HW News - AMD Fixing USB Drop-Out, New AMD CPUs, NVIDIA Crypto Lawsuit
Posted on March 13, 2021
This round of HW News brings in a ton of news from several corners of the industry. We have news of Nvidia winning the cryptocurrency lawsuit it found itself tangled in, which is likely some good news for both Nvidia and AMD as both companies become more involved within the cryptocurrency mining segment.
There’s also news regarding AMD’s upcoming Epyc Milan CPUs and, also on the AMD front, updates for the Ryzen USB drop-out issue. Speaking of Microsoft, the company recently reported that it has completed the acquisition of ZeniMax, which will give it access to some of the most popular gaming IPs in history.
In mid-February, AMD posted on the AMD sub-reddit that it was aware of an intermittent USB connectivity issue and that it’d be working on the fix. Now, AMD has posted an update stating that it’s working on an updated AGESA binary to fix the issue.
AMD’s post states that the company “believes it has isolated the root cause and developed a solution that addresses a range of reported symptoms,” and in that list, AMD includes “USB port dropout, USB 2.0 audio crackling with DAC and AMP combos, and USB/PCIe Gen 4 exclusion.
AMD says AGESA version 18.104.22.168 will fix the update. For anyone unaware, AGESA is a collective name for the binary that AMD provides its motherboard partners for inclusion in BIOS. It is up to the motherboard makers to ultimately include AGESA updates in BIOS updates posted on their sites, but it’s rare that a manufacturer skips inclusion except in cases of deprecated boards.
AMD said: “Customers can expect downloadable BIOSes containing AGESA 22.214.171.124 to begin with beta updates in early April. The exact update schedule for your system will depend on the test and implementation schedule for your vendor and specific motherboard model. If you continue to experience intermittent USB connectivity issues after updating your system to AGESA 126.96.36.199, we encourage you to download the standalone AMD Bug Report Tool and open a ticket with AMD Customer Support.”
At the end of 2019, Nvidia found itself in the middle of a class action lawsuit over the alleged act of misleading investors and obfuscating how much of its revenue at the time was related to cryptocurrency mining. The lawsuit claimed that Nvidia was masking over $1B in crypto mining revenue as gaming revenue, and that in turn allegedly misled investors into believing Nvidia's booming gaming segment was capable of sustaining that growth once the crypto bubble burst.
Of course, that’s not exactly how it panned-out. Both Nvidia and AMD suffered from a glut of GPU inventory in retail channels, saw stock prices drop, and got to deal with a slew of angry investors after the last cryptocurrency craze dried up.
As of Tuesday, March 9, 2021, a U.S. District Court Judge officially tossed the case out, netting Nvidia an important win. The fulcrum of the case was the inability of plaintiffs to provide evidence that NVIDIA misled investors in 2017 and 2018.
It was absolutely no secret that GeForce cards were -- and still are -- an extremely popular fare among miners, and that shouldn’t have been a secret to investors, either. That fact seems to have set the tone for the case, and any future inventor suits involving Nvidia, AMD, or crypto will likely have to try harder than that.
08:13 | AMD: Third Generation Epyc Coming March 15
AMD sent out news that it’s planning to host a digital launch event for its upcoming third generation of Epyc processors, which have long gone by the “Milan” codename internally. According to AMD’s press release, the event will feature the usual presentations from high-ranking AMD executives, including CEO Dr. Lisa Su and CTO Mark Papermaster, among others.
Milan is set to succeed AMD’s current line of Epyc Rome processors, and will be socket and pin compatible with not only Rome, but the earlier Naples platform as well. Not a whole lot is known about AMD’s Milan cores, other than they’ll be based on Zen 3, which should afford a notable IPC gain by itself. Milan will also use one of TSMC’s N7 process nodes, and will feature up to 64 cores and 128 PCIe lanes. As usual, AMD is targeting one and two-socket systems with Epyc Milan.
We’ll know more on Monday, March 15, 8 a.m. PT / 11 a.m. ET.
Samsung announced its first entry-level NVMe 980 SSD, which brings about a few notables for Samsung. This is also Samsung’s first SSD to make use of a DRAM-less controller. Samsung is billing the 980 SSD as something of a successor to the 970 Evo, but has dropped the three-letter-suffix branding.
As of this writing, there’s already a number of reviews available, but we’ll briefly list the specs Samsung has provided. The 980 SSD uses a PCIe 3.0 x4, NVMe 1.4 interface, and uses 128L TLC NAND Flash. As previously stated, the SSD uses a DRAM-less controller, instead relying on the NVMe Host Memory Buffer (HMB) feature to connect the SSD to a very small portion of system DRAM.
