Microsoft has unveiled Project Solara, a chip-to-cloud platform designed to power a new generation of "agent-first" enterprise devices — hardware designed to run AI agents instead of traditional apps. Announced at the Microsoft Build 2026 Developer Conference on the 2nd of June, the platform, developed by Microsoft's Applied Sciences Group, features a lightweight edge OS called the Microsoft Device Ecosystem Platform (MDEP). Interestingly, the OS is built on the Android Open Source Project (AOSP) rather than Windows.

MDEP is paired with Azure-hosted agent services and persistent cloud-based state, meaning devices act as interfaces to AI agents running across Microsoft's cloud infrastructure rather than as fully self-contained computers. Together, the software stack forms what Microsoft describes as a chip-to-cloud architecture for enterprise AI devices, combining cloud-hosted agents with centralized security, management, and orchestration capabilities.

"The 'operating system' is liminal, transcending the device and the cloud. The system brings a lightweight window to the edge, where the agent manifests and where the state, via Azure, can encompass a constellation of specialized devices,” explained Steven Bathiche, Microsoft's Corporate Vice President and Technical Fellow in the Applied Sciences Group.

To populate that ecosystem with hardware, Microsoft has partnered with Qualcomm and MediaTek as its first silicon partners — Qualcomm for portable and wearable form factors and MediaTek for stationary devices. The company has no plans to manufacture end products itself. Instead, the company is releasing reference designs for OEMs to build from, alongside an "approved chipsets" requirement that gives Microsoft certification-level control over which hardware qualifies for the platform, similar to Google's GMS certification model for Android.

To demonstrate Solara, Microsoft unveiled two concept reference designs built on the platform. A stationary desk-mounted AI hub built around MediaTek IoT silicon and a wearable AI badge powered by Qualcomm hardware. The desktop companion features a display, a camera, a UWB (ultra-wideband) presence sensor that handles automatic login and lock, dual far-field mics, and two USB-C ports. Connected to an external display, the device can double as a Windows 365 cloud PC client.

Meanwhile, the wearable badge is equipped with a touchscreen, Hello for Business fingerprint sensor, far-field high-SNR microphone array, side-facing camera, and 5G, WiFi, Bluetooth, and GNSS connectivity — targeting front-line workers such as nurses, retail staff, and field workers. Microsoft confirmed that both devices are intended as reference designs for OEM partners rather than retail products.

Central to the platform is what Microsoft calls just-in-time UI — an adaptive interface layer that allows a single agent to render appropriately across different screen sizes and input modalities without requiring developers to rebuild the experience for each device. Microsoft positions this on a spectrum between conventional responsive design and fully generative UI, where AI constructs interfaces dynamically with no predefined structure; Solara currently targets the middle ground, prioritizing consistency while avoiding per-device redesign overhead.

“The same agent can render a custom experience on multiple screen sizes and modalities with little or no additional work from the developer. For us, that is the first proof point: a path to specialized devices without requiring developers to rebuild the experience from scratch each time,” said Bathiche.

Another notable aspect of Solara is Microsoft's decision to build MDEP on Android rather than Windows. AOSP scales naturally to the lightweight, constrained hardware that wearables and embedded devices run on — something Windows, with its memory and processing overhead, was never designed to do. It also sidesteps the application compatibility expectations that come with Windows. Because Solara devices are built around cloud-hosted agents rather than traditional software, Microsoft can optimize the platform for dedicated AI hardware without carrying decades of legacy PC baggage.

To manage multiple agents running simultaneously, Microsoft is also working on an agent dispatcher and agent task manager — components that automatically surface or activate the right agent based on context, rather than requiring users to launch each one manually. Neither component is shipping yet. Early agent integrations include Dragon Copilot for healthcare workflows and GitHub Copilot for developer task tracking — both exploring how persistent, context-aware agents behave differently on dedicated hardware than they do inside a browser or IDE.

The platform appears to be aimed at enterprise buyers in retail, healthcare, and field service sectors, where dedicated agent hardware makes more sense than repurposing a smartphone. Microsoft has already lined up pilots with Best Buy, CVS Health, Levi's, Target, and AccuWeather, with broader OEM deployment targeted across healthcare, hospitality, financial services, legal, and industrial verticals.

A bipartisan bill passed by the U.S. House of Representatives aims to curb China's access to advanced AI chips by way of renting offshore data centers. The bill extends export controls on potent AI chips from the likes of AMD and Nvidia to cloud computing, a loophole that companies in China have allegedly used previously to access the hardware.

As reported by The Information, the Select Committee on the CCP announced the passing of the Remote Access Security Act on Monday. In a statement, the committee stated that the new bill "modernizes the Export Control Reform Act by expanding federal authority to restrict foreign adversaries’ ability to access technologies, including AI chips, remotely through cloud computing services."

In November of last year, multiple reports emerged claiming that Chinese companies were finding ways to access Nvidia's latest Blackwell chips by exploiting a cloud computing loophole. According to an investigation by the Wall Street Journal, INF Tech, a Shanghai-based startup, had allegedly gained access to 2,300 banned Nvidia AI GPUs by renting a server in Indonesia. The 32 Nvidia GB200 servers were rented from an Indonesian telecommunications company in a deal thought to be worth $100 million. Perhaps more concerning for U.S. lawmakers, the GPUs were allegedly bought after the company had secured the Chinese outfit as a customer.

Similarly, both Alibaba and ByteDance were accused of training their respective Qwen and Doubao LLMs using Nvidia chips, accessed by renting data centers in Southeast Asia.

The fast-moving world of AI chip exports is a cat-and-mouse affair. On the one hand, Washington continues to try to block Chinese access to its most potent AI hardware while also relaxing some restrictions, like allowing the sale of Nvidia H200 chips to approved customers. On the other hand, some Chinese companies continue to try to skirt export controls to obtain the best AI hardware available, both over the cloud and through physical means. Beijing's tack has shifted more towards encouraging the use of homegrown chips where possible. So, whilst H200 export approval could be a boon to Nvidia, China has dragged its feet a little and still needs to approve H200 imports before big companies like the aforementioned Alibaba and ByteDance get the green light.

While H200 exports are now approved, heftier Blackwell-based chips remain firmly off limits, as will Nvidia's new Vera Rubin architecture. It's remote access to this hardware that the Remote Access Security Act aims to curb. In a statement, Chairman and cosponsor of the legislation, John Moolenaar, said, "The CCP’s AI ambitions are being fueled by its access to American chips housed in data centers located outside of China. "This bill brings our laws into the digital age and makes it clear that cloud compute is subject to U.S. export control law, just like physical chips. Closing these loopholes will strengthen U.S. national security and protect American innovation.”

Nvidia has reorganized its cloud computing group, further scaling back its ambitions to operate a public cloud service that would directly compete with Amazon Web Services, according to reporting by The Information. The changes include folding the DGX Cloud business into Nvidia’s core engineering structure, under the leadership of SVP Dwight Diercks, who oversees software engineering.

The decision to scale back comes as Nvidia continues to post record revenue from data center GPUs, while facing growing pressure to balance internal platform ambitions against the interests of its largest customers. The restructuring comes after it was announced in September that Nvidia would stop trying to compete with AWS and Azure.

Nvidia is not exiting cloud infrastructure altogether, but is narrowing its scope. Instead of selling GPU compute as a service under its own brand, the company is repositioning DGX Cloud as an internal platform for its engineers, with a focus on chip demand and AI model development.

Preventing channel conflict

DGX Cloud was introduced in early 2023 as Nvidia’s attempt to abstract its flagship DGX systems into a managed service. Hosted initially on infrastructure provided by AWS, Google Cloud, Oracle Cloud, and Microsoft Azure, the service offered dedicated H100-based clusters with Nvidia’s full software stack preinstalled. It offered a straightforward, attractive pitch to enterprise customers, allowing them to rent Nvidia’s preferred AI platform without building their own data centers.

In practice, the model proved difficult to scale. Pricing was high compared to commodity GPU instances, integration with existing cloud tooling was uneven, and support responsibilities were split between Nvidia and its hosting partners. Customers running DGX Cloud across multiple providers faced operational complexity, while hyperscalers themselves were rapidly cutting prices on H100 capacity and rolling out their own managed AI services.

Against that backdrop, Nvidia has now folded the DGX Cloud team into its broader engineering organization under engineering leadership. The group’s remit has shifted toward internal use, including AI model development, software validation, and pre-silicon and post-silicon testing of new GPU platforms.

This makes a lot of sense given that Nvidia’s largest customers are the very companies it would have been competing against should DGX Cloud have continued operating as it was. AWS, Microsoft, Google, and other cloud providers account for a significant share of Nvidia’s data center revenue. Running a first-party cloud service risked creating channel conflict at a time when those customers are committing billions of dollars to Nvidia hardware.

Economics and partner friction

Operating a competitive cloud platform also requires sustained capital expenditure on facilities, networking, power, and operations. Nvidia’s advantage lies in silicon, systems, and software, not running global data center fleets. While DGX Cloud leveraged partner infrastructure, Nvidia still bore the cost of support, platform engineering, and customer acquisition without controlling the underlying economics.

Considering that hyperscalers have strong incentives to differentiate above Nvidia’s hardware layer, DGX Cloud as a first-party provider fundamentally didn’t make sense. AWS continues to invest in Trainium and Inferentia accelerators; Google is pushing TPUs; and Microsoft is expanding its Maia program, which recently partnered with Intel Foundry to build chips. Even so, all of them remain deeply dependent on Nvidia GPUs for leading-edge AI workloads. For Nvidia, preserving that dependency is understandably worth more than capturing cloud service revenue.

There were also signs of strain in GPU supply dynamics. Nvidia has increasingly acted as both supplier and customer, leasing back large volumes of GPU capacity from cloud providers and specialized operators — in one case, 18,000 GPUs to the tune of $1.5 billion over four years.

Long-term commitments to buy unused capacity ensure Nvidia has access to compute for internal development while helping partners justify aggressive data center expansion. This arrangement becomes harder to justify if Nvidia is simultaneously selling competing cloud services.

By stepping back and restructuring DGX under its engineering business, Nvidia is simplifying those relationships. DGX Cloud becomes a tool that helps partners deploy Nvidia hardware more effectively, rather than a product that competes with their own offerings. That alignment is particularly important as Nvidia prepares successive GPU generations that will demand even tighter hardware and software coordination.

What does this mean for Nvidia’s cloud strategy?

The reorganization does not necessarily mean there will be reduced investment in cloud-adjacent technologies. On the contrary, Nvidia continues to expand its software stack across inference, orchestration, networking, and systems management. Platforms such as CUDA, TensorRT, and Nvidia’s inference frameworks are designed to run everywhere hyperscalers operate, and DGX Cloud now functions as a proving ground for those technologies.

Ultimately, everyone’s a winner here. Hyperscalers benefit from the neutering of what could have become a major source of friction, and Nvidia benefits from maintaining its position as the go-to partner of those hyperscalers rather than a competitor, even as cloud providers continue exploring in-house silicon to manage costs and supply risk. It also demonstrates to enterprise customers that Nvidia’s primary role is to enable and power AI infrastructure rather than operate it.

There’s no ignoring how quickly Nvidia’s priorities have evolved. What looked, two years ago, like an effort to move up the stack into cloud services now looks more like an experiment that informed a more focused strategy. DGX Cloud is still kicking, but as infrastructure for Nvidia’s own engineers and a bridge to its partners, not as an effort to become the next big cloud platform.

Google Cloud just announced that Anthropic has signed a major deal that will expand its use of the search giant’s TPU chips for training future Claude models. According to the press release, this will allow the AI company to use up to a million TPUs from Google, as well as access other cloud services that it offers, giving Anthropic over a gigawatt of compute capacity by 2026.

The company has already been using Google Cloud’s services since 2023, when it first signed a strategic partnership to use the latter’s infrastructure to train and run its models. Anthropic has since used Google Cloud’s platforms, including Vertex AI and Google Cloud Marketplace, to offer its Claude models to customers, including design and collaboration platform Figma, cybersecurity firm Palo Alto Networks, and AI code editor Cursor, among others.

“Anthropic and Google have a longstanding partnership, and this latest expansion will help us continue to grow the compute we need to define the frontier of AI,” Anthropic CFO Krishna Rao said. “Our customers — from Fortune 500 companies to AI-native startups — depend on Claude for their most important work, and this expanded capacity ensures we can meet our exponentially growing demand while keeping our models a the cutting edge of the industry.”

It makes sense for Anthropic to stick with its current provider, Google Cloud, when expanding its capacity, because its models and the engineers behind them are already used to Google’s system. Aside from that, it also offers great price-performance and efficiency, which will help with its $26 billion revenue target for 2026.

“Anthropic’s choice to significantly expand its usage of TPUs reflects the strong price-performance and efficiency its teams have seen with TPUs for several years,” said Google Cloud CEO Thomas Kurian. “We are continuing to innovate and drive further efficiencies and increased capacity of our TPUs, building on our already mature AI accelerator portfolio, including our seventh-generation TPU, Ironwood.”

The Claude developer is seemingly taking a different route from its biggest competitors, including OpenAI and xAI, which are investing billions in their own infrastructure. For example, the former, aside from trying to develop its own chip, plans to build massive data centers in the U.S. and across the globe through Stargate. On the other hand, Elon Musk has already spent a lot of money on his Memphis Supercluster, with plans to spend billions more on an even bigger AI data center project, even though the company is expected to burn through $13 billion this year.

Anthropic’s move to stay away from massive infrastructure investments means reducing its risk of laying out a lot of cash for hardware that will take years to materialize. However, that also puts it at the mercy of hardware providers, making it harder for the company to scale and making it prone to price fluctuations, especially if it does not have long-term contracts in place that will protect it from such instability. But with the massive data center build-out happening all over the world, it seems that this is a risk that the company may be willing to take, so it can focus more on its core products.

The first Nvidia H100 GPUs are heading into space next month. This out-of-this-world news comes as AI factory company, Crusoe, reveals its plans to become “the first public cloud operator to run workloads in outer space.” The orbital Crusoe Cloud deployment will be powered by Nvidia H100 GPUs and AI accelerators in strategic partner Starcloud’s AI data center satellites.

Why have AI data centers in space?

Crusoe and Starcloud share an energy-first vision of an AI data center infrastructure that leverages solar power in space. The rays of energy-rich light emitted by the sun are unencumbered by the Earth’s atmosphere in the Starcloud’s orbit. Thus, companies that establish these kinds of resources in space can enjoy “almost unlimited, low-cost renewable energy.”

Estimates suggest space-based solar-powered data centers will benefit from 10x lower energy costs (including launch cost) vs those on Earth.

As you can see in our embedded image, space allows for massive solar panel arrays, with no concerns about land use. Moreover, operating such systems puts no strain on the Earth’s energy grid. However, we would be interested to learn more about GPU cooling in space-based data centers, given that there is no air for the Earth-like traditional convection of heat, but also very little heat. Nvidia claims Starcloud will use the vacuum of deep space as "an infinite heat sink."

Crusoe and Starcloud

Crusoe has a long history of co-locating its compute infrastructure near novel energy sources. Thus, GPUs in space are just a logical step for the company, albeit quite a large one. Cully Cavness, co-founder, president and COO of Crusoe, sai,d "By partnering with Starcloud, we will extend our energy-first approach from Earth to the next frontier: outer space."

Starcloud is a Redmond-based Nvidia Inception program backed company. It is focused on building data centers in space, scalable to gigawatt capacities, making it a natural fit for Crusoe’s ambitions.

“Having Crusoe as the foundational cloud provider on our platform is a perfect alignment of vision and execution,” said Philip Johnston, CEO of Starcloud. “Crusoe's expertise in building rugged, efficient, and scalable computing solutions makes them the ideal partner to pioneer this new era.” Johnston also underlined the possibilities of unlocking cloud computing power in space for research, discovery, and innovation.

The partners seem to have even more ambitious plans, going forward, with regard to scaling. However, at this stage, it is rather premature to talk of plans more astronomical than those already sketched out from now until early 2027.

Space data center timescales

Starcloud will be the first to launch Nvidia H100s into space in November 2025. Crusoe will deploy its Crusoe Cloud service on a Starcloud satellite scheduled to launch in late 2026. Subsequently, Crusoe plans to offer limited GPU capacity from space by early 2027, thus “pioneering a new paradigm for AI factories.”

Although the U.S. government restricts sales of advanced processors for AI and HPC to China-based entities, these entities can use cloud services from companies like Amazon Web Services (AWS) and Microsoft to access advanced U.S. processors, reports Reuters. Although the U.S. government mulls restricting availability of American cloud services for AI and HPC to entities in China and other sanctioned countries, for now they can bypass export bans by using AWS, Azure, and the like.

Chinese entities use AWS and Azure to access Nvidia's A100 and H100 processors

More than 50 public tender documents reviewed by Reuters reveal that at least 11 Chinese entities have pursued restricted U.S. technologies through cloud services over the past year. In fact, as many as four of these entities identified AWS as their cloud provider, although they accessed these services via Chinese intermediaries instead of directly from AWS.

One notable example is Shenzhen University, which spent approximately ¥200,000 (around $28,000) on an AWS account to access cloud servers powered by Nvidia A100 and H100 processors to work on an unspecified project. These GPUs are specifically banned for export to China by the U.S., yet the university acquired access to them through an intermediary, Yunda Technology.

Similarly, Zhejiang Lab, which is developing its own large-language model (LLM) called GeoGPT, considered spending ¥184,000 (around $25,800) on AWS cloud computing services citing insufficient computing power from Alibaba's Alicloud. However, the lab did not make the purchase for an unknown reason, and it also did not disclose how it ultimately met its compute performance requirements.

In March, the Suzhou Institute of Advanced Research at the University of Science and Technology of China (USTC) announced a tender to rent 500 cloud servers, each equipped with eight Nvidia A100 chips, for an undisclosed purpose. By April, Hefei Advanced Computing Center Operation Management had provided the servers, though the cloud service provider remained unknown. In May, USTC was placed on the U.S. 'Entity List' due to its use of U.S. technology for quantum computing with potential military and nuclear applications.

Also in April, Sichuan University announced in a tender document its plan to develop a generative AI platform, for which it intended to acquire 40 million Microsoft Azure OpenAI tokens. By May, procurement records indicated that Sichuan Province Xuedong Technology was the supplier of these tokens. Azure OpenAI operates under Microsoft's Azure policies and is therefore has no direct business ties to OpenAI.

AWS and other companies involved have stated that they comply with all applicable U.S. laws. Following inquiries from Reuters, AWS updated and removed certain promotional posts on its Chinese-language platforms, reflecting heightened scrutiny over its services in China.

U.S. moving closer to ban cloud access to GPUs for Chinese entities

Last October, the U.S. government first said that it was mulling restrictions on Chinese entities' access to U.S.-based cloud computing services that offer high-performance hardware for AI and HPC, amid concerns over China's use of the technology to boost its military potential. The U.S. Secretary of Commerce Gina Raimondo reiterated the plan in January.

