you can have an AI do it for you?
The post reads like ChatGPT wrote it. It’s got all the hallmarks: a lead that sounds like it’s written to satisfy a high school English class grading rubric, bullet points that walk through the essay’s talking points, and calls to action that focus on the non-controversial. At the end of the essay, Eliot offers a final piece of advice for those with an angry turkey-day guest who just wants to argue.
“A last resort might be to ask the person to go somewhere that offers solitude at your event and have them argue with generative AI,” he says. “Have the person engage in their heated argument with AI. They can do this until the cows come home. It might allow them to vent their anger. The AI can take it, don’t worry about that. Once they’ve done all their chirping and whirling, they can rejoin the group if they are going to henceforth be peaceful and thankful.”
I like to think (and the sooner the better) of a cybernetic meadow where people who would exile difficult people into a room to battle with generative AI are themselves exiled to a land of AI-generated LinkedIn posts. Let the Eliots of the world retreat from the complexities of life into a land of corporate speak and ChatGPT-led interactions.
Give me the meat of human interaction. I want the fights over politics with difficult relatives, the anger and sadness of genuine human conversation, and all the joys and pains that come with it. Let the anodyne world of LLMs live on LinkedIn. Do not bring it into your life or your home.
How great would it be if we could extract resources from our waste products? Or even better, raise them from the ashes? Scientists in the United States have suggested doing exactly that to boost the nation’s supply of rare earth elements.
Researchers, co-led by Bridget Scanlon of The University of Texas at Austin (UT), have found that up to 11 million tons of rare earth elements could be extracted from coal ash in the U.S., a waste product of coal burning. That’s almost eight times the amount of rare earth elements currently in domestic reserves. Their findings, detailed in a September 17 study in theInternational Journal of Coal Science & Technology, highlight that this approach could significantly reinforce national supplies without the need for further mining.
“This really exemplifies the ‘trash to treasure’ mantra,” Scanlon said in a UT statement.
Rare earth elements are 17 elements crucial to many technologies, including smartphones, flat-screen TVs, computer monitors, batteries, magnets, offshore wind turbines, and solar panels. The U.S. imports most of its rare earth element supply from abroad, with 75% coming from China, according to the statement. The new study, however, suggests that coal ash in the U.S. could supply $8.4 billion worth of rare earth elements.
“There’s huge volumes of this stuff all over the country,” said Davin Bagdonas of the University of Wyoming, who also participated in the study. “And the upfront process of extracting the (mineral host) is already taken care of for us.”
Various aspects, like place of origin, determine the amount of rare earth elements in a particular coal ash supply as well as how much of it can be extracted, as detailed in the study. For instance, coal ash from the Appalachian Basin has a higher amount of rare earth elements than coal ash from the Powder River Basin, though a higher percentage of rare earth elements can be extracted from the latter than from the former.
In general, there are lower amounts of rare earth elements in coal ash than in geological deposits, but the researchers point out an obvious advantage: The U.S. has large quantities of coal ash within its own borders. From the total coal ash the U.S. produced from 1985 to 2021, 1.873 billion tons could be recovered from disposal locations such as landfills and ponds.
“The idea of getting rare earth elements out of tailings (mining by-products) just makes a lot of sense. It’s a common-sense approach,” said Chris Young, the chief strategy officer at Element USA, a company that extracts minerals from aluminum industry waste. “The challenge is to convert that common-sense approach to an economic approach.”
In fact, researchers are still testing the viability of this method. It remains to be seen whether they’ll be able to make use of the (literal) ashes of the past.
Raspberry Pi has announced the Pico 2 W, a wireless version of its Pico 2 microcontroller board built for hobbyists and industrial applications. At $7, it’s a relatively inexpensive way to control electronic devices like smart home gadgets and robots. With the new version, users will be able to securely link to remote sources to send and receive data, either via Bluetooth 5.2 or Wi-Fi 802.11n.
As with the Pico 2, the wireless variant is built around the RP2350 microcontroller built in-house by Raspberry Pi. it offers more speed and memory than the original RP2040 chip, along with a security model built around Arm’s TrustZone for Cortex-M. Users can program it using C, C++ and MicroPython, and choose between Arm Cortex-M33 or RISC-V cores.
There are many potential use cases for the Pico 2 W, like smart home control that can link to external devices (plugs, lights, etc.) over Wi-Fi, robotics and science experiments. It’s now available for hobbyists from a variety of sources at $7.00 for the board only, $21 for a basic kit or $31 for the starter kit. You can see them here when you select the Pico 2 W option at the bottom of the page.
This article originally appeared on Engadget at https://ift.tt/EkoHx0L
The race is over and we have lost. As far as lithium-ion battery technology goes, the Chinese have won. They set their sights on a product that they could excel at and now they own it.
CATL Qilin Li-ion battery with 620-mile (1,000 km) range
China has pursued a position of supremacy in the lithium-ion battery space since 2001, when the country made it a cornerstone of its Five Year Plan. After “inviting” their many joint venture partners into China and learning how to properly manufacture vehicles, there was a realization that they would not be able to out-innovate the Americans and Europeans when it came to internal combustion engine (ICE) vehicles.
