Not only US.

To my understanding, Chinese GPU compute products are nowhere close to Nvidia (more than 5 year behind, if not more, in an apples to apples comparison).

Samsung’s troubles become evident with HBM3E. While SK hynix and Micron ramped 8-high and 12-high HBM3E for customers, Samsung struggled to get its 12-high stacks qualified. It reportedly took 18 months and several attempts to meet Nvidia’s quality and performance criteria for HBM3E. By Q3 2025, Samsung had finally cleared Nvidia’s validation, with its 5th-generation HBM3E 12-stack passing all tests.

Until now, Samsung HBM has appeared only in AMD’s MI300-series accelerators. However, with Nvidia’s certification, the company has agreed to purchase between 30,000 and 50,000 units of 12-high HBM3E for use in liquid-cooled AI servers. Samsung’s HBM3E is also shipping for AMD’s accelerators as of mid-2025.

A bit reminiscent of their fight against TSMC in the semiconductor fabrication space. If Intel does have return to form (which is a big if), Samsung might once again fight themselves in third place.

I love the irony of this statement. I wonder if the Palantir CEO recognizes that a similar statement could be made about him, his company and Huang/Nvidia.

For multi-core I would agree, but it seems like a viable benchmark. ST is critical for things like day-to-day application use, gaming and so on…

This flagship chip will reportedly be used only in premium models, like cars made by Hyundai’s sub-brand Genesis.

Hyundai reportedly chose an 8nm process for its chip due to its cost-effectiveness while still offering comparable performance to 5nm chips. The chip will be used in several cars across Genesis, Hyundai, and Kia brands. The 5nm chip will only be used in high-end cars.

Seems like a relatively small market, a quick web search suggests Hyundai sold 230 K Genesis brand cars in 2024 (a small number for their 5 nm order). I will speculate that Samsung had to offer Hyundai a very good deal as automotive is a highly competitive, low margin business. I highly doubt they would be willing to pay a premium.

That being said, it does looks like Samsung is trying to build out a business focusing on older (but still relatively modern) nodes. I wouldn’t be surprised if in the next 5-10 years, 7/8 nm and 5 nm become mainline nodes for many industries that don’t need to be on the leading edge of semiconductor fabrication.

Why do you say that? GB seems like a relatively balanced benchmark.

I also like Cinebench, but I find it’s better for specific use cases. I have my own DIY CPU gaming benchmark via running an old single thread game (Cities in Motion 1) with free look and custom map size that stretches the limits of the engine, but that a personal thing.

Not an expert on Apple’s CPU, but we are looking at single-thread results, I believe single thread results do not really scale across Apple’s computer CPU brand portfolio.

AFAIK, Geekbench scores are extremely CPU-specific and are not really relevant for GPU compute performance. We would need a different set of benchmarks for that.

A 6 year depreciation schedule seems unrealistically long for a GPU.

Even in gaming terms (I know this is completely different use case), a 2080S from 2019, a high end SKU, would struggle with many modern games at 1440p and higher. A profession streamer would be unlikely to use a 2080S.

Then there is the question of incentives. An objective look at American technology and VC suggests they are far closer to criminal organizations than their treatment by media and US institutions would imply. They very much can be expected to engage in what is essentially accounting fraud.

The original title “Apple M5 chip smashes Snapdragon X2 Elite in early single-thread benchmarks — single core scores rival Intel’s Core Ultra 9 285K and beat AMD’s 9950X3D, teasing multi-core potential of future variants” is misleading.

GB6 ST results:

• Apple M5: 4,263 (MacBook)
• Snapdragon X2 Elite: 4,080 (the result is likely misleading, as Qualcomm likes to post early results that can never be replicated in real world products, see the first X Elite results)

That being said 4,263 verses 4,080 is a mere 4.3% uplift, within the margin of error. I don’t how other people approach benchmarks, but I consider anything below 5% to be irrelevant. You want at least high single digit uplift or more realistically double digit uplift to notice a difference.

Tomshardware recently released a premium subscription. That’s fair, I think the best option is to pay directly for news sources. But, if you want people to pay you directly you must avoid these sort of scam-like, sensationalist headlines and show a measure of respect for your paying audience.

Considering the immense corruption inherent to modern commercial institutions, it’s probably for the best.

I think it’s our responsibility as voters and members of the community to have police forces that are respected, well paid and also who have allegiance to the people at large.

Getting there, that’s the real challenge.

According to Flock’s announcement, its Ring partnership allows local law enforcement members to use Flock software “to send a direct post in the Ring Neighbors app with details about the investigation and request voluntary assistance.” Requests must include “specific location and timeframe of the incident, a unique investigation code, and details about what is being investigated,” and users can look at the requests anonymously, Flock said.

“Any footage a Ring customer chooses to submit will be securely packaged by Flock and shared directly with the requesting local public safety agency through the FlockOS or Flock Nova platform,” the announcement reads.

In principle, allowing police to place requests for security camera data is not necessarily a bad thing. There can be legitimate use cases for such functionality. But this needs to be done without sketchy companies like Amazon or Flock being involved and broad legal safeguards for this process (e.g. requests can only be placed for serious crimes).

Faber is now best known for proposing the so-called “Forever Mouse” concept – a device with constantly evolving firmware features built on base hardware that customers would purchase only once. According to Faber, users would happily pay monthly or annual subscriptions to access these software-based upgrades. However, no concrete business plans have yet been announced to bring the idea to market.

It’s fascinating how myopic, vapid and delusional some CEOs can be.

Even a superficial analysis of this idea easily shows that the math simply does not work out (and that’s why they never brought the idea to the market).

I believe one the Chinese OEMs made a phone exactly like that. The camera extended from the top when needed.

Unfortunately, I don’t remember the name of the OEM or even the model brand.

EDIT: Another poster linked some models:

https://www.deepspecs.com/top-pop-up-camera-mobiles-in-world/

With all due respect, I think China would love a fully proprietary approach in the vein of American companies.

Their current use of open source (which is arguably not all encompassing) feels like a tactical decision more so than a strategic choice.

“Moving assembly is rather easy, but [shifting production] down to the component level is radical and very challenging, especially with a timeframe like 2026,” said another executive at a Microsoft supplier with knowledge of the situation. “We need to see how this ambition will be carried out.”

I am honestly surprised they are trying to shift all component production as well. That does sound extremely challenging. More business for countries such as Vietnam, Malaysia and Thailand I guess.

A much needed win for Intel’s foundry, but this feels more like a “testing the waters” project.

Amazon notes that X-Energy’s SMRs should be smaller, faster to deploy, and cheaper to operate than conventional pressurized water reactors. This is a common argument in support of the miniaturized nuclear power plants, but it’s worth noting that the tech hasn’t actually been proven out. In fact, higher-than-expected operating costs have already doomed one early SMR project.

The whole things sounds like a US oligarch’s pet project that’s driven by “vibes” as opposed to a conservative business plan and technology viability analysis.

This is pretty good news for those of who like to keep local collections of media.

In many ways, you can think of a datacenter’s use of hard drives as the ultimate test for a hard drive—you’re keeping a hard drive on and spinning for the max amount of hours, and often the amount of times you read/write files is well over what you’d ever see as a consumer. Industry trend-wise, drives are getting bigger, which means that oftentimes, folks are buying fewer of them. Reporting on how these drives perform in a data center environment, then, can give you more confidence that whatever drive you’re buying is a good investment.

Depends on the consumer. My x2 HDDs (7.27 TB total) have seen at least ~765 TB in reads and ~60 TB in writes since Dec 2021. True number is likely somewhat higher, especially for writes.