The U.S. government is now moving to close this regulatory gap. Legislation introduced in Congress would give the Commerce Department the authority to regulate remote access to U.S. processors through cloud services. The Commerce Department has also proposed new rules that would require cloud providers to verify the identity of users training large AI models and report their activities. However, it is unclear when and if any laws will be passed.

Big loophole in any future laws

However, the U.S. cannot prevent Chinese entities from accessing cloud services offered by non-American companies. The only way to do so would probably be to restrict sales of AI and HPC processors to prominent cloud providers in Europe and the Middle East, which would harm many American companies, including Nvidia. As a result, instead of using American cloud services, Chinese companies could simply switch to alternatives in Europe and the Middle East

Microsoft is asking nearly 10% of its China-based workforce to move to the U.S., Ireland, Australia, or New Zealand. According to Reuters, most affected workers are Chinese engineers working in the company’s machine learning and cloud computing divisions — key battlegrounds that the U.S. and China are vying for in their bid to gain technological supremacy.

This announcement came soon after the White House’s press release announcing increased tariffs on high-tech Chinese goods and can be seen as Microsoft’s reaction to keep the U.S. government happy. However, a Microsoft spokesperson said in an emailed statement to Reuters that this was just a routine business move.

“Providing internal opportunities is a regular part of managing our global business. As part of this process, we shared an optional internal transfer opportunity with a subset of employees,” Microsoft said in its email. The spokesperson also added that “Microsoft remains committed to China and will continue to operate there and other markets.”

Although the company did not confirm how many employees received the transfer opportunity, internal sources say it affects 700 to 800 people. This is a substantial number of Microsoft’s China-based workforce. The Global Times reported in 2022 that Microsoft already had more than 9,000 full-time workers and was expected to hire over a thousand more in the following year.

There is no telling how many offers will be accepted, especially as this is optional for the employees. But even if a small number only accepts Microsoft’s proposal, this is a potential brain drain of China’s top talent, especially given their highly niche specialties. After all, many former Microsoft employees eventually became leaders in top Chinese tech firms, including Baidu and TikTok-owner ByteDance.

Microsoft’s heavy presence in China has invited scrutiny from American lawmakers. In a March 2024 letter to U.S. Commerce Secretary Gina Raimondo, U.S. Representative Pat Fallon said, “What we seek to understand is if and how Microsoft’s broad usage across the U.S. federal government, close ties to [the People’s Republic of China]’s government and compliance with intrusive PRC laws threatens U.S. national and economic security.”

He then adds, “No U.S. company should be playing a role in supporting the Chinese government. It is critical that any such efforts be stopped, and that broader Chinese operations be carefully scrutinized.” (via the New York Post)

We cannot confirm whether the transfer offer is indeed a routine business decision or a move designed to keep American pressure off Microsoft’s China operations. However, with the emails coming so soon after the White House’s recent moves to up the ante in the Sino-American chip war, we cannot help but feel that Microsoft is taking steps to slowly move away from China and avoid U.S. sanctions.

A report published by the Wall Street Journal on Monday surfaces an accusation that Nvidia might be willing to delay data center GPU orders if it becomes aware of a customer looking for greener pastures. Thus, rival AI chipmaker Groq says that fearful customers are secretive about acquiring or designing AI acceleration technology for fear of retaliatory shipment delays. This stands in contrast to Nvidia's own statements on the matter, with the company saying that it is trying to allocate supply fairly and to offer customers alternative access to compute while they wait for their shipments. 

It would be a pity if something were to happen to your pending GPU order...

Fear of being found out is so strong that it isn’t uncommon for people to deny they have had meetings with rival AI accelerator chip firms, Jonathan Ross, CEO of rival chip startup Groq, told the WSJ.

“A lot of people that we meet with say that if Nvidia were to hear that we were meeting, they would disavow it,” Ross told the WSJ. “The problem is you have to pay Nvidia a year in advance, and you may get your hardware in a year, or it may take longer, and it’s, ‘Aw shucks, you’re buying from someone else, and I guess it’s going to take a little longer.’”

If true, this revelation has echoes of Intel using strong-arm tactics in the early noughties to disincentivize PC-making partners from offering AMD-inside systems. Elsewhere in the WSJ report, it's made clear that Nvidia is enjoying its dominance of the booming AI market, with an estimated 80% of companies using AI acceleration reliant upon the green team's hardware.

We aren’t competitors, honest

The WSJ report even seems to hint that tech goliaths like Microsoft, Google, and Amazon are making their own AI accelerators, yet insisting they aren’t planning to become Nvidia competitors.

In contrast to the statement from Ross, Nvidia CEO Jensen Huang is reported to have told an analyst that he tries to be fair in allocating GPUs to companies and will avoid selling to companies that won’t immediately use the accelerators.

Huang also indicated that he will introduce GPU customers to cloud providers to tide them over as they wait for GPU orders to be fulfilled. Incidentally, Nvidia invested heavily into a cloud company called CoreWeave, which operates large data centers stuffed with a significant fleet of Nvidia GPUs and rents out computing power.

Groq clearly isn't a disinterested third party observing all of this from afar. Quite the opposite in fact. And Nvidia is undoubtedly the 800 pound gorilla of the AI world right now. What isn't clear is whether there's any proof of malfeasance on the part of Nvidia, or whether anyone would dare to come forward with said proof if it exists.

Demand for GPUs is unquestionably ramping up again as prices eclipse $70,000 per GPU in some China locales (that's for a data center H800 GPU), and leaders in the U.S. computing industry are taking to social media to complain that cloud-based GPU resources are fully booked and GPU hardware supplies are all reserved for the year ahead. Drastic times call for drastic measures, and we are beginning to see evidence that GPUs that should be heading to home desktop PC rigs are instead being snapped up by the AI industry players.

Naturally, gamers and enthusiasts will be worried about a repeat of the cryptomining craze, which decimated consumer GPU supplies: Are the crypto-bros of old destined to be replaced by the AI-bros — snapping up our precious gaming GPUs?

The first solid evidence of AI-focused businesses buying up consumer GPUs comes from a boast Tweeted by iconic IT hacker and entrepreneur George Hotz (AKA geohot).

Above, if you expand the tweet, you can see Hotz, who is currently leading an automated AI-driven driving assistance business called Comma AI, talk about buying up cases of AMD gaming GPUs.

The image shows a partly unpacked purported "7.38 PFLOPS of compute." We can clearly spy an XFX MERC 310 GPU in the photo, and the box reads "AMD Radeon RX 7900 XTX." That's an XFX design based around AMD's flagship consumer GPU and the second-placed Best Graphics Cards for Gaming in 2023 in our frequently updated roundup.

Our calculations from the Hotz statement and image are that there are 60 GPUs in this "7.38 PFLOPS" batch (based on FP16 performance). Each of these RX 7900 XTX cards costs $979.99 on Amazon, at the time of writing. Thus, Hotz apparently just splashed out around $60,000 on this modest stack of GPU power.

Obviously this is a mere drop in the ocean, as the Comma AI founder and president says that there are "exaFLOPS more to come." For reference, one exaFLOPS is 1,000 petaFLOPS of performance. If we assume that Hotz aims to buy up to 7.38 exaFLOPS of GPU compute (1,000 x the first batch), he may have budgeted ~$60 million for this consumer graphics card buying spree. Or perhaps only ~$20 million, if he's only after around 2 exaFLOPS of AI compute.

One interesting aside with the above purchase of AMD GPUs is that certain AI workloads can use a lot of memory. While paying nearly $1,000 per GPU for an AMD 7900 XTX that packs 'only' 122.88 teraFLOPS of FP16 number crunching prowess might seem odd, given the RTX 4060 costs $300 and can provide around 121 teraFLOPS of FP16 compute (with sparsity), the AMD GPU provides three times as much VRAM. So, if the workloads need memory, you'd need three RTX 4060 cards or two RTX 4070 cards to get the same 24GB — plus the servers to hold all those GPUs.

The Tweeted statement from George Hotz isn't the first warning sign we have seen of AI-industry demand for GPUs heating up. A few days ago, we reported on the rapid inflation of GPUs in China. Fears of further US sanctions on China are thought to have helped stoke up prices of accelerators like the A800 and H800 GPUs. A few weeks earlier, the price of an A800 jumped 20% almost overnight due to impending sanction fears.

Given other recent indicators, we can clearly see that the rise of AI is having a profound impact on the supply of compute-focused GPUs. Now the only question is how far that insatiable demand will bleed over to gaming GPUs, too.

IN yet another sign of the increasing scarcity of Ipv4 addresses, citing a pricing increase of 300% for new addresses, Amazon Web Services (AWS) will introduce a charge of $0.005 per IP per hour for all public IPv4 addresses starting from February 1, 2024. New charges will apply to Virtual Private Cloud (VPC), Amazon Global Accelerator, and AWS Site-to-site VPN tunnel services using public IPv4 addresses. The advance notice was given to AWS customers just ahead of the weekend via a blog post by the Chief Evangelist of the firm, Jeff Barr.

The main reason behind AWS' introduction of charges is the purported 300% rise in costs in acquiring a single public IPv4 address over the last five years (due to scarcity). It is also highlighted in the blog that there is a need to conserve IPv4 addresses and modernize by accelerating moves to IPv6 (no hourly IP charges).

So, $0.005 seems like a very small charge, but remember that this is an hourly charge and will apply to every public IPv4 address allocated in your account, whether attached to an AWS service or not. A quick calculation shows that for one IPv4 address, the new charging amount will be $43.80 per annum. The chart below should help you see where the charges apply and what exactly is new.

Amazon is also introducing a new Public IP Insights tool for users to audit their use of public IPv4 addresses. It hopes the free tool guides users to see, sort, filter, and learn more about each of the public IPv4 addresses they are using. It will also help users see where updates to their apps might minimize the effects of new charges. The AWS blog has previously provided guidance on moving services to IPv6.

To provide some background to the new charges AWS is advising its customers to be ready for, let's look a bit closer at IPv6, the destination to which customers are being nudged. IPv6 was designed to replace IPv4 for the provision of internet addresses, as the latter is limited to 4.3 billion devices by nature of its 32-bit addressing scheme. IPv6 was created in 1998 with 128-bit addressing to provide approximately 340 undecillion addresses. That ought to be enough for anybody.

IPv6 wasn't doesn't just provide an extension to the number of addresses available; it's not just like adding digits to phone number lengths. It is also claimed to handle packets more efficiently, improve performance, increase security, and more. Google charts IPv6 adoption among its users, and it reckons just over 42% of the internet uses IPv6 right now. Meanwhile, most of the world is thought to have run out of IPv4 addresses sometime between 2011 and 2018, with almost all 'new' addresses put into use today being IPv4 addresses that are abandoned or have been sold for re-use.

Windows offers multiple built-in methods to back up your PC so you don't lose your precious data files in the event of a hardware failure or accidental erasure. But choosing a backup plan without knowing its pros and cons is a recipe for disaster. Not all backup methods on Windows are created equal, and each has a specific use case. 

If you want to keep a backup of your entire drive so that you can restore everything – the OS, the apps, the settings and the data – onto a replacement disk, see our article on how to make a full system image backup in Windows. However, for regular backups, it’s often preferable to keep copies of your data files only. 

Microsoft bundles three types of backup solutions for a Windows PC: Backup & Restore, File History, and OneDrive (Cloud Backup). The first two do not require an internet connection, while cloud backup with OneDrive needs an active internet connection to work properly. We will discuss each in detail to help you clearly understand them. Let's begin.

Back Up Your PC Using File History

Microsoft introduced File History in Windows 8 to offer a slightly better backup solution for personal files. Unlike the old Backup and Restore feature (see below), it can preserve older versions of files so you can go back to an earlier draft of a document or an earlier version of your code. By default, File History only saves the data in the User's folder, including Downloads, Videos, Documents, Music, Desktop, OneDrive files, and more folders. So, all the folders that contain your personal data on the C: drive are safe and backed up every hour.

You must use an external hard disk for File History because it automatically backs up files every hour. So, you will have multiple backup versions of personal files, which requires more storage space. 

Unlike the Backup and Restore feature, File History stores the files in an unencrypted format. It doesn't create a big single archive file and stores and creates folders. So, you can use File Explorer to view, copy, and modify any file or folder inside the FileHistory folder on the external hard drive. Or, you can use the Restore personal files option on the Control Panel's File History page.

To use File History on your Windows PC, repeat the following steps:

1. Navigate to Control Panel->System and Security->File History.

2. Click the Turn On button to enable File History. It will begin copying the files to your external hard drive and will do so every hour.

You can change this schedule by clicking Advanced Options to alter the backup frequency and deleting old copies.

3. Click the Run now button to force create a file backup anytime.

File History is great for preserving the data in the C:\users\ folder but ignores the data on other locations and drives. You can manually add folders to the Library folder using File Explorer to force File History to back up those too.

Back Up Your PC Using OneDrive

Microsoft's OneDrive is now deeply integrated into Windows 11. You can back up the User folder and your system preferences, settings, and apps using a Microsoft Account on your Windows PC. This approach will help you save your crucial files in the cloud and sync them across all the devices signed in with the same Microsoft Account. 

It is also useful while reinstalling Windows or setting up a new PC because you can sign in with the same Microsoft Account. The setup will configure Windows to use all the previous settings and preferences, and you can reinstall all the Microsoft Store apps in one click. 

To use OneDrive on your Windows PC, repeat the following steps:

1. Launch OneDrive. You can do this by searching "OneDrive" in the Start menu and pressing the Enter key.

2. Sign in using your OneDrive email address or Create a new Account.

3. Click Next.

4. Enable the toggles of all folders you want OneDrive to back up. Click the Start backup button.

By default, it adds only the Documents, Desktop, and Pictures folders, but you can add more folders afterward. Wait for OneDrive to upload all the folders. Click Next repeatedly.

5. Click the Open my OneDrive folder button to view all the uploaded files of OneDrive in File Explorer. If you want to add more folders, drag and drop any folder to the OneDrive folder.

OneDrive is useful because you rely on cloud storage and can access your files with or without a Windows PC. But it requires a decent internet connection with ample bandwidth to upload files. It cannot save your installed programs and has a meager 5GB storage limit in the free tier. You can upgrade to get 1TB of cloud storage, but compared to the other two Windows backup options, you will have to pay for it.

Back Up Your PC Using Backup and Restore

Backup and Restore is a legacy backup option in Windows which still exists in the Control Panel. It was designed for Windows 7, but still is available and works in Windows 11. You can use it to either back up your data folders or create a full system image you can restore from if your boot drive fails. However, we recommend using third-party disk imaging software for a complete image and employing Backup and Restore for your files – perhaps all the content of your Document, Photos, Video and Music libraries.

Backup and Restore lets you save the data from all internal drives on an external drive (perhaps one of the best external SSDs) or a USB Flash Drive. The latter will only be useful for small files with a slow copying speed. We recommend you use an external hard drive that can accommodate bigger backups. 

Another great thing about the Backup and Restore option is that you can schedule automatic backups. It detects changes in files and saves the newer ones as well. Backup and Restore will create three files: a Windows backup file, a mediaID.bin file, and a "WindowsImageBackup" folder. The size of the Windows backup file depends on the data stored in C and other drives and can be more than 3 GB. The "WindowsImageBackup" folder is about 11 GB. 

You can view and restore the contents stored inside the backup file using the Restore option on the Control Panel's Backup and Restore page. Restoring a particular file, folder, or even a drive is possible. However, you cannot view or modify the files and folders inside the backup file directly using File Explorer due to the encrypted nature of the backup.

To use Backup and Restore on your Windows PC, connect the external hard disk to your PC and repeat the following steps:

1. Open Control Panel. The easiest way to do so is by searching for "Control Panel" using the Start menu or Windows Search and clicking on the first relevant search result.

2. Navigate to System and Security->Backup and Restore (Windows 7).

3. Click the Set up backup option.

4. Select the external hard drive name from the list and click Next.

5. Click the Let me choose radio button to personally select all the files you want to backup from the C drive and other drives. Click Next.

6. Select the C drive and other drives. Keep the Include a system image of drives: EFI System Partition, (C:), Windows RE tools checkbox selected, and click Next.

7. Click the Save settings and run backup button.

By default, Windows will run an automated backup every Sunday at 7:00 PM. We recommend you go with the default settings unless you handle a lot of new important files daily. To change this schedule, click the Change schedule button and pick the backup's preferred day, frequency, and time.

Always keep the external hard drive connected to your PC all the time or on the scheduled backup date, never to miss an automatic update. You can restore individual files or folders whenever you need using the Backup and Restore section in the Control Panel. If your PC encounters a major issue, you can use the system image backup to restore all its drivers, files, and installed programs in a few clicks. 

The legacy Backup and Restore feature with scheduled backups is a decent option if you run into an issue. File History is useful for preserving personal files and saving multiple versions. Lastly, OneDrive paired with appropriate Windows 11 backup settings can benefit while reinstalling Windows 11.

Intel's project Endgame, a network-based solution that aimed to bring the processing power of cloud based Intel ARC GPUs to local systems, seems to be on an indefinite hold. Announced by Intel through Twitter, the news means that the timeline for users looking to increase their available GPU grunt through the cloud has become even more blurry than it already was. Hat-tip to Bionic_Squash.

Project Endgame was meant to be available in Beta form by the end of 2022, but that time has come and gone and Intel's accelerated computing dreams seem to have faded away. Unfortunately, the tweet from Intel didn't provide any updates on the how or the why of Project Endgame's apparent hold.

Project Endgame was announced by Intel back in 2022, at a time when Raja Koduri still served as Intel’s Senior Vice President and General Manager of Intel Graphics group (Raja himself having been poached by Intel from a similar position at AMD). At the time, Raja showcased the potential of the service by piloting a low-power laptop running Epic Games’ impressive "Matrix Awakens" demo.

With that particular demo's performance demands, however, frame-rates on the low-power local computer were low and choppy; a situation rapidly corrected by activating Endgame and its "continuous compute" option, which enabled Raja's laptop to leverage a network-connected ARC GPU to accelerate the workload.

While Intel didn't confirm the scenarios where Endgame would be a feasible way to access more performance, it's believed that the company wanted to extend it from general computing to even Edge Computing and IoT - allowing even low-power, remote installations to process graphics and compute workloads (including gaming, generative AI, video encoding, and similar) so long as their network had the necessary bandwidth.

Endgame could be a truly invisible performance boost option; unlike services like GeForce Now, which generate a cloud-based OS instance where all the processing happens, Endgame would allow users to bring ARC's performance to their own installations. The project's cryptic delay is unfortunate, but considering its potential, it's likely that Intel will eventually push forward with the project - whether it remains an Endgame or not.

Apple is reportedly joining the AI rat race. According to a report from Bloomberg's Apple whisperer, Mark Gurman, the Mac and iPhone builder is working on an artificial intelligence framework and its own chatbot.

