HW News - NZXT "Safety Issue," GPU Availability, AMD MI100 GPU , NVIDIA A100 80GB
Posted on November 30, 2020
It’s been another insanely busy two weeks for us here at GN, sandwiched between various product launches. Of course, this week has been anchored by the arrival of AMD’s RX 6000-series RDNA 2 GPUs. As ever, you can find our RX 6800XT and RX 6800 reviews, as well as our usual teardowns, on our YouTube channel. We’ll briefly recap them below, but you’ll need to watch the reviews for the full scope and context.
Outside of consumer GPUs, there’s also new GPUs from both AMD and Nvidia this week aimed at HPC and supercomputing. There’s also news of a new security co-processor from Microsoft, developed in collaboration with partners such as AMD, Intel, and Qualcomm. There’s also an interesting prospect of using supercomputers to help fuel progress in silicon manufacturing.
As usual, article and video embed follow below.
01:06 | GN Store Special for Reforestation Project Charity
We wanted to announce a special partnership with Eden Reforestation Projects via the GN Store for the end of the year. The past few years, we’ve partnered with Eden Reforestation Projects to help employ people below the poverty line to plant trees and reforest land that has been destroyed for logging. The Reforestation Projects’ mission is to reforest deforested land, help replenish lost habitat for wildlife, and to employ impoverished people with long-term jobs dedicated to forestry and maintenance and protection of those forests.
Just like the last few years, we’ll be doing two things for the group this year: First, until December 18th, we will be planting 10 trees via Eden Reforestation Projects for every item purchased on the store. That means that if you order two mouse mats, we’ll work with them to plant 20 trees. If you do a mouse mat, a modmat, and a shirt, we’ll work with them on 30 trees. That includes all product sales -- so even if you back-order an item, like our Large Modmats or our mouse mats, we’ll still give to the organization. Now, as always, we encourage only buying stuff -- even if it’s from us -- if you actually have a use for it. If you don’t really have a use for our items, it would defeat the purpose of our environmental assistance to buy only to plant trees. If you’d like to just donate directly to Eden Reforestation Projects and skip our store, we’re completely fine with that. We’re to match up to $1500 of viewer donations in total. In the past year alone, GN and its viewers have contributed toward the planting of approaching 200,000 trees via Eden Reforestation Projects, which has extremely low overhead. We’ve also spoken with the organization on attrition rate for trees -- which ends up being negative within just a few years, because the forest can start replenishing itself -- and we’ve spoken with them about species diversity, which it is careful about to ensure a single disease can’t wipe out the entire ecosystem. We like Eden Reforestation Projects because, just like we do with our content, they have a very scientific approach to the problems of deforestation, habitat loss, and provision of jobs.
While this special is going on, feel free to either donate directly or pick up one of GN’s critically acclaimed products. Our Wireframe Mouse Mats are in super high demand, but we have a huge order arriving in the next few weeks. These are a desk-sized surface replete with wireframe design coolers, SSDs, video cards, motherboards, and more, and also have a high-quality stitched blue border for long-term endurance, alongside a custom blue rubber underside that was difficult to source. These mouse mats are also easily washable by hand with detergent and a sink. You might also find our brand new Bar Runner Bar Mats interesting, which are in stock and shipping now. We’ve been impressed by all the home bar uses that our viewers have tweeted at us. Of course, there’s also the Large Modmat anti-static work surface, or the numerous shirts, hoodies, and glassware items on store.gamersnexus.net.
One final quick recap before we move on: In the EVGA charity stream with Jay that we recapped in a separate video, the final numbers came in for the contributions: GN contributed just over $1600 to EVGA’s charity of choice, St. Jude Children’s Hospital. In total, that brings the EVGA, GN, and community donations to over $5400 for St. Jude. For the competition, Jay and GN did dogs vs. cats for the theme, so EVGA contributed $2500 to Cat Angels, our local shelter, for our first-place victory, and they contributed $2000 to Jay’s shelter and rehab choice of Husky Haven in LA. We contributed another $500 to bring the prizes equal to each other, since Jay put up such a fierce fight. Anyway, you can check out the recap for more info. We like to do these types of things toward the end of the year every year, just to try and bring it to a close on a more positive note.
