HW News - Intel Gaming GPU, AMD Wants US Fab & More Linux, StarLink Scares ISPs
Posted on February 12, 2021
This week comes with more industry news as it relates to semiconductor manufacturing and supply chain expansion in the US. Additionally, as cryptocurrency prices heat up, miners are turning to laptops in the wake of persistent GPU shortages. Elsewhere, Raja Koduri is back with another Intel Xe Graphics tease, Intel is suing a former employee, and we have a GTX 1050 Ti rumor.
At GN, we’ve made our return to case reviews, as we recently looked at the be quiet! Silent Base 802 and MSI’s MPG Sekira 500X. We also took a look at PS5 vs. Xbox Series X airflow, where we used Schlieren photography to gauge air patterns and movement.
01:18 | AMD, NVIDIA Want US-Based Manufacturing
Executive leadership from all three major silicon designers and manufacturers recently joined forces to issue a letter addressed to US Government leadership, notably including Lisa Su of AMD, Bob Swan (outgoing) of Intel, and Jensen Huang or his executive team. Joining all of these CEOs together very nearly creates a Voltron of the computer world’s executives. In the letter, the silicon design and manufacturing companies sought financial support and “substantial funding” from the US Government “in the form of grants or tax credits,” further informing the government that “costs of inaction are high.”
As for why these hundred billion-plus dollar companies want assistance, it’s because they’re interested in bringing more semiconductor manufacturing into the US. Silicon fabrication plants are highly sophisticated facilities -- Intel has published many factory floor footage showing how its silicon wafers are processed, for instance -- and are generally skilled labor jobs. Because silicon is at the heart of military and government devices, and because those silicon parts are increasingly being made outside of the US, the companies are trying to urge the US to help with domestic manufacturing.
Of course, this would also benefit the companies involved: They score more government contracts for chips and can also help fund multi-billion dollar fabrication facilities, which would be useful for product distribution and distribution of fab space consumption.
The executives note that the US currently holds 12% of silicon fabrication capabilities, a 25-point drop from the 90s. The letter said the following:
“As a result, the U.S. is uncompetitive in attracting investments in new fab construction and our technology leadership is at risk in the race for preeminence in the technologies of the future, including artificial intelligence, 5G/6G, and quantum computing.”
More manufacturing in general would probably be a good thing, regardless of where it ends up. The past year has demonstrated that any major demand event renders nearly every industry unable to meet that demand, spanning outside of PC DIY and into even automotive. Computers are in everything now. In somewhat related news, TSMC noted this week that it’s working on a $9 billion investment in Japan for a new offshoot fabrication plant.
In a recent report by Michael Larabel at Phoronix -- a site we’re happy to recommend for its Linux-specific benchmarks -- it was reported that AMD is growing its team of Linux developers. AMD currently has job listings for a Linux Kernel development manager, described on the AMD careers page as a mentor to a team of developers to enable design and implementation of “Linux operating system features supporting AMD CPUs.” The job lists responsibilities as including management of an engineering team “in the area of Linux kernel and KVM virtualization development.”
Other job listings on AMD’s jobs page include Linux Engineering for client devices, defined as “a key engineer on our team [who] configures and tunes Linux distribution packages to take advantage of AMD technologies on customer platforms.” They say that this role would participate in planning next generation CPUs and APUs, too.
Additionally notable, the jobs page is seeking a Linux Architect in the client group, with the job role requiring collaboration with silicon architects to plan features and platform needs.
For us, that’s just interesting because it’s more focus on Linux. This could be useful for continuing AMD’s push into server, where it saw a market share increases versus Intel, but as many of these listings are for the client group, it’s likely this is at least partially attached to OEMs and portables roll-out.
08:06 | Cyberpunk, Witcher Source Code Allegedly Sold
If you missed it, CD Projekt Red recently posted a notice that unauthorized access to its servers resulted in a ransomware attack on its software. This included Cyberpunk 2077, The Witcher, Gwent, and financial documents, including unreleased game code. This was posted by CDPR’s twitter account. CDPR also noted that user data was unaffected.
In the few days since, the source code dump was allegedly sold for $7 million to a direct buyer and exclusively. We wonder whether this might be a consortium of investors, CDPR itself via a different company, or some other interested agency, as a leak of this capacity could be damaging in the wrong hands.
In what largely appears as an effort to circumvent shortages for add-in graphics cards, miners are buying laptops for their mobile GPU variants and deploying them in laptop mining farms. Over the last week or so, reports have streamed in regarding cryptocurrency miners in China buying up RTX 30-series laptops in bulk for the express purpose of mining Ethereum. Etherum, like Bitcoin, has been on the rise for months now, spurring many prospective miners to snap up as much hashing horsepower as possible.
Weibo user BTCer and twitter user Harukaze posted photos and images showing laptops propped up like tents, stacked on shelves and tables, and some units have even been partially dismantled to allow for a semi open-air environment. We also noticed that several of the laptops -- shown precariously standing in a room with innocent wallpaper depicting dolphins -- had their batteries removed. We assume this is a fire safety choice given the explosive nature of compromised lithium-ion. The laptops also share just two visible surge protectors, with one alone hosting about 5-6 laptops.
