HW News - Definitely-Real Intel Arc GPU, Right to Repair Laws, & Apple M1 Vulnerabilities
Posted on June 13, 2022
Hardware news this week is flooded with silicon updates, but also features good news on the Right to Repair front.
We'll be recapping AMD's delidded Zen 4 CPU (Ryzen 7000), Intel's singular Arc GPU that "shipped," NVIDIA's first HBM3 GPU, and Apple's M1 vulnerability. Plenty more below!
01:04 | GN Hourglass Shirts Shipping, XOC Board
- Ad-lib, show renders from Andrew and front shots from existing b-roll
- 10% Kramden donation promo is active
- XOC board auction
03:51 | AM5 Ryzen Delidded and IHS Shown
Arachnophobes be warned that AMD’s new IHS resembles an 8-legged monster. Extreme overclocker HiCookie posted a photo of this CPU on facebook, and after that, a few of the overclockers currently working on these CPUs got in touch with us to talk about their experiences so far.
The IHS was previously attached to a 3-die CPU, with the two core dies along the right edge and a massive I/O die near the center. The IHS looks thicker than typically here, likely to accommodate the loss of PGA pins in order to allow the same overall z-height to maintain compatibility with AM4 coolers. Around the edge, a few dots of silicone adhesive indicate that the IHS will be closer to the CPU cores than CPUs with thicker silicone adhesive that we’ve had to scrape off in the past to maximize thermal improvements. More silicone adhesive creates more of a gap between the die and the IHS, meaning more solder or paste, and reduces efficacy of the IHS. The AMD AM5 IHS is looking good so far.
The IHS is comprised of a nickel-plated copper with a gold contact patch, where we then see (likely) indium solder contacting the dies. There isn’t a protruding surface like we’ve seen in some past AMD CPUs, like the APUs, and instead it’s just one flat surface.
The overclocking.com images help to illustrate capacitor placement across the substrate as well, helping to show why AMD did the spider-legged IHS. Of course, they could have just done a big square, but we’ve been told there’s a reason for the existing design -- and we’re still waiting to hear from AMD’s engineers officially on what that is.
Source: https://overclocking.com/le-ryzen-7000-delid-un-petit-tour-et-disparait/
06:42 | Intel Arc GPU Spotted at IEM
Intel displayed a definitely-real-and-functional Arc GPU that is definitely not only a mechanical sample at Intel Extreme Masters Dallas, an event at Dreamhack. As far as we're aware, Intel did not officially disclose which GPU was on display: the unit in question was imprisoned, eternally rotating, in an acrylic case. We think this is probably symbolic for Intel’s ability to hit its release targets, but it’s hard to be certain. It could also represent the anguish of PC builders as they await a competitor in the market.
The design mostly matches the render shown in Intel's March 30th video, but with the addition of two power connectors: 1x 8-pin and 1x 6-pin. That's the most concrete information to come out of this display, other than the fact that Intel has managed to get at least one GPU cooler made, and maybe even a PCB.
According to attendees, there were some functional Arc demo stations, but the display card stole the show. Intel Discord community member Tuanies posted some pictures and footage from the show floor, leading to some questionable dimensional math to approximate the GPU size based upon the assumed dimensions of Tuanies’ head. Arc desktop cards are expected to arrive this summer.
Primary source:
Intel Insiders discord (has videos + pictures) starting here https://discord.com/channels/554824368740630529/940688440402530364/982359289144418345
https://www.youtube.com/watch?v=n6LSJcV5Vco&ab_channel=IntelGraphics
https://twitter.com/theBryceIsRt/status/1532775229356969988
https://www.youtube.com/watch?v=rx3n7J-xOck
08:26 | NY “Digital Fair Repair Act” Passes
There’s a big development in the world of Right to Repair. The New York State legislature has passed Bill S4104A, officially known as the “Digital Fair Repair Act”, with votes of 59 to 4 in the Senate and 145 to 1 in the Assembly. Now it only requires the Governor’s signature to pass into law. If signed, it will be the first consumer electronics repair law in the United States. The bill will go into effect one year after it is signed into law. The Digital Fair Repair Act will require OEMs to offer diagnostic and repair information for electronic parts and equipment, like phones and computers, sold in New York state.