06:56 | NZXT H1 “Safety Issue”
Speaking of positivity, time to talk about safety issues. NZXT posted a thread on its sub-reddit entitled “H1 Safety Issue.” The post reads as follows:
“NZXT has identified a potential safety issue with H1 cases and we are working with the US Consumer Products Safety Commission along with the proper global authorities to notify our customers and provide them with a solution.”
“While we believe this issue only impacts a small percentage of cases, we are playing it safe and have paused sales of the H1 and are developing a simple-to-use repair kit that H1 owners can install themselves without having to ship their cases anywhere.”
NZXT recommends emailing [email protected] to get assistance with this matter. We have already launched a few probes into this and expect to have more information soon. We’ve also officially reached-out to the NZXT rep who posted this thread and, at time of filming, are awaiting a response.
Initial research shows that the Amazon product page for the H1 has been deleted, making all associated user reviews inaccessible. The Newegg page has one review visible, which indicates that the user had a fire start in association with the PCIe riser. Our assumption is that this was a short due to unintended contact somewhere. We had to use https://www.centralcomputer.com/nzxt-h1+&cd=9&hl=en&ct=clnk&gl=us&client=firefox-b-1-d to pull up the update page for retailer “Central Computers,” which has since been pulled. The cached result says the following:
“Message from our NZXT rep: We do need to halt sales due to the fact that a top screw has caused 1 case screw on the case to spark and possibly catch fire. We do have a fix for this, but we would need to halt sales for this product until we are able to get you guys the screw per units that you do have. If you have already purchased the unit and are experiencing the issue, please contact the NZXT customer service team (https://www.nzxt.com/customer-support) directly to get the screw and receive a tutorial on how to fix the issue.”
At present, it sounds like NZXT is taking the necessary steps to remedy this issue; however, we would have liked to see the company be more forthcoming with the issue in its initial reddit post, because now users are uncertain whether they can safely continue using their systems or not. We’d assume that, if it’s a short caused by a screw and a riser, it’s probably fine to use if it hasn’t already shorted. That said, since we don’t have any actual information from NZXT directly, we can’t commit to that and would advise that you contact NZXT ASAP to get more information on whether or not you can continue to safely operate your computer. Additional transparency on the issue would have been nice to reduce potential for panic.
In the interim, we have launched probes into the matter in a way which NZXT will not be able to detect as coming from us, so we should have more information soon. By the time this news video goes up, we should have already gotten the info we need and can follow-up after.
Source: https://www.reddit.com/r/NZXT/comments/k26whj/h1_safety_issue/
11:31 | Partners on Supply & Availability
We’ve heard from a few board partners over the past week regarding GPU availability leading into 2021. We’ll keep this one short.
On the NVIDIA side, partners are telling us that restocks look better in the last two weeks of December than before it, but cautioned that they still don’t anticipate demand to be met. From one major NVIDIA partner, we heard that the largest restock to date will be just before Christmas. That doesn’t mean you’ll be able to get a card, but that’s the news we got. We asked around and were told that it’s a mix of impossibly high demand and supply that’s having trouble keeping up. Although initial supply was similar to previous launches, our current understanding from partners is that yields remain bad, which is affecting NVIDIA’s ability to ramp as expected.
On the AMD side, a major AMD partner warned us to be careful not to break any cards during tear-down videos or overclocking, highlighting that we likely wouldn’t be able to get a replacement from that partner until end of December or early January. Both NVIDIA and AMD are having trouble meeting demand, but AMD in particular is dealing with extra challenges: All of AMD’s major silicon products, aside from the I/O dies on Ryzen, come from 7nm TSMC silicon. That includes the PlayStation 5, Xbox Series X, and obviously the
13:33 | AMD RX 6000-Series Review Recap
Headlining this week was AMD’s highly anticipated RDNA 2 GPUs. We looked at both the RX 6800 XT and the RX 6800, as well as doing a teardown of both cards. We’re currently working on follow up coverage for AMD Smart Access Memory (SAM), as well as our pressure maps and cold plate flatness testing. As we stated, the sheer depth of these reviews meant we had to draw a line somewhere. Additionally, this generation of GPUs has been incredibly complex, so these reviews have been some of our most dense. That said, we’re simply offering a brief recap here; we’re not gonna attempt to boil down our entire testing matrix into a TL;DR. You’ll still need to check out our videos and subsequent coverage.
