En retard, mais pas le dernier

La prochaine version du navigateur, prévue pour le 22 juillet, prendra en charge une API graphique dont le développement est le fruit d’un effort collectif. Bien que Chrome en soit doté depuis longtemps, l’arrivée de WebGPU dans Firefox reste un petit évènement.

« Après des années de développement, nous allons lancer WebGPU sur Windows dans Firefox 141 ! », s’est exclamé hier l’équipe Mozilla GFX. Cette API, qui prend la suite de WebGL, fournit un accès beaucoup plus direct aux capacités des GPU. À la clé, des améliorations importantes de performances, aussi bien sur les calculs graphiques que sur d’autres liés à l’IA, dont l’inférence.

Millefeuille graphique

Cette interface de programmation est le fruit d’un effort conjoint au sein du W3C et réunissant des entreprises comme Mozilla, Apple, Intel et Microsoft, l’ensemble étant largement tracté par Google. L’objectif était de donner aux navigateurs un accès plus moderne au matériel, dans le sillage des API de bas niveau apparues sur les différentes plateformes. WebGPU dépend d’ailleurs de ces dernières, en fournissant un lot de capacités, selon son implémentation bien sûr.

L’ensemble n’est pas si simple. Sur toute plateforme, on trouve ainsi le pilote responsable de l’exploitation du GPU. Charge à lui d’exposer de transmettre les instructions à la puce. Au-dessus, l’API graphique native (Vulkan sur Linux par exemple) est une autre brique essentielle. Dans un système d’exploitation, elle expose les capacités, dans lesquelles les applications vont venir piocher. Puis vient WebGPU dans le navigateur ou, plus précisément, son implémentation. C’est elle qui créera des appareils logiques (logiciels) pour chaque application en ayant besoin.

Si vous utilisez Chrome ou un navigateur basé sur Chromium, vous pouvez déjà observer les possibilités de WebGPU via le site dédié. Chrome fournit en effet cette capacité depuis deux ans. Pourquoi tout ce temps chez Mozilla ? On ne sait pas exactement, mais Google y a consacré davantage de moyens, la plateforme web représentant le cœur de ses activités. Apple, bien qu’ayant participé au développement de l’API, ne l’intègrera que dans Safari 26 cet automne.

Il reste « beaucoup de travail »

Selon Mozilla, WebGPU est une API « vaste et complexe ». Les effets ont été concentrés assez logiquement sur les fonctions les plus évidentes, afin « que les applications et démonstrations WebGPU à haute visibilité fonctionnent sans problème ». Selon Mozilla, tout devrait donc bien se passer dans la plupart des cas.

L’équipe explique également qu’il reste « beaucoup de travail », aussi bien sur les performances que la conformité avec la spécification. Par exemple, le navigateur utilise une communication inter-processus sans tampon pour transmettre les requêtes à la sandbox du GPU. Ce problème a déjà été corrigé, mais la solution ne sera déployée que dans Firefox 142, avec des gains significatifs de performances.

En outre, le navigateur ne dispose pas d’un moyen moderne de savoir quand un GPU a terminé une opération et introduit des intervalles pour vérifier, ce qui entraine des latences. Les développeurs se penchent actuellement sur le problème et explorent diverses solutions. De même, Firefox ne prend pas encore en charge la méthode importExternalTexture de WebGPU, qui permet au GPU la lecture vidéo décompressée directement depuis le décodeur.

Windows d’abord, les autres d’ici la fin de l’année

Le support de WebGPU ne sera également disponible que pour Windows lorsque Firefox 141 sera disponible le 22 juillet. Une question de priorité pour Mozilla : c’est là que se trouve l’écrasante majorité des utilisateurs. L’équipe précise cependant que des versions Mac et Linux sont prévues « dans les mois à venir » et qu’il est possible de les tester dans le canal Nightly du navigateur. Le support sera aussi étendu à Android.

Enfin, Mozilla indique que l’implémentation de WebGPU dans Firefox est basée sur WGPU, un projet indépendant et écrit en Rust. Il permet d’offrir une interface unifiée pour exposer les capacités sous-jacentes des API bas niveau en fonction de la plateforme utilisée : Direct3D 12 sur Windows, Metal sur macOS et Vulkan sur Linux. Mozilla contribue activement au projet, évoque une communauté très vivante et invite les personnes intéressées à se pencher sur WGPU pour participer à son développement.