This led to major government support for the development of an EV battery industry starting in 2009. This was something that the “foreigners” were not pursuing. From 2009 to 2023, the Chinese government poured a substantial $230 billion into both batteries and EVs. This took the form of inexpensive land, tax breaks, and other incentives. Top Chinese battery producers like CATL, BYD, CALB, and Gotion have reaped the benefits and dominate the battery market, in China and elsewhere.
In addition to the manufacturing of EV batteries, China has gained control of the entire EV battery supply chain. This includes materials found on its home turf as well as supplies on other continents.
Here’s one example: Partially or completely Chinese-owned firms will produce over 90% of Africa’s entire lithium supply for the next ten years! To make things worse, China’s EV battery production capacity already exceeds world demand by around 400%.
This helps to explain how the Chinese can price their EVs so low – all of the materials going into the battery have been subsidized by their government. This gives them a pricing advantage, which when combined with the industry’s overcapacity, now has them shipping their EVs all over the world to various export markets.
BMW Li-ion battery plant, Woodruff, South Carolina, USA
While it’s pretty much game over as far as liquid electrolyte lithium-ion batteries go, we should not give up. We should be building our own lithium-ion battery plants and supply chains so that we can provide our EV industry with a stable source of batteries that cannot be cut off for political reasons.
The Inflation Reduction Act has already spurred a massive investment in American-based lithium-ion battery plants, making that goal a reality. Regardless, we should realize that lithium-ion batteries are reaching their performance limits and it’s time to go beyond them.
Moving forward, the real action is at the next level of EV battery development. And that’s solid-state batteries, for which we have not yet ceded development to the Chinese. It’s the best way to preserve our auto industry for the future.
Solid-state batteries will solve many of our EV problems
The next generation of solid-state EV batteries are the answer to many troubling issues we must live with in today’s electric vehicles. These solid-state batteries will charge faster, have more energy density and thus be lighter, and will be much safer than today’s lithium-ion cells, eliminating the possibility of thermal runaway. Even better, solid-state batteries need no graphite, which China has near-total control of.
Just imagine an EV with 1,000 miles of range, a five-minute charging time, normal weight, and low fire hazard. That would solve most of our EV adoption problems right there!
We need to get on the solid-state bandwagon now, while there’s still time
In order to reap a commercial advantage from the development of solid-state EV batteries, we must provide more funding for R&D, accelerate the commercialization of products that come out of the lab, and provide a protected environment (such as military-related projects) in which these batteries can be made ready for mass production as soon as possible.
The next race for solid-state batteries is on and we are not the only ones running in it. In addition to a Chinese-sponsored consortium including battery maker CATL and automaker BYD, Japan’s Toyota, Nissan, and Honda, and Korea’s Samsung are also hotly pursuing solid-state batteries. The big challenges facing all solid-state battery developers are making them at large scale while bringing the cost down.
Here in the U.S., the largest firm involved in solid-state batteries is QuantumScape, who already has a deal with the Volkswagen Group. But there’s not much time – Samsung is planning for 2027 production, while Toyota and Nissan are shooting for 2028. Honda is building a solid-state battery demonstration production line, with production planned sometime during the final half of the decade.
It’s now or never. The Chinese have eaten our lunch where lithium-ion batteries are concerned, but we still have a chance to grab the lead in the solid-state battery race. The clock is ticking…
In 1845, somewhere between inventing a system for detonating explosives by electricity and the refillable fountain pen, Robert Thomson, a Scottish engineer and entrepreneur, patented the first pneumatic tire—a then wondrous, now everyday item that has been gradually evolving ever since.
Now, in the era of electric vehicles, they are more in focus then ever before. On the one hand, while passenger safety remains a priority, the right tires can have a significant effect on efficiency—and thereby the range of your EV—but on the other they’re also a source of noise and pollution.
Gear Newsletter: Reviews, Guides, and Deals
Upgrade your life with our buying guides, deals, and how-to guides, all tested by experts.
Since the traditional global tire market is worth well over $200 billion, and 2.5 billion tires are sold a year worldwide, manufacturers are rubbing their hands at the coming death of pure combustion cars, and gearing up for a battle to fashion the ideal balance of eco credentials, performance, and efficiency that will create the perfect EV tire. Whoever wins will secure quite the prize.
Rolling Resistance or Longevity?
Range optimization has been the primary concern so far. According to Michelin, the efficiency difference between good and bad tires can be as much as 7 percent. Better tires reduce rolling resistance, meaning a car will coast further before coming to a stop. It will therefore need less energy to travel the same distance. A 7 percent increase in efficiency will give an EV that much more range—so, if it could go 300 miles with a poor tire, it will travel 321 miles with a good one.
“There are several tire components that can influence rolling resistance,” says Thomas Wanka, principal technology development engineer at Continental, a company that has been exploring EV tire design through its association with electric motorsport series Extreme E. “These include the rubber compound and the tread.”
Manufacturers are experimenting with nanomaterials in their tires, such as nanocarbon and nanosilica, to improve performance, traction, and durability. There is also research into bio-based alternative compounds such as guayule and dandelion rubber.
You can reduce rolling resistance by reducing tread depth, but this also means the tire won’t last so long and produces increased noise. Continental, however, thinks it has the answer. “We have developed special soft rubber compounds that allow us to reduce rolling resistance and noise at the same time without sacrificing mileage,” says Wanka.
