Johnathan Smith

Smithsonian warns that dismantling orbiter for relocation is history in the wrecking

How would you move Space Shuttle Discovery from Virginia to Texas? The White House Office of Management and Budget asked NASA and the Smithsonian Institution and the response was to dismantle it.…

Samsung and SK hynix have agreed to supply undiced DRAM wafers to OpenAI’s Stargate project, which could consume up to 900,000 wafers per month, or about 40% of global DRAM output, as Stargate is gearing up for unprecedented expansion.

The Norwegian browser company Opera is shipping Opera Neon to the first users today: it's a browser built to not only let you browse the web, but to also use agentic AI to act for you and with you as you browse and work on complex projects. Opera Neon moves beyond a simple AI chat to execute tasks, create code, and deliver outcomes directly within the browser experience.

"We built Opera Neon for ourselves - and for everyone who uses AI extensively in their day-to-day" said Krystian Kolondra, EVP Browsers at Opera. "Today, we're welcoming the first users who will help shape the future of agentic browsing with us".

Alleges bias and security problems

US President Donald Trump has demanded Microsoft fire its recently appointed head of global affairs Lisa Monaco.…

In early 2023, RTINGS launched one of the most ambitious TV reliability tests to date, placing 100 LCD and OLED televisions under extreme conditions to simulate a decade of use.

The higher the wage level, the more entries they'll get in the draw

The H-1B lottery in the US is being tipped heavily in favor of high-wage earners under a long-awaited rule proposal unveiled on Tuesday.…

Protected content in some Blu-ray and DVD applications broken

Microsoft has added another entry to its growing list of problematic updates in the Windows Hall of Shame, this time causing Digital TV and Blu-ray applications to stutter and freeze when playing protected content.…

RFK Jr.'s reshaped ACIP vaccine panel re-did a vote from yesterday on the MMRV vaccine and scrapped plans for another vote on the hepatitis B birth dose.

(Image credit: Elijah Nouvelage)

Teams with UAV operator Flytrex for service that moves meals in minutes

Flying pigs may soon be on their way to some US households, after rideshare and food delivery behemoth Uber teamed with drone operator Flytrex for food delivery services.…

Intel has clarified its position regarding its graphics roadmap after announcing a major partnership with Nvidia. While the two companies will cooperate on developing x86 processors aimed at both consumer PCs and large-scale data centers, Intel stressed that this collaboration does not signal an end to its own GPU ...

Microsoft reminds holdouts they've got less than a month before the update tap runs dry

Start the countdown! For any administrators living under a rock, Microsoft has posted another warning that Windows 10 22H2 will reach end of servicing on October 14.…

The group was chosen by Health Secretary Robert F. Kennedy Jr. amid controversy over their credentials. It's changing guidance for vaccines for measles, mumps, rubella and chickenpox.

(Image credit: Elijah Nouvelage)

NVIDIA's surprise $5 billion investment in Intel today came with an unexpected revelation - the two companies have been quietly working together for almost a year on fusing x86 CPUs with RTX and data center GPUs through NVLink. The result? Actual system-on-chip designs that could finally break the PCIe bottleneck that's been holding back AI servers. NVIDIA will handle the heavy lifting on design and manufacturing of these hybrid chips, integrating NVIDIA's NVLink directly into Intel's x86 silicon. It's basically the same approach NVIDIA already uses with their Vera processors (Arm + Blackwell GPUs), except now they're doing it with Intel's x86 cores instead of custom Arm designs. Anyone who's worked with current GPU servers knows the pain points. PCIe connections between CPUs and GPUs create bandwidth choke points, add latency, and make memory management a nightmare for AI workloads. These new chips bypass all that with direct GPU-CPU communication and shared memory pools.

The target market isn't just data centers either. Intel mentioned both server and client applications, which suggests we might see this tech trickle down to gaming laptops and workstations eventually. For now though, the focus is clearly on machine learning clusters and HPC installations where PCIe bandwidth is already maxed out. AMD won't be thrilled about this development. They've been pushing their own CPU-GPU integration story, but this Intel-NVIDIA combo could leapfrog their efforts entirely. The manufacturing question remains murky though. When pressed about using Intel's fabs for production, Intel CEO Lip-Bu Tan gave a diplomatic non-answer about "perfecting the process" first. Reading between the lines, TSMC will probably keep making the actual chips for both companies, at least initially. Jensen said that basically for the start, NVIDIA will buy a CPU chip then sell a unified CPU plus GPU chiplet.