Gurman's anonymous sources detail a key piece of technology, dubbed "Ajax," which serves as a foundation for building large language models like those that power ChatGPT. Using Ajax, a small engineering team at Apple also built a chatbot that some internally are allegedly calling "Apple GPT." Ajax was reportedly developed in 2022 to "unify machine learning development at Apple," while the chatbot in use with internal teams seems to be a more recent development. Ajax is built on Google Jax and runs on Google Cloud, Bloomberg reports.

Apple has been noticeably silent on artificial intelligence. The company didn't refer to AI at all at its Worldwide Developers Conference, instead referring to advancements in machine learning. Many users have noted that Siri, Apple's voice assistant first released in 2011, isn't as advanced as competitors like Amazon Alexa or the Google Assistant. 

Bloomberg's report paints Apple as being uneasy about how fast generative AI tools have grown into the mainstream and is worried "about missing a potentially paramount shift in how devices operate." Apple's head of machine learning and AI, John Giannandera, and the senior vice president of software engineering, Craig Federighi, are leading the charge.

Apple Maps, search and Siri have seen some AI improvements based on Ajax already, Gurman reports. The question about the so-called "Apple GPT" (definitely not a final name) is in what way it could potentially show up in consumer projects.

Right now, that chatbot, which is only available through a web interface, is just being used internally by a small team, and apparently requires employees need to be rubber-stamped by the higher ups to get access. Additionally, Apple isn't allowing output from the bot to be used for projects that will be consumer-facing.  However, Apple employees apparently are using it for prototyping, summarizing text and responding to questions.

What Apple's bot doesn't do, yet, is anything new, according to Gurman's sources. It's similar to Bing Chat, ChatGPT and their ilk, but doesn't have any differentiators.

It's unclear where Apple will use this AI tech in consumer-facing work, if at all. Gurman notes that Siri is the obvious choice, though Apple has pointed out that AI has ended up in areas like crash detection and fall detection. But Siri, with a focus on privacy, hasn't been able to surpass its predecessors. Smartening it up with AI may make it a bit more helpful.

Apple appears to be taking AI slower than others companies, like OpenAI, Microsoft, and Google, which are largely testing with open betas. That matches Apple's approach in many product categories, in which it waits for them to mature before swooping in with a refined product.

CEO Tim Cook himself has tried ChatGPT, but also warned about the issue for bias and misinformation. It will be interesting to see how Apple tackles those issues.

Following rumors that the U.S. government may ban sales of on high-end compute GPUs to China, the price of of Nvidia's A800 chips on the Chinese market has surged in a big way. The price jumped by a whopping 20% in just two weeks, according to a report from DigiTimes.

The standard price of Nvidia's A800 compute GPU in the PCIecard form-factor was around CNY90,000 (US$12,400) per unit just two weeks ago. Right now, the price is approaching CNY110,000 (US$15,000) per unit, which is a 20% increase.

The issue of inflated chip prices, alongside a pressing shortage of computational power in China, have drawn attention at the World Artificial Intelligence Conference 2023. Acquiring high-end AI chips via official routes in China has become increasingly difficult. And as a result, Chinese technology firms that need computing power have resorted to utilizing cloud computing services provided by companies like Amazon AWS and Microsoft Azure. These services are typically used for large language model training tasks, with data centers situated in Singapore or China.

Chen Pei, vice president of Vibranium Consulting, stated that leasing cloud GPU computational power is notably costlier than constructing own GPU computing clusters, with prices for large cloud providers hovering around $2 to $3 per hour per GPU. Echoing this, Sun Jin from CloudWalk Technology, a Chinese AI and computer vision firm, disclosed that Chinese companies are often left with no option but to bear the high costs of leasing cloud computing power. This comes despite these costs being 50-100% higher than establishing their own datacenters.

However, the ability of Chinese vendors to continue using U.S.-based cloud computing services such as AWS and Azure in the future is uncertain, as this will depend on the regulatory steps taken by the U.S. government, which could involve the necessity for licenses or even a complete ban. 

China's computing industry faces multiple significant challenges, including the time-consuming process of building AI computing clusters, difficulties in procuring high-end AI chips from abroad, and a shortage of domestically produced AI chips, such as Biren's BR104 and BR100. In light of the potential limitations on accessing European and U.S. cloud computing power, there is a rising expectation for Chinese semiconductor manufacturers to focus solely on refining their chip manufacturing processes and to make progressive strides in software innovation.

Inflection AI, a new startup found by the former head of deep mind and backed by Microsoft and Nvidia, last week raised $1.3 billion from industry heavyweights in cash and cloud credit. It appears the company will use the money to build a supercomputer cluster powered by as many as 22,000 of Nvidia's H100 compute GPUs, which will have peak theoretical compute power performance that is comparable to that of the Frontier supercomputer.

"We will be building a cluster of around 22,000 H100s," said Mustafa Suleyman, the founder of DeepMind and a co-founder of Inflection AI, reports Reuters. "This is approximately three times more compute than what was used to train all of GPT-4. Speed and scale are what's going to really enable us to build a differentiated product."

A cluster powered by 22,000 Nvidia H100 compute GPUs is theoretically capable of 1.474 exaflops of FP64 performance — that's using the Tensor cores. With general FP64 code running on the CUDA cores, the peak throughput is only half as high: 0.737 FP64 exaflops. Meanwhile, the world's fastest supercomputer, Frontier, has peak compute performance of 1.813 FP64 exaflops (double that to 3.626 exaflops for matrix operations). That puts the planned new computer at second place for now, though it may drop to fourth after El Capitan and Aurora come fully online.

While FP64 performance is important for many scientific workloads, this system will likely be much faster for AI-oriented tasks. The peak FP16/BF16 throughput is 43.5 exaflops, and double that to 87.1 exaflops for FP8 throughput. The Frontier supercomputer powered by 37,888 of AMD's Instinct MI250X has peak BF16/FP16 throughput of 14.5 exaflops.

The cost of the cluster is unknown, but keeping in mind that Nvidia's H100 compute GPUs retail for over $30,000 per unit, we expect the GPUs for the cluster to cost hundreds of millions of dollars. Add in all the rack servers and other hardware and that would account for most of the $1.3 billion in funding.

Inflection AI is currently valuated at around $4 billion, about one year after its foundation. Its only current product is a generational AI chatbot called Pi, short for personal intelligence. Pi is designed to serve as an AI-powered personal assistant with generative AI technology akin to ChatGPT that will support planning, scheduling, and information gathering. This allows Pi to communicate with users via dialogue, making it possible for people to ask queries and offer feedback. Among other things, Inflection AI has outlined specific user experience objectives for Pi, such as offering emotional support.

At present, Inflection AI operates a cluster based on 3,584 Nvidia H100 compute GPUs in Microsoft Azure cloud. The proposed supercomputing cluster would offer roughly six times the performance of the current cloud-based solution.

Reuters reports that the Biden administration is considering tightening its export restrictions to China even further, following China's legislative move that allows it to essentially bottleneck exports of rare earth metals - gallium and germanium) required for semiconductor fabrication. Except now, the dance isn't about hardware or tech exports: it's actually about access to US-based cloud computing capabilities (and especially those with Artificial Intelligence hardware). This is but the latest move in the continuously-escalating economic and logistical tensions, which have historically resulted in higher pricing for hardware components as friction is added to the worldwide supply-chain.

Following years of less-than-stellar results from the US' imposed technological export rules to China, the US is now seemingly looking to cut into another escape route used by China to acquire access to the latest and greatest processing power: cloud computing. If you can't acquire the latest chips for your own data center, you can always purchase access to them in a cloud environment; that's exactly the scenario the White House wants to see end by forcing cloud computing providers such as Microsoft, Google, Amazon and others to seek a licence with the US government in order to serve chinese customers. As usual, the U.S. Department of Commerce will oversee the execution of this added restriction, which is expected to be implemented in the coming weeks.

China was always in an extremely strong position to execute its plans to achieve technological independence from the West due to it controlling around 55% of the world's rare earth output (as of 2020). While talks have been had regarding the need to procure other supply chain routes other than China, that's easier said than done. It requires not only finding economically viable rare-earth deposits outside China's influence, but also the build-up of supporting infrastructure. That's no small feat to achieve; and China's power mostly comes from the fact that around 85% of the world's rare earth metals have to go through processing facilities in the People's Republic of China.

That leveraging power was finally wielded by China yesterday, as it imposed restrictions on exports for gallium and germanium (germanium being one of the breakout materials that could power next-gen semiconductors). Being the world's largest producer (and stock holder) for both rare metals, it's not like there are other sourcing alternatives.

For China, it's a win-win: the country doesn't have the technological know-how to fully explore semiconductor designs that make use of gallium or germanium, but they can certainly cut off access to materials and R&D for those that can. In the end, China likely wouldn't have been able to purchase the silicon whose production it's now bottlenecking anyway (due to the technological export restrictions), so the country can safely leverage its monopolistic position to stem the required minerals without losing much in the tech arms race itself. Smartly, China isn't bottlenecking essential, run-of-the-mill materials that could impact its own semiconductor needs; it only aims to smother the bleeding-edge.

There's seemingly no end to the back and forth between China and the US as both superpowers fight for global hegemony. As friction is added, however, pricing of impacted products is only expected to rise. There are still many routes both countries could take in this particular fight, but in the end, it seems like China has a slight upper-hand: that's what you get with any sort of monopoly (whether it's a "deserved" one or not).

Microsoft is officially rolling out its new AI assistant Windows Copilot to Windows 11 Insiders today with Dev Channel Build 23493. The new assistant was just announced last month and will now be available for the most hardcore Windows users to try out. According to Microsoft, Windows Copilot will be a controlled feature rollout, and will require the use of Windows 11 build 23493 and Edge version 115.0.1901.150 for a chance to play with the new feature.

Copilot is a new AI assistant that in many ways will succeed the ChatGPT-powered Bing Chat assistant built into Microsoft Edge right now. Essentially what Microsoft is doing with Windows Copilot is taking its browser-based assistant and turning it into a proper desktop application, with full integration inside the Windows 11 desktop. Though, based on the Edge browser requirements, Copilot appears to be an Edge-powered service that is not fully baked into Windows 11 (yet).

Once enabled, Windows Copilot can be activated by pressing WIN + C or by clicking the new Copilot icon on the taskbar. From there the AI assistant will pop up from a side window allowing you to interact with the application without disturbing any of your open windows. Like Bing Chat, Windows Copilot allows you to ask it any question you want, about any subject or topic, and it will try its best to give you a satisfactory answer.

In the future, Microsoft will be expanding Copilot to be much more than just a search box. Eventually, it will receive support for Bing Chat plugins, a new feature called Hybrid AI loop for AI development, a new home productivity environment for developers — called Dev Home, and more.

Other Features Added To This Latest Dev Channel Build

Windows Copilot is just one of many features being added to this latest Windows 11 dev build. Another major addition is a new home page in the Settings app that will isolate all of your most commonly used settings in one area. The new home page includes settings for Bluetooth devices, Xbox services, Microsoft 365, personalization, account recovery, and cloud storage. There will also be a recommended settings card that will show you your most used settings.

Another update is added native support for open-source archive file formats including, .rar, .7z, .tar, and more. With native support for these formats, you can open these file types natively in File Explorer without installing third party file browsers such as 7-Zip or WinRAR.

A few other additions include new options for Snap Layouts, an enhanced volume mixer in Quick Settings, better app restoration, and more. For more details be sure to check out the the full patch notes here.

Oracle this week said that it had ported its Oracle Database 19c Enterprise Edition, the current long-term support release of Oracle Database, to Ampere's Altra processors that use the Arm instruction set architecture (ISA). The move marks a milestone both for Ampere and Arm ISA as Oracle is one of the most widely used enterprise software suites.

Separately, the company said that eventually it might ditch x86-based instances running Database on processors from AMD and Intel from its data centers in favor of instances enabled by Ampere CPUs. Oracle hopes that by tailoring its Database software for Ampere's single-thread CPUs, it will tangibly increase performance efficiency of its data centers. Furthermore, Ampere, in which Oracle is a lead investor, could also implement tweaks into CPUs to better run Oracle's Database.

"It is a major commitment to move to a new supplier. We've moved to a new architecture and we've moved to new supplier," Larry Ellison, Oracle's founder, said at an event hosted by Ampere, reports Reuters. "We think that this is the future. The old Intel x86 architecture, after many decades in the market, is reaching its limit."

Oracle Database 19c Enterprise Edition is now certified for work on Ampere Altra-based servers for both on-premises deployments and in the cloud by subscribing to Oracle Database Service using OCI Ampere A1 compute instances enabled by Ampere Altra on Oracle Cloud Infrastructure (OCI). It will continue to be offered on AMD and Intel-powered instances probably for years to come. But with Arm-based Ampere servers, Oracle hopes to offer "highly economical price points."

Oracle's OCI Ampere A1 can be used in a flexible VM format ranging from 1 to 57 CPU cores, each with 8GB of memory (maxing out at 456GB), 1 Gbps of network bandwidth for each CPU core (up to a total of 40 Gbps per VM).

Oracle's Database enterprise database management software is used by large businesses, banks, government agencies, retailers, and manufacturers for running online transaction processing (OLTP), data warehousing (DW) and mixed (OLTP & DW) workloads. The software has been in development since 1979 and supports a wide variety of hardware and software platforms, including IBM's Mainframe and Power-bases systems, x86 CPUs from AMD and Intel, Sun's SPARC processors, Intel's IA64 (Itanium) chips, and now Arm-based Ampere Altra SoCs.

"Today's announcement highlights the broad architectural shift across the market to Ampere processors that meet the demands of both modern cloud and on-premises environments," said Jeff Wittich, Chief Product Officer of Ampere. "With the Ampere Altra family of processors, customers of the world's most popular database — Oracle Database — now have a high-performance, energy efficient architecture built with sustainability in mind for organizations of all sizes."

While running Database on energy-efficient cloud-native Ampere's Altra CPUs promises to make a lot of economic sense for Oracle, OCI still needs to offer high-performance computing (HPC), and Dense-IO instances for those who need maximum performance and ultra-fast local storage. So, this week the company announced OCI Compute E5 HPC and OCI Compute E5 Dense-IO instanced based on AMD's 4th Generation EPYC processors with their vast core counts and rich I/O capabilities.

With so many cloud-enabled services allowing you to stream music, movies, and even games from just about any device (or browser), why can’t we do the same for operating systems? Microsoft has already dabbled into this foray with Windows 365, its cloud-based Windows virtual machine offering aimed at businesses.

With Windows 365, each user is given their own custom Windows virtual machine hosted in the Windows 365 service. Since it’s a cloud-based service, each Windows virtual machine can be accessed using a web browser or an app on Windows and Mac devices. Even more flexibility is afforded with support for iOS/iPadOS and Android devices. 

So, what if Microsoft extended the capabilities it currently bills to businesses on a per-user, per-month basis to general consumers? That is precisely what the Redmond, Washington-based company is envisioning, according to internal documents [PDF] made public thanks to the FTC vs. Microsoft hearing currently taking place. While the court case revolves around the FTC’s attempt to block Microsoft’s proposed $68.7 billion acquisition of Activision-Blizzard, ancillary details about the company’s business units have also come to light.

As first reported by The Verge, an internal Microsoft presentation entitled “Modern Life Strategies and Priorities” talks about the company’s ambition to “empower people to make the most of their time.” Microsoft also talks about its goal to serve in excess of 1.5 billion people daily across its software platforms. 

The presentation, dated June 2022, also reveals that one of Microsoft’s long-term goals is to use the foundation it created with Windows 365 to “enable a full Windows operating system streamed from the cloud to any device.” By shifting Windows to the cloud, Microsoft says it will leverage the “power of the cloud and client to enable improved AI-powered services and full roaming of people’s digital experience.”

The idea of a fully “in the cloud” Windows for consumers may have seemed far-fetched a decade ago, but Microsoft has already eased its customers into such a transition. Office has been one of Microsoft’s big tentpole applications for decades in corporate and home settings. With Office 365, Microsoft moved from a monolithic software application that saw a new release every one to two years to a software-as-a-service model where customers pay a monthly or yearly subscription to access cloud-backed Office apps like Word, Excel and PowerPoint. 

This subscription model means reliable, recurring revenue for Microsoft versus a one-time purchase that customers may choose to keep for years without upgrading. Transitioning Windows to this model seemed inevitable for a large, mature company like Microsoft that is perennially tasked with adopting clever ways to increase revenue.

Perhaps one of the first steps in Microsoft’s Windows 11 in the cloud model is Windows 365 Boot. This feature was announced a month ago and allows users to boot directly into an Azure-based cloud PC instance without first logging into their local Windows 11 install. Windows 365 Boot is currently available in Windows 11 22H2. 

Elsewhere in the slide, Microsoft talked about its desire to bolster both its Windows software and Surface hardware business by investing heavily in silicon partnerships. Microsoft already partners with Qualcomm on custom Arm chips for its Surface Pro X convertibles, and the company is actively recruiting senior project engineers for the “Microsoft Silicon Team.”

The event has concluded, and you can see our overview with the live blog below, However, here are links to our deeper coverage of each topic:

AMD Expands MI300 With GPU-Only Model, Eight-GPU Platform with 1.5TB of HBM3

AMD EPYC Genoa-X Weilds 1.3 GB of L3 Cache, 96 Cores

AMD Details EPYC Bergamo CPUs With 128 Zen 4C Cores, Available Now

AMD Intros Ryzen 7000 Pro Mobile and Desktop Chips, AI Comes to Pro Series

AMD is holding its Data Center and AI Technology Premiere today, June 13, 2022, at 10 am PT here in San Francisco -- which is now. We're here to cover the event live and bring you the news as it happens as AMD CEO Lisa Su takes to the stage to reveal AMD's new AI-focused silicon.

AMD has already said that it will reveal its EPYC Bergamo chips at the event. These chips come with up to 128 cores, an innovation that's enabled by the company's new 'Zen 4c' efficiency cores. These new cores are optimized for density through several techniques, yet unlike Intel's competing efficiency cores, retain support for the chips' full feature set.

AMD is also expected to announce its Instinct MI300 accelerators. This data center APU blends a total of 13 chiplets, many of them 3D-stacked, to create a chip with twenty-four Zen 4 CPU cores fused with a CDNA 3 graphics engine and 8 stacks of HBM3. Overall the chip weighs in with 146 billion transistors, making it the largest chip AMD has pressed into production. This chip is designed to compete with Nvidia's Grace Hopper. 

Other expected announcements include the debut of the company's Genoa-X processors, which use 3D-Stacked L3 cache to boost performance in technical workloads, much like the existing Milan-X processors. We also expect news about the company's first teleco-optimized chips, Sienna, and perhaps an update on the company's next-gen Zen 5 'Turin' data center chips.

