Johnathan Smith

A group of flight sim enthusiasts bought an old Boeing 747 cockpit to convert it into the ultimate flight sim setup.

Western Digital Blue and Red HDDs from 2020 that use SMR technology are experiencing enough failures to prompt an investigation from WD itself. These same drives, which included SMR without telling customers, resulted in a class action lawsuit against WD in 2021.

Bajaj Auto has revealed plans to cut workforce by over 50% across several areas at KTM, including motorsport, as it inches closer to completing a takeover of the Austrian manufacturer.
Bajaj’s managing director Rajiv Bajaj stated in an interview that it intends to halve expenses across various departments in a move that will directly affect KTM and Tech3’s MotoGP teams.
“This is really ...Keep reading

After a whooping cough outbreak killed two infants, Louisiana health officials waited months to officially alert physicians or do public outreach. That's not the typical public health response.

(Image credit: Rebecca Blackwell)

Why monitor staff through phones or cameras when Bezos' boxshifter can strap surveillance to their heads?

Amazon is testing AI-powered smart glasses to help its drivers get from their vans to customers' doorsteps.…

Ebola is one of the nasty viruses that can hide in the body even after a patient recovers and tests negative. It can reemerge and trigger a new outbreak years later. How do they survive? And how can they be kayoed?

When your best engineers log off for good, don’t be surprised when the cloud forgets how DNS works

column  "It's always DNS" is a long-standing sysadmin saw, and with good reason: a disproportionate number of outages are at their heart DNS issues. And so today, as AWS is still repairing its downed cloud as this article goes to press, it becomes clear that the culprit is once again DNS. But if you or I know this, AWS certainly does.…

Millions of Americans use weed to treat chronic pain, but there's little high quality research on whether it works. New findings suggest it can be effective for low back pain, on par with opioids.

(Image credit: Tinnakorn Jorruang/iStockphoto)

It seems like Microsoft has encountered a significant issue with the latest Windows 11 25H2 October update, KB5066835. The company confirmed that this update disrupts mouse and keyboard functionality within the Windows Recovery Environment (WinRE), making them unresponsive and unusable. As a result, the WinRE feature is completely inoperative. WinRE is a built-in troubleshooting toolkit included with Windows. It's intended to assist users when their computer encounters startup problems or system issues. WinRE activates automatically when Windows crashes or fails to boot properly, but users can also access it manually to utilize various repair tools.

However, with the current problem affecting keyboard and mouse input, WinRE is essentially ineffective. Microsoft stated that "the USB keyboard and mouse continue to work normally within the Windows operating system," and assured users that they are "working to release a solution to resolve this issue in the coming days. More information will be provided when it is available." This is yet another incident related to the recent Windows 11 updates, which have previously caused localhost issues. The list of Windows 11 problems continues to grow as the latest updates are released. Microsoft maintains a Windows 11 version 25H2 known issues and notifications website that provides status updates on the latest problems, including this one.

Update October 21 16:05 UTC: Microsoft has issued a hot fix. The company claims that "this out-of-band (OOB) update includes quality improvements. This update is cumulative and includes security fixes and improvements from the October 14, 2025, security update (KB5066835)..." which you can check out here.

French Interior Minister Laurent Nuñez called the heist "a major, highly organized operation" that lasted just seven minutes. Authorities were still drawing up the value of the stolen items.

(Image credit: Alexander Turnbull)

Hello you fine Internet folks!

Today we are looking at AMD’s largest client APU to date, Strix Halo. This is an APU designed to be a true all-in-one mobile processor, able to handle high end CPU and GPU workloads without compromise. Offering a TDP range of 55W to 120W, the chip targets a far higher power envelope compared to standard Strix Point, but eschews the need for dedicated graphics.

To get y’all all caught up on the history and specifications of this APU, AMD first announced Strix Halo at CES 2025 earlier this year to much fanfare. Strix Halo is AMD’s first chiplet APU in the consumer market with AMD using Strix Halo as a bit of a show piece for what both CPU and GPU performance can look like with a sufficiently large APU.

AMD’s Strix Halo can be equipped with dual 8 core Zen 5 CCDs for a total of 16 cores that feature the same 512b FPU as the desktop parts. This is a change from the more mainstream and monolithic Strix Point APU which has “double-pumped” 256b FPUs similar to Zen 4 for use with AVX512 code. What is similar to the more mainstream Strix Point is the same 5.1GHz max boost clock which is a 600MHz deficit compared to the desktop flagship Zen 5 CPU, the Ryzen 9 9950X.