En retard, mais pas le dernier

La prochaine version du navigateur, prévue pour le 22 juillet, prendra en charge une API graphique dont le développement est le fruit d’un effort collectif. Bien que Chrome en soit doté depuis longtemps, l’arrivée de WebGPU dans Firefox reste un petit évènement.

« Après des années de développement, nous allons lancer WebGPU sur Windows dans Firefox 141 ! », s’est exclamé hier l’équipe Mozilla GFX. Cette API, qui prend la suite de WebGL, fournit un accès beaucoup plus direct aux capacités des GPU. À la clé, des améliorations importantes de performances, aussi bien sur les calculs graphiques que sur d’autres liés à l’IA, dont l’inférence.

Millefeuille graphique

Cette interface de programmation est le fruit d’un effort conjoint au sein du W3C et réunissant des entreprises comme Mozilla, Apple, Intel et Microsoft, l’ensemble étant largement tracté par Google. L’objectif était de donner aux navigateurs un accès plus moderne au matériel, dans le sillage des API de bas niveau apparues sur les différentes plateformes. WebGPU dépend d’ailleurs de ces dernières, en fournissant un lot de capacités, selon son implémentation bien sûr.

L’ensemble n’est pas si simple. Sur toute plateforme, on trouve ainsi le pilote responsable de l’exploitation du GPU. Charge à lui d’exposer de transmettre les instructions à la puce. Au-dessus, l’API graphique native (Vulkan sur Linux par exemple) est une autre brique essentielle. Dans un système d’exploitation, elle expose les capacités, dans lesquelles les applications vont venir piocher. Puis vient WebGPU dans le navigateur ou, plus précisément, son implémentation. C’est elle qui créera des appareils logiques (logiciels) pour chaque application en ayant besoin.

Si vous utilisez Chrome ou un navigateur basé sur Chromium, vous pouvez déjà observer les possibilités de WebGPU via le site dédié. Chrome fournit en effet cette capacité depuis deux ans. Pourquoi tout ce temps chez Mozilla ? On ne sait pas exactement, mais Google y a consacré davantage de moyens, la plateforme web représentant le cœur de ses activités. Apple, bien qu’ayant participé au développement de l’API, ne l’intègrera que dans Safari 26 cet automne.

Il reste « beaucoup de travail »

Selon Mozilla, WebGPU est une API « vaste et complexe ». Les effets ont été concentrés assez logiquement sur les fonctions les plus évidentes, afin « que les applications et démonstrations WebGPU à haute visibilité fonctionnent sans problème ». Selon Mozilla, tout devrait donc bien se passer dans la plupart des cas.

L’équipe explique également qu’il reste « beaucoup de travail », aussi bien sur les performances que la conformité avec la spécification. Par exemple, le navigateur utilise une communication inter-processus sans tampon pour transmettre les requêtes à la sandbox du GPU. Ce problème a déjà été corrigé, mais la solution ne sera déployée que dans Firefox 142, avec des gains significatifs de performances.

En outre, le navigateur ne dispose pas d’un moyen moderne de savoir quand un GPU a terminé une opération et introduit des intervalles pour vérifier, ce qui entraine des latences. Les développeurs se penchent actuellement sur le problème et explorent diverses solutions. De même, Firefox ne prend pas encore en charge la méthode importExternalTexture de WebGPU, qui permet au GPU la lecture vidéo décompressée directement depuis le décodeur.

Windows d’abord, les autres d’ici la fin de l’année

Le support de WebGPU ne sera également disponible que pour Windows lorsque Firefox 141 sera disponible le 22 juillet. Une question de priorité pour Mozilla : c’est là que se trouve l’écrasante majorité des utilisateurs. L’équipe précise cependant que des versions Mac et Linux sont prévues « dans les mois à venir » et qu’il est possible de les tester dans le canal Nightly du navigateur. Le support sera aussi étendu à Android.

Enfin, Mozilla indique que l’implémentation de WebGPU dans Firefox est basée sur WGPU, un projet indépendant et écrit en Rust. Il permet d’offrir une interface unifiée pour exposer les capacités sous-jacentes des API bas niveau en fonction de la plateforme utilisée : Direct3D 12 sur Windows, Metal sur macOS et Vulkan sur Linux. Mozilla contribue activement au projet, évoque une communauté très vivante et invite les personnes intéressées à se pencher sur WGPU pour participer à son développement.

