This week, we have news from System76, makers of Linux-based workstations and servers, as the company introduces its newest Thelio Mira desktop. We’re also seeing reports that NZXT could be getting into the display market, based on the finding of an NZXT job listing.
Elsewhere, we have AMD news on a couple of different fronts: AMD unwrapped its latest Epyc server CPUs, and the company is also moving up the food chain as it relates to TSMC’s customer base. We’re also following up on a couple of previous stories involving Google and Grand Theft Auto Online.
Intel recently tweeted a teaser to its Xe HPG gaming architecture, intended for use in discrete GPUs in the near future. While we all wait for Intel to save us from silicon shortages, the company has given indicators that there may be a cypher challenge (no, not the rap kind) to decrypt its messages and learn more about the Xe HPG hardware. NVIDIA somewhat famously had a cryptographic set of packages and webpages back for the GTX 10 series launch (the “Order of 10”), and if we’re lucky, Intel will do something similarly fun.
The teaser video had a binary string hidden in the bottom corner. If you don’t want to hear what that string decoded to (so that you can solve it yourself), skip to the next story.
The string was an 8x4-digit string and decoded to 35 (.) 160 (.) 237 (.) 208, which, when entered as an IP address into a web browser, directs to an Intel server and reveals a web address of “xehpg.intel.com,” which says “welcome to the Xe HPG Scavenger Hunt” and has a launch date of March 26, 2021. Now, to be clear, that’s the launch date of the scavenger hunt. We’ll keep you posted on what happens with it. We’ve participated in some of these in the past and, when not too corny and done cryptically enough, they can be a lot of fun. Here’s hoping Intel has something promising to deliver to the current GPU landscape. They will never have a better chance than right now.
In speaking with industry sources at major board partners and PC part manufacturers, we’ve received word that the current expectation of easing supply shortages isn’t until the distant future: A few major board partners reported to GN that their guidance from NVIDIA (specifically) has indicated more ability to meet demand in Q3 of 2021. Current guidance suggests that AMD’s CPU silicon has already become more prevalent (as we reported last week, via PC World), and that the higher-end CPUs should become more readily available into mid-year this year. The 5800X and 5600X have already been available for a few days at a time lately. No word on AMD GPU shortages yet, from our contacts, but it does sound like Q3 is the current target for more constant supply of NVIDIA GPUs. That’s further than we’d like, and keep in mind it is just industry guidance, but that’s the current news.
07:15 | NVIDIA’s Crypto Self-Own
NVIDIA wants to celebrate. NVIDIA was cleared by a Federal Judge of the crypto lawsuit last week, whereupon investors claimed NVIDIA underreported its mining revenue and overreported gaming, and so the company is celebrating by hacking its own drivers.
NVIDIA’s RTX 3060 launched with both firmware and driver locks in place, forming what the company described as a “handshake” between the two to obstruct Ethereum mining hash rates. The RTX 3060s were still able to mine other cryptocurrencies at the unobstructed rate, and some reports suggested that this meant that the block had been bypassed, but in fact, Ethereum was still limited.
At least, until NVIDIA hacked itself.
Tired of waiting for reddit to crack the solution, NVIDIA launched beta driver version 470.05 ahead of the game-ready launch, and in so doing, accidentally fully unlocked Ethereum mining performance. We’d assume that the company had a code branch at some point just before launch to fork protection into the drivers and firmware, and perhaps it wasn’t grafted into the beta branches. Either way, PC Watch, Computer Base, and Hardware Luxx have confirmed that the beta driver did, in fact, remove the block. As would be obvious, this screw-up can’t be unscrewed: Miners could just opt to run older drivers permanently.
As promised last week, AMD took the lid off of its latest Epyc Milan CPUs. AMD’s new Epyc 7003-series will be crowned by the 64C/128T Epyc 7763, which also runs a base frequency of 2.45GHz and a boost clock of 3.50GHz. All in all, the new Epyc Milan 7003 lineup will consist of 19 SKUs and will target AMD’s usual 1P and 2P server customers.
One of AMD’s biggest strengths with Epyc is its simplified and non-price-segmented feature set. As in, everything AMD has to offer with Epyc is on tap: all SKUs offer 128 PCIe 4.0 lanes, 8-channel memory support, and DDR4-3200 support. Based on Zen 3, Epyc Milan will also bring a tangible IPC uplift and improved boosting algorithms to extract higher frequencies from the silicon.
As with other Zen 3 CPUs, Epyc Milan is manufactured on TSMC’s N7 node and will use AMD's tried and true chiplet topology, allowing for up to nine chiplets (CCDs) to be stitched together. As usual, Zen 3 CCDs span up to eight cores, and are afforded access to 32MB of L3 cache. Milan will use the SP3 socket, and will be backwards compatible with Rome and Naples.
AMD noted that its partners like AWS, Dell, HPE, Google Cloud, Microsoft Azure, and more will all be adopting 3rd-gen Epyc for their own Epyc 7003-based designs.