In terms of raserization performance, AMD has some of its most competitive GPUs in years. The RX 6800 (non-XT) is especially compelling, as it almost universally outperforms the RTX 3070 in rasterization. The RX 6800 XT competes directly with the RTX 3080, and at resolutions such as 1080p and 1440p, often leads in rasterization; 4K tends to be hit or miss.
When looking at ray tracing performance, however, Nvidia is simply far ahead of AMD. This is AMD’s first foray into hardware-based ray tracing, whereas Nvidia has had some time to refine its approach. If ray tracing is important to you, we’d suggest seeking out other reviews, as it wasn’t a focal point in our coverage. Also, if ray tracing is a must-have feature, you’d simply be better served with Nvidia’s GPUs at the moment.
Source: GN
16:35 | AMD Instinct MI100 Accelerator
AMD also has a new GPU that isn’t for consumers. Rather, it’s the newest entry in AMD’s Instinct line of HPC accelerators. The Instinct MI100 is AMD’s first GPU to use the CDNA architecture, which is the compute-focused counterpart to RDNA. The MI100 is a 7nm GPU that looks to succeed last gen’s MI50, and according to AMD, is the first GPU to exceed 10 TFLOPS in FP64 compute.
The MI100 features 120 CUs (7,680 Stream Processors), 32GB of HBM2 memory, 1.23 TB/s of memory bandwidth, a 300W TDP, and a PCIe 4.0 interface. The GPU also boasts support for AMD’s second-gen Infinity Fabric. AMD also announced ROCm 4.0, its open-source developer platform for HPC and AI. ROCm 4.0 supports a new, open-source compiler and optimizations for OpenMP® 5.0, PyTorch, and Tensorflow.
AMD announced that, so far, OEM and ODM partners like Dell, Gigabyte, HPE, and Supermicro are all expected to have MI100-based systems out by the end of the year.
17:55 | Microsoft and Silicon Partners Roll Out Pluton Processor
In the wake of CPU security flaws like Spectre and Meltdown -- just to name a couple of the most infamous in recent years -- CPU makers (especially Intel) have had to move fast to provide both firmware and hardware level mitigations, often at the cost of aggregate performance. Still, many security researchers have called for CPU makers to pivot, and overhaul the way chips are designed, with a greater focus on security.
Interestingly, Microsoft seems to think it has a solution of sorts: The Pluton security processor. Pluton isn’t exactly new; in fact, it debuted back in 2013 inside Microsoft's Xbox One console. The chip acted as a hardware-level root of trust, and helped prevent physical attacks, hacks, and prevented the console from running pirated games, as well as providing a path for recovery in the event of a software bug.
The overall idea with Pluton is the same, except this time it will be built into chips from Intel, AMD, and Qualcomm. The updated Pluton processor was a collaboration between Microsoft and the aforementioned CPU makers as well. At a high level, the Pluton doesn’t sound radically different from Apple’s T2 chip, sans maybe locking down the device. Pluton will act in conjunction with the CPU’s TPM (Trusted Platform Module) and act as a shared root of trust at the CPU level, and is aimed at securing the communication path between the CPU and TPM.
David Weston, Microsoft’s Director of Enterprise and OS Security, had this to say:
“The Pluton design removes the potential for that communication channel to be attacked by building security directly into the CPU. Windows PCs using the Pluton architecture will first emulate a TPM that works with the existing TPM specifications and APIs, which will allow customers to immediately benefit from enhanced security for Windows features that rely on TPMs like BitLocker and System Guard. Windows devices with Pluton will use the Pluton security processor to protect credentials, user identities, encryption keys, and personal data. None of this information can be removed from Pluton even if an attacker has installed malware or has complete physical possession of the PC.
“This is accomplished by storing sensitive data like encryption keys securely within the Pluton processor, which is isolated from the rest of the system, helping to ensure that emerging attack techniques, like speculative execution, cannot access key material. Pluton also provides the unique Secure Hardware Cryptography Key (SHACK) technology that helps ensure keys are never exposed outside of the protected hardware, even to the Pluton firmware itself, providing an unprecedented level of security for Windows customers.”
Additionally, Microsoft and its silicon partners envision the Pluton processor as a means to reduce the overall attack surface, as well as eliminate certain attack vectors all together.