Après la Nintendo Switch 2, au tour des Mac ! CD PROJEKT RED annonce en effet que Cyberpunk 2077 sera disponible dès demain sur Mac, sans toutefois donner les recommandations à propos des configurations requises. Mais ne nous arrêtons pas à ce genre de détail, car il y a plus important : si on possède déjà le jeu sur GOG, Epic ou Steam, l'ajout de la version Mac se fait automatiquement ! Pas besoin de repasser à la caisse, sauf si on possède le jeu de base et qu'on souhaite s'offrir Phantom Liberty. Et oui, on pourra débuter sa partie sur PC et la continuer sur son Mac pendant une réunion au bureau. […]

Lire la suite

An anonymous reader quotes a report from Ars Technica: For some years now, detractors of battery electric vehicles have held up hydrogen as a clean fuel panacea. That sometimes refers to hydrogen combustion engines, but more often, it's hydrogen fuel cell electric vehicles, or FCEVs. Both promise motoring with only water emitted from the vehicles' exhausts. It's just that hydrogen actually kinda sucks as a fuel, and automaker Stellantis announced today that it is ending the development of its light-, medium- and heavy-duty FCEVs, which were meant to go into production later this year. Hydrogen's main selling point is that it's faster to fill a tank with the stuff than it is to recharge a lithium-ion battery. So it's a seductive alternative that suggests a driver can keep all the convenience of their gasoline engine with none of the climate change-causing side effects. But in reality, that's pretty far from true. [...] Between the high development costs and the fact that FCEVs only sell with strong incentives, the decision was made to cancel the production of hydrogen vans in France and Poland. Stellantis says there will be no job losses at its factories and that R&D staff will be put to work on other projects. "In a context where the Company is mobilizing to respond to demanding CO2 regulations in Europe, Stellantis has decided to discontinue its hydrogen fuel cell technology development program," said Jean-Philippe Imparato, chief operating officer for Enlarged Europe. "The hydrogen market remains a niche segment, with no prospects of mid-term economic sustainability. We must make clear and responsible choices to ensure our competitiveness and meet the expectations of our customers with our electric and hybrid passenger and light commercial vehicles offensive."

Read more of this story at Slashdot.

Longtime Slashdot reader Qbertino writes: Heise, a German IT news publisher, reports (English version via Google Translate) that the German state of Brandenburg is getting the world's tallest wind turbine, with an overall height of 300 meters (approximately 365 meters including rotor blades), designed to capture so-called third-level winds at higher altitudes. The article also includes a short 3D animation illustrating the construction and its size relative to standard modern wind turbines. The wind turbine uses a dual-framework base instead of a traditional closed tower to access stronger high-altitude winds, aiming to match offshore energy output while keeping onshore operating costs. According to Heise, the prototype could lead to the installation of up to 1,000 units across Germany -- fitting seamlessly between existing wind farms without needing extra land.

Read more of this story at Slashdot.

Intel is laying off more than 5,000 employees across four states, according to updated Worker Adjustment and Retraining Notification filings. From a report: Most of the cuts are happening in California and Oregon. Intel more than doubled its layoff estimates for Santa Clara and Folsom to a total of 1,935 affected employees, according to California WARN filings. The cuts began taking place in Folsom on July 11, and in Santa Clara on July 15.

Read more of this story at Slashdot.

Scale AI is laying off 14% of its workforce and 500 contractors as part of a major restructuring just weeks after Meta bought a 49% stake and absorbed its CEO into a new superintelligence lab. The Verge reports: Jason Droege, CEO of Scale AI, sent an email to all Scale employees today, which was viewed by The Verge. Droege said he plans to restructure several parts of Scale's generative AI business and organize it from 16 pods to "the five most impactful": code, languages, experts, experimental, and audio. The company will also reorganize its go-to-market team into a single "demand generation" team that will have four pods, each covering a specific set of customers. "The reasons for these changes are straightforward: we ramped up our GenAI capacity too quickly over the past year," Droege wrote. "While that felt like the right decision at the time, it's clear this approach created inefficiencies and redundancies. We created too many layers, excessive bureaucracy, and unhelpful confusion about the team's mission. Shifts in market demand also required us to re-examine our plans and refine our approach." Droege said that he believes the changes to the company will make it more able to adapt to market shifts, serve existing customers, and win back customers that have "slowed down" work with Scale. He also said that the company would deprioritize generative AI projects with less growth potential. "We remain a well-resourced, well-funded company," he wrote. Scale's generative AI business unit will have an all-hands meeting tomorrow, followed by a company-wide meeting on July 18th. Osborne said that Scale plans to increase investment and hire hundreds of new employees in areas like enterprise, public sector, and international public sector, in the second half of 2025 and that severance has been paid out to impacted roles. "We're streamlining our data business to help us move faster and deliver even better data solutions to our GenAI customers," he said.