NVIDIA and Intel today announced one of the largest industry collaborations in recent years, jointly developing multiple generations of consumer PC and data center products, utilizing the best design teams from both companies. According to the latest announcement, NVIDIA is investing $5 billion at $23.28 per share, making NVIDIA one of Intel's largest shareholders, with a 4.9% stake in the total company. The goal of this collaboration is split into two verticals. One—and perhaps the most interesting one—is the integration of NVIDIA RTX GPUs inside Intel x86 System-on-Chips (SoCs), which Intel's own Arc GPUs previously powered. Intel once even considered tapping AMD with its "Kaby Lake G" design, which never progressed beyond generation one. Now, NVIDIA's RTX GPUs will become the standard for integrated graphics, powering millions of laptops, handheld devices, and possibly even desktop processors.

The second pillar of this collaboration is the development of custom x86 CPUs for NVIDIA, which will be integrated into NVIDIA's AI infrastructure platforms. Spanning multiple verticals such as DGX workstations, HGX servers, and NVIDIA SuperPODs. NVIDIA currently uses a mix of its self-developed Arm-based Grace/Vera CPUs, as well as x86 CPUs in some HGX systems. For now, this has served NVIDIA well, and its data center roadmap indicates that NVIDIA will continue to ship these custom CPUs through its product family. However, with custom x86 CPUs fine-tuned for NVIDIA, Intel will gain significant share in the AI training and inference infrastructure, where NVIDIA previously completely bypassed both x86 vendors with custom Arm designs. For example, Intel's designs, such as Clearwater Forest Xeons, are coming soon with up to 288 "Darkmont" E-cores on a massive chiplet package using the 18A node. This should be a perfect companion for NVIDIA's accelerators, and we could even see some customized versions that offer lower core counts but higher frequencies, all within NVIDIA's SuperPODs with hundreds of GPUs.

Health Secretary Robert F. Kennedy Jr. picks more new vaccine advisers to the Centers for Disease Control and Prevention, days before a two-day meeting to consider COVID and hepatitis B shots.

(Image credit: Ben Hendren)

Chinese authorities have accused Nvidia of violating anti-monopoly laws, threatening the US chipmaker with hefty fines. However, with trade negotiations between America and China hinging on access to high-end hardware like Nvidia's, this may be a warning shot to encourage more favorable trading terms.

A combat sim creator built an air combat sim gaming rig that moves with you and shoots guns, missiles, and chaff.

If an employer asks you do to this, demand a trial run so you can learn the rules of this strange new world

A startup called Job Bolt has created AI avatars that conduct job interviews. The Register couldn't help but give it a try and can report that it's an unnerving experience.…

Agency wants a single private integrator to herd 74K bits of kit with only 40% of the funding so far

Get ready to start flying American skies with a renewed sense of confidence, at least eventually, as the Federal Aviation Administration has finally decided to start soliciting ideas for an overhaul of the US' antiquated air traffic control systems. In classic Trump administration style, the FAA wants a single private-sector integrator to run the overhaul, with the public footing the bill. …

Though the FDA narrowed criteria for the shots, many people still want them, to avoid illness and protect vulnerable family members. Some are turned down at the pharmacy or have to jump through hoops.

(Image credit: Therese McRae; Stephan Neidenbach; and Jason Mitton)

TCL China Star Optoelectronics Technology (TCL CSOT), a global leader in display technologies, is set to unveil its groundbreaking display technologies at IFA 2025. Under its premium APEX display technology brand, the company will showcase its pioneering Inkjet Printing OLED (IJP OLED) technology and a curated product portfolio, demonstrating its commitment to creating sustainable, user-centric visual experiences that blend cutting-edge performance with environmental responsibility.

"Our presence at IFA 2025 is a testament to our vision for transforming visual technology," said Jun Zhao, Chief Executive Officer of TCL CSOT. "We're not just creating advanced displays - we're developing solutions that enhance how people work, create, and connect while maintaining our dedication to sustainability and user well-being."