The Samsung 980 SSD will come in three capacities: 250GB, 500GB, and 1TB. The sequential read/write speeds are up to 3,500 / 3,000 MB/s, depending on model capacity. Likewise, random read/write speeds are up to 500K IOPS / 480K IOPS. Endurance comes in at 600TBW (1TB), 300TBW (500GB), and 150TBW (250GB). Each model is covered with a 5-year warranty, and prices start at $50 for 250GB, $70 for 500GB, and $130 for 1TB.
According to the most recent report from Gartner, PC growth and shipments are currently riding a 10-year high. For the fourth quarter of 2020, total PC shipments totaled 79.4 million units, which Gartner notes as a 10.7% increase QoQ. Annually, PC shipments totaled 275 million units for 2020, which Gartner says is 4.8% increase YoY and the most growth the PC market has seen in 10 years.
“The worldwide PC market saw a strong finish to 2020, recording a third consecutive quarter of year over year growth, although there continued to be supply shortages due to this high demand. Robust consumer PC demand again drove sales, particularly in regions where governments maintain stay-at-home orders as the pandemic persists. Prior to 2020, consumers had been shifting to a phone-first focus, yet the pandemic reversed this trend. PCs have resurfaced as an essential device as consumers, including younger children, are relying on them for work, school, socializing and being entertained from their homes,” said Mikako Kitagawa, research director at Gartner.
As usual, Lenovo, HP, and Dell remain the top three PC vendors. However, Gartner noted that all vendors, such as Asus, Apple, Acer, and others managed to grab more market share throughout 2020. And while Gartner didn’t include sales of Chromebooks in its research, it does note that 2020 was a record year for Chromebook sales, mostly due to huge demand from the North American educational market.
Gartner expects this growth and demand to remain for at least the first half of 2021, but beyond that, the sustainability will depend on how many current trends remain in place, such as virtual learning and remote work.
13:33 | Samsung’s Texas Fab Shutdown Impacting SSD Controller Supply
We’re starting to get a better picture of what the semiconductor plant shutdowns in Texas will mean for both the companies affected and the market at large. Digitimes is reporting that the disruption in production at Samsung’s Texas facilities will affect its PCIe SSD controller supply and output.
“Samsung Electronics has notified clients that production of its PCIe SSDs is being disrupted by tight supplies of controller chips, with the production unlikely to restore until May, according to market sources,” says Digitimes.
As Tom’s Hardware notes, Samsung doesn’t produce the NAND Flash for its SSDs in Texas, but it does produce the chips that power its controllers there. According to Digitimes, Samsung has reported that up to 75% of its PCIe SSD controller production will be affected this month. As of this writing, the Austin American-Statesman is reporting that Samsung’s Texas fab is still offline.
A brief update on Ryzen 5000 availability, as it seems to possibly be hitting a turning point. Several outlets have been monitoring retailers’ stock on Ryzen 5000 CPUs, and PCWorld reported that Amazon has been able to keep stock of the Ryzen 7 5800X and the Ryzen 5 5600X, and at MSRP prices to boot.
As noted by PCWorld’s Gordon Mah Ung, the higher-end Ryzen 9 5900X and Ryzen 9 5950X continue to be elusive, coming in and out of stock sporadically. Additionally, Gordon Mah Ung reached out to AMD for an update on stock and inventory, and received the following positive news.
“We have shipped a significant volume of Ryzen 5000 series CPUs and are shipping out additional stock on a daily basis. We are also refreshing stock of AMD Ryzen 5000 Series processors and Radeon RX 6000 Series graphics cards on AMD.com on a weekly basis, giving gamers and enthusiasts a direct option to purchase the latest Ryzen CPUs and Radeon GPUs at the suggested etail and retail price.”
16:23 | Intel Building Custom Silicon for DARPA/DPRIVE Program
Intel announced that it is partnering with Defense Advanced Research Projects Agency (DARPA) as a part of the Data Protection in Virtual Environments (DPRIVE) program. DARPA recently introduced its DPRIVE program aimed at developing a custom hardware accelerator for Fully Homomorphic Encryption (FHE) compute.
Fully Homomorphic Encryption (FHE) is the process of running calculations and computations on fully encrypted data or datasets, meaning the data never has to be decrypted, and is much less susceptible to compromise or exposure. FHE has huge implications for the future of machine learning, AI, and neural networks. And while FHE isn’t altogether new, it also isn’t viable with current hardware.