Moving to the 3rd die on a Strix Halo package, a RDNA 3.5 iGPU takes up the majority of the SoC die with 40 compute units, 32MB of Infinity Cache, and a boost clock of up to 2.9GHz placing raw compute capability somewhere between the RX 7600 XT and RX 7700.

To feed this chip, AMD has equipped Strix Halo with a 256b LPDDR5X-8000 memory bus, which provides up to 256GB/s shared between all of the components. This is slightly lower than the 288GB/s available to the RX 7600 XT but is much higher than any other APU we have tested.

Acknowledgments

A massive thank you to both Asus and HP for sending over a ROG Flow Z13 (2025) and a ZBook Ultra G1a 14” for testing which were both equipped with an AMD Ryzen AI Max+ 395. All of the gaming tests were done on the Flow Z13 due to that being a more gaming focused device and all of the microbenchmarking was done on the ZBook Ultra G1a.

Memory Subsystem from the CPU’s Perspective

Starting with the memory latency from Zen 5’s perspective, we see that the latency difference between Strix Point and Strix Halo is negligible with Strix Point at ~128ns of memory latency and Strix Halo at ~123ns of memory latency. However, as you can see the CPU does not have access to the 32MB of Infinity Cache on the IO die. This behavior was confirmed by Mahesh Subramony during our interview about Strix Halo at CES 2025.

While the 123ns DRAM latency seen here is quite good for a mobile part, desktop processors like our 9950X fare much better at 75-80ns.

Moving on to memory bandwidth, we see Strix Halo fall into a category of its own of the SoCs we have tested.

When doing read-modify-add operations across both CCDs, the 16 Zen 5 cores can pull over 175GB/s of bandwidth from the memory with reads being no slouch at 124GB/s across both CCDs.

However, looking at the bandwidth of a single CCD and just like the desktop CPUs a single Strix-Halo CCD only has a 32 byte per cycle read link to the IO die. And just like the desktop chips, the chip to chip link runs at ~2000MHz, which caps out the single CCD read at 64GB/s. Unlike the desktop chips, the write link is 32 bytes per cycle and we are seeing about 43GB/s for the write bandwidth. That brings the total theoretical single CCD bandwidth to 128GB/s and the observed bandwidth is just over 103GB/s.

CPU’s Performance

The performance of Strix Halo’s CPU packs quite a bit more of a punch than Strix Point’s CPU.

Strix Halo’s CPU can match a last generation desktop flagship CPU, the 7950X, in Integer performance despite a 11.7% clock speed delta. And nearly matches AMD current desktop flagship CPU, the 9950X, in Floating Point performance again with a 11.7% clock speed deficit.

Looking at the SPEC CPU 2017 Integer subtests and while Strix Halo can’t quite match the desktop 9950X, likely due to the higher memory latency of Strix Halo’s LPDDR5X bus, it does get close in a number of subtests.

Moving to the FP subtests and the story is similar to the Integer subtests but Strix Halo can get even closer to the 9950X and even beat it in the fotonik3d subtest.

Memory from the GPU’s Perspective

Moving to the GPU side of things and this is where Strix Halo really shines. The laptop we used as a comparison to Strix Halo was the HP Omen Transcend 14 2025 with a 5070M equipped which maxed out at about 75 Watts for the GPU.

Strix Halo has over double the memory bandwidth of any of the other mobile SoCs that we have tested. However, the RTX 5070 Mobile does have about 50% more memory bandwidth compared to Strix Halo.

Looking at the caches of Strix Halo, the Infinity Cache, AKA MALL, is able to deliver over 40% higher bandwidth compared to the 5070M’s L2 while having 33% more capacity. Plus Strix Halo has a 4MB L2 which is capable of providing 2.5TB/s of bandwidth to the GPU.

Moving to the latency, the more complex cache layout of Strix Halo does give it a latency advantage after the 128KB with Strix Halo’s L2 being significantly lower latency than the 5070M’s L2 and the larger 32MB MALL that Strix Halo has a similar latency to the 5070M’s L2. And Strix Halo’s memory latency is about 35% lower than the 5070M’s memory latency.

The GPU’s Compute Throughput

Looking at the floating point throughput, we see that Strix Halo unsurprisingly has about 2.5x the throughput of Strix Point considering it has about two and a half times the number of Compute Units. Strix Halo oftentimes can match or even pull ahead of the 5070 Mobile in terms of throughput. I will note that the FP16 results for the 5070 Mobile are half of what I would expect; the FP16:FP32 ratio for the 5070 Mobile should be 1:1 so I am not positive about what is going on there.