Reports of widespread SSD failures following Microsoft’s August Windows 11 update have taken another turn. According to PCDIY!, a Chinese Facebook community focused on PC hardware, the issue was traced back to engineering firmware on certain Phison-based SSDs.

A district judge in Virginia was specially tapped to oversee the unusual case after DOJ named all 15 federal district court judges in Maryland as defendants in a lawsuit related to deportations.

(Image credit: Michael M. Santiago)

As stated on August 18, Phison was made aware of the KB5063878 and KB5062660 updates on Windows 11 that potentially impacted several storage devices, including some supported by Phison. In response, Phison dedicated over 4,500 cumulative testing hours to the drives reported as potentially impacted and conducted more than 2,200 test cycles. We were unable to reproduce the reported issue, and no partners or customers have reported that the issue affected their drives at this time.

Phison remains committed to the highest standards of reliability and continues to closely monitor the situation in collaboration with our industry partners.

Okay, who has the gold leaf, paint, and #FFD700?

US government websites are getting an aesthetic and functional overhaul under a Trump executive order and a new "America by Design" initiative headed by a Silicon Valley veteran and DOGE insider. …

Intel used Hot Chips to pull back the curtain on "Clearwater Forest", its next-generation E-core Xeon. The company has paired major architectural upgrades with its newest 18A process and advanced 3D Foveros packaging paired with EMIB. The chip replaces "Sierra Forest" and is built around the new "Darkmont" efficiency cores. Intel says Clearwater Forest will ship in a massive chiplet package using the 18A node for the compute pieces while relying on Foveros Direct 3D for stacking. The company is pitching this design as a showcase of what its foundry and packaging teams can now deliver.

Under the hood, the Darkmont E-core leans heavily on a wider, more parallel front end and a beefed-up out-of-order engine. Intel disclosed a 64 KB instruction cache per core and an expanded decoder that can handle more instructions per cycle than the previous "Crestmont" E-core design. Reorder and allocation structures have evolved, allocation units have been increased, and the out-of-order window has been enlarged to accommodate more in-flight work. Execution resources were also multiplied, with many more execution ports for integer and vector math, allowing the cores to sustain much higher parallel throughput.

Liquid cooling is a familiar concept to PC enthusiasts, and has a long history in enterprise compute as well. Recently, liquid cooling has taken an increasing role in datacenters, amid increasing power draw and correspondingly high heat output from the latest chips. Machine learning in particular has an insatiable appetite for power and cooling. Google notes that water has a thermal conductivity about 4000 times that of air, making it an attractive solution to deal with the cooling demands associated with the current AI boom. Their talk at Hot Chips 2025 focuses on datacenter-level cooling for their TPUs, which are machine learning accelerators.

Google’s foray into liquid cooled TPUs took form in 2018 after some experimentation and iteration. The company has continued to develop and advance their cooling designs since. Their current liquid cooling solution is designed for datacenter scale, with liquid cooling loops spanning racks rather than being contained within servers. Racks of six CDUs, or Coolant Distribution Units, perform a role analogous to the radiator+pump combo in an enthusiast water cooling loop. The CDUs use flexible hoses and quick disconnect couplings to ease maintenance and reduce tolerance requirements. A CDU rack can provide adequate cooling capacity with five CDUs active, allowing maintenance on one unit without downtime.

CDUs exchange heat between coolant liquid and the facility-level water supply. The two liquid supplies don’t mix, and the CDUs only move heat between the two pools of liquid. Coolant liquid from the CDUs pass through manifolds that distribute the coolant to TPU servers. TPU chips are hooked up in series in the loop, which naturally means some chips will get hotter liquid that has already passed other chips in the loop. Cooling capacity is budgeted based on the requirements of the last chip in each loop.

Google uses a split-flow cold plate, which they found to perform better than a traditional straight-through configuration. To further optimize cooling, Google employs another techniques with parallels in the enthusiast world. TPUv4 switches to a bare-die setup compared to TPUv3’s lidded one. That mirrors delidding in the enthusiast world, where PC builders with especially strong stomachs remove heatspreaders (lids) to gain the higher heat transfer efficiency that bare die cooling can provide. For Google’s part, TPUv4 needs such an approach because it has 1.6x higher power draw compared to TPUv3.

