HW News - GPU Smuggling, AMD Zen 3 Vulnerability, 1080 Ti Fake Rumors
Posted on April 15, 2021
This week, we have headlining news regarding AMD research that outlines some security implications regarding Predictive Store Forwarding on Zen 3 CPUs. For the time being, it mostly seems to be a non-issue, unlike the Spectre variants that have hit Intel CPUs hard. Also in AMD news is another patent filed that expands on AMD’s research into GPU chiplets, and is seemingly an improvement over the previous patent we saw earlier this year.
There’s also news regarding a massive haul of smuggled GPUs that were seized in Hong Kong, some discussion on Lenovo using AMD’s PSB feature on its servers and workstations, some commentary on a potential X570S chipset, and more.
At GN, we recently benchmarked Intel’s UHD 750 and UHD 630 IGPs and compared them to the low-end Nvidia GT 1030. We also reviewed the Arctic Freezer 50 CPU cooler, and detailed some of its pressure and assembly weaknesses.
Video embed and article follow below, as usual.
06:47 | Potential AMD Zen 3 Side Channel Exploit
In a new whitepaper published by AMD, the company details some potential security vulnerabilities regarding Predictive Store Forwarding (PSF), which is a feature on Zen 3 CPUs. PSF works by attempting to predict the dependency between a load and a store, and then speculatively executes instructions based on that prediction. It’s similar in nature to other CPU technologies aimed at improving execution performance, such as branch prediction and speculative execution.
“In typical code, PSF provides a performance benefit by speculating on the load result and allowing later instructions to begin execution sooner than they otherwise would be able to. Most of the time, the PSF prediction is accurate. However, there are cases where the prediction may not be accurate and cause incorrect CPUspeculation,” say AMD in its whitepaper.
In the case of inaccurate speculation, the CPU is supposed to flush the incorrect results from the CPU pipeline. AMD illustrates a couple different ways in which inaccurate prediction can occur, as well as detailing some limitations regarding speculation on Zen 3 CPUs. AMD then goes on to outline what it believes are the primary security concerns.
“Because PSF speculation is limited to the current program context, the impact of bad PSF speculation is similar to that of speculative store bypass (e.g., Spectre v4). In both cases, security concern arises if code exists that implements some kind of security control which can be bypassed when the CPU speculates incorrectly. This may occur if a program (such as a web browser) hosts pieces of untrusted code and the untrusted code is able to influence how the CPU speculates in other regions in a way that results in data leakage. This is similar to the security risk with other Spectre-type attacks,” according to AMD.
Furthermore, AMD warns that software that uses isolation, or sand-boxing, may be vulnerable. All told, AMD believes that the amount of code that may be vulnerable to PSF’s security implications is low, though AMD does suggest disabling the feature if users are concerned.
While it was thought that disabling PSF would carry a significant performance hit, early testing done by Phoronix finds that any performance penalty seems to be negligible. In most cases, Phoronix found that workloads showed no more than a 1% or 2% difference in performance with the feature disabled. Additionally, AMD has also made patches available to the Linux kernel that allow PSF to be disabled.
For the time being, it seems this issue isn’t as big a concern as some of the other x86 speculative execution vulnerabilities, and that’s a good thing.
Source: https://www.amd.com/system/files/documents/security-analysis-predictive-store-forwarding.pdf
Testing Update: https://www.phoronix.com/scan.php?page=article&item=amd-zen3-psf&num=1
Patch Update: https://www.phoronix.com/scan.php?page=news_item&px=AMD-PSF-Mitigation-Linux
09:37 | GTX 1080 Ti Revival Rumors Are False
Just as a quick update: There was a story circulating this past week about the GTX 1080 Ti allegedly being revived, particularly by EVGA, as a means to quell some of the shortage concerns. We called EVGA and asked if the 1080 Ti was getting remade, and the official answer was “no, it is not being remade. The 1080 Ti is no longer in production.” The unofficial answer when we first posed the question was “HAHA! No.” Sorry to shoot that one down for 1080 Ti fans.