We're now seated and ready for the show to begin in less than ten minutes. 

AMD CEO Lisa Su has come on stage to introduce the company's new products, noting that she will introduce a range of new products including CPUs and GPUs. 

Lisa Su is outlining AMD's progress with its EPYC processors, particularly in the cloud with instances available worldwide. 

Lisa Su touts that AMD EPYC Genoa offers 1.8x the performance of Intel's competing processors in cloud workloads, and 1.9X faster in enterprise workloads. 

The vast majority of AI runs on CPUs, and AMD says it has a commanding lead in performance over competing Xeon 8490H, offering 1.9X more performance. Su also touted a 1.9X efficiency advantage.

Here we can see AMD's AI benchmarks relative to Intel's Sapphire Rapids Xeon.

Dave Brown, the VP of AWS's EC2, came on stage to talk about the cost savings and performance advantages of using AMD's instances in its cloud. He provided several examples of customers that benefited from the AMD instances, with workloads spanning from HPC to standard general-purpose workloads. 

Amazon announced that it is building new instances with AWS Nitro and the fourth-generation EPYC Genoa processors. The EC2 M7a instances are available in preview today, offering 50% more performance than M6a instances. AWS says they offer the highest performance of the AWS x86 offerings. 

AMD will also use the EC2 M7a instances for its own internal workloads as well, including for chip-designing EDA software.  

AMD also announced that Oracle with have Genoa E5 instances available in July.

Lisa Su has now transitioned to talking about cloud-native processors, explaining that they are throughput-oriented and require the highest end density and efficiency. Bergamo is the entry for this market, and uses up to 128 cores per socket with a consistent x86 ISA support. The chip has 83 billion transistors and offers the highest vCPU density available. 

The Zen 4c core offers higher density than standard Zen 4 cores, yet maintains 100% software compatibility. AMD optimized the cache hierarchy, among other trimmings, for a savings of 35% on the die area. The CCD core chiplet is the only change. 

Here is the die breakdown. 

The core is 35% smaller than standard Zen 4 cores. 

Here is a diagram of the chip package. 

Bergamo is shipping now to AMD's cloud customers. AMD also shared the following performance benchmarks. 

A Meta representative joined Lisa Su on the stage to talk about the company's use of AMD's EPYC processors for its infrastructure. Meta is also open-sourcing its AMD-powered server designs. 

Meta says that it has learned that it can rely upon AMD for both chip supply and a strong roadmap that it delivers on schedule. Meta plans to use Bergamo, which offers 2.5X more performance than the previous-gen Milan chips, for its infrastructure. Meta will also use Bergamo for its storage platforms. 

Dan McNamara, AMD's SVP and GM of the Server Business Unit, has come to the stage to introduce two new products. Genoa-X will add more than 1 GB of L3 cache with 96 cores. 

Gen0a-X is available now. Four SKUs, 16 to 96 cores. SP5 socket compatibility, so it will work with existing EPYC platforms. 

McNamara showed performance benchmarks of Genoa-X against Intel's 80 core Xeon. 

Here we can see a comparison of Genoa-X against an Intel Xeon with the same number of cores. 

A Microsoft representative joined McNamara on the stage to show Azure HPC performance benchmarks. In just four years, Azure has seen a 4X improvement in performance with the EPYC processors. 

Azure announced the general availability of its new HBv4 and HX-series instances with Genoa-X, and new HBv3 instances. Azure also provided benchmarks to show the performance gains, which top out at 5.7X gains 

AMD's Sienna is optimized for Telco and Edge workloads but comes to market in the second half of the year. 

AMD's Forrest Norrod, MD's executive vice president and general manager of the Data Center Solutions Business Group, has come to the stage to share information about how the data center is evolving. 

Citadel Securities has joined Norrod on the stage to talk about their shift in workloads to AMD's processors, powering a 35% increase in performance. They use over a million concurrent AMD cores. 

Citadel also uses AMD's Xilinx FPGAs for its work in financial markets with its high frequency trading platform. It also uses AMD's low-latency solarflare networking.

AMD purchased Pensando to acquire DPU technology. Norrod explained how AMD is using these devices to reduce networking overhead in the data center. 

AMD's P4 DPU offloads networking overhead and improves server manageability. 

AMD's Pensando SmartNICs are an integral part of the new data center architectures. 

The next step? Integrating P4 DPU offload into the network switch itself, thus providing services at the rack level. This comes as the Smart Switch they've developed with Aruba Networks. 

Lisa Su has come back to the stage to talk about AMD's broad AI silicon portfolio, including the Instinct MI300

Lisa Su outlined the massive market opportunity for the AI market driven by large language models (LLMs), causing the TAM to grow to around $150 billion. 

AMD Instinct GPUs are already powering many of the world's fastest supercomputers. 

AMD President Victor Peng came to the stage to talk about the company's efforts around developing the software ecosystem. That's an important facet, as Nvidia's CUDA software has proven to be a moat. AMD plans to use an 'Open, Proven, and Ready' philosophy for its AI software ecosystem development, which Peng is in charge of. 

Peng showed some of AMD's latest hardware efforts. 

AMD's ROCm is a complete set of libraries and tools for its optimized AI software stack. Unlike the proprietary CUDA, this is an open platform.

AMD is continually optimizing the ROCm suite. 

PyTorch is one of the most popular AI frameworks in the industry, and they've joined Peng on the stage to talk about their collaboration with ROCm. The new PyTorch 2.0 is nearly twice as fast as the previous version. AMD is one of the founding members of the PyTorch Foundation. 

Here are details of PyTorch 2.0. 

AMD is shifting to talking about AI models, with Hugging Face joining Peng on the stage. AMD and Hugging Face announced a new partnership, optimizing their models for AMD CPUs, GPUs, and other AI hardware. 

Lisa Su has returned to the stage, and now we expect to learn about the biggest announcement of the show: The Instinct MI300. This is for training larger models, like LLMs behind the current AI revolution. 

SU is talking about the Instinct roadmap, and how the company previewed the MI300 with the CDNA 3 GPU architecture paired with 24 Zen 4 CPU cores, tied to 128GB of HBM3. This gives 8x more performance and 5x higher efficiency than the MI250. 

146 billion transistors across 13 chiplets. 

There will be a GPU-only MI300, the MI300X. This chip is optimized for LLMs. this delivers 192GB of HBM3, 5.2 TB/s of bandwidth, and 896 GB/s of Infinity Fabric Bandwidth. 

And here's new chip. 153 billion transistors all in one package with 12 5nm chiplets. 

MI300X offers 2.4X HBM density than the Nvidia H100 and 1.6X HBM bandwidth than the H100, meaning that AMD can run larger models than Nvidia's chips. 

Lisa Su conducted a demo of the MI300X running a Hugging Face AI model. The LLM wrote a poem about San Francisco, where the event is taking place. This is the first time a model this large has been run on a single GPU. A single MI300X can run a model up to 80 billion parameters. 

This allows fewer GPUs for large language models, thus delivering cost savings. 

SU also announced the AMD Instinct Platform, which has 8 MI300X in an industry-standard OCP design, offering a total of 1.5TB of HBM3 memory. 

MI300A, the CPU+GPU model, is sampling now. The MI300X and 8-GPU Instinct Platform will sample in the third quarter, and launch in the fourth quarter. 

Lisa Su wrapped up the presentation. Here's a few more wrap up slides. Stay tuned for our ongoing coverage over the coming hours. 

The event has concluded, and you can see our overview with the live blog below, However, here are links to our deeper coverage of each topic:

AMD Expands MI300 With GPU-Only Model, Eight-GPU Platform with 1.5TB of HBM3

AMD EPYC Genoa-X Weilds 1.3 GB of L3 Cache, 96 Cores

AMD Details EPYC Bergamo CPUs With 128 Zen 4C Cores, Available Now 

AMD Intros Ryzen 7000 Pro Mobile and Desktop Chips, AI Comes to Pro Series

Cloud computing has taken a step further with the introduction of Microsoft’s Windows 365 Boot. Early access to the new platform will enable users to test the ability to boot into an Azure-based cloud PC interface. Instead of accessing their Windows account locally from the hard drive, users can access cloud-based desktop profiles enabling multiple users to access their accounts from one machine.

According to the announcement, the new cloud-based interface is accessible via Windows 11 version 22H2 for certain users. The idea is to enable the cloud PC login screen initially when the machine boots so users will have the option of accessing their cloud-based profile immediately, without having to launch a separate application.

Microsoft explains that no additional steps will be necessary to access the Windows 365 Cloud PC. Once the user logs into their account, the virtual desktop will be immediately accessible. This environment will likely be most beneficial to offices and other corporate use cases in which users will need PC access from multiple locations.

The new system is not yet available to everyone and it’s not clear when it will be made widely available. There are a few requirements for early access: users must be a part of the Windows Insider Program with Microsoft Intune Administrator rights. Those interested in testing out the new cloud-based login interface will also need a Windows 365 Cloud PC license.

Microsoft has provided thorough details about how to enable Windows 365 Boot as well as what the login process will be like once implemented. When Windows 365 Boot is enabled, the lock screen on the PC will be a default Windows 11 screen. Users will be prompted to log into their Microsoft account. Once authenticated, the PC will load the Cloud PC. When the user logs out, they or another user can log into their Microsoft account to access their Cloud PC.

You can read more about how to access the new Windows 365 Boot application on the Microsoft website where the announcement was unveiled.

Microsoft's Build developer conference is all about AI. The 2023 show, which is taking place both online and in person, is being headlined by AI, Bing Chat, and new tools for developers using Windows, GitHub, and the Windows Terminal. Every change, big or small, is being touched by artificial intelligence.

There's plenty of developments for the cloud, like AI in Azure, as well as a new analytics system that Microsoft is calling Fabric. But here are the AI announcements Microsoft is making at Build that you're most likely to see coming to tools you use soon:

AI Plugins

At Build, Microsoft said that it will use the same plugin standard as its partners at OpenAI use for ChatGPT. This should allow developers to make plugins that are interoperable between ChatGPT, Bing Chat, Dynamics 365 Copilot, and Microsoft 365 Copilot.

Microsoft is also announcing support for new plugins for Bing Chat. Newly announced add-ons include those for Instacart, Kayak, Redfin, Zillow, Klarna, and many others, alongside OpenTable and WolframAlpha. Communications head Frank Shaw told the press that as part of Microsoft 365 Copilot, there will be more than 50 plugins announced at Build, including from Atlassian, Adobe, Thomas Reuters and Mural, with "thousands" of plugins by time Copilot it hits general availability.

Copilot coming to Edge, Windows 11

Microsoft is bringing some of its Copilot features (which are currently in a private preview in MIcrosoft 365) to the Edge browser.

Microsoft 365 Copilot already uses large language models to answer requests from Microsoft 365 apps such as Excel, Word, Teams, Outlook, and PowerPoint. Now, the Copilot will also work with Edge, so people will be able to use the AI in conjunction with their web browser. Perhaps they'll get a PDF that opens in the browser, and users can ask Edge to summarize it.

Windows 11 will use Bing Chat along with third-party plugins and Microsoft's own. Windows Copilot will let people change settings, open apps or perform other actions with AI, while also mixing in Bing Chat to answer questions from the web. It will be able to be parked aside in a panel for easy access.

Media Provenance

At Google I/O, Google discussed features that would secretly mark images and videos made by AI so they could be identified as such. At Build, Microsoft is following suit with "media provenance" abilities being added to Bing Image Creator and Microsoft Designer.

Microsoft's tech will use cryptographic techniques to "mark and sign AI-generated content with metadata about its origin." The company says that this signifier will be in accordance with  standards set by the Coalition for content Provenance and Authenticity (C2PA), of which Microsoft is a co-founder. Microsoft says it will start by supporting "major images and video formats" when the initiative ramps up in the "coming months."

Microsoft Store Gets an AI Hub

The Microsoft Store will be filled to the brim with AI. The Microsoft Store will get a section specifically for AI apps and experiences. Developers can submit their AI-focused apps to be featured.

 AI is also coming to review summaries. Reviews in the app store will be scanned by AI, which will compile them and recapitulate. Additionally, developers will be able to fill their listings with AI-generated keyword suggestions to improve search engine optimization.

Dev Home

Beyond its AI tools, Microsoft also doubled down on Windows as a tool for developers with a new feature called Dev Home. Available in preview, it lets users set up a machine, connect to GitHub and build dashboards and other extensions, as well as use GitHub widgets and track CPU and GPU performance.

Microsoft is also adding WinGet configuration to the Windows Package Manager, taking care of setup and, according to Microsoft, dropping setup time from days to just hours. WinGet's configuration file will ensure you get the right version of software packages and frameworks.

There's also a new storage volume, Dev Drive, which Mifcrosoft says is based on the Resilient File System (ReFS) and offers a 30% improvement in build times, partially due to a "performance mode" in Windows Defender. Microsoft hasn't said much about this new mode, other than that  this "new performance mode is more secure for developer workloads than folder or process exclusions, providing a solution that balances security with performance.  

But of course, dev tools are also getting the AI treatment. Windows Terminal is gaining support for GitHub Copilot X, and will be able to use an "experimental chat experience" to use language both inline and in chat to learn more about errors, get recommendations and to perform actions throughout Terminal. Microsoft says it also wants to add GitHub Copilot-style AI in other developer tools, such as WinDBG.

We'll likely learn more about all of these features during the Microsoft Build keynotes. CEO Satya Nadella will start the show today at 9:00 a.m. PT / 12 p.m. ET. 

Intel today announced its Agilex 7 M-Series family of Field-Programmable Gate Array (FPGA) products, doubling down on its revitalized 10nm SuperFin fabrication node. Citing rising market demand for FPGA solutions as co-processors for Networking, Data Center, High Performance Computing (HPC)Cloud computing and other applications, Intel promises increased flexibility (mainly due to FPGA's inherently programmable nature) and higher scaling capabilities than ever before. Intel's Agilex 7 FPGAs introduce a new chiplet as part of their heterogeneous multi-die architecture, the R-Tile, which is responsible for delivering the latest connectivity technologies - namely PCIe 5.0 and CXL support - in hardware-accelerated, hard-coded IP blocks.

The new heterogeneous R-Tile chiplet is the star of the show for Intel, allowing it to claim the title of the only FPGA family that carries certification for the full PCI-SIG 5.0 x16 data rate. Xilinx, which is now folded into AMD, would be an example of another leading-edge FPGA developer, so there's the feeling of a race won for Intel here. 

Interestingly, Intel seems to be doubling down on a separation between FPGA and CPU products. Perhaps that's a result of its earlier inroads into integration, of which only silicon ghost stories remain. AMD seems to be much more confident it's cracked the case: the company is seemingly looking to incorporate FPGA capabilities into the company's EPYC CPUs as early as this year. Now, that doesn't cement a departure between both companies - AMD may be also looking at a chiplet-like integration, whether vertically via 3D stacking or by incorporating segregated FGPA-specific IP in a separate chip.

The general idea of FPGAs is their inherent flexibility, which allows developers to quickly iterate on circuit arrangement and processing blocks to adapt the FPGA for the specific workloads at hand. Being more specialized hardware, FPGAs can be used to accelerate workloads that aren't CPU-dependent, thus freeing up valuable CPU resources for their specific tasks (such as spinning out VMs in a cloud-based installation) instead of tapping into their lower power efficiency (the cost of their generalized processing capabilities).

Intel's R-Tile essentially brings the hardware-accelerated IP blocks responsible for handling the PCIe 5.0 and CXL 1.1/2.0 protocols into its Agilex 7 FPGA family. That  should bring about significant improvements in power efficiency and data throughput, which are important elements to lower Total Cost of Ownership (TCO) for high-performance installations. But with these choices, there are always trade-offs: Intel is adding yet another fixed-function hardware block onto a product whose desirability lies in it being programmable. Programmable die area is the FPGA buyer's mantra, after all.

It also bears mentioning that as a product, R-Tile is specifically saying "I'm here to reduce the load on your CPUs while delivering increased performance". But another answer to that need isn't about moving CPU functions towards an FPGA; it's to merely increase the number of available CPU resources. And that could be done with more CPUs (which might make sense to some installers) or additional CPU cores. Intel's Agilex 7 M-Series is specifically marketed to Intel's 4th Gen Scalable Xeons, after all, and those aren't the core-count kings in any shape.

Intel's bet is that the questions above have an answer, and that it knows the answer: that's why it introduced Agilex 7. Intel's answer is that consumers want to remove CPU overhead by moving it onto an FPGA packaging. They want the best performance/watt possible (one of the highest contributors to high TCO cost), so they'll move things to where they're fastest. Luckily, this move benefits Intel in other ways. For the company, it's a matter of its own efficiency - and as such, a matter of cost. 

That's where Intel's embedded multi-die interconnect bridge (EMIB) really shines. As the proverbial "glue" holding the disparate processing blocks together, EMIB allows Intel to further separate IP blocks at the manufacturing level, increasing die efficiency and reducing the overall cost-per-wafer (and resulting cost-per-chip) equation. 

For the consumer, it also theoretically lowers costs: Intel (and the industry's) dream is to be able to mix and match different hardware IP blocks (from the same vendor, or even multiple vendors and manufacturing processes), meaning that customers need only pay for the silicon they'll actually use, with the specs they'll actually use. In a sense, this would make every chip an FPGA.

With all that in mind, it's possible today's Agilex 7 with R-Tile is as much a new FPGA product as it is a new staple on Intel Foundry Services' (IFS) catalog. Either way, it moves Intel in the direction it wants - and needs - to go. And that's just good business.

Ampere this week introduced its AmpereOne processors for cloud datacenters that happen to be the industry's first general-purpose CPUs with up to 132 that can be used for AI inference.

The new chips consume more power than their predecessors — Ampere Altra (which will remain in Ampere's stable for at least a while) — but the company claims that despite of higher power consumption, its processors with up to 192 cores provide higher computational density than CPUs from AMD and Intel. Some of those performance claims can be controversial.

192 Custom Cloud Native Cores

Ampere's AmpereOne processors features 136 ­– 192 cores (as opposed to 32 to 128 cores for Ampere Altra) running at up to 3.0 GHz that are based on the company's proprietary implementation of the Armv8.6+ instruction set architecture (featuring two 128-bit vector units that support FP16, BF16, INT16, and INT8 formats) that are equipped with a 2MB of 8-way set associativity L2 cache per core (up from 1MB) and are interconnected using a mech network with 64 home nodes and directory-based snoop filter. In addition to L1 and L2 caches, the SoC also has a 64MB system level cache. The new CPUs are rated for 200W – 350W depending on exact SKU, up from 40W – 180W for the Ampere Altra. 

The company claims that its new cores are further optimized for cloud and AI workloads and feature 'power and are efficient' instructions per clock (IPC) gains, which probably means higher IPC (compared to Arm's Neoverse N1 used for Altra) without a tangible increase in power consumption and die area. Speaking of die area, Ampere does not disclose it, but says that the AmpereOne is made on one of TSMC's 5nm-class process technology. 