Moving to the integer throughput and we see the 5070 Mobile soundly pulling ahead of the Radeon 8060S.

GPU Performance

Looking at the GPU performance, we see Strix Halo once again shine, with a staggering level of performance available for an iGPU, courtesy of the large CU count paired with relatively high memory bandwidth. Our comparison suite includes several recent iGPU’s from Intel/AMD, along with the newest generation RTX 5070 Mobile @ 75W to act as a reference for mid to high-range laptop dedicated graphics, and the antiquated GTX 1050 as a reference for budget dedicated graphics.

Looking at Fluid X3D for a compute-heavy workload, we can see the Radeon 8060S absolutely obliterates the other iGPU’s from Intel/AMD, putting itself firmly in a class above. The 5070 is no slouch though, and still holds a substantial 64.1% lead largely due to the higher memory bandwidth afforded to the 5070M.

Switching to gaming workloads with Cyberpunk 2077, we start with a benchmark conducted while on battery power. The gap with other iGPU’s is still wide, but now the 5070M is limited to 55W and exhibits 7.5% worse performance at 1080p low settings when compared to the Radeon 8060S.

Finally, moving to wall power and allowing both the Radeon 8060S and 5070M to access the full power limit in CP2077, we can see that the 8060S still pulls ahead at 1080p low by 2.5%, while at 1440p medium we see a reversal, with the 5070M commanding an 8.3% lead. Overall the two provide a comparable experience in Cyberpunk 2077, with changes in settings or power limits adjusting the lead between the two. This is a seriously impressive turnaround for an iGPU working against dedicated graphics, and demonstrates the versatility of the chip in workloads like gaming, where iGPU’s have traditionally struggled.

Conclusion

Strix Halo follows in the footsteps of many other companies in the goal of designing an SoC for desktop and laptop usage that is truly all encompassing. The CPU and GPU performance is truly a class above standard low power laptop chips, and is even able to compete with larger systems boasting dedicated graphics. CPU performance is especially impressive with a comparable showing to the desktop Zen 5 CPUs. GPU performance is comparable to mid range dedicated graphics, while still offering the efficiency and integration of an iGPU. High end dedicated graphics still have a place above Strix Halo, but the versatility of this design for smaller form factor devices is class leading.

However, this is not to say that Strix Halo is perfect. I was hoping to have a section dedicated to the ML performance of Strix Halo in this article, however AMD only just released preview support for Strix Halo in the ROCm 7.0.2 release which came out about a week ago from time of publication. As a result of the long delay between the launch of Strix Halo and the release of ROCm 7.0.2, the ML performance will have to wait until a future article.

However, putting aside ROCm, Strix Halo is a very, very cool piece of technology and I would love to see successors to Strix Halo with newer CPU and GPU IP and possibly even larger memory buses similar to Apple’s Max and Ultra series of SoCs with 512b and 1024b memory respectively. AMD has a formula for building bigger APUs with Strix Halo, which opens the door to a lot of interesting hardware possibilities in the future.

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.

Intel has secured a significant order from Microsoft, putting trust from external customers into Intel Foundry. According to Charlie Demerjian from SemiAccurate—and shared by X user @Jukanlosreve—Microsoft will manufacture its Maia 3 chip codenamed "Griffin" at Intel Foundry using the 18A or 18A-P process. This 18A-P node improves upon Intel's 18A by incorporating the RibbonFET and PowerVia technologies, offering enhanced performance and energy efficiency. Compare to the regular 18A node, improvements include newly designed low-threshold voltage components, optimized elements to reduce leakage, and refined ribbon width specifications, all aimed at boosting performance-per-watt metrics. This is particularly important for AI accelerators used in data centers, which require high efficiency due to the large number of units involved.

Microsoft could relocate the production of future Maia accelerators, following Maia 3, to Intel Foundry. If the Maia 3 project is successful, they plan to continue with Intel's more advanced nodes. As a reminder, Microsoft's first-generation Maia 100 chips were produced using TSMC's N5 process with CoWoS-S interposer technology. These chips feature an 820 mm² die size and operate at a 500 W TDP, with a maximum design power of 700 W. They include 64 GB of HBM2E memory providing 1.8 TB/s bandwidth and 500 MB of L1/L2 cache. The chips achieve peak tensor performance of 3 PetaOPS at 6-bit precision, 1.5 PetaOPS at 9-bit, and 0.8 PetaFLOPS for BF16. Connectivity options include 600 GB/s backend network bandwidth via twelve 400GbE ports and 32 GB/s host bandwidth through PCIe Gen 5 x8.