10:52 | Hong Kong Customs Seized Smuggled GPUs
Hong Kong news station TVB News reported on the Hong Kong Customs and Excise Department seizing a cache of 300 “unidentified” graphics cards, which was part of a much larger haul that was intercepted from a Chinese smuggling ring. The smuggling run included other technology products, with TVB primarily showing cell phones, video cards, and trays upon trays of RAM. We’d be open to shipping the Hong Kong customs department one of our modmats so that they can identify if it’s DDR4 or DDR3.
Hong Kong customs said that the phones, RAM, and GPUs were worth approximately $2M HKD, or just over $257,000 USD. All of the items were found on a fishing boat, we’re assuming destined for ports in Shenzhen or mainland China. Other illegal items included shark fins and other endangered sealife.
We noticed that a few of the video cards looked like they might have display outputs, but it does appear that most are lacking display and might be part of the NVIDIA CMP HX line.
Last month, serial leaker @momomo_us on Twitter spotted a Palit CMP 30HX overseas, going for $724.
Original source: https://www.youtube.com/watch?v=0grAY6kgexY
English source: https://www.tomshardware.com/news/gpu-smugglers-300-nvidia-cmp-30hx-gpus
13:42 | New AMD Patent Expands Upon GPU Chiplets
AMD is continuing its foray into non-monolithic chip design -- both in terms of actual manufacturing and R&D. A new AMD patent expands on a previous patent AMD filed regarding GPU chiplets, with one of the key differences being an active bridge versus a passive bridge. In the past, NVIDIA has published white papers about a future of multi-chip modules, or MCMs, as an alternative approach to GPUs than the traditional monolithic designs. Monolithic silicon is more expensive to make and has yield challenges, but chiplets face other challenges, like chip-to-chip latency.
In the context of the patent, the bridge is the means in which the GPU chiplets are interconnected and coupled to resources, such as L3 or another form of cache. AMD’s first GPU chiplet patent called for a passive bridge to connect the GPU chiplets. Back in January, AMD explained it like this:
“The GPU chiplet array includes the first GPU chiplet communicably coupled to the CPU via a bus and a second GPU chiplet communicably coupled to the first GPU chiplet via a passive crosslink. The passive crosslink is a passive interposer die dedicated for inter-chiplet communications and partitions systems-on-a-chip (SoC) functionality into smaller functional chiplet groupings.”
AMD has refined its approach with the second patent. In AMD’s first patent, it proposed a passive interposer-based bridge, whereas now it’s proposing an active silicon-based bridge.
“The GPU chiplet array includes the first GPU chiplet communicably coupled to the CPU via a bus and a second GPU chiplet communicably coupled to the first GPU chiplet via an active bridge chiplet. The active bridge chiplet is an active silicon die that bridges GPU chiplets and allows partitioning of systems-on-a-chip (SoC) functionality into smaller functional chiplet groupings,” AMD explains in the patent.
Other aspects remain the same, such as connecting all chiplets to an L3 cache that is cache coherent and unified across all GPU chiplets. It seems the L3 cache is actually located on the active bridge itself, and seemingly functions similar to that of AMD’s Infinity Cache. AMD’s patents mention a communications bus, and that’s likely some iteration of AMD’s Infinity Fabric interconnect.
Another key aspect is how the active bridge approach would mimic a monolithic die GPU, which would have certain benefits from a developer or programming standpoint.
“Accordingly, as discussed herein, an active bridge chiplet deploys monolithic GPU functionality using a set of interconnected GPU chiplets in a manner that makes the GPU chiplet implementation appear as a traditional monolithic GPU from a programmer model/developer perspective. The scalable data fabric of one GPU chiplet is able to access the lower level cache(s) on the active bridge chiplet in nearly the same time as to access the lower level cache on its same chiplet, and thus allows the GPU chiplets to maintain cache coherency without requiring additional inter-chiplet coherency protocols. This low-latency, inter-chiplet cache coherency in turn enables the chiplet-based system to operate as a monolithic GPU from the software developer's perspective, and thus avoids chiplet-specific considerations on the part of a programmer or developer.”
Source: https://www.freepatentsonline.com/y2021/0097013.html
18:25 | Confirmed: AMD X570 Passive Chipset
A rumor making the rounds recently involves several motherboard listings filed to the Eurasian Economic Commission (EEC). The listings belong to Gigabyte, and denote a “S” suffix at the end of the X570 chipset. We won’t dig too much into the rumor, mostly because we haven’t been able to substantiate any of it on our own. Furthermore, it’s worth remembering that EEC listings are largely meaningless.