Specifically, the bill states that the OEM will make available any documentation, parts, and tools required for the diagnosis, maintenance, or repair of digital electronic equipment sold or used in the state of New York. The OEMs can fulfill this either directly or through an authorized repair provider. This will be available to both independent repair shops and owners of the devices for actual, reasonable costs - not more than they would charge an authorized repair partner.
The bill also goes to some length to prevent OEMs from making things harder than they need to be. The OEMs can’t make it to where the tools require authorization or internet access for use. Documentation must be made available at no cost unless physical copies are requested. Also, OEMs can’t restrict parts supply via pricing or usage terms to make it unreasonably difficult to obtain the parts. Any security locks or features must be able to be reset if disabled during the repair. Only time will tell if OEMs will find loopholes in the language, but they’ll definitely try.
There are some significant exclusions in what the bill covers, like: emergency communications equipment, home appliances, cars, medical devices, and heavy off-road equipment (like for farming or mining no, not that mining]). As a note, cars are actually covered under a 2014 agreement by the auto industry to respect a 2012 Massachusetts automotive Right to Repair law nation-wide.
A large amount of credit is owed to the Repair Preservation Group founded by Louis Rossmann, who has been championing the issue since at least 2016. You can find out more about the initiative and follow updates at fighttorepair.org and on Twitter @fighttorepair. We at GN are really interested in this issue as a whole. As consumers should be able to extend the life of the things we buy and keep unnecessary waste down. It’s critical economically and environmentally.
Primary Source: https://www.nysenate.gov/legislation/bills/2021/s4104/amendment/a
Secondary Source: https://twitter.com/fighttorepair
12:58 | Arctic P14 Slim 140mm Fan
Arctic has launched its new P14 Slim PWM PST fan. The Arctic P14 is pressure-optimized and 16mm thick, down from the standard 25mm, with a range of 150 RPM to 1800 RPM. Speed control is via PWM and the motor uses a fluid dynamic bearing. The fan’s cable has a splitter built-in to daisy-chain multiple fans together, all feeding back to a single fan header. You’d have to watch out for total current draw eventually, but with a rating of 0.19 amps, you should be able to get ~5 fans on a standard 1-amp header.
Arctic is shipping shorter radiator screws to accommodate the shorter fan body, and relatedly, the P14 Slim uses standard 140mm mounting holes. Most slim 140mm fans available come with 120mm mounting holes closer to the fan blade area.
This is an underserved niche mostly appreciated by small-form-factor users, but it’s still good to have the option out there. The P14 Slim PWM is currently available on Amazon for $13.
Primary Source: https://www.arctic.de/us/P14-Slim-PWM-PST/ACFAN00268A
14:45 | SK Hynix Begins HBM3 Production, Ships to NVIDIA
SK hynix announced the start of mass production of HBM3 (High Bandwidth Memory 3). SK hynix had previously announced development of HBM3 back in October of 2021 and has taken seven months to get to this stage.
NVIDIA is the first customer using HBM3, starting with the Hopper-based H100 compute accelerator modules that are intended to be used in high-performance data center applications. The H100 is expected to start shipping in the third quarter of this year. After that, SK hynix will rampup production to keep up with demand.
HBM3 offers a few advantages over the previous generation HBM2E. Density is boosted by way of stacking the dies up to 12-high, versus HBM2E’s maximum of 8-high. It seems this higher stacking is made possible by lower heat output per die. SK hynix claims a 1.8 times bandwidth increase, bringing it up to about 819GB per second, versus 410 to 460 GB per second for HBM2E. Big words like “gigabyte” can be confusing and hurt the brain, so SK Hynix has helpfully let us know what that means in more everyday terminology.
At 819GB/s, that means you could transfer “163 Full-HD movies every second.” Full-HD Movies Per Second, or FHDMPS, has fast become the standard for measuring bandwidth.