Read more of this story at Slashdot.

Longtime Slashdot reader bobdevine shares a report from OSTechNix: For the first time, Linux has officially broken the 5% desktop market share barrier in the United States of America! It's a huge milestone for open-source and our fantastic Linux community. While many might think of Linux as a niche choice, this new data shows a significant shift is happening. According to the latest StatCounter Global Stats for June 2025, Linux now holds 5.03% of the desktop operating system market share in the United United States of America. This is fantastic news! [...] One truly satisfying detail for me? Linux has finally surpassed the "Unknown" category in the USA! It shows that our growth is clear and recognized. "It took eight years to go from 1% to 2% (by April 2021), then just 2.2 years to reach 3% (June 2023), and a mere 0.7 years to hit 4% (February 2024)," notes the report. "Now, here we are, at over 5% in the USA! This exponential growth suggests that we're on a promising upward trend."

Read more of this story at Slashdot.

An anonymous reader quotes a report from TechCrunch: Security researchers say Chinese authorities are using a new type of malware to extract data from seized phones, allowing them to obtain text messages -- including from chat apps such as Signal -- images, location histories, audio recordings, contacts, and more. In a report shared exclusively with TechCrunch, mobile cybersecurity company Lookout detailed the hacking tool called Massistant, which the company said was developed by Chinese tech giant Xiamen Meiya Pico. Massistant, according to Lookout, is Android software used for the forensic extraction of data from mobile phones, meaning the authorities using it need to have physical access to those devices. While Lookout doesn't know for sure which Chinese police agencies are using the tool, its use is assumed widespread, which means Chinese residents, as well as travelers to China, should be aware of the tool's existence and the risks it poses. [...] The good news ... is that Massistant leaves evidence of its compromise on the seized device, meaning users can potentially identify and delete the malware, either because the hacking tool appears as an app, or can be found and deleted using more sophisticated tools such as the Android Debug Bridge, a command line tool that lets a user connect to a device through their computer. The bad news is that at the time of installing Massistant, the damage is done, and authorities already have the person's data. "It's a big concern. I think anybody who's traveling in the region needs to be aware that the device that they bring into the country could very well be confiscated and anything that's on it could be collected," said Kristina Balaam, a researcher at Lookout who analyzed the malware. "I think it's something everybody should be aware of if they're traveling in the region."

Read more of this story at Slashdot.

Steam has begun banning games that violate the payment rules of banks and card networks, targeting adult content in particular -- especially titles with extreme or controversial themes. Engadget reports: The new clause states that "content that may violate the rules and standards set forth by Steam's payment processors and related card networks and banks, or internet network providers" is not allowed and could result in removal from the platform. In other words, if credit card companies get mad about something, they could actually have the power to ban a game. The clause goes on to say that this will affect "certain kinds of adult-only content." This has likely already resulted in many games being pulled off the platform. SteamDB doesn't give a reason for these removals, but the timing does match up.

Read more of this story at Slashdot.

OpenAI has added Google Cloud as a provider for ChatGPT and its API, expanding beyond Microsoft to address growing demand for computing power. CNBC reports: OpenAI has added Google to a list of suppliers, specifying that ChatGPT and its application programming interface will use the Google Cloud Platform, as well as Microsoft, CoreWeave and Oracle. The announcement amounts to a win for Google, whose cloud unit is younger and smaller than Amazon's and Microsoft's. Google also has cloud business with Anthropic, which was established by former OpenAI executives. The Google infrastructure will run in the U.S., Japan, the Netherlands, Norway and the United Kingdom.

Read more of this story at Slashdot.

An anonymous reader quotes a report from Android Authority: Italian YouTuber Once Were Nerd covers a variety of retro gaming topics, but his reviews of ANBERNIC devices appear to be the straw that broke the camel's back. According to the video [here], customs enforcement officers from the Guardia di Finanza showed up at his home and office on April 15 with a search warrant to investigate promotion of pirated copyrighted materials. They seized a variety of ANBERNIC, Powkiddy, and TrimUI gaming handhelds from his collection. In total, more than 30 consoles were taken. The creator, assuming he didn't do anything wrong, complied with demands, providing full transcripts of his conversations and chats with gaming handheld manufacturers. The officers also took his phone, promising to return it in a few days. It was returned two months later, on June 15. According to the video, officials are not required to disclose what exactly the charges are or who has brought them until the initial investigation is complete under Italian law. At that point, the case is either dismissed or goes to trial. The complaint specifically mentions reproduction of copyrighted material from Nintendo and Sony, but the case may originate from the agency itself. However, in the meantime officials have the option to shut down his channel, even before proving any wrongdoing. This is a scary prospect for any creator who has spent years building a channel, and unlike YouTube copyright strikes, there's likely no remedy. Currently, officials contest that his reviews of ANBERNIC devices like the RG Slide, which often, but not always, ship with microSD cards filled with copyrighted ROMs, are punishable under Article 171 ter of the Italian Copyright Law. This law, which was originally written in 1941, allows for a maximum punishment of 15,000 euros (or 30 million Italian Lira, since the law pre-dates the Euro) and three years of jail time.