Beyond moving heat off chips, liquid cooling helps reduce cooling-related power requirements. Google found the power consumption of liquid cooling pumps was less than 5% of fan power associated with an air cooling solution. Because Google uses water-to-water heat transfer to get heat off the coolant, the bulk of cooling power comes from pumps. Enthusiast setups don’t realize this potential advantage because they use liquid to move heat from chips to a more optimal location for air cooling. Usually, that’s a radiator with a fan mounted at a case intake. In theory, a PC builder could try exchanging heat to a toilet tank, allowing for highly efficient cooling with each flush. But fan and pump power in a water cooled PC is quite low in absolute terms, especially compared to the high RPM fans typically found in a server. Thus there’s little reason for an enthusiast to tackle cooling-related power.

Finally, maintenance is a concern. PC enthusiasts are familiar with water cooling challenges like microbial growth and leak risk. The same concerns apply to datacenter-scale water cooling solutions. Both worlds share some mitigating measures, like quick disconnect fittings. But datacenters have to address those challenges without taking down a substantial portion of compute capacity. PC builders with an open loop have to power down their machine before draining the loop and replacing components. Google’s extra CDU, mentioned earlier, allows maintenance with zero downtime. But Google takes measures beyond that, because a small problem with one machine can translate to a maintenance nightmare at scale.

Google extensively validates components with leak testing, uses alerting systems to discover problems like leaks, and takes preventative measures like scheduled maintenance and filtration. They also have a clear set of protocols to respond to alerts and issues, allowing a large workforce to respond to problems in a consistent fashion. It’s a far cry from the ad-hoc measures enthusiasts take to maintain their water cooling setups.

All in all, the recent rise in datacenter liquid cooling represents a fascinating crossover with high end enthusiast PCs. Both worlds are driven to liquid cooling by its incredibly efficiency when moving heat. Both must contend with similar problems. But datacenter approahces diverge from there, and driven by scale and reliability requirements. Plenty of water cooling hardware is already visible on the first day of Hot Chips 2025. Nvidia has a GB300 server on display, with external water cooling connections clearly visible. It also uses flexible tubing, and curiously deploys fans as well.

Rebellions AI, a South Korean company demo-ing a new ML accelerator, also has a water cooling setup. Their new “REBEL Quad” chip will eventually use air cooling in a PCIe card. But their demo uses a chiller and a cool looking water block, and it’s quite the sight. Zooming back up, all signs point to water cooling being here to stay as long as datacenter cooling requirements continue to intensify. The AI boom almost guarantees that will be the case.

If you like the content then consider heading over to the Patreon or PayPal if you want to toss a few bucks to Chips and Cheese. Also consider joining the Discord.

Are tech giants getting nervous? They should be

Opinion  There tend to be three AI camps. 1) AI is the greatest thing since sliced bread and will transform the world. 2) AI is the spawn of the Devil and will destroy civilization as we know it. And 3) "Write an A-Level paper on the themes in Shakespeare's Romeo and Juliet."…

But the court, in its emergency-docket order, left in place by a 5-4 order a lower court ruling that threw out National Institutes of Health memos that enforced the administration's policies.

(Image credit: Andrew Harnik)

According to the latest report from Bloomberg, the U.S. government is currently discussing taking a 10% stake in Intel Corp. Late last week, we reported on Intel's CEO Lip-Bu Tan's visit to the White House to discuss the future of Intel as America's most important semiconductor manufacturer. And today, parts of the deal seem to be near completion, according to sources close to Bloomberg. Reportedly, some or all of Intel's CHIPS Act grants are going to be converted by the federal government into an equity stake in the company, which could help fuel further Intel Foundry expansion and R&D investments. Intel's CHIPS Act funding has reportedly not yet been reflected in the balance sheet, as the company has issued warnings in the past that these grants have been slow to arrive. However, under the new initiative, the U.S. government now plans to use this grant funding as a way to get equity in Intel Corp and become its largest shareholder with a 10% share.

Historically, Intel has been awarded roughly $10.9 billion in CHIPS Act grants for commercial and military applications. However, that funding hasn't materialized yet. What the U.S. now plans is a 10% stake, roughly worth $10.5 billion, which would practically mean the same thing, with voting power over company decisions as its largest stakeholder.