We don’t know when or if we’ll see HBM3 in any consumer graphics cards. There’s a lot of expense in an HBM solution versus GDDR options and it has a somewhat rocky track record in gaming cards. AMD’s Radeon VII used it in 2019, and NVIDIA’s most recent use in “consumer” cards was the Titan V. HBM has strong use cases in density and latency reduction, as it gets the memory as close to the die as possible (without being 3D-stacked), but it’s expensive. That makes it hard to justify on gaming cards.
Primary Sources: https://news.skhynix.com/sk-hynix-to-supply-industrys-first-hbm3-dram-to-nvidia/
https://product.skhynix.com/products/dram/hbm/hbm3.go
17:40 | Phanteks G360A Released
The new Phanteks G360A launched on June 8th. This is a refresh of the existing P360A, a budget enclosure which we like well enough, but has been overshadowed for its entire existence by the already-inexpensive P400A Digital. The P-to-G update for the 360A could hint at future updates for the rest of the Eclipse lineup, making this an especially interesting release.
Phanteks provided us a list of changes, but started off by noting that the internal chassis remains the same. Externally, the front panel has been updated to match the P500A, making it look less like another retrofit of the ancient P400 and more like it was built as an airflow case from the ground-up. More significantly, the case now has a removable 360mm fan and radiator bracket at the front, as well as a 360mm mount on top of the case (not usable simultaneously). That brings total capacity up to 7x 120mm fans, but without increasing the overall dimensions of the case.
We have a review sample on hand, so stay tuned for our own measurements and opinions. The G360A is currently available for $100 with three preinstalled 120mm fans–we're optimistic about this one.
Primary source:
https://phanteks.com/Eclipse-G360A.html
Source: Phanteks to GN
19:18 | Apple M1 CPU Vulnerability Discovered
A group of researchers at MIT’s Computer Science and AI Lab (CSAIL) found another vulnerability in the M1 series of ARM-built processors. They call it PACMAN, and the official website is aptly dubbed pacmanattack.com.
According to pacmanattack.com, the exploit isn’t currently “being used in the wild.”
Like Spectre, which the MIT team based its research on, PACMAN executes in the speculative area of CPU memory, which makes it difficult to track or log. The research team stated that the exploit can be executed remotely.
As far as a fix goes, the MIT group said that the M1's hardware vulnerabilities can't be patched with software, and they believe the vulnerability could impact current and future ARM-based devices that employ PAC in its current implementation. Apple (M1 and mobile devices), Samsung, and Qualcomm were specifically mentioned. To that end, their number 1 countermeasure was for hardware companies to modify future microarchitecture.
The fix suggested for the masses was to patch any memory corruption vulnerabilities by keeping OS and software patched. The website says: “As long as you keep your software up to date, there’s no need to worry. PACMAN is an exploitation technique- on its own it cannot compromise your system. While the hardware mechanisms used by PACMAN cannot be patched with software features, memory corruption bugs can be.”
The catchy PACMAN name comes from a hardware-level security technique called Pointer Authentication Code which is present in the M1 chips and other Qualcomm and Samsung chips. The security feature was built to protect CPU memory space from an attacker who’s already gained memory access th rough a bug. PAC makes it more difficult to inject malicious code into a device’s memory by checking memory pointer integrity against a cryptographic hash. If the hashes don’t match up during verification, the apps crash to prevent undesired code execution.
The MIT group’s PACMAN attack avoids the crashing by brute force guessing the correct hash in speculative processing space. If the hash is guessed correctly, PACMAN uses a microarchitectural side channel to pass the fact that the guess is correct back. Summarized, the exploit requires an existing memory corruption to inject PACMAN, abuses speculative processing to avoid crashing, and utilizes hardware side channels to confirm correctly guessed hashes. It needs both hardware and software to pull the attack off. From there, the researchers say that operating system kernel attacks are possible.