Although Ampere does not reveal all the details about its AmpereOne core, it says that they feature a highly accurate L1 data prefetcher (reduces latency, ensures that the CPU spends less time waiting for data, and reduces system power consumption by minimizing memory accesses), refined branch misprediction recovery (the sooner the CPU can detect a branch misprediction and recover, it will reduce latency, and will waste less power), and sophisticated memory disambiguation (increases IPC, minimizes pipeline stalls, maximizes out-of-order execution, lowers latency, and improves handling of multiple read/write requests in virtualized environments). 

While the list of AmpereOne core architecture improvements does not seem too long on paper, these things can indeed improve performance significantly and they required a lot of research to be made (i.e., which things slowdown performance of a cloud datacenter CPU the most?) and a lot of work to implement them efficiently.

Advanced Security and I/O

Since the AmpereOne SoC is aimed at cloud datacenters, it is equipped with appropriate I/O, which includes eight DDR5 channels for up to 16 modules supporting up to 8TB of memory per socket, 128 lanes of PCIe Gen5 with 32 controllers and x4 bifurcation.

Datacenters also require certain reliability, availability, serviceability (RAS), and security features. To that end, the SoC fully supports ECC memory, single key memory encryption, memory tagging, secure virtualization, and nested virtualization, just to name a few of them. In addition, AmpereOne has numerous security capabilities like crypto and entropy accelerators, speculative side channel attack mitigation, ROP/JOP attack mitigation, and so on.

Curious Benchmark Results

Without any doubts, Ampere's AmpereOne SoC is an impressive piece of silicon designed to handle cloud workloads and featuring 192 general-purpose cores, the industry's first. Yet, to prove its points, Ampere uses rather curious benchmark results.

Ampere sees compute density of its AmpereOne as its main advantage. The company claims that one 42U 16.5kW rack filled with 192-core AmpereOne SoC-based 1S machines can support up to 7926 virtual machines, whereas a rack based on AMD's 96-core EPYC 9654 'Genoa' can handle 2496 VMs and a rack powered by Intel's 56-core Xeon Scalable 8480+ 'Sapphire Rapids' CPUs can handle 1680 VMs. This comparison makes a lot of sense in the 16.5kW power budget. 

But 42U rack power density is rising and exascalers like AWS, Google, and Microsoft are ready for this particularly for their performance-demanding workloads. Based on a survey from UpTimeInstitute in 2020, we can say that 16% of companies deployed typical 42U rakcs with rack power density from 20kW to over 50kW. By now, the number of typical deployments with 20kW racks has increased, not decreased, as the latest and previous-generation CPUs from AMD increased their TDPs compared to their predecessors.

When it comes to performance, Ampere demonstrates the advantages of its 160-core AmpereOne-based system with 512GB of memory running Generative AI (stable diffusion) and AI Recommenders (DLRM) against systems based on AMD's 96-core EPYC 9654 CPU with 256GB of memory (meaning that it worked in an eight-channel mode, not 12-channel mode that is supported by Genoa). Ampere-based machines produced 2.3X more frame/s for generative AI and over 2X more queries/s for AI recommendations. 

In this case Ampere compared performance of its systems crunching data with an FP16 precision, whereas AMD-based machines computed with an FP32 precision, which is not an apples-to-apples comparison. Furthermore, many FP16 workloads are now run on GPUs rather than on CPUs and massively-parallel GPUs tend to offer spectacular results with generative AI and AI recommendations workloads.

Summary

Ampere's AmpereOne are the industry-first general-purpose CPUs with up to 192 cores, which certainly deserves a lot of respect. These CPUs also feature robust I/O capabilities, advanced security features, and promise improved instructions per clock (IPC) gains. Also they can run AI workloads with FP16, BF16, FP8, and INT8 precision. 

But the company chose to use rather controversial methods to prove its points when it comes to benchmark results, which casts some shadow on its achievements. That said, it will be particularly interesting to see independent test results of AmpereOne-based servers.

Google kicked off Google I/O this afternoon by talking for more than an hour about its numerous advances in artificial intelligence.  The company discussed its new PaLM 2 large language model (LLM) for generative AI, which powers the Bard chatbot tool. This is a foundational pillar for adding AI-infused features across Google's product portfolio, including Google Maps, Google Photos, and Gmail (among others).

With that in mind, there is a need for some serious horsepower in the cloud to power models in the wild, as millions (and eventually billions) of users send requests for operations as mundane as removing a person lingering in the background of a picture to composing an entire email for you based on a short text prompt. That's where Google's new A3 GPU supercomputer comes into focus. Google says the new A3 supercomputers are "purpose-built to train and serve the most demanding AI models that power today's generative AI and large language model innovation" while delivering 26 exaFlops of AI performance.

Each A3 supercomputer is packed with 4th generation Intel Xeon Scalable processors backed by 2TB of DDR5-4800 memory. But the real "brains" of the operation come from the eight Nvidia H100 "Hopper" GPUs, which have access to 3.6 TBps of bisectional bandwidth by leveraging NVLink 4.0 and NVSwitch.

According to Google, A3 represents the first production-level deployment of its GPU-to-GPU data interface, which allows for sharing data at 200 Gbps while bypassing the host CPU. This interface, which Google calls the Infrastructure Processing Unit (IPU), results in a 10x uplift in available network bandwidth for A3 virtual machines (VM) compared to A2 VMs.

"Google Cloud's A3 VMs, powered by next-generation NVIDIA H100 GPUs, will accelerate training and serving of generative AI applications," said Ian Buck, VP for hyperscale and high-performance computing at NVIDIA. "On the heels of Google Cloud's recently launched G2 instances, we're proud to continue our work with Google Cloud to help transform enterprises around the world with purpose-built AI infrastructure." 

If your business wants to leverage A3 virtual machines, the only way to gain access is by filling out Google's A3 Preview Interest Form to join the Early Access Program. But as Google clearly states, plugging in your information doesn't guarantee a spot in the program.

In March, Western Digital identified a network security incident in which both customer and company data were compromised. Western Digital notified those affected so they could take action to secure their private data as much as possible. Since this security breach, certain online functions have been halted. This week we’ve received an update on the situation along with plans to restore online activity.

According to Western Digital, an unknown third party accessed a Western Digital database containing customer information, including not only their names, addresses, and phone numbers but also billing details such as credit card information, although the numbers were encrypted. The customers impacted by this breach were contacted by Western Digital immediately so they could take action to help protect their private information.

A reader submitted to us a copy of the notification sent from Western Digital pertaining to the matter.

Within a few days of the incident, on April 2nd, Western Digital made a public announcement about the matter and initiated a response to rectify the situation. This involved a plan to completely shut down certain services from public internet access, including the online web store. Many services have since been restored, but the online store is still shut down and cannot be used to process orders.

Despite the ongoing effort to limit online activity, Western Digital says that physical factories have been operating the entire time to help deliver products. The only impact has been for online services. So while new orders currently cannot be placed, existing orders can still be processed and delivered.

Western Digital confirmed that My Cloud services were completely restored by April 13th. The online store, however, is still shut down. However, users can expect account access to be restored by the week of May 15th. In the meantime, online orders will not be processed to ensure customer data can be appropriately protected.

If you want to read more about the incident, its impact, and Western Digital’s response, you can find the full press release on the official Western Digital website.

Microsoft is working on its own artificial intelligence chip, internally called "Athena," and has been working on the project since 2019, according to a report from The Information. The initial version of the chip is planned to use Taiwan Semiconductor's (TSMC) 5-nanometer process, though there will likely be multiple generations of chips as part of the project.

The Information suggests that Microsoft's goal is at least partly to save money on buying chips from suppliers like Nvidia.

Microsoft has poured billions into OpenAI, which makes ChatGPT, to advance in a fast-growing race to make products using artificial intelligence and large-language models (LLMs). The Athena chips are reportedly designed to train LLMs and similar software, including inferring from data the models acquire in training.

The chips, according to The Information's sources, are currently being used among a small subset of employees at Microsoft and OpenAI. Microsoft could potentially make the chips more widely available "as early as next year," but it's unclear whether the chip will be opened up to Azure customers.

Many who are training LLMs, whether locally or in the cloud, have defaulted to Nvidia's offerings, including powerful graphics cards that have proven to be the go-to for any company without its own chips. Dylan Patel of the resarch firm SemiAnalysis told The Information that "Athena, if competitive, could reduce the cost per chip by a third when compared with Nvidia’s offerings." If Microsoft is going to add AI capabilities beyond Bing Chat to its Office products and GitHub, it could save a ton.

Microsoft isn't the first to make its own chips, of course. Google, Amazon and Meta are already among the tech behemoths that have invested in in-house silicon.

That's not to say Microsoft sees Athena as replacing Nvidia entirely. In March, the two companies partnered on Nvidia's DGX Cloud specifically for AI supercomputing and the Nvidia Omniverse cloud, both using Azure and support for Microsoft 365. Though the report suggests Athena could be a "touchy subject," Microsoft reportedly doesn't see its chip as a "broad replacement for Nvidia's products."

If Microsoft more widely produces the Athena chips next year, it will no longer be on TSMC's cutting-edge nodes. But it's no surprise to see that Microsoft, with its multi-billion investment in OpenAI and early (occasionally reckless) entry into chat bots in Bing Search and beyond, is looking for ways to further development.

Elon Musk has confirmed that his companies Tesla and Twitter were buying tons of GPUs when asked to confirm whether he was building up Twitters compute prowess to develop a generative artificial intelligence project. Meanwhile, the Financial Times reports that Musk's AI venture will be a separate entity from his other companies, but it could use Twitter content for training.

Elon Musk's AI project, which he began exploring earlier this year, is reportedly separate from his other companies, but could potentially use Twitter content as data to train its language model and tap into Tesla's computing resources, according to Financial Times. This somewhat contradicts the earlier report which claimed that the AI project would be a part of Twitter. 

To build up the new project, Musk is recruiting engineers from top AI companies, including DeepMind, and has already brought on Igor Babuschkin from DeepMind and approximately half a dozen of other AI specialists.

Musk is also reportedly negotiating with various SpaceX and Tesla investors about the possibility of funding his latest AI endeavor, according to an individual with firsthand knowledged about the talks, which may confirm that the project is not set to be a part of Twitter.

In a recent Twitter Spaces interview, Musk was asked about a report claiming that Twitter had procured approximately 10,000 of Nvidia compute GPUs. Musk acknowledged this stating that everyone, including Tesla and Twitter, are buying GPUs for compute and AI these days. This is true as both Microsoft and Oracle have acquired tens of thousands of Nvidia's A100 and H100 GPUs in the recent quarters for their AI and cloud services. 

"It seems like everyone and their dog is buying GPUs at this point," Musk said. "Twitter and Tesla are certainly buying GPUs."

Nvidia's latest H100 GPUs for AI and high-performance computing (HPC) are quite expensive. CDW sells Nvidia's H100 PCIe card with 80GB of HBM2e memory for as much as $30,603 per unit. On Ebay, these things sell for over $40,000 per unit if one wants this product fast. 

Recently Nvidia launched its even more powerful H100 NVL product that bridges two H100 PCIe cards with 96GB of HBM3 memory on each one for an ultimate dual-GPU 188GB solution designed specifically for training of large language models. This product will certainly cost well above $30,000 per unit, though it is unclear at which price Nvidia sells such units to customers buying tens of thousands of boards for their LLM projects.

Meanwhile, the exact position of the AI team in Musk's corporate empire remains unclear. The renowned entrepreneur established a company called X.AI on March 9th, Financial Times reported citing business records from Nevada. Meanwhile, he recently changed the name of Twitter in the company's records to X Corp., which may be a part of his plot to build an 'everything app' under the 'X' brand. Musk is currently the sole director of X.AI, while Jared Birchall, who happens to manage Musk's wealth, is listed as its secretary. 

The rapid progress of OpenAI's ChatGPT, which Elon Musk co-founded in 2015 but no longer is involved with, reportedly inspired him to explore the idea of a rival company. Meanwhile, this new AI venture is expected to be a separate entity from his other companies possibly to ensure that this new project will not be limited by Tesla's or Twitter's frameworks.

Sending files to and from an Android device can be irksome at times. Sure we can use Bluetooth file transfer, which is slow, but reliable. Or we can email the files to ourselves, save them to a cloud service etc. Is there a better way to do it?

Android has just released a beta for Nearby Share, a Windows app that enables quick and easy file transfer between Android and Windows devices. The best thing is that we only need to install the Windows app. Android has featured Nearby Share since Android 6 and it seamlessly integrates into the Share feature found in many apps.

In this how to, we will install Nearby Share on a Windows 10 PC and learn how to use it to send files to and from an Android smartphone.

For This Project you will need

• Android phone connected to Wi-Fi with Bluetooth active
• PC running Windows 10 or 11

1.  Open a browser to this page and click “Download Nearby Share”.

2. Install Nearby Share. You will find it in the download folder where you double click the executable called “BetterTogetherSetup.exe”. Installation took mere seconds on our machine.

3. Sign in to your Google account.

4. Set the name of your PC, this is the name that will be shown to devices.

5. Click on Receiving.

6. Configure Receiving to your preferred option. If you are just wanting to share files between your devices, select “Your Devices”. Other choices include sharing to your contact list and everyone. Both of these have security implications, so exercise caution.

7. Click Done to finish setup.

Nearby Share is now ready to send and receive files.

8. Open Settings >> System and search for Nearby. We found these extra steps useful in ensuring that our Windows PC was correctly talking to our Android device.

9. Enable “Share content with a nearby device by using Bluetooth and Wi-Fi”, set it to “My devices only” and check that the download location is convenient.

10. Scroll down and configure share across devices to use your devices.

Sending a File to Another Device

1.  Drop a file or folder onto the Nearby Share app. Alternatively, manually select the files / folder via a dialog.

2. Select your device. This will trigger the computer to send the file / folder to the device.

In a few moments your phone will confirm the transfer and the file is ready for use on your Android device.

Sending to a Windows Machine

1. Select the file that you want to send on your Android device. This can be a photo via the Photos app, or any other file via the Files app. We chose a photo to send to our PC.

2. Click Share and select Nearby Share. You may need to click on More to see a fuller list of options.

3. Connect to Wi-Fi and select your Windows PC to start the transfer.

4. Click Done once the transfer is complete.

The Nearby Share app will confirm the transfer, saved to your default folder. In our case it saved the image to Downloads.

Loongson, a Chinese fabless chipmaker, has launched the new 3D5000 processor for data centers and cloud computing. MyDrivers reported that Loongson claims its 32-core domestic chips deliver 4X higher performance than rival Arm processors.

The 3D5000 still leverages LoongArch, Loongson's homemade instruction set architecture (ISA) from 2020. The chipmaker was previously a firm believer in MIPS. However, Loongson eventually built LoongArch from the ground up with the sole objective of not relying on foreign technology to develop its processors. LoongArch is a RISC (reduced instruction set computer) ISA, similar to MIPS or RISC-V.

The 3D5000 arrives with 32 LA464 cores running at 2 GHz. The 32-core processor has 64MB of L3 cache, supports eight-channel DDR4-3200 ECC memory, and up to five HyperTransport (HT) 3.0 interfaces. It also supports dynamic frequency and voltage adjustments. Officially, the 3D5000 has a 300W TDP; however, Loongson stated that the conventional power consumption is around 150W. That's roughly 5W per core.

The 3D5000 flaunts a chiplet design since Loongson has glued together two 16-core 3C5000 processors. Loongson developed the 3C5000 server part to compete with AMD's Zen and Zen+ architectures. The latest 3D5000, which measures 75.4 x 58.5 x 7.1mm,  slides into a custom LGA4129 socket. 

The processor supports 2P and 4P configurations; therefore, Loongson has launched the 7A2000 bridge chip to manage the communication between the processors and other components. As per the chip designer, the 7A2000 is up to 400% faster than the previous generation. Furthermore, with the help of the 7A2000, there's a possibility to scale up to 128 cores per motherboard.

According to Loongson's provided numbers, the 3D5000 scores over 425 points in SPEC CPU 2006, a depreciated benchmark replaced with the newer SPEC CPU 2017 version. The 3D5000 also delivers over 1 TFLOPs of FP64 performance, up to 4X higher than regular Arm cores. Meanwhile, the processor's stream performance with eight channels of DDR4-3200 memory crosses the 50GB mark.

While performance isn't the 3D5000's strong suit, security is. The 32-core processor allegedly has a custom-made mechanism to defend against vulnerabilities such as Meltdown or Spectre. The chip also has its Trusted Platform Module (TPM), so it doesn't rely on an external solution. In addition, according to MyDrivers' report, the 3D5000 also supports a secret national algorithm with an embedded security module that seemingly delivers excellent encryption and decryption efficiency higher than 5 Gbps.

In addition to the 3D5000 and 7A2000, Loongson also announced the 2K050, the company's baseboard management controller (BMC). The 2K050 features LA264 cores at 500 MHz, integrated 2D GDP, 32-bit DDR3 support, and outputs at a 1080p (1920x1080) resolution at 60 Hz.

Loongson's 3D5000 is no match for AMD's EPYC Genoa or Intel's Sapphire Rapids Xeon processors. It was never about beating the foreign competition but pushing for self-sufficiency. Unfortunately, with the ongoing U.S. sanctions, Chinese companies have no means to secure chipmaking tools originating from the U.S. In addition, the U.S. Department of Commerce recently blacklisted Loongson, which likely derailed some of the company's plans.

Ampere has quietly launched its Altra developers kit aimed at software creators for cloud data centers. Along with Dev Kit featuring the company's system-on-chips with up to 80 cores, the Ampere also offers a pre-built workstation running its 128-core SoC, according to Joe Speed, the company’s edge computing chief. An unexpected twist is that the workstation can run Windows and even has driver support for Nvidia's GeForce RTX graphics cards. 

The Ampere Altra Developer Platform (AADP) is a prototyping system for general embedded applications, but it can obviously be used for building software for the cloud. The machine can use a variety of add-in boards, including Nvidia's GeForce RTX cards. What is a bit surprising is that it can run Windows, making it perhaps the most powerful Arm-based machine that runs the consumer-oriented Microsoft operating system. 

Given AADP's Windows support, nothing stops software makers from trying to port their performance applications to high-performance multi-core Arm hardware and see what kind of performance they can get out of Ampere's SoC. 

The AADP uses Ampere's Altra Dev Kit, which includes a COM-HPC module powered by an Arm Neoverse N1-based architecture. The module has 32/64/80 Arm v8.2 64-bit cores that operate at frequencies of up to 1.70 GHz, 2.20 GHz, and 2.60 GHz. The motherboard supports up to 768 GB of DR4 memory and fully exploits the I/O capabilities of Ampere's 128-core SoC, which offers three x16 and two x4 PCIe slots as well as a couple of M.2 slots for SSDs.  