RTX 30-GPUs have been especially popular among Ethereum miners, but even the mobile SKUs fare pretty well. A chinese content creator made a video where she took her RTX 3060 equipped laptop to a local Starbucks, and within two hours, had allegedly mined enough Ethereum to pay for a cup of coffee. If it’s Starbucks, we’d assume that cup was about the size of a thimble.
It seems most of the laptops being bought in China are equipped with RTX 3070 or 3060 GPUs, and many of the laptops seem to have come from Hasee (Shénzhōu Diànnǎo). Hasee is one of China’s largest computer manufacturers, and offers a number of laptop models with RTX 30-series GPUs.
13:38 | Gelid Protection Bracket For Ryzen CPUs
We just bought Gelid’s new “CPU Protection Bracket,” used with socket AM4 CPUs. AMD’s PGA packaging is prone to having CPUs get sucked up by the cooler on removal, and the new bracket slides over the AM4 substrate to hold it down even under that suction. The bracket doesn’t cover the IHS, although it does limit compatibility with some coolers, and it also means that there’s an extra layer in the mounting kits that could pose problems with the fit or pressure.
While the bracket will work with any AM4 CPU, this bracket only (officially) works with Gelid’s AM4 mounting kits and hardware.
15:33 | Intel Sues Former Employee for Trade Secret Theft
According to Intel and court documents filed in a US District Court in Portland, Oregon, Intel is suing a former employee by the name of Dr. Varun Gupta. According to court documents, Gupta was a former product marketing engineer who worked at Intel for 10 years, before departing for Microsoft in January 2020.
The suit alleges that Gupta had access to confidential Intel files during his tenure with the company, and that Gupta transferred some 3,900 files from his Intel issued laptop to a pair of thumb drives on his last day. The files in question are related to how Intel customizes and prices its various Xeon CPUs.
“Throughout the course of his employment, Gupta had direct access to documents containing Intel confidential information and trade secrets related to Intel’s pricing structure and strategies, definition of the parameters, and manufacturing capabilities for customized Xeon processors, among other Intel confidential information and trade secrets,” says Intel.
Intel alleges that such information is commercially damaging, as it would allow Microsoft -- or any company that becomes privy to it -- to leverage an unfair advantage in negotiating prices for Xeon products.
“As a purchaser of Intel’s processors, Microsoft’s rational objective would be to negotiate the most advanced technical features at the lowest achievable price. Gupta knew Intel’s confidential technical and manufacturing capabilities, customized product offerings, marketing strategies, and pricing for Xeon processors, and used that confidential information and trade secrets to gain an unfair advantage over Intel in the negotiations concerning product specifications and pricing.”
For his part, Gupta seems to partially deny the claims, alleging that only one USB drive exists, and that the drive was surrendered to Microsoft for analysis. Intel notes in its filings that Microsoft has aided in the investigation and forensic analysis, and that Gupta has accessed the stolen files multiple times throughout his employment at Microsoft.
It seems Intel is seeking no less than $75,000 in damages, plus its legal fees. Intel is also looking to bar Gupta from using any of Intel’s confidential information.
18:42 | Raja Koduri Teases Xe HPG Running 3DMark Shading Test
As we’ve recently mentioned, teasing Intel’s upcoming in-house GPU development has been Raja Koduri’s thing lately. Just a couple of weeks ago, Koduri teased a die shot for the 7nm Xe HPC that we believe will eventually take the form of Ponte Vecchio. Koduri is back at it, this time teasing Intel’s Xe HPG, albeit in less flattering form.
The latest showcase was Intel’s Xe HPG enthusiast gaming GPU running an unreleased test from 3DMark, featuring a new mesh shader feature. We suspect this is the Xe-HPG DG2 variant, which would be the dGPU counterpart to the lower-end DG1.
There isn’t much to glean from the tweet, other than Intel’s testing and development for Xe HPG must be well underway if it’s already being tested on 3DMark. This could bode well for rumors of a late 2021 or early 2022 release. Speculation isn’t worth much, but talk online lately seems to suggest Intel targeting performance somewhere between Nvidia’s RTX 3070 and AMD’s Radeon RX 6800.
The GPU should come with current marquee features like ray-tracing support and GDDR6 memory, and we know it will be manufactured outside of Intel, likely on one of TSMC or Samsung’s nodes.
Koduri’s post shows the upper and lower deck of the silicon, demonstrating Intel’s multi-layered packaging approach to its new silicon. Some marked sections show 7nm and 10nm SuperFin silicon side-by-side.
20:16 | Intel Rocket Lake-S Not Compatible With B460 or H410
Intel’s Rocket Lake-S desktop CPU is expected in March and aims to compete with AMD in desktop gaming. Rocket Lake-S will use a backported Cypress Cove core architecture and is set to be Intel’s last batch of 14nm, Skylake-derived silicon, as Alder Lake-S will use Intel’s 10nm SuperFin.