Update October 18, 17:30 UTC: Charlie Demerjian confirmed that the name of the chip is actually a third generation Maia 3 accelerator codenamed "Griffin."

Recovery feature lets trusted contacts help you get back in when other methods fail

The latest security feature for Gmail enables users to recover their accounts with a little help from their friends.…

It’s finally time to say goodbye to Windows 10.Microsoft has confirmed that support for the operating system officially ends on October 14, 2025, closing out almost ten years of service.

A blast in rural Tennessee that leveled an explosives plant and was felt for miles around killed 16 people and left no survivors, authorities said.

(Image credit: Obed Lamy)

Viva Insights turns AI guzzling into a leaderboard

Microsoft is adding Copilot adoption benchmarks to Viva Insights, a tool that lets managers monitor teams to spot those that are gulping down the AI Kool-Aid fastest.…

When sellers collude through a computer algorithm, that doesn't make it right

California companies that use algorithms to fix the prices of their products and services could now face stiff antitrust penalties if they continue to do so. …

The soap opera that is Nvidia's GB10/N1X goes on with a pathetic publicity stunt.
Read more ▶

---

The post Nvidia Pulls A PR Stunt For The GB10/N1X ‘Release’ appeared first on Semiaccurate.

Intel officially revealed the Core Ultra 300 “Panther Lake” processor architecture at its 2025 Annual Technology Tour in Arizona, offering a technical deep dive into its next-generation SoC design.

Since revealing Stargate in January, Altman and friends have brought about 200 MW online - they'll need at least 16 GW to claim their red and green prize

Comment  AMD has issued OpenAI a warrant for roughly 10 percent of its stock. In exchange, the AI model giant will work with its partners (such as Oracle) to deploy up to 6 gigawatts' worth of AMD GPUs.…

Synology has backtracked on one of its most unpopular decisions in years. After seeing NAS sales plummet in 2025, the company has decided to lift restrictions that forced users to buy its own Synology hard drives.

Qualcomm Technologies, Inc. today announced its agreement to acquire Arduino, a premier open-source hardware and software company. The transaction accelerates Qualcomm Technologies' strategy to empower developers by facilitating access to its unmatched portfolio of edge technologies and products. This acquisition builds on the Company's recent integrations of Edge Impulse and Foundries.io, reinforcing its commitment to delivering a full-stack edge platform that spans hardware, software, and cloud services. The closing of this transaction is subject to regulatory approval and other customary closing conditions.

By combining Qualcomm Technologies' leading‑edge processing, graphics, computer vision, and AI with Arduino's simplicity, affordability, and community, the Company is poised to supercharge developer productivity across industries. Arduino will preserve its open approach and community spirit while unlocking a full‑stack platform for modern development—with Arduino UNO Q as the first step.

OpenAI has secured up to 6 gigawatts of AMD GPU compute in a landmark supply agreement that could see the ChatGPT maker take a 10% stake in AMD.

AMD and OpenAI have announced a multibillion-dollar partnership to collaborate on AI data centers powered by AMD chips.

Kessler syndrome is bad; atmospheric incineration may be worse, says astrophysicist Jonathan McDowell

If you had to guess how many Starlink satellites burn up in Earth's atmosphere on an average day, how many would you pick? This isn't a trick question - SpaceX is deorbiting about one or two satellites daily, and that number is only going to grow. …

Two groups are calling for new leadership at HHS after Health Secretary Robert F. Kennedy Jr.'s actions on substance abuse treatment and mental health medications, among other issues.

(Image credit: Win McNamee)

The Centers for Disease Control and Prevention accepted a controversial recommendation from outside vaccine advisers to tighten guidelines for the COVID vaccine.

(Image credit: FREDERIC J. BROWN)

A few days ago, a financial note went around saying AMD is going to fab chips at Intel Foundry.
Read more ▶

---

The post Is AMD fabbing at Intel Foundry? appeared first on Semiaccurate.

Access to the COVID-19 vaccines remains difficult because of an unusual and unexplained delay by the Centers for Disease Control and Prevention in accepting recommendations from its advisers.

(Image credit: Joseph Prezioso/AFP)

What looks like an oversized zeppelin is actually a flying power plant and possibly the first step into a new era of power generation. China’s latest airship project, the “S1500” from Beijing Sawes Energy Technology, completed a historic test flight in the Xinjiang desert in September 2025 – and achieved an output of over one […]

Source