When it comes to supporting Nvidia's GeForce RTX consumer graphics cards, this is something unexpected as Ampere's SoCs are aimed primarily at cloud and edge computing applications, whereas Nvidia's GeForce RTX is designed primarily for gamers and content creators (when used with Studio drivers). Yet, the support does not come as completely unsurprised as, at the end of the day, Ampere and Nvidia work together on the data center front in general and the gaming data center part of it in particular. 

Anyhow, for now, Ampere's AADP workstation is available for orders starting at $3,250 for a 32-core version with 32 GB of DDR4 and $5,658 for a 128-core version with 128 GB of memory. Meanwhile, the devkit itself starts at $2,003.

Out of the box, Windows 11 is a resource-hungry operating system that gulps up tons of RAM and CPU cycles, leaving older or underpowered systems unable to run it. The official system requirements demand not only TPM security and secure boot, but also at least 4GB of RAM, a dual-core processor and a 64GB SSD.  But, with the release of Tiny11 Builder, you can make a lightweight Windows 11 install that gets around these requirements and runs well on PCs with limited memory and processing power. 

Tiny11, a custom image of Windows 11 from Twitter user NTDEV solves these issues by removing unnecessary libraries, apps and services so that the system can run with just 2GB of RAM, take up a mere 12GB of disk space and not check for any of the system requirements. However, up until now, to use Tiny11, you had to download the entire operating system from NTDEV's cloud storage drive, which meant you didn't have a choice of Windows 11 version and it was awkward downloading the whole OS from someone other than Microsoft.

Now, NTDEV has released Tiny11 Builder, a free tool that lets you turn the latest Windows 11 ISO (installer file) into one which installs a light version of the OS. In just a few simple steps, you can grab Windows 11 Home, Pro, Education, Enterprise and then use Tiny11 Builder to remove the cruft from it, before you even install. NTDEV says that, right now, only the latest builds 22621.1265 and 22621.525 are officially supported. However, this may remove some extraneous stuff from pre-release builds. 

I used Tiny11 Builder to create a stripped-down build of Windows 11 Home that I installed on both a Virtual Machine and a decade-old, sluggish ThinkPad X131e. The VM was blazing fast and the ThinkPad, which was so slow under Windows 10 that letters would appear seconds after I typed them, was actually usable. 

Below, I'll show you how to create your own lightweight Windows 11 ISO using Tiny11 Builder. If you have a Raspberry Pi, also see our tutorial on how to install Tiny11 on your Raspberry Pi.

How to Make a Lightweight Windows 11 Image with Tiny11

1. Download Tiny11 Builder from Github and unzip it to its own folder.

2. Download a Windows 11 ISO from Microsoft (or using UUP Dump). For now, only builds 22621.525 and 22621.1265 are officially supported. When I downloaded the ISO from Microsoft and when I created it using the media creation tool, my build number was 22621.525. For more details on getting an ISO, see our article on how to download a Windows 11 ISO.

3.  Mount the ISO. You can do this by right clicking on it and selecting Mount. 

File Explorer should open, showing you what drive letter now represents the ISO, which will act as if it were a physical drive.

4. Find your ISO's Windows build number (if you don't know for sure). You can do this by launching an elevated command prompt and entering dism /Get-WimInfo /WimFile: [DRIVELETTER]:\sources\install.wim /index:1 where [DRIVELETTER] is replaced with the drive letter of your mounted ISO. If that doesn't work, dism /Get-WimInfo /WimFile:[DRIVELETTER]:\sources\install.esd /index:1. For example, if your ISO is mounted as drive letter F, the two commands to try would be these.

dism /Get-WimInfo /WimFile:F:\sources\install.wim /index:1dism /Get-WimInfo /WimFile:F:\sources\install.esd /index:1

The reason you might have to try both is that the information is stored either in a file called install.wim or install.esd. In my case, install.wim worked with an ISO I had downloaded directly from Microsoft while install.esd worked when I used the Media Creation Tool to build my ISO.

When the command is successful, it will output a list of information about the ISO which includes the build number. Look for it under "version" and "ServicePack Build"

5. Launch tiny11 creator.bat or tiny11 creator 22621.525.bat as an administrator, depending on which build your ISO is.You can do this by right clicking on the icon and selecting “Run as administrator.”

6. Click “More info” and Run anyway if Windows Smart screen warns you about running the program.

Also, click Yes if warned by Windows User Account Control. A command prompt window appears.

7. Enter the drive letter the ISO is mounted to and hit Enter.  In our case, it was D. 

Depending on whether your ISO file contains just one version of Windows 11 or many, you may see a list of different Windows “images,” including Windows 11 Home, Windows 11 Pro and Windows 11 Pro Education. 

8. Enter the image number for the Windows version you want and hit Enter. In our case, we hit “1” for Windows 11 Home. If there are no images listed, choose 1 for the default.

You’ll have to wait a few minutes while the software writes your ISO file. 

When the process is complete, you will have a file called tiny11.iso in the same folder as the Tiny11 Creator.

You can now “burn” this ISO file to a USB Flash drive so you can install Tiny 11 on a PC. Or you can use it to install Windows 11 on a virtual machine. If you don't know how to make a boot disk from an ISO file, see our story on how to do a clean install of Windows 11.

This Tiny11 version of Windows should install on any PC and will not hold you to any of the system requirements, including the TPM requirement, secure boot requirement and RAM requirement. More importantly, it will only use about 2GB of RAM out of the box. Obviously, as you add your own software, the system will use more resources.

Based on my experience, I can say that, while Tiny11 will run on a system with just 2GB of RAM, it does better on one that has at least 4GB. You'll want that headroom for any apps you're running within the OS.

Installing a Browser in Tiny11

It's important to note that Edge is not installed, though you may see a blank Edge icon on the desktop. So, if you want to install a web browser, you will need to either copy the installer from another PC, use Microsoft Store or use the winget command at the command line.  Entering winget install -e --id Google.Chrome will download and install Chrome.

What's Removed from Windows 11 in Tiny11?

According to NTDEV, the following apps and features are removed from Windows 11 in Tiny11. You can, of course, always add them back using the Windows Store or the Turn Windows Features On and Off menu in Control Panel.

• Clipchamp
• News
• Weather
• Xbox (Xbox identity provider is still there)
• GetHelp
• GetStarted
• Office Hub
• Solitaire
• PeopleApp
• PowerAutomate
• ToDo
• Alarms
• Mail and Calendar
• Feedback Hub
• Maps
• Sound Recorder
• Your Phone
• Media Player
• QuickAssist
• Internet Explorer
• LA57 support
• OCR for en-us
• Speech support
• TTS for en-us
• Media Player Legacy
• Tablet PC Math
• Wallpapers
• Edge
• OneDrive

Unfortunately Teams is not removed, but you can go into the Apps menu and uninstall it yourself.

AMD has been steadily gaining datacenter CPU market share since 2017 — when it launched its EPYC processor for servers — and is expected to command over 20% of shipments this year, according to DigiTimes Research. Meanwhile, various developers of Arm-based system-on-chips (SoC) for datacenters are projected to occupy around 8% of the market. While Intel will maintain the lead, its share will decline significantly.

AMD Is Gaining Share

There are several factors that will allow AMD to continue grabbing market share from Intel this year, according to Frank Kung from DigiTimes Research. These include higher core count of AMD's latest EPYC processors compared to range-topping Intel's Xeon Scalable (96 vs 60), lower prices of EPYC CPUs when compared to Intel Xeon Scalable products with similar core count, and better supply of AMD's latest EPYC CPUs made at TSMC when compared to Intel's Xeon Scalable produced in house. 

Although not mentioned directly by DigiTimes Research, there are two other major factors driving adoption of AMD's EPYC platform: introduction of new processors without major delays and better performance compared to the latest offerings from Intel as the latter arrived considerably later than they were supposed to. 

Late last year AMD introduced its EPYC 'Genoa' processors with up to 96 cores based on the Zen 4 microarchitecture for mainstream servers and this platform will ramp up gradually in 2023. Later this year the company plans to release its EPYC 'Bergamo' CPUs with up to 128 cores featuring the Zen 4c architecture for cloud datacenters, which will further improve positions of AMD as Intel is only set to ship its Sierra Forest CPUs for cloud workloads in 2024. Also, AMD plans to introduce its EPYC 'Siena' processors for communications market this year.

Arm Is Getting Stronger

There is a fundamental headwind for Arm-based SoCs when it comes to gaining server market share — software compatibility. Performance wise, these SoCs can compete against x86 processors, but if they cannot run programs, they will not be adopted.  

Nonetheless, Arm-powered SoCs from companies like Amazon, Ampere, Alibaba, and Marvell increased their share from 3.5% in 2021 to 6.8% in 2022, according to estimates by DigiTimes Research. Analysts expect such processors to command 8.1% of the server CPU market as new players (such as Nvidia) enter the scene and established players (such as Ampere) roll-out their new offerings. 

Given how strong Nvidia's compute GPUs are in AI and HPC fields, its Arm-based processors for servers are poised to grab a significant share of the market. 

As Arm-based datacenter SoCs gain software compatibility with popular applications, they will also gain market share. As a result, only time will tell what market share such processors will control several years down the road. According to the analyst from DigiTimes Research, it is almost guaranteed that Arm-based SoCs will continue to gain share in datacenter and edge computing servers.

When Apple moved to its own silicon, its Mac computers lost a feature called Boot Camp, which let people dual boot between macOS and other operating systems, like Windows. Now, Microsoft is officially sanctioning running Windows 11 on Macs with M1 and M2 processors. Today, it published a new support page entitled "Options for using Windows 11 with Mac computers with Apple M1 and M2 chips."

The options to do so aren't necessarily new — Microsoft is pushing Windows 365 Cloud PCs or by using Parallels Desktop 18 — but it's the first time we've seen Microsoft provide Windows guidance to Mac users since Boot Camp went away. In general, it seems to be pointing it only toward business users.

Windows 365 Cloud PCs are currently available in business and enterprise plans, starting at $31 per user per month. In this case, you can stream Windows 11 to a Mac (or any device) through the software-as-a-service. The benefit of this is that Windows 365 offers nested virtualization, allowing you to run emulators and get full compatibility with features that require virtualization, like running Android apps from the Amazon app store.

Using Parallels Desktop 18 may be more accessible for most people, but the standard edition for home and student use starts at $99.99 a year, with professional and business edition subscription services at $119.99 and $149.99 per year. Microsoft says it's authorized for running Arm versions of Windows 11 Pro and Windows 11 Enterprise.

Windows 11 on Arm has many of the same restrictions on a Mac as it does on PCs designed for Windows. "The Arm version of Windows 11 has limitations that can impact your ability to use various types of hardware, games, and apps, including those that rely on DirectX 12 or OpenGL3.3 or greater," the support page reads. In other words, don't expect to use your M2 Max's GPU cores for gaming with Windows 11 on Arm.

But going the Parallels route also doesn't allow for nested virtualization, so you won't be able to run the Windows Subsystem for Android or Linux (to run Android apps from the Amazon app store or run a GNU environment in Windows), or the Windows Sandbox or Virtualization-based security (VBS) to separate some memory from the OS.

Additionally, 32-bit Arm apps from the Microsoft Store in Windows don't work on Macs with M1 and M2 processors (Apple stopped supporting 32-bit apps with macOS Catalina). But Microsoft says that 32-bit Arm apps "are in the process of being deprecated for all Arm versions of Windows." Ideally, you would run 64-bit Arm apps on Mac, but you can also emulate x64 or x86 apps.

We don't expect that this news means that Microsoft will also make the Arm versions of Windows 11 more available to the public. Microsoft explicitly told me that "Parallels Desktop is currently the only authorized option that supports Windows 11 in VMs on Macs." Parallels has been used to run Windows 11 for quite some time on M-series Macs, however, and installing Windows 11 is a simple part of the app's setup.

It isn't clear why Microsoft is now publishing this page, other than to make it explicit which methods of getting Windows 11 on a Mac are officially sanctioned. But hey, if it helps someone find what they're looking for, we're all for it. It's not as exciting as the prospect of dual-booting macOS and Windows 11 on Arm, though. We can only dream.

The first picture of Intel's LGA7529 socket for its upcoming codename 'Sierra Forest' processors powered by energy-efficient cores has been published by hardware leaker @YuuKi_AnS. The CPUs will use small Atom-class energy-efficient cores, and it will plug into a massive 7529-pin socket.

Intel's Xeon 'Sierra Forest' processor and the Birch Stream platform are designed for high-density cloud datacenters. These are for running workloads that benefit mostly from core count rather than from single-threaded performance. To that end, the core count of Sierra Forest system-on-chips will be significantly higher than regular Xeon processors based on high-performance cores, such as Emerald Rapids as well as Granite Rapids.

To feed those cores with data and power, Sierra Forest will use loads of pins, which explains the LGA7529 form-factor with a massive number of pins. Meanwhile, it does not look like Intel's LGA7529 socket is significantly bigger than the company's LGA4677 for Xeon Scalable 'Sapphire Rapids' processors and their successors, at least if we compare sizes of both sockets to the DDR5 memory slots located nearby. 

Unfortunately, the published image doesn't reveal many new details about Sierra Forest CPUs, such as the number of cores or memory channels. Still, if there are LGA7529 motherboards in the wild, it means server makers are likely (or will soon be) testing the future CPUs.

Intel's Xeon 'Sierra Forest' processor is set to be made on Intel 3 process technology and is set to become available in 2024.

Microsoft is stepping up its investment in the artificial intelligence realm, and today announced it is expanding its partnership with OpenAI. According to Microsoft, this will be the third phase in its tie-up with OpenAI, representing a "multi-billion dollar investment" in its effort to be at the forefront of all things AI.

"We formed our partnership with OpenAI around a shared ambition to responsibly advance cutting-edge AI research and democratize AI as a new technology platform," said Microsoft CEO Satya Nadella. "In this next phase of our partnership, developers and organizations across industries will have access to the best AI infrastructure, models, and toolchain with Azure to build and run their applications."

From now on, Microsoft's Azure cloud infrastructure will be used exclusively to power OpenAI workloads, API services, research, and backend systems. Microsoft will also use OpenAI to enhance its products to create "new categories of digital experiences." Those experiences will leverage OpenAI's ChatGPT, a conversational AI routine that can be used to write articles, poetry, and even debug code. 

The Information reported earlier this month that Microsoft planned to fuse ChatGPT with its Bing search engine to become more competitive with Google (or, more specifically, the Google Knowledge Graph). Perhaps even more interesting is the rumor that claims Microsoft intends to bring the GPT text-generation model to Microsoft Word and Outlook. This would potentially let AI write emails for you based on input commands, or flesh out an article you're writing using information siphoned from the web. 

On that note, we used ChatGPT (using the GPT-3 language model) last week to inquire about how to build a PC. Unfortunately, despite ChatGPT's impressive capabilities, its advice on building a PC turned out to be pretty bad. Its instructions were grammatically correct, but would have likely resulted in the destruction of the CPU if followed by a tech neophyte. 

Microsoft's investment in AI should not come as a surprise to anyone. When Nadella announced that the company would lay off 10,000 employees this year, he noted that "the next major wave of computing is being born with advances in AI, as we're turning the world's most advanced models into a new computing platform." 

OpenAI was co-founded in 2015 by a group of tech luminaries, including Sam Altman, Peter Thiel, and Elon Musk. Its mission is to "ensure that artificial general intelligence benefits all of humanity," and the company's continued partnership with Microsoft will hopefully advance those efforts. 

"The past three years of our partnership have been great," said Altman, who serves as OpenAI's CEO. "Microsoft shares our values and we are excited to continue our independent research and work toward creating advanced AI that benefits everyone."

Multiple sources, including Bloomberg and Sky News, confirm that Microsoft will announce wide-ranging layoffs across "a number of engineering divisions." Microsoft currently employs over 221,000 workers globally, and this newest round of cuts could see that figure slashed by 5 percent. If true, more than 10,000 workers will, unfortunately, lose their jobs. 

The official announcement of the layoffs could come as early as Wednesday. Microsoft's hiring exploded during the pandemic as more people worked from home and relied on the company's software products and cloud services. In September 2022, CEO Satya Nadella confirmed that the company hired 50,000 people between 2020 and 2022. Now, as the global economy slows and fears of a lingering recession materialize, Microsoft is focused on cutting costs.

"We did have a lot of acceleration during the pandemic, and there's some amount of normalization of that demand," said in an interview with CNBC earlier this month. "[There are] clearly many, many challenges right now around the world. For us as a global company, we're not going to be immune from what's happening in the macro. We'll also have to get our own operational focus to make sure our expenses are in line with our revenue growth." 

We should note that Microsoft isn't the only big tech firm faced with reducing its workforce. For example, Facebook's parent company Meta eliminated 11,000 jobs in 2022, and Amazon revealed two weeks ago that it would lay off 18,000 employees during 2023 (nearly double what earlier reports had suggested). Likewise, Salesforce announced this month that it would lay off 10 percent of its workforce (roughly 8,000 employees).

According to Microsoft's June 30th regulatory filing, 122,000 of its full-time employees are based in the United States. The remaining 99,000 work from the company's various offices around the globe. It is likely that Microsoft's U.S. workforce will take the brunt of the looming cuts. 

Microsoft will update investors on its financial performance on January 24th. During fiscal Q1 2023, Microsoft reported an 11 percent increase in revenue to $50.1 billion (an 11 percent increase year-over-year) and net income of $17.6 billion (a 14 percent decrease year-over-year).

"In a world facing increasing headwinds, digital technology is the ultimate tailwind," said Nadella at the time. "We're focused on helping our customers do more with less, while investing in secular growth areas and managing our cost structure in a disciplined way."

Microsoft rebranded Office 365 to Microsoft 365 in Oct. 2022, part of its effort to slowly take the long-running "Office" brand out of the limelight. At the time, the company didn't make any significant changes to the subscription tiers or features offered. But that will change later this month with the addition of a new Microsoft 365 Basic tier, which will cost $1.99/month ($19.99/year).

Microsoft 365 Basic slots between Microsoft 365 (free) and Microsoft 365 Personal ($6.99/month). It replaces the previous 100GB OneDrive storage subscription and adds some perks. Microsoft says customers can expect these features with the new tier:

• 100GB cloud storage
• Works on Windows, macOS, iOS, and Android
• Web and mobile versions of Word, Excel, PowerPoint, OneNote, OneDrive, and more
• Ad-free Outlook web and mobile email and calendar with advanced security (data encryption, suspicious link cheer, malware scanning for attachments)
• Microsoft technical support

The extra functionality seems like a nice upgrade for users currently getting by with just the 100GB OneDrive storage tier. In addition, Microsoft says it will add "even better advanced security features" later this year, including shared links that are password-protected, and ransomware recovery.