That said, Gigabyte has listed a total of eight X570S motherboards. The prevailing theory is that the S is to designate a passively cooled chipset, rather than an actively cooled chipset with a fan. Some are getting carried away that this coincides with another rumor regarding an alleged Zen 3+ refresh, but that’s a real stretch.
Earlier this year, AMD shipped new AGESA microcode that offered wider support for passively cooled X570 boards. After nearly two years, it’s entirely possible that AMD has honed the chipset enough and offset the power requirements for PCIe 4.0 enough that a chipset fan is no longer needed.
We reached-out to motherboard partners to ask about this new rumor. The partners we spoke with noted that the “S” designation might be a Gigabyte naming decision and isn’t necessarily the chipset name, but that there is a different piece of silicon for passive-ready X570 chipsets. This was a silent refresh and improvement to remove the active cooling need, and our current understanding is that it’s new silicon.
Source: https://twitter.com/KOMACHI_ENSAKA/status/1379349765024968705
19:42 | US Govt. to Hold Summit to Discuss Shortages
The continued semiconductor shortage is gaining an increasing amount of attention from the US Government. Previously, the White House signed an executive order meant to address the silicon shortage by proposing a review of the supply chain for key semiconductors. The executive order was preceded by a letter penned by multiple silicon designers, including Intel, NVIDIA, and AMD, all calling for an expansion to US-based semiconductor manufacturing.
Additionally, the White House recently called for $50B to help bolster the semiconductor industry, with funds earmarked for manufacturing incentives, research and design, as well as creating a new National Semiconductor Technology Center. On April 12, the US Government will host a summit to discuss the global semiconductor shortage and will include Samsung, General Motors, and GlobalFoundries, alongside Intel boss Pat Gelsinger.
In addition to automakers and chip manufacturers, National Security Advisor Jake Sullivan and National Economic Council Director Brian Deese will host the meeting, and will discuss the shortage that has impacted multiple industries at this point.
Semiconductor manufacturing has become a political football for countries over the past year. Most of the leading semiconductor manufacturing is split between two companies: TSMC and Samsung. Increasing capacity at foundries is measured in months to years, not days or weeks. And there’s no amount of sitting around a table talking about it that will speed that process up.
Frankly, there’s nothing that can be done that’s going to have any immediate, large impact on this shortage. It will take several years before we see dividends from the current government initiatives in any country, given the slow nature of fabrication plants.
Lenovo Among Server OEMs Using AMD PSB
Something that’s been drumming-up some polarizing takes, especially in the second-hand server market, is AMD’s Platform Secure Boot (PSB), a part of its Secure Processor platform. AMD’s Secure Processor platform is a handful of security measures that are driving AMD’s hardware-level security. Hardware-based security is getting an increasingly bigger push, and most silicon designers have some form of it by now. Intel has its SGX, Apple has the T1 chip, and Microsoft is working with multiple silicon designers on the Pluton security chip, which we’ve discussed before.
While more security, especially in the wake of Spectre, Meltdown, etc., isn’t a bad thing, these features have slowly revealed a side-effect: vendor-locked hardware that complicates repairability or the second-hand hardware ecosystem -- or both.
AMD’s Platform Secure Boot feature makes use of an embedded Arm Cortex-A5 SoC that resides among the other chiplets in AMD’s Epyc CPUs. This SoC runs its own isolated OS/kernel, and serves as a hardware-level root of trust. There’s a lot of technical information that goes along with that, but essentially what PSB allows is for an OEM to use an Epyc CPU and a PSB-enabled motherboard and bind the CPU with an OEM’s firmware/BIOS. The embedded SoC will look for cryptographically signed BIOS code, and should it not find the key, the system won’t boot.
OEMs don’t have to use PSB exactly like that, or at all. For instance, HPE uses its own BMC microcontroller on its server motherboards to establish a chain of trust with AMD’s Epyc CPUs (or any other CPU, for that matter), which doesn’t lock a CPU to HPE platforms. A CPU that’s vendor-locked and won’t boot in another vendor’s server effectively kills the option of reselling a CPU, hurts the typically healthy gray market for server CPUs, and ultimately contributes to the growing e-waste pile.