As Intel is wont to do, it’s already axing support for lower tier motherboards as it relates to Rocket Lake-S compatibility. An updated support document regarding BIOS updates for upcoming 11th-Gen Intel CPUs states that B460 and H410 chipsets will not see support for the new chips, but Z490 has already been known to support them.
The document also addresses that Z490 and H470-based motherboards will likely require a BIOS update in order to boot with Rocket Lake-S CPUs, which is fairly routine when new processors get dropped into existing motherboards.
As for Rocket Lake-S support, it seems the upgrade path for owners of B460 or H410 is to either move to Z490 or H470, or shell out for a 500-series board. Both options are a dead-end, considering Alder Lake will introduce the new LGA1700 socket.
21:36 | Samsung Expanding US Semiconductor Manufacturing
Samsung is exploring avenues for expanding its manufacturing in the US, according to documents filed in various potential host states, including Texas, Arizona, and New York. Samsung already has one manufacturing facility stateside: its S2 fab in Austin, Texas. However, the S2 plant mostly deals with older process nodes. All of Samsung’s leading manufacturing processes are handled in South Korea.
Over the last few years, the US has been slowly cranking up pressure on semiconductor manufacturers to bring more production into the states (and vice versa).
TSMC has already announced plans for a $12 billion plant in Arizona. TSMC’s future Arizona location will initially focus on 5nm, but will likely also see TSMC’s future nodes as well, being TSMC is already pumping out 5nm chips. If all goes to plans, TSMC’s Arizona plant will come online in 2024.
Samsung doesn’t want to be left out, and so is planning expand its North American operations. Right now, it seems Samsung is in the courting phase, as it chooses which state will offer it the most resources and financial incentives. According to The Austin American-Statesman, Samsung is seeking incentives and subsidized breaks worth more than $1 billion from Texas.
In return, Samsung would build a 7-million square foot facility, estimated to create 1,800 jobs. The new fab would be built in close proximity to Samsung’s existing S2 fab and cost more than $17 billion. According to documents filed with Texas, Samsung is looking for a state that can offer access to talent, an existing semiconductor manufacturing ecosystem (i.e,. Supply chains), speed to market, and a strong public-private partnership. Austin, Texas would seem a good enough choice, though Samsung knows there are more than a few states that would welcome Samsung’s expansion deal. According to these documents, Samsung’s current plans would not move any R&D to the US and would only consist of manufacturing.
"Based on various factors, we are reviewing the possibility of expanding our semiconductor manufacturing facilities. A number of global locations, including multiple candidates within the U.S., are under consideration. However, no decision has been made at this time,” Samsung told The Austin-American Statesman.
Samsung is calling the expansion Project Silicon Silver, and it appears Samsung is loosely targeting a Q4 2024 completion date, assuming it decides to partner with Texas. No word on exactly what process technology will be used at the new site.
24:39 | Other News: Chrome Ditches Old CPUs, ISPs Are Afraid of StarLink
In other news this week, Chrome is ditching support for old CPUs, and unsurprisingly, current ISPs are afraid of StarLink.
Chromium developers have issued a notification that Google Chrome will assume support for SSE3 (Supplemental Streaming SIMD Extensions 3) in x86 Chrome builds, starting with Chrome 89. What this means is that if you have an especially old machine with a CPU that supports SSE2 but not SSE3, Chrome will no longer work on that machine.
As the change document notes, SSE2 was introduced on Intel x86 CPUs in 2000, and on AMD x86 CPUs in 2003. Furthermore, SSE3 was introduced on Intel CPUs in 2003, and on AMD CPUs in 2005. So, if you’re using a CPU older than 2005, Chrome 89 will crash. The document also notes that until such time that SSE3 is officially required, Chrome will warn impacted users.
Moving on, ISPs are becoming increasingly more alarmed about StarLink, SpaceX’s satellite internet service that is currently in public beta. In a study that was commissioned by Fiber Broadband Association (FBA) and The Rural Broadband Association, it’s alleged that StarLink will fall short FCC deployment goals as required by Rural Digital Opportunity Fund (RDOF) funding and the FCC.
StarLink was awarded almost $886M from the RDOF, paid out over 10 years, to connect 642,925 homes and businesses in 35 states with 100Mbps download and 20Mbps upload speeds. The funding is dependent on certain milestones, and the report alleges that StarLink will run into a capacity shortfall, or fail to deliver enough bandwidth, by 2028. The report claims that given certain assumptions and model scenarios, 56-57% of RDOF subscribers may not receive enough bandwidth to avoid service degradation during peak hours.
The report urges the FCC to consider whether or not StarLink should receive RDOF funding, based on how likely the service is to achieve deployment milestones and service quality. Obviously, traditional ISPs (cable, DSL, Fiber) would love to see the RDOF money go back into the pool. As ArsTechnica points out, traditional ISPs routinely fall short of FCC/RDOF requirements -- Centurylink and Frontier being two recent examples.
The FCC will review the study, but it seems highly unlikely that the report will convince the FCC to hesitate giving money to what may be the best option rural users have had in years.