While the inclusion of Office is welcome, we should note that this tier only includes the basic, web-based versions of Microsoft's popular productivity apps, which are available for free. To access the "premium" desktop versions of Word, Excel, PowerPoint, and OneNote, you'll have to step up to Microsoft 365 Personal.

Microsoft was quick to point out that Jan. 29, 2023 marks the 10th anniversary of Office 365. The high-end Office 365 Home Premium subscription initially included only 20GB of cloud storage. Today, the Microsoft 365 Personal tier offers 1TB of OneDrive storage. In addition, the flagship Microsoft 365 Family tier for consumers ($9.99/month) supports up to six people, each with their own 1TB allotment of OneDrive storage.

Microsoft 365 Basic will be available starting Jan. 30. Current 100GB OneDrive subscribers will automatically be shifted to the newer, more feature-packed tier.

Amazon Web Services hosts some of the world's most valuable data for commercial companies in its cloud, but when Russia invaded Ukraine earlier this year, it faced challenges that it probably had not faced before. AWS had to secure vital government data hosted in Ukraine and then bring all the digital services the country offered back online, reports Business Insider.

It was not particularly easy to move petabytes of crucial data using secure Internet channels, as Ukraine was heavily bombed and usage of the web increased significantly. To get physical copies of Ukraine's critical infrastructure and economic information data and preserve it, AWS shipped custom-built Snowball Edge solid-state drives to Poland and then drove them to Ukraine.

Unfortunately, AWS does not show the Snowball Edge suitcase-sized SSDs used to physically transfer data from Ukraine to safety. We know that datacenter Snowball Edge machines offer up to 80TB of usable storage space, and we presume these are all-flash arrays of extreme storage density that are also designed to be extremely reliable.

AWS currently hosts over 10PB (petabytes) of vital data from 27 Ukrainian ministries, 18 Ukrainian universities, the largest remote learning K–12 school (serving hundreds of thousands of displaced children), and dozens of various private sector companies, the cloud giant said. There are now 61 government data migrations to AWS with more to come.

PrivatBank, the largest bank in the country, has migrated 270 applications and 4PB of client data residing on 3,500 Ukraine-based servers to AWS. Interestingly, PrivatBank plans to keep using the AWS infrastructure even when the war is over, which is a departure from banks' usual approach to use on-premises machines to run their mission critical software.

Digitization of these data and services ensures the continuity of the Ukrainian state and allows people to access information they need from whenever they are. This data is important even today when students want to continue their education elsewhere, but it will also be crucial tomorrow, when the country is rebuilt. "Government in a box" servers and storage allows Ukraine to avoid the chaos associated with war and post-war periods, something that's hard to overstate.

"We used to assume that this is just how it is in war — everything gets destroyed and you have to rebuild from nothing," said Liam Maxwell, director of the Government Transformation team in the AWS public sector business and a volunteer in AWS’s efforts to aid Ukraine. "But by migrating to the safety and security of the cloud, the government and its citizen services prevail."

Chinese high-tech companies such as Alibaba's T-Head are unable to buy advanced CPUs and GPUs from American companies such as AMD, Intel, and Nvidia due to the recently imposed export control rules. Chinese entities are also apparently unable to license leading-edge CPU IP from Arm because of the same rules and the Wassenaar arrangement. 

According to a report from Financial Times, Chinese companies cannot get access to Arm's advanced Neoverse V-series CPU IP due to export control rules of the U.S. and U.K. as well as the Wassenaar arrangement on export controls for dual-use goods and technologies and conventional arms. Arm's Neoverse V1 and Neoverse V2 CPU core IP is designed to enable high-performance computing applications, including supercomputers. 

Arm does not sell processors, but licenses designs that power system-on-chips which can be used to build everything from smartphones to supercomputers. The recently set U.S. export regulations prohibit shipping American technologies that enable supercomputers with performance of over 100 FP64 PetaFLOPS or over 200 FP32 PetaFLOPS within a 41,600 cubic feet (1178 cubic meters) or smaller envelope to China. Arm's high-end Neoverse V-series designs are considered an American technology, and, as a result, Arm cannot license its designs to Chinese entities. 

The Arm Neoverse V-series designs can are aimed at HPC processors for supercomputers and and can be used to develop nuclear weapons — so they are subject to both the recently-imposed U.S. technology export rules as well as the Wassenaar arrangement. 

China's high-tech titan Alibaba made a splash when it unveiled its 128-core Yitian 710 server processor for cloud workloads last year and even set an integer performance records with the chip, a result which eventually was expelled from rankings due to lack of availability. But Alibaba could not use Arm's Neoverse V for its SoC. Instead, it used an undisclosed Armv9 core, which is either its own design or a custom implementation of an Arm Cortex core based on the latest instruction set from the British company — possibly a Neoverse N-series, if we had to guess.

"We feel that the western world sees us as second-class people," an engineer from Alibaba's T-Head told Financial Times. "They won't sell good products to us even if we have money." 

Alibaba's engineers and managers have reason to worry. Amazon Web Services has been using Arm's Neoverse V-series cores in its Graviton SoCs for some time and therefore had an advantage over Alibaba's cloud services.  

A source close to Arm told Financial Times that the company was working with Alibaba and other customers from China to find solutions that would enable them to meet their performance goals while also complying with both the Wassenaar arrangement and the latest U.S. export regulations.

Intel has officially revealed its Intel On Demand program that will activate select accelerators and features of the company's upcoming Xeon Scalable Sapphire Rapids processor. The new pay-as-you-go program will allow Intel to reduce the number of SKUs it ships while still capitalizing on the technologies it has to offer. Furthermore, its clients will be able to upgrade their machines without replacing actual hardware or offering additional services to their clients.

Intel's upcoming Intel's 4th Generation Xeon Scalable Sapphire Rapids processors are equipped with various special-purpose accelerators and security technologies that all customers do not need at all times. To offer such end-users additional flexibility regarding investments, Intel will deliver them to buy its CPUs with those capabilities disabled but turn them on if they are needed at some point. The Software Defined Silicon (SDSi) technology will also allow Intel to sell fewer CPU models and then enable its clients or partners to activate certain features if needed (to use them on-prem or offer them as a service).

The list of technologies that Intel wants to make available on demand includes Software Guard Extensions, Dynamic Load Balancer (DLB), Intel Data Streaming Accelerator (DSA), Intel In-Memory Analytics Accelerator (IAA), Intel In-Memory Analytics Accelerator, and Intel QuickAssist Technology (QAT) to accelerate specific workloads.

Since Intel's On Demand technologies are aimed at entirely different workloads, very few customers will need all of them at once. But as they scale their data centers, they may require some of them, which is when the On Demand capability comes into play. Meanwhile, some of Intel's customers will offer those capabilities as a service (e.g., for cloud and co-hosting machines). In contrast, others will contribute to activating them on servers installed on-premise.

The formal rollout of the Intel On Demand program leaves more questions than answers. We do not know how much Intel plans to charge to activate certain features or how much its clients will want to start them 'as a service.' We know that companies like H3C, HPE, Inspur, Lenovo, Supermicro, PhoenixNAP, and Variscale will be a part of the On Demand program.

For now, Intel's On Demand program is reserved for servers, and we would expect it to remain a prerogative of Xeon platforms. Meanwhile, back in the day, Intel offered software upgrades for its desktop processors to make them run faster. Unfortunately, that program faced criticism as Intel essentially crippled its perfectly fine processors. As a result, some might think the On Demand program mimics the ill-fated Intel Upgrade Service. Still, keeping in mind that the server world behaves differently than the client PC world and that we do not know the terms of Intel's On Demand, we would not draw parallels here until we know all the details.

Environmental concerns have become increasingly entrenched in the collective consciousness - whether through climate activists' decisions on what pieces of art deserve to have soup thrown at them or through the participation (and notable absences) from the United Nations Climate Change Conference. In the meantime, and as spotted by The Register, the European Commission (EC) is exploring radical ideas for reducing the climate impact of data centers and other HPC (High-Performance Computing) deployments. Concepts include completely installing them outside the Earth's atmosphere, within the deep black of space.

The EC-led feasibility study ASCEND (short for Advanced Space Cloud for European Net zero emission and Data sovereignty) is part of the EU's "Horizon Europe" initiative and is bolstered by a €2 million budget. It also counts on several partners from the environmental, cloud computing, and space technology fields spearheaded by Thales Alenia Space, a joint venture between European aerospace and defense companies Thales and Leonardo. According to Thales Alenia Space, the primary motivator for the feasibility study doesn't pertain to performance, longevity, or energy efficiency. Instead, those elements have been wrapped within the bigger environmental picture of dealing with running emissions from operational data centers.

It's easy to see how space-borne data centers could be a boon for the environment. Due to them being outside Earth's atmosphere, emissions relating to their operation would no longer have an impact on our planet. It'd be impossible to reduce this value to zero (outside carbon compensation programs linked to new data centers) because most components would still have to be manufactured within our "pale blue dot" of a planet. But even as companies increase their hardware products' power efficiency generation after generation, the increasing performance requirements for the latest HPC hardware from any of the prominent hardware vendors have led to an ever-growing energetic (and environmental) footprint. That is precisely what ASCEND aims to tackle.

The idea, then, would be to build data centers that solar panels could entirely power: they already achieve higher efficiencies outside our atmosphere. They could provide power in the "hundreds of megawatts" range. These orbiting data centers would connect to Earth using optical links, which would do the heavy lifting in transmitting information from orbit. According to Thales Alenia Space, Europe has already mastered the underlying technologies that could make a feasible deployment scenario such as this.

Of course, the concern with atmospheric emissions doesn't disappear completely when we start putting data centers in space. CO2 emissions from the building and deploying spaceships that could carry the data center payload would add to the data centers' overall carbon footprint, possibly neutering any environmental advantages from being operated in space. And, of course, this also plays out with improvements in payload capacity from the latest (and future) rocket technology: one SpaceX Falcon Heavy could carry a much more significant payload in one go than multiple Falcon launches could ever aspire to.

ASCEND's first order of business, then, is to calculate whether or not it would make more sense to "yeet" data centers into space rather than build more of them on Earth's surface (or even, perhaps, in the deep seas as Microsoft successfully explored through its project Natick).

But even if ASCEND finds out that rocket launches would turn the environmental concerns on their head, there's always the option to... slingshot them? At least, that is the idea of SlingShot, a private company that's been tapped by NASA for trials of its launch system. SlingShot does away with fossil-fuel-heavy rockets and flings cargo out into space by attaching a payload onto one end of a giant spinning arm powered by electric motors. By making the spinning arm rotate at around 450 rotations per minute, datacenter-borne payloads could be shot into space at speeds reaching 8,000 Km/h. That's certainly one way of controlling carbon emissions.

Whether via rocket or other exotic propulsion means, ASCEND's second order of business is to understand whether data center payloads can be put to work after being carried by rocket and deployed into space - we have to remember that anything leaving our atmosphere faces immense pressures that could wreak havoc with anything but the most resilient hardware installations. And even if they can be put to work, there are many other concerns, such as maintenance, upkeep, the fast pace of obsolescence for data centers, and the ability to decommission these space-faring installations. We have enough "space junk" in Earth's orbit without several decommissioned data centers being added to that tally.

All in all, this is still an exploratory study and one that's likely to be revisited as technology progresses. Launch capability, cost, and efficiency have improved immensely since we first went "Ad Astra," The same is true of computing in general. So we'd say it's a question of time until ASCEND's solution (or another party's) fits the problem it's trying to solve - if not today, then in one of our tomorrows.

Nvidia may have fired the first salvo in the latest next-generation GPU battle with the RTX 4090. But the card's impressive performance aside, things haven't gone particularly smoothly, with melting power adapters and the "unlaunching" of the 12GB RTX 4080. Now it's time for AMD's return volley, in the form of its RDNA 3 architecture and Radeon 7000 cards which will vie for a slot in your next PC, as well as placement on our Best Graphics Cards page. 

PowerColor gave us a tease yesterday, with its Tweet of a glowing Red Devil logo, surrounded by what is likely an angular backplate. But at AMD's "together we advance_gaming" event in Las Vas Vegas, which will be live-streamed on YouTube, the company is set to announce its next generation of graphics cards, promising "new levels of performance, efficiency and functionality to gamers and creators."

Our team is on the ground at the event, as well as following along back at the office with a mechanical keyboard and a comfy 4K display. Stay tuned as we find out what AMD has in store and how competitive its new cards will be as it faces stiff headwinds from rising production costs, Intel's own Arc lineup, and Nvidia's advancements, like DLSS3 and third-generation ray tracing hardware. Time will tell where Radeon 7000 lands on our GPU hierarchy. But after years of extremely high prices and hard-to-find hardware, it's exciting just to know we'll soon have new (and hopefully more affordable) gaming hardware to run our favorite games.

AMD's event is set to start at 1 p.m. PT / 4 p.m. ET, so keep it locked here for the latest updates.

Lisa Su takes the stage, welcomes everyone, and we're off and running. Seems AMD is, as always, quite busy.


"We're in the midst of introducing actually four new architectures to the market across more than a dozen products… We started with the Ryzen 7000 launch in September… Next week, we're going to bring Zen  to high-end server processors. And then soon after that to notebook PCs. And we also have our Zen 4 core optimized for Cloud Native Computing and our next-gen DNA architecture that we're going to launch in 2023. But of course, today, it's all about gaming."

"When you look through 2025, we're expecting 3.5 billion gamers in the world."

There's a lot to aim for when your customer base is half of the planet. 

Talking about chiplets, and this is the first chiplet-based gaming GPU. Like with Ryzen 7000, this offers a modular approach. Graphics compute die is 5nm, containing "all of the high performance compute in the GPU." Sharder, display engine, and a new media engine. The other die is memory cache, built on a 6nm process. This includes the memory subsystem's GDDR6 controllers and up to 96 megabytes of infinity cache.

"As a result, we have just incredible capability with more than 61 teraflops of compute… We also have the world's best chiplet interconnect to enable 5.3 terabytes per second of peak bandwidth."

24GB of GDDR6. "When you put that together, we have 58 billion transistors in this design, which gives us just an incredible amount of gaming performance." 

54% gen over gen improvement in performance per watt.

Radeon RX 7900 XTX with 24GB and 7900 with 20GB officially announced. Hopefully, they won't have to unlaunch one of those. 

Going a bit deeper into the RDNA 3 compute die. New display engine and dual media engine. 165% more transistors per square mm.

"We've used 5nm to the fullest."

Dual issue SIMD units and a 2.7x increase in AI performance.

50% more performance per compute unit.

Displayport 2.1! You know they like sticking that to Nvidia. Up to 165 Hz at 8K. Jarred needs a new monitor. 

Scott Herkelman talks performance. RTX 7900 XTX is up to 1.7 times faster than previous flagship. 355W, 2.3GHz clock, and it doesn't look TOO big. 

7900XT, 84 CUs, 2GHz clock and 20GB of DDR6. "There's no need for a new power adapter. We made it as easy as pulling out your old card and putting in a new one."

Scott is of course calling out DisplayPort 2.1 and other things aimed at making Nvdia look bad. Teased a Samsung 8K monitor, which we should see at CES early next year. 

DirectX ray tracing coming to Halo Infinite. DXR coming with season 3.

Talking games now, which of course all looks great here. But time and testing will tell how things look and run compared to the competition. 

FSR3, to of course compete with DLSS3. I sense Jarred will be busy testing and comparing soon. As always, so much depends on how widespread these features are and, superficially, if they're offered on the games you want to play.

Frank Azor takes the state to talk more about software and the overall experience. AMD Adrenalin is "your hub for just about everything."

AMD HYPR-RX mode to combine AMD's software features to provide a simple, one-click experience so you can just sit down and get in the game. Coming early next year.

RDNA 3 Media engine promises up to 7x speed improvements. Another interesting thing to test. 

AMD Advantage desktops. A new way for AMD to push their key components together. 

Scott is back. Seems like we are about to find out pricing and availability.

December 13th, 2022 for both cards.
$899 for the XT and $999 for the XTX. Depending on how the cards compare to the 4080 and 4090, Nvidia may have to rethink pricing. 

For all those who haven't jumped on an RTX 4090 yet, these cards at the very least will give gamers something to think about. No need for a brand-new PSU or a four-pin adapter with possible electrical connector issues, and these cards look like they'll fit in a lot more cases. If the 7900 cards are at all competitive and, at least as importantly, if they stay at those prices and there is enough stock, these cards will likely sell well.

But as much as I like the below-$1,000 prices, both AMD and Nvidia are (for now at least), only catering to the extreme high end of the market. I hope to see some new and more reasonably affordable cards by the time CES comes around again early next year.

Hey, I can dream. 

The Decentralized Storage Alliance was born Tuesday — promising cooperation between major technology players to facilitate the transition between Web2 and Web3. The major thrust of this initiative, indicated by its name, is in evolving decentralized storage, driving awareness of this technology, and promoting its adoption. Major technology partners include AMD, Seagate, Ernst & Young, Protocol Labs, and the Filecoin Foundation.

AMD's role in the Web3 tech coven is leveraging its expertise in high-performance and adaptive processor technologies — combining CPUs, GPUs, FPGAs, Adaptive SoCs, and deep software integration. This expertise is expected to be instrumental in the future success of decentralized storage.

Seagate is a well-known provider of physical storage technology, and has been even before the days of Web1. Its solutions range from home users to businesses and even to the cloud providers (Web2), which this newer technology hopes to displace. It's wise for Seagate to get onboard and be a guiding member of this alliance from the beginning if — as expected — decentralized storage does take off one day.

Investments and accountancy firm Ernst & Young will take on the role of guiding and developing secure blockchain technology for storage, to benefit its customers. 

The above trio, alongside alliance founders Protocol Labs and the Filecoin Foundation, aim to set up a widely-adopted foundation for decentralized storage by embarking on the following tasks: 

• Developing standard specifications and reference architectures that address the unique needs of enterprise companies.
• Providing access to education materials, technical resources, and best practices.
• Improving the process of onboarding data to decentralized storage networks and making it easier for new data centers to onboard to the network.
• Enabling the creation of impactful Working Groups that solve specific issues with the transition to decentralized storage technologies and Web3.

Why would you be interested in decentralized storage? This storage technology promises a Web3 version of the cloud — offering great efficiency and robust security, as well as a "significantly lower" cost. Its decentralized nature means that uses of this storage tech aren't locked in to a provider — they will meet data sovereignty requirements, and will benefit from other advantages inherent in the blockchain.

Filecoin is the largest provider of this type of service: Filecoin says it hosts 240 PiB of data, which is approximately 65,000 Wikipedia's worth of data. Its customers number in the thousands, and include UC Berkeley, USC's Shoah Foundation, and the University of Utah. It seems like the Decentralized Storage Alliance is off to a good start.

Dropbox, the cloud storage provider, has announced it has been the target of a phishing attack that successfully accessed its private GitHub repos. GitHub was able to quickly notify Dropbox of the attack, and no customer data or passwords were affected.