“A computation that would take a millisecond to complete on a standard laptop would take weeks to compute on a conventional server running FHE today,” said DARPA program manager, Tom Rondeau.
For its part, Intel plans to develop a custom ASIC accelerator, as well as developing the underlying IP blocks and software stack that will eventually lead to an SoC design to enable better FHE performance in the cloud. Intel states that once fully realized, the ASIC design could improve FHE workload execution performance by five orders of magnitude, compared to existing CPU-based systems.
Microsoft will also be part of the DPRIVE program, providing the cloud computing ecosystem and infrastructure. The DPRIVE program is set to span several years and phases, and Intel and Microsoft will be working with global standards bodies in an effort to develop international standards for FHE.
Microsoft has announced that it has officially closed on its acquisition of ZeniMax Media, which will see Microsoft bringing in all of ZeniMax’s subsidiaries -- including historic studios like Bethesda Softworks and iD Software. Other development studios include ZeniMax Online Studios, Arkane, MachineGames, Tango Gameworks, Alpha Dog, and Roundhouse Studios.
Unsurprisingly, Microsoft's Phil Spencer confirmed that some new Bethesda games in the future “will be exclusive to Xbox and PC players.” Of course, Spencer stopped just short of saying exactly which future games would be exclusive. Microsoft has clarified that existing exclusive contracts (like Deathloop and Ghostwire: Tokyo for the PS5) will be honored, and that games for other systems will still be supported.
However, Spencer also noted that “This is the next step in building an industry-leading first party studios team, a commitment we have to our Xbox community.” From day one, it was obvious Microsoft was looking to significantly bolster its ability to produce first party/exclusive content for Xbox, and this statement from Spence subtly reaffirms that.
19:56 | AMD Now Accounts for More Than 50% of Puget Systems’ Builds
Puget Systems, builders of custom and often high-end workstations and servers, released some interesting sales data as it pertains to systems that the company has built and shipped. According to the data, AMD CPUs now account for more than 50% of machines built at Puget Systems.
Puget Systems notes that it quietly dropped AMD CPUs from its build options back in 2015, with the reason being Intel offering better performing chips and a declining interest in AMD CPUs from customers. Today, Puget Systems data suggests that Ryzen 3000 was the turning point, as 2019 saw a significant uptick in AMD-based builds at Puget.
Puget’s graph shows 100% Intel utilization until about October of 2017, which is similar to sales data we previously published from our affiliate sales information. The AMD representation remained under 10% until about late 2019 to early 2020, and within 1 year since that mark, Puget saw AMD climb to over 50% of its sales. Puget noted that Threadripper was responsible for a big part of the uptick, and that makes sense: Puget’s market tends to be more focused on professionals, which includes users more likely to need HEDT components.
We’ve been discussing GAAFETs and how they’re set to eventually replace FinFETs in the next few years, as shrinking process and interconnects get more complicated. TSMC, Intel, and Samsung are all working on some version of GAAFET transistors, though Samsung recently previewed what’s possible with its own multi-bridge channel FET (MBCFET), which makes use of nanosheets.
As a reminder, nanosheets are flexible in that they can be built either wide or narrow to augment channel width, leading to either higher performing or lower-power designs. At International Solid-State Circuits Virtual Conference 2021, Samsung Vice President, Taejoong Song said that “We have used FinFET transistors for about a decade, however at 3nm we are using a gate all around transistor.” Song went on to say that the new GAAFET transistors will provide “high speed, low power, and small area."
Again, the distinct advantage held by nanosheets is the flexibility of the “Weff”, or the width of the channel in a transistor. Samsung’s research with nanosheets in potential next generation SRAM (SRAM is predominantly used for the cache on CPUs) highlights this flexibility. Samsung used different width adjustments with SRAM in an effort to reduce the amount of voltage needed to switch the state of an SRAM cell.
“SRAM’s six transistors can be divided into three pairs: the pass gates, the pull ups, and the pull downs. In a FinFET design, the Weff of all three types would be equal. But with nanosheet devices, the Samsung team was free to make alterations. In one they made the pass gates and the pull downs wider. In another, they made the pass gate wider and the pull down narrower,” explains IEEE Spectrum.
The end result was that by using various channel widths afforded by nanosheets, Samsung was able to produce SRAM with a write margin with 230mv less voltage. As we’ve reported previously, Samsung plans to more widely incorporate its multi-bridge channel FETs (MBCFET) at 3nm.