We’ve mentioned System76 here before, namely for the fact that it’s a manufacturer specializing in Linux-based systems. System76 offers a number of laptops, desktops, mini-PCs, and servers all running a Linux distro. System76 recently trotted-out its latest desktop design, the Thelio Mira desktop.
Like the rest of its catalog, System76’s Thelio Mira ships with Linux -- either Ubuntu 20.04 LTS or System76’s own Ubuntu-based Pop!_OS 20.10 or Pop!_OS 20.04 LTS. The Thelio Mira is aimed at users looking for a compact workstation, and can be configured with AMD’s Ryzen 3000 or 5000 CPUs and Nvidia GPUs. System76 is offering a GT 1030 at the bottom-end, and up to two Quadro RTX cards at the high-end, with select RTX 30-series cards being offered in between.
Elsewhere, Thelio Mira will be configurable with up to 128GB of dual-channel ECC DRR4 memory and up to 36TB of storage, with multiple M.2 PCIe Gen4 NVMe slots and 2.5” bays on tap.
Thelio Mira also comes with System76’s Thelio IO daughterboard, which is an open-source I/O board that moves chassis and fan controls from the motherboard to the daughterboard, which aligns with System76’s goal of moving functionality away from proprietary and closed-source hardware. System76 also contends that this allows for more granular control over fan, storage, and power. The board uses an embedded Microchip ATmega32U4 for the controller, and the design files are hosted on Github, so anyone can iterate or improve on them.
A recent job posting suggests that NZXT is looking to get its feet wet in the ever-crowded display market. The listing was caught by OC3D.net, and calls for a “Display/Monitor Engineering Lead.” Furthermore, according to the listing, anyone assuming this position “will be responsible for the development and qualification of Flat Panel Displays for desktop monitors and portable products.”
The listing is seeking applicants that have experience with multiple display technologies, a MS or BS degree in Electrical Engineering, 15+ years experience in developing display products, and proven experience in taking 5 or more products from concept to production.
NZXT is only the latest computer hardware company looking to get into displays. MSI and Cooler Master have already taken the plunge, and Corsair posted a similar job listing on Indeed calling for a “Products Manager” for displays back in 2018.
15:59 | Rockstar Listens, Will Implement User-Created GTA:Online Fix
We recently reported on a programmer and GTA Online player, going by the name of t0st, improving the game’s sluggish loading times with a custom DLL. At the time, there was speculation as to how serious Rockstar would take the findings, but surprisingly, Rockstar has confirmed that what t0st found is accurate.
"After a thorough investigation, we can confirm that player t0st did, in fact, reveal an aspect of the game code related to load times for the PC version of GTA Online that could be improved. As a result of these investigations, we have made some changes that will be implemented in a forthcoming title update,” said Rockstar.
As of March 16, Rockstar has rolled out a patch that implements the findings of t0st, and if the GTA Online subreddit is anything to go by, the results are significant. Additionally, t0st confirmed that they also received a $10K award through Rockstar’s bug bounty program. According to t0st, Rockstar made a specific exception with the award, as its bug bounties are usually only for security related bugs.
As expected, Qualcomm has closed its acquisition of silicon/CPU design company Nuvia. Qualcomm shelled out $1.4B to acquire Nuvia, and it seems that will afford it the opportunity to take on Apple and Arm, and even take another shot at the server market. Qualcomm’s Snapdragon SoCs have lived in the shadow of Apple’s A-series SoCs for years now, and Qualcomm’s custom designs have never strayed too far from vanilla Arm designs.
Now, Qualcomm has wasted no time in stating its immediate intent to address high-performance laptops. “The first Qualcomm Snapdragon platforms to feature Qualcomm Technologies’ new internally designed CPUs are expected to sample in the second half of 2022 and will be designed for high performance ultraportable laptops,” said Qualcomm in its press release.
Additionally, Qualcomm will leverage its Nuvia acquisition to design in-house silicon for “flagship smartphones, laptops, and digital cockpits, as well as Advanced Driver Assistance Systems, extended reality, and infrastructure networking solutions.”
As a reminder, Nuvia was founded by former Apple A-series architects John Bruno, Manu Gulati, and Gerard Williams III. Nuvia never actually released a product before getting Qualcomm’s attention, but the company was working on custom silicon for servers and HPC applications. It was no doubt the engineering pedigree of Nuvia’s founders that got Qualcomm’s interest, and as noted by Qualcomm’s press release, Gerard Williams is now serving as Qualcomm’s SVP of Engineering.
19:41 | Google Lawsuit Over Incognito Mode Continues
In June of 2020, we discussed a $5B class-action lawsuit being slapped against Google for the alleged tracking of users while in Google’s “private” incognito browsing mode. The lawsuit is leaning on Google violating the Federal Wiretap Act and intentionally deceiving users.