The data breach took place on October 13, with Dropbox becoming aware that things were amiss the next day. The attackers impersonated the CircleCI integration and delivery platform that can be logged into using GitHub credentials, bombarding Dropbox staff with realistic-looking phishing emails. Many of them were blocked by Dropbox’s internal systems, but some got through - enough, it seems, for at least one employee to visit a fake CircleCI login page, enter their GitHub credentials, and use a hardware authentication key to pass a one-time password to the malicious site.

This allowed the attacker into Dropbox’s private Github area, from where they copied 130 code repositories. Data accessed includes, according to Dropbox’s statement: “...some credentials—primarily, API keys—used by Dropbox developers. [It] also included a few thousand names and email addresses belonging to Dropbox employees, current and past customers, sales leads, and vendors.”  Then later: “These repositories included our own copies of third-party libraries slightly modified for use by Dropbox, internal prototypes, and some tools and configuration files used by the security team. Importantly, they did not include code for our core apps or infrastructure. Access to those repositories is even more limited and strictly controlled.” 

Back in September, GitHub warned its users in a blog post about attacks targeted at CircleCI, noting that “If the threat actor successfully steals GitHub user account credentials, they may quickly create GitHub personal access tokens (PATs), authorize OAuth applications, or add SSH keys to the account in order to preserve access in the event that the user changes their password.”

Dropbox was able to cut off the attackers’ access on the same day it found out about the intrusion, and believes the risk to customers is minimal. The company is also upgrading its multi-factor authentication method to WebAuthn—a change already in progress when the attack happened. 

Cloud game streaming services have been quite a popular topic in the game industry in the past 15 – 20 years, as the perspective of playing a high-end game on a mediocre device with a good Internet connection sounds very alluring. But after numerous major game companies have invested hundreds of millions in their cloud gaming platforms, Microsoft says that the technology is still in its infancy and its prime time is nowhere close.

At least, this is what the software giant wrote in response to the U.K.'s Competition and Market Authority this week. Among the significant concerns CMA has regarding Microsoft's proposed acquisition of Activision Blizzard are withholding popular games like Call of Duty from competing platforms (namely Sony's PlayStation) and subsequent monopolization of the emerging market of cloud game streaming services. As a result, CMA launched an in-depth investigation of the proposed transaction earlier this month to learn more about the matter.

"This is a new and immature technology which the CMA has recognized faces significant challenges, particularly on mobile devices," Microsoft wrote in its 33-page response to CMA's concerns (via PC Gamer). "Microsoft agree that in future cloud gaming services may mean that hardware distinctions will become less important. However, the reality is that today cloud gaming remains in its infancy and unproven as a consumer proposition."

Back in the 2000s and for the better part of the 2010s, cloud game streaming services suffered from major technology limitations, starting from the imperfection of data centers and servers on the cloud side and a slow Internet connection on the client side. By now, many technological limitations have been largely overcome. Yet, local rendering still provides the best and most consistent visual quality and the lowest input lag, which is particularly important for avid gamers and something most gamers know.

Since Microsoft, Sony, and Nvidia continue to invest hefty sums of money in improving their cloud gaming services and providing an experience that is on par with that offered by local PCs or consoles, cloud gaming services are still not quite there. Meanwhile, they can enable playing PC or console games using smartphones, a type of experience that has not been available before and which is something that numerous companies are putting their money on (e.g., Qualcomm, Razer, and Verizon). But to compete against traditional games, streaming services will need to offer visual quality, latencies, and loading times comparable to those of downloadable titles.

"While this may grow, particularly on mobile devices, adoption is not expected to be rapid as it requires a significant change in consumer behaviour," Microsoft explained. "Gamers care about subject matter, storylines, graphical performance, speed (e.g., loading times and latencies), mechanics, game selection, and game cost. […] Streaming services therefore need to compete effectively with downloadable gaming options across these metrics if they are to grow."

Because cloud game streaming services are not popular among consumers, Microsoft says that it is not in its interest to harm competing for game streaming services or withhold popular Activision Blizzard games from rivals as it is interested in promoting cloud gaming in general.

"Consumer adoption of cloud gaming remains low," Microsoft said. "Harming or degrading rival services would significantly set-back adoption of this technology – protecting market-leading incumbents (i.e., Sony on console, Apple and Google on mobile, as well as Steam on PC). […] Instead, [Microsoft Xbox's] incentive is to encourage the widespread adoption of cloud gaming technologies by as many providers as possible to encourage the major shift in consumer behaviour required for cloud gaming to succeed."

Cloud gaming has the potential to bring games to devices incapable of rendering high-end titles locally, which will expand the global gaming market. Many technologies required for competitive cloud gaming platforms are already here, so the question is when cloud gaming will become a mainstream phenomenon. Yet, keep in mind that right now, Microsoft needs to convince the CMA and other regulators of two things: that they should not be concerned about cloud gaming today as it is an immature technology, but that the proposed takeover of Activision Blizzard will enable Microsoft to increase adoption of the technology.

Oracle on Tuesday announced plans to deploy tens of thousands of Nvidia's top-of-the-range A100 and H100 compute GPUs to its Oracle Cloud Infrastructure (OCI). The A100 and H100 GPUs will be available for Oracle's cloud customers for their AI workloads enabled by Nvidia's AI software. The deal's exact terms remain behind closed doors, but we are talking about a transaction worth hundreds of millions of dollars.

The new collaboration between Nvidia and Oracle will make AI training, computer vision, data processing, deep learning inference, and simulation available to all enterprise customers. They will not have to invest hefty sums into deploying their data centers with Nvidia's expensive compute GPUs. Oracle already offers OCI clients access to high-performance computing instances and will now provide them with various AI capabilities.

Enterprise clients of Oracle's OCI will be able to access all of Nvidia's AI platforms, including the following:

• AI Enterprise — a set of engines that can be used for AI model training, compute vision, conversational AI, data processing, recommender systems, and simulation, among others.
• RAPIDS — acceleration for Apache Spark data processing on the OCI Data Flow fully-managed Apache Spark service, including bare metal instances like BM.GPU.GM4.8 with A100 Tensor Core GPUs.
• Clara — medical imaging, genomics, natural language processing, and drug discovery (coming soon).

Nvidia's A100 and H100 compute GPUs are pretty expensive. Even previous-generation A100 compute GPUs cost $10,000 to $15,000 depending on the exact configuration, and the next-generation H100 products promise to be even more costly. So while Oracle is unlikely to buy Nvidia's compute GPUs for retail prices, it still pays a premium for Nvidia's hardware and software.

Keeping in mind that we are talking about tens of thousands of computing GPUs along with Nvidia's NVLink switches and possibly data processing units, we would expect the deal between Nvidia and Oracle to be worth hundreds of millions of dollars, which is quite an unprecedented contract. In any case, working with Oracle is vital for Nvidia as shortly it will not be able to sell compute CPUs to customers in China.

"Our expanded alliance with Nvidia will deliver the best of both companies' expertise to help customers across industries — from healthcare and manufacturing to telecommunications and financial services — overcome the multitude of challenges they face," said Safra Catz, CEO of Oracle.

Microsoft is preparing to retire or at least deemphasize its decades old Office brand in favor of Microsoft 365 branded cloud productivity tools. Users should expect to see the changes permeate throughout the Microsoft cloud ecosystem in the next few weeks. Meanwhile, PC stalwarts using offline standalone Office suites and applications will probably not notice any difference.

The news was secreted within Microsoft’s announcement of a new Microsoft 365 app on Wednesday. On that page you will find a FAQ which helpfully spells out what is changing with regards to Microsoft 365 and Office brands, when changes will be pushed through and what it means to users of either branded products / suites. It also explains that “over the last couple years, Microsoft 365 has evolved into our flagship productivity suite,” and this is seemingly behind the decision to transition Office.com, the Office mobile app, and the Office app for Windows into Microsoft 365 branded products.

The key thing for many users of Microsoft's flagship office productivity tools, is that they are not going away. Microsoft says that “as part of Microsoft 365 you will continue to get access to apps like Word, Excel, PowerPoint and Outlook.” Also, Office 365 subscribers in business / enterprises will not be moved to Microsoft 365 plans. Importantly to many readers, those seeking the most recent offline / perpetual Office editions will still have the opportunity to buy apps as one-time purchases via Microsoft’s Office 2021 and Office LTSC plans, too.

While the branding change is a pretty big deal, once you read all the small print, it isn’t very impactful. However, the same could be said of any branding or naming change. The most affected will be personal consumers, students, and small home offices which currently use the Microsoft 365 cloud-powered productivity platform via mobile and web-apps. Existing users of these cloud office tools won’t have to do anything; existing accounts, profiles, subscriptions, and files won’t be disrupted, even if you are accessing them from the new app with new branding and icon. In a way, the naming changes help emphasize the difference between the cloud-powered and the perpetual Office releases; with the online version highlighting collaborative tools like Teams and OneDrive being used alongside more traditional office productivity apps in the new hybrid working world.

Expect to see many of the above changes arrive with the rebranding of the unified Office app to Microsoft 365 in November, with the whole rebranding effort expected to be completed by January 2023.

Google has announced a new idea in its search for success in the lucrative gaming market. Today, Google and a trio of hardware partners launched “the world’s first laptops built for cloud gaming,” also known as Gaming Chromebooks. With Google Stadia’s death sentence certain, the cloud gaming services that the new offerings will embrace are Nvidia’s GeForce Now (preinstalled), Microsoft’s  Xbox Cloud Gaming (web app Beta now ready), and Amazon Luna (U.S. mainland only for now).

The reasoning behind Gaming Chromebooks is that Google has now established Chromebooks as the go-to devices for those seeking “fast, secure and easy-to-use” computing. With compact and robust hardware now more readily available than ever, Google and its partners want to extend the Chromebook universe with gaming features and fast, easy access to “cutting-edge graphics through the cloud.” Google boasts that Gaming Chromebook compatibility with the aforementioned services means owners could have a library of up to 1,500 games at their fingertips.

Regular readers will be aware that one of the most significant benefits of cloud gaming is that it doesn’t require powerful hardware. However, that isn’t to say that any old device will excel at the job – game streaming undoubtedly benefits from faster device networking, and brighter and faster screens. These points seem to have influenced the hardware recipe adopted by Acer, Asus, and Lenovo.

The three Gaming Chromebook launch devices are the new Acer Chromebook 516 GE, the Asus Chromebook Vibe CX55 Flip, and the Lenovo Ideapad Gaming Chromebook from Lenovo, which are all expected to be available this month. Google offers this trio via its Chromebook web store, and you can learn more about each device there. A quick check of the specs of these three reveals quite a varied CPU/memory configuration – they have a mix of Core i5 and i7 CPUs, but RAM quotas vary widely from 4GB to 16GB. However, they all have 120 Hz+ screens, with two models sporting 16-inch 1600p displays while the other has a 1080p touch panel. Two have anti-ghosting technology keyboards, but they all have RGB backlighting capabilities. Pricing varies from $399 to $699.

Another interesting commonality we see with these launch devices is that if the RGB gaming keyboards aren’t set to stun, they won’t look out of place in a corporate or business environment. Moreover, they all seem to have benefitted from some minimal and sleek design. Remember, without particularly beefy CPUs and GPUs on board, these are all pretty slim portables.

In addition to the  laptop hardware, or should we say Gaming Chromebook hardware, Google lists a number of certified for Chromebook gaming peripherals on its new ‘discover gaming’ microsite.

It is easy to be cynical about Google starting a new project, particularly with Stadia in the rearview mirror. In our report on the death of Stadia we mentioned that KilledByGoogle reckons Google's game streaming service was the 274th project shuttered by the company. Logically, such a fate shouldn’t face Gaming Chromebooks unless all Chromebooks get the chop. Prerequisites such as faster networking and more responsive displays are already becoming mainstream, as are keyboard backlighting implementations. Therefore, it could be possible in the not too distant future for Google to claim all Chromebooks are Gaming Chromebooks.

The Secure Shell Protocol (SSH) is perhaps the most well-known means to make a secure connection between a client machine (your laptop, phone or desktop) and a remote server in an office, data center or in your home network. You’ll likely use SSH if you want to get to the command line on your web hosting service or a headless Raspberry Pi. SSH is available in some form for nearly every operating system, and often it is integrated into the OS.

Most servers give you a choice of connecting to SSH via a password or via SSH keys, which are more secure. The SSH key method uses cryptographically-generated public and private keys to create an encrypted connection between devices.

Our public key is stored on the remote machine and a private key is stored on our machine. The two SSH keys are required to make a secure connection. Keys can also be used with passphrases to add another level of security, but they can also be used without, for example in automated processes.

In this how to we will learn how to create SSH keys using PuTTY, the most popular SSH client, and at the Windows Command Prompt / Linux Terminal.

Preparing the Remote Server for SSH Keys

Our remote machine can be in a data center run by a web hosting service, our office or home. Typically Linux servers such as VPS and cloud hosting will have SSH running by default, using passwords for secure logins. If this is not the case, you will need to enable SSH via the control panel for your VPS / cloud service. If you are using a home server, then it is possible that it may not be installed. If that is the case, follow these steps before moving onwards.

1. Open a terminal and check for a running SSH service on the machine. If the SSH service is running it will return Active: active (running).

sudo service ssh status

2. On the physical server, open a terminal and install OpenSSH Server. You will need to be sat in front of the machine to issue these commands.

sudo apt updatesudo apt install openssh-server

3. Start the SSH service.

sudo service ssh start

4. In your home directory create a hidden directory called .ssh.

mkdir .ssh

5. Close the connection by pressing CTRL+D or typing exit and pressing Enter.

Using PuTTY to Connect to a Remote Server Using SSH Keys

PuTTY is an application to create and manage SSH and serial connections to devices. PuTTY comes with its own key generator application and in this part of the how to we will create a public for our remote server and private key on our trusted device.

1. Download and install PuTTY. Windows does have its own command prompt, which can be used with SSH, but PuTTY is by far the most accessible means to open an SSH connection.

2. Search for puttygen and open the application.

3. Click on RSA and set the bits to 4096. Click Generate to create a key.

4. Move the mouse around the blank area of the dialog to generate a random seed for the key.

5. Create a passphrase for the key. This is advised for SSH keys that will be used in interactive sessions.

6. Save the public key as id_rsa_putty.pub to a folder called .ssh.

7. Save the private key as id_rsa_putty.ppk. The ppk file is Putty’s own private key format.

8. Highlight the public key and copy the text.

Copying the Public Key to the Remote Server

1. Launch PuTTY.

2. Enter the IP address or hostname for your remote server and click Open.

3. When prompted, enter your username and password for the remote server. Note that the password is not shown. This is a security feature to prevent “shoulder surfing”.

4. Using the nano text editor, create a new file called authorized_keys in the .ssh directory.

5. Right click and paste the public key (created in PuTTYGen) into the blank file. Save by hitting CTRL+X,then Y and Enter.

6. Log out of the SSH session by pressing CTRL + D.

7. Re-Open PuTTY and go to SSH >> Auth.

8. Select the Putty Private Key (ppk) that we just created.

9. Scroll back to Session, enter the hostname / IP address for your server and click Open to start a connection.

10. Enter your username, then the passphrase for your key. Press Enter to login.

You are now in control of the remote server, using a Linux terminal working in the Linux filesystem. To close a connection use CTRL + D or click on the X to close the window.

How To Create an SSH Key Pair via the Command Prompt / Terminal

Creating an SSH key pair from the command prompt / Linux terminal can be done with just one command and a few questions to answer. We set up the public and private keys on our client machine, copying the public key to the remote server. These steps apply to the Windows command prompt and the Linux terminal.

1. Open a Command Prompt by pressing the Windows key and search for CMD. Press Enter to run.

2. Use the ssh-keygen command to create a SSH key using the RSA key type, with 4096 bits.

ssh-keygen -t rsa -b 4096

3. When prompted to name your key, press Enter. This will save the private and public key to the .ssh directory for your named account. For example our keys were saved to C:\Users\lespo\.ssh. Linux users, the keys will save to .ssh in your home directory \home\user\.ssh

4. When prompted, give your key a passphrase as an extra level of security. A passphrase is an additional security step for SSH keys that will be used by real users (interactively). A passphrase is not needed if the SSH connection will be used in an automated script.

5. Change directory to the location of your SSH keys. Here we assume that you are in your named account. For example C:\Users\lespo is ours. Under Linux it would be /home/les/.ssh

cd .ssh

6. List the files in the directory. There should be id_rsa and id_rsa.pub.

Windows CommanddirLinux Commandls

Copying the Public Key to the Remote Server

The public key is stored on our remote server, and it interacts with the private key on our trusted machine to form a secure connection. In order to get the public key to our server we need to securely copy (scp) the file across.

1. In a Command Prompt use the scp command to securely copy the id_rsa.pub to your home directory on the remote server. You will need to know the IP address or hostname of the remote computer. In our example we copied the file to testuser@192.168.0.10:/home/testuser/

scp id_rsa.pub user@hostname:/home/username

2. SSH into the remote computer.

ssh user@hostname

3. Verify that the id_rsa.pub file is present in your home directory.

ls *.pub

4. Copy the contents of the file into a new file in the .ssh directory. Using the cat command we send the contents to the file, authorized_keys using a pipe that appends the data to the file (>>).

cat id_rsa.pub >> .ssh/authorized_keys

5. Close the SSH connection by pressing CTRL + D or by typing exit.

6. Reconnect via SSH to the remote computer. If you created a passphrase for your SSH key, you will be prompted for it.

Using SSH Keys on Another Machine

Reusing your private SSH key is possible, but it isn’t the best security practice. Losing a laptop with the key means that you will need to regenerate your keys. Best practice would be to generate a new key pair for each device that wishes to connect. Appending the public key to the authrozied_keys file on the server. That said, it is relatively easy to reuse a private key across multiple devices.

Reusing a PuTTY Private Key

PuTTY stores the private key as a PPK key and this file is all we need for a machine to connect to a remote server using our public key.

1. Go to your .ssh folder and copy the PPK file to a USB stick. Keep this USB stick safe as it can be used by anyone to gain access to your server.

2. Insert the USB stick into another computer.

3. Create an .ssh folder in your home directory. For windows this would be C:\Users\username\.ssh and Linux /home/user/.ssh.

4. Copy the PPK file to the .ssh directory.

5. Follow from Step 7 of Copying the Public Key to the Remote Server to login.

Reusing a Command Prompt / Terminal Private Key

1.  Go to your .ssh folder and copy the private key file to a USB stick. Keep this USB stick safe as it can be used by anyone to gain access to your server.

2. Insert the USB stick into another computer.

3. Create an .ssh folder in your home directory. For windows this would be C:\Users\username\.ssh and Linux /home/user/.ssh.

4. Copy the private key file to the .ssh directory.

5. Start an SSH session to the remote computer. If you created a passphrase for your SSH key, you will be prompted for it. 

ssh user@hostname