All that said, ServeTheHome has been able to confirm that Lenovo is among the server OEMs actively using AMD’s PSB feature, and Lenovo also seems to be using PSB in its workstation machines that use AMD’s Threadripper Pro CPUs. In addition to Lenovo, Dell EMC has also been using AMD’s PSB in its server products as well. We bring this up not in an attempt to shame any companies using the feature, but to instead bring it to users’ attention who may be unaware of the practice.
Source: https://www.servethehome.com/lenovo-is-using-amd-psb-to-vendor-lock-amd-cpus/
23:01 | Epic Games Loot Box Settlement Deadlines
Not long ago, Epic Games settled its class action lawsuit regarding randomized loot boxes, or “loot llamas'' in the case of Fortnite, in its games. The $26.5M settlement will see Epic Games hand out tens of millions of dollars worth of in-game credits for affected games; V-Bucks for Fortnite or Rocket League Credits for Rocket League. In fact, Epic started depositing those credits to accounts not long after the settlement was approved.
However, the $26.5M was set aside to settle claims for additional compensation. For valid claims, Epic will pay an additional $50 in cash, or up to 13,500 V-Bucks or 13,000 Rocket League Credits. While this news is a few weeks old at this point, we bring it up because several deadlines to make claims are approaching this month (April).
April 12, 2021 marks the deadline to object, or exclude yourself from the settlement altogether. April 26, 2021 is the current deadline to file a claim, while the final approval hearing takes place on May 6, 2021.
Source: https://www.epiclootboxsettlement.com/
24:09 | AMD Releases RX 6800 XT Midnight Black GPU
AMD recently, and rather quietly, released a special “Midnight Black” edition of its RX 6800 XT graphics card. The special edition is relatively unnoteworthy, sans the fact that it took zero seconds to both make the internet mad and sell out of the card.
The card is a mostly all-black version of AMD’s standard RX 6800 XT, and according to AMD, was created in response to community feedback and popular demand. The card was also released as first dibs for AMD Red Team Community Members, as those members received advance notice of the card’s arrival.
“Based on community feedback and popular demand, we have created a select quantity of AMD Radeon RX 6800 XT (Midnight Black) graphics cards featuring the same great performance of the widely popular AMD Radeon RX 6800 XT. This is an exclusive advance notice to members of the AMD Red Team community and this offer has limited availability, while supplies last,” says AMD.
Source: https://hothardware.com/news/amd-launches-radeon-rx-6800-xt-midnight-black-edition
25:15 | Intel Inadvertently Uses MacBook Pro in Ad
Intel’s marketing team is out in force promoting “The world’s best processor on a thin and light laptop,” according to Intel’s latest advertisement. The “best processor” Intel is referring to here is its Tiger Lake i7-1185G7; however, there’s a problem with the thin and light laptop in the picture: It’s a MacBook Pro.
Now, MacBooks have used Intel’s chips in the past, sure, but Apple famously dditched Intel processors for its own line of Arm-based Apple Silicon M-series processors last year. That said, there’s no way the MacBook Pro in the picture is using one of Intel’s latest x86 mobile chips.
Intel is using a stock image from Getty Images, and while we could probably have a discussion on $250 billion companies using stock images for advertisements, the practice is a common one. If one doesn’t look too close, you’d likely never know the machine was in fact a Mac. Give it more than a cursory glance, and you’ll notice things like the Touch bar and Magic mouse.
Furthermore, visiting Getty Images shows the same “Millennial man playing computer game on laptop at home” at several different angles, and further confirms the MacBook Pro. This blunder probably wouldn’t be as funny had Intel not recently gone on its anti-Mac and M1 campaign, as well as using its own questionable first-party benchmarks to try and cast the M1 in a less than favorable light.
Source: https://www.techspot.com/news/89233-intel-accidentally-uses-macbook-pro-ad-new-processor.html
Host, Additional Writing: Steve Burke
Editorial: Eric Hamilton
Video: Andrew Coleman, Keegan Gallick