21:07 | Nvidia A100 80GB, DGX Station A100
At the SC20 supercomputing show, Nvidia announced a pair of its next products aimed at HPC and AI: the A100 80GB and the DGX Station A100. Nvidia’s A100 is its flagship accelerator, and the primary addition here is the doubling of the framebuffer -- up to 80GB from the previous 40GB. Nvidia also somewhat overhauled the memory subsystem a bit, as the A100 80GB will use newer 3.2Gbps per pin HBM2e memory, and offer a memory bandwidth of 2.0TB/sec. The A100 80GB will exist alongside the A100 40GB, and will be aimed at customers with dense workloads, where local memory capacity could offer performance improvements.
As ever, the focal point for Nvidia and its A100 accelerators is its HGX and DGX platforms. The A100 80GB will come in various configurations for HGX customers, and Nvidia is offering the DGX A100 640GB and DGX Station A100 to DGX customers. The DGX Station A100 is interesting, as it’s a server-class workstation that uses an AMD Epyc CPU and a refrigerant-based cooling system.
No word on whether Nvidia will offer the A100 80GB in a PCIe card, as it does with the A100 40GB.
22:57 | RTX 3000-Series Supply Issues to Persist Into 2021
In a recent earnings call, Nvidia’s top brass fielded the usual slew of questions from analysts and investors, and also offered a somewhat vague glimpse into its supply constrained GPU future.
In talking about Ampere’s successful ramp and launch, Nvidia’s CFO, Colette Kress said the following:
“While we had anticipated strong demand, it exceeded even our bullish expectations. Given industry-wide capacity constraints and long cycle times, it may take a few more months for product availability to catch up with demand,” said Kress.
There’s two main points here we’ll address. First, Kress mentions cycle times. Cycle time is the time it takes to process a wafer from start to finish, and is dependent on several factors -- the size of the wafer, layers, process node, etc. Different fabs have different cycle times; some higher, some lower. Cycle time can be measured in a couple different ways, but generally speaking, fabs are 3-4 months out. Which means Samsung, the fab currently manufacturing Nvidia’s RTX 30-series GPUs, can’t exactly be agile in terms of supply and demand issues.
That’s where Kress’ “a few more months” comment really stings, because given cycle times and insatiable demand, it will likely be well into 2021 before supply stabilizes. We also don’t expect AMD’s situation with its RX 6000 cards to be any better.
25:32 | Intel Could Use Machine Learning and Supercomputer for Future Manufacturing
At SC20 supercomputing show, it seems Intel and Argonne National Laboratories are intending to work together to help accelerate the future of chip manufacturing at Intel. As originally reported by AnandTech’s Dr. Ian Cutress, Intel’s Trish Damkroger, and Prof. Rick Stevens, Argonne National Laboratory’s Associate Lab Director talked briefly about how Intel and Argonne’s collaboration could evolve beyond the Aurora Supercomputer.
Stevens stated that Argonne was interested in using AI, machine learning, and simulation to help optimize and solve problems with the current state of silicon manufacturing. Stevens also mentioned that there would be some “announcements” coming in the next few months about how this may be done.
Additionally, Cutress has some thoughts and insight of his own to offer, so check out the video linked above.
Source: https://twitter.com/IanCutress/status/1329109482044059649
26:28 | AMD Wafer Supply Predictably Eaten by Consoles
Perhaps unsurprisingly, one of the many reasons AMD is facing supply shortages of CPUs and GPUs to meet demand is thanks to the initial console demand. Microsoft and Sony have high volume production targets to meet for holiday season, and that meant that the majority of AMD’s 7nm wafer supply is currently going to PlayStation 5 and Xbox Series X consoles. In news from Gong1shang1 shi2bao4, or Commercial Times Taiwan, AMD has reportedly allocated 120,000 of its 7nm wafers to consoles. Further compounding the issue of DIY availability is the fact that AMD is still fulfilling component orders for OEMs and SIs, leaving the DIY PC market to fight over the scraps.
Logically, it’s safe to assume that console demand will be diminished after the holidays end, so we may see some additional supply available for PC component manufacturing after the initial console launch quarter.
Source: https://ctee.com.tw/news/tech/375133.html
Editorial: Eric Hamilton
Host, Additional Reporting: Steve Burke
Video: Keegan Gallick, Andrew Coleman