Read more of this story at Slashdot.

In addition to releasing Mesa 25.2-rc1 with its many new features to test, Mesa release manager Eric Engestrom today released Mesa 25.1.6 as the newest bi-weekly stable point release for last quarter's series...

The abstract of a paper featured on NBER: Most U.S. innovation output originates from firms that operate R&D facilities across multiple local markets. We study how this geographic structure influences aggregate innovation and growth, and whether it is socially optimal. First, we develop an endogenous growth model featuring multi-market innovative firms that generate knowledge spillovers to geographically proximate firms. In equilibrium, firms may operate in too few or too many local markets, depending on how sensitive are the local spillovers they generate to their local size. Second, to quantify these effects, we link the model to data on firms' R&D locations, patents, and citation networks. Using an event-study design, we show that firms' spatial expansion increases spillovers to other firms and estimate how these spillovers depend on a firm's local footprint. Our estimates imply that U.S. innovative firms operate in too few markets relative to the social optimum. Third, using quantitative counterfactuals, we find that policies promoting broader spatial scope yield larger welfare gains than standard R&D subsidies. Moreover, unlike R&D subsidies, such policies can also reduce regional inequality.

Read more of this story at Slashdot.

VMware has notified partners that its current channel program will end, replacing it with an invitation-only system that significantly reduces the number of authorized partners. Partners not invited to the new VMware Cloud Service Provider program would have received non-renewal notices on July 15, 2025, and can continue transactions only until October 31, 2025, after which they may service existing contracts through their current terms. The company is also ending its White Label program on October 31, 2025. The changes mark the second major partner program overhaul in 18 months, following Broadcom's January 2024 decision to terminate partners operating VMware-powered clouds with fewer than 3,500 processor cores.

Read more of this story at Slashdot.

BrianFagioli writes: Meta has partnered with Amrize and the University of Illinois to develop an "AI-optimized" concrete mix that cuts carbon by 35% for its new data center.

Read more of this story at Slashdot.

An anonymous reader shares a report: Google will now let everyone in the US call local businesses using AI. The feature, which is now available in Search, allows you to use AI for pricing or availability information without having to talk on the phone. Google first started testing this feature in January, and it's still only available for certain kinds of businesses, like pet groomers, dry cleaners, and auto shops. When you search for one of these services, like a pet groomer, Google will display a new "have AI check pricing" prompt beneath the business listing.

Read more of this story at Slashdot.

Earlier this year when the Framework 13 was updated for the AMD Ryzen AI 300 series I ran benchmarks looking at the performance gains across different Linux distributions with Debian 13, Clear Linux, and CachyOS being the outstanding performers for that Strix Point hardware. With the recent launch of the Framework 12 2-in-1 laptop powered by Intel Raptor Lake you may be wondering what Linux distributions have the edge there. Here is an eight-way comparison of different Linux operating systems on the Framework 12 with Intel Core i5 1334U with the likes of Arch Linux, CachyOS, Clear Linux, Debian, Fedora, openSUSE, and Ubuntu.

Vous êtes en train d'envisager l'achat d'un écran 27 pouces QHD ? L'OLED vous tente comme bien des gens, mais pas à n'importe quel prix et vous vous dites que vous pourriez craquer à condition que ce prix soit inférieur à 500 € ? Depuis plusieurs semaines, surtout en cette période de soldes d'été, l...

Amazon.com marked its 30th anniversary Wednesday, three decades after Jeff Bezos launched the company as an online bookstore promising "one million titles" from Seattle. The e-commerce giant began in 1995 with Bezos, his then-wife MacKenzie Scott, and seven employees. The company now employs 1.5 million people and carries a market capitalization exceeding $2 trillion. Amazon has expanded from books into groceries through its $13.7 billion Whole Foods acquisition, cloud computing via Amazon Web Services, and entertainment with Prime Video.

Read more of this story at Slashdot.