These findings were published this week in advance of the International Symposium on Computer Architecture where the information will be presented, but the team has been working with Apple since 2021, so Apple isn’t getting blindsided by the news. The research team plans to file for a Common Vulnerabilities and Exposures (CVE) number soon.
Source 1: http://pacmanattack.com/
Source 2: https://pacmanattack.com/paper.pdf
Source: https://techcrunch.com/2022/06/10/apple-m1-unpatchable-flaw/
22:16 | Arc A730M Benchmarks Surfacing
About a month ago, Intel Arc A730M laptops appeared on the Chinese retail site JD.com. The A730M is towards the high end of the publicly-known Arc A-Series product stack, containing 384 EUs to the top-of-the-line A770M's 512 EUs. Now, we learn the listing for the $1267 Machenike(?) Dawn Discovery 2022 with i7-12700H and A730M graphics was the real deal, as a Weibo user has posted TimeSpy and game benchmark results from a system matching those specs.
Starting with the 3DMark result, the laptop achieved a graphics score of 10107 in the standard TimeSpy test. According to UL's site, that puts it roughly between a mobile RTX 2080 and a mobile RTX 3070. It's impossible to judge a GPU based on a single result from a single laptop, but that's a promising start for a company that hasn't put out a dGPU since 1998.
As for games, the user posted a small selection of prefab benchmark results: Metro Exodus, F1 2020, and Assassin's Creed: Odyssey, each at 1080p and 1440p. We're not going to make any irresponsible comparisons to other hardware, but if you want to make your own, the settings used are in the screenshots. Metro averaged 55FPS at 1440p and F1 2020 averaged 95FPS, while Odyssey struggled even at 1080p with a 38FPS average. It's a safe bet that Intel still has some work to do: the user also posted that Shadow of the Tomb Raider wouldn't even launch.
Preorders for A730M laptops outside of Asian markets (Australia and NZ) went live a week ago.
Primary source: https://weibo.com/3219724922/LwwomE5uU
Primary source: https://weibo.com/3219724922/Lwvhfcmee
KG used: https://twitter.com/9550pro/status/1534001508932431872
KG used: https://twitter.com/9550pro/status/1533807671278592002/photo/1
Comparisons: https://benchmarks.ul.com/compare/best-gpus?amount=0&sortBy=SCORE&reverseOrder=true&types=MOBILE&minRating=0
Original product listing ~1 month ago: https://www.tomshardware.com/news/intel-arc-a730m-powered-laptop-surfaces-with-dollar1200-price-tag
24:30 | FSR 2.0 is Here
FidelityFX Super Resolution 2.0 officially arrived a month ago with an update to a single game: DEATHLOOP. FSR is an upscaling method, basically AMD's equivalent of DLSS, although it's more open and doesn't rely on AMD-specific hardware. The 2.0 update largely concerns temporal data: frames are now upscaled using data from previous frames to increase quality.
As usual with new features like this, the queue of early adopters is a mixture of a few heavy hitters and then smaller titles to flesh out the list. First up after DEATHLOOP are Farming Simulator 2022 and God of War (2018), both of which now have FSR 2.0 available as an option.
AMD offers some performance numbers, but these are of limited use without visual comparisons to accompany them. AMD claims that "Looking at the new AMD graphics powerhouse for 1440p gaming, the AMD Radeon™ RX 6750 XT GPU, FSR 2.0 lets you play God of War at 4K with no compromises at “Ultra” graphics quality." That's optimistic: AMD's own chart shows the 6750 XT averaging 63FPS at FSR Performance, which (at 4K) upscales from a 1080p source image. We'll likely do our own feature piece once FSR 2.0 has been patched into a few more games.
Looking over the list of upcoming FSR 2.0 titles, Microsoft Flight Simulator and Hitman 3 catch our eye as games that could benefit from FPS boosts. A total of 19 games have committed to using FSR 2.0, and more are likely to follow as the tech works its way into game development pipelines.
Writers: Steve Burke, Jeremy Clayton, Patrick Lathan, Patrick Stone
Video: Keegan Gallick