For its part, Google tried to have the case dismissed on the basis that Chrome support documents make it clear that “activity might still be visible.” According to Bloomberg and ArsTechnica, Google was unable to convince the courts and the lawsuit will move forward.
"The court concludes that Google did not notify users that Google engages in the alleged data collection while the user is in private browsing mode,” according to US District Judge Lucy Koh. As usual, Google disputes this claim, and has vowed to “vigorously” defend itself. The lawsuit seeks at least $5 billion in damages, up to $5,000 per affected user. The suit is retroactively targeting any user that has used Chrome Incognito mode since June 1, 2016.
A few years ago, Micron and Intel announced a joint venture to develop 3d Xpoint memory, which was concepted as something of an evolution of NAND Flash and DRAM, with theoretical higher endurance, performance, and density. Intel branded its version of 3D Xpoint as Optane, while Micron went on to brand its own 3D Xpoint products as QuantX. However, none of Micron’s QuantX products ever shipped, with the exception of Micron’s X100 SSD that shipped to select datacenter customers.
In the intervening years, Intel and Micron dissolved their 3D Xpoint partnership, and Intel has even scaled back its own Optane development and ambitions, primarily focusing on datacenter and enterprise applications. Now, Micron has decided to abandon the 3D Xpoint business entirely.
“Micron has now determined that there is insufficient market validation to justify the ongoing high levels of investments required to successfully commercialize 3D XPoint at scale to address the evolving memory and storage needs of its customers,” says Micron. “In line with this new strategic focus, Micron is engaged in discussions for the sale of its Lehi, Utah fab currently dedicated to 3D XPoint production. The company aims to reach a sale agreement within calendar year 2021.”
Micron also made it clear that going forward, it will focus on developing memory and storage products that support the emerging Compute Express Link (CXL) interface. This is something that Intel probably isn’t keen on getting onboard with, as Optane is proprietary and exclusive to Intel’s CPUs.
With Micron exiting the 3D Xpoint business, there was some speculation on what that would mean for Intel, as Micron has been producing the memory for Intel’s Optane products. Tom’s Hardware spoke to both Micron and Intel, and received the following statements:
"Micron is committed to completing our obligations under the existing wafer supply agreement with Intel, which currently goes through the end of CY21, and Micron intends to retain all of its intellectual property associated with 3D XPoint,” says Micron (via Tom’s Hardware).
"Micron's announcement doesn't change our strategy for Intel Optane or our ability to supply Intel Optane products to our customers,” says Intel (via Tom’s Hardware).
23:01 | AMD to Become TSMC’s Second-Largest Customer
New reports detailing the capital expenditure (CapEx) spending of Samsung and TSMC suggest that Samsung will not be displacing TSMC as the industry leader anytime soon, though Samsung does appear to be closing the gap in overall manufacturing prowess and technology.
Back in November of 2020, Bloomberg reported that Samsung planned to spend $116 billion to bolster its manufacturing capabilities in an effort to catch up to TSMC by 2022. This includes expanding the development of GAAFET transistors, which we’ve been discussinga lot lately. As we’ve said previously, Samsung intends to use GAAFETs, or its trademarked Multi-Bridge Channel FETs (MBCFET) at 3nm, while TSMC will still be using FinFETs. At the Institute of Electrical and Electronics Engineer’s recent International Solid-State Circuits Conference (that would be the IEEE’s ISSCC for acronym enthusiasts), Samsung recently demonstrated the performance and power efficiency improvements that it could accomplish with GAAFETs and SRAM.
Seeking Alpha reports that TSMC currently holds an advantage over Samsung in both production capacity and revenue, as well as a broader customer base for contract chip making. TSMC has an overall 2.3x greater capacity than Samsung. Samsung has been slowly closing the gap, in 2016 TSMC’s capacity advantage was 3.3 times what Samsung could produce. Smaller nodes show an even greater advantage for TSMC. In 2021, TSMC’s capacity for 7nm is projected to be between five and six times that of Samsung, and for 5nm, TSMC is projected to have four times the capacity of Samsung. TSMC is also looking to begin 3nm production this year, something Samsung is not prepared for yet.
Furthermore, TSMC currently boasts a broader client base than Samsung. Industry estimates state that as much as 60% of Samsung’s capacity is allocated for its own devices, like Exynos SoCs for Samsung smartphones. Meanwhile, TSMC serves 460 customers, including Apple, Hi-Silicon, AMD, Nvidia, Broadcom, Qualcomm, and Intel, among others.
Speaking of customers, one highlight of the reports is that AMD is on track to become TSMC’s second largest customer for 2021, accounting for an estimated 9.2% of TSMC’s 2021 revenue. AMD will still be far and away from TSMC’s top customer in Apple, which will account for 25.4% of TSMC’s 2021 revenue. However, AMD will be ahead of other prominent customers like Nvidia, Qualcomm, and Broadcom.