00:53 | Jay vs. Steve Stream & GN Store Update
This news item is exclusive to the video. Today, Saturday 8/8, at 12PM Pacific / 3PM Eastern (US), we will be competing versus JayzTwoCents in a livestream to troubleshoot a PC on LinusTechTips’ channel, hosted by Linus. Check out the news video for basics, and don’t miss it!
01:54 | RTX 3000 in September, 2070 Super Not Dead
There was a rumor propagated earlier this week that the RTX 2070 Super is getting discontinued. We followed-up with some board partners and learned that at least a few of them are still ordering RTX 2070 Super GPUs from NVIDIA, and that their models are still getting manufactured. In that sense, it’s not discontinued yet. It probably will be soon, but the rumor is incorrect and partners are still buying GPUs from NVIDIA. That said, RTX 3000 should be announced on September 9th, according to board partner discussions with GamersNexus, so it’s reasonable to assume that stock of the current cards will start to dwindle as manufacturers try to clear it out.
04:11 | Update: Mindfactory NVIDIA & AMD GPU RMA Rates
Mindfactory is one of the more prominent German retailers serving the market, and the retailer regularly makes data available regarding its business. Mindfactory recently published its RMA rates for NVIDIA and AMD GPUs. While this data only represents a glimpse into the overall market, and should in no way be taken as conclusive, it’s usually interesting nonetheless, as these data sets often highlight certain patterns.
A couple of important caveats to address, first. Mindfactory is updating this sheet every 5 minutes, so the numbers are changing. Second, they did not disclose the timeline for which it collected this data -- we’d guess it’s been collected over the last year, but that is just a guess. Secondly, Mindfactory didn’t disclose the reason for RMAs, so this data can’t be taken as empirical for trying to determine failure rate for a specific model or vendor. For instance, it’s entirely possible AMD’s RX 5000 RMA rates have been affected by the poor driver support early on, which isn’t the same as failure.
Overall, Mindfactory is reporting the RMA numbers for what appears to be a total of 172,650 GPUs; 50,440 belonging to AMD, and the remaining 122,210 being Nvidia’s. Earlier reports indicated 40,000 units and 76,000 units, which was accurate to the numbers provided at the time, but is no longer accurate. Mindfactory released the data before completing it, which gave a somewhat irresponsible picture of marketshare that painted AMD in a much more favorable light than reality. According to the data, as of time of writing, Mindfactory saw a total of 3016 RMAs for Nvidia and 1625 for AMD. Given that NVIDIA’s sales are higher, this makes sense.
The data encompasses most of each company’s current product stack, with data representing the GTX 1660 Ti at the bottom of the stack for Nvidia, and going all the way up to the RTX 2080 Ti. Looking at the data, the RTX 2080 Ti has the single highest RMA rate for any GPU model, with an RMA rate hovering just over 5%. The cost may factor into this, as people might be less willing to wait around for a fix on a $1000 card.
For AMD, Mindfactory’s RMA data encompsess data from the RX 5500 XT on up to the RX 5700 XT. It looks like the RMA rate for the RX 5500 XT is fairly low, and gets progressively higher as we move up the stack to the RX 5600 XT and RX 5700 XT. Overall, AMD’s RX 5500 XT has the lowest RMA rate at 1.1%, but also has the lowest sales figures. For Nvidia, its GTX 1660 Ti has the lowest RMA rate at 1.2%.
Mindfactory’s data also has numbers for specific AIB models and partners as well, although not all RMA quotes from all models and vendors were available.
Primary source: https://docs.google.com/spreadsheets/d/e/2PACX-1vQbqvpHU0z6oK9HaCRbDPhkEoq5OA32mRGysyDZYhFsAk2kwKie-DaKplFyco7vwlw3ansFpjNstrpG/pubhtml
07:23 | Arm Co-Founder Doesn’t Welcome NVIDIA
There’s at least one person who doesn’t like the idea of Nvidia owning Arm, and it just so happens that this is the same person who co-founded Arm. Arm co-founder Hermann Hauser recently made it clear in a BBC interview that he wasn’t for the deal.
"It's one of the fundamental assumptions of the ARM business model that it can sell to everybody," Hauser said to BBC.
"The one saving grace about Softbank was that it wasn't a chip company, and retained ARM's neutrality. If it becomes part of Nvidia, most of the licensees are competitors of Nvidia, and will of course then look for an alternative to ARM."
Hauser was also against Softbank acquiring Arm in 2016, but Softbank vowed to keep Arm’s focus in Cambridge and increase employment there, which he fears won’t be the case should Nvidia get a hold of Arm. "It will become one of the Nvidia divisions, and all the decisions will be made in America, no longer in Cambridge,” said Hauser.
Hauser states he would like to see Arm become an independent UK company again, and that the UK Government should move to make that happen. "The great opportunity that the cash needs of Softbank presents is to bring ARM back home and take it public, with the support of the British government," said Hauser.
Hauser’s comments come as Nvidia is reportedly in advanced discussions to acquire Arm; however, such a deal would welcome the scrutinous gaze of lawmakers and regulators. So, it remains to be seen how likely an Arm acquisition by Nvidia is.
Primary source: https://www.bbc.com/news/technology-53637463
Secondary source: https://www.tomshardware.com/news/arm-cofounder-nvidia-owning-arm-would-be-a-disaster
09:58 | Samsung, TSMC, Foxconn Potentially Interested in Arm
Nvidia isn’t the only silicon designer throwing in to acquire Arm. It appears as if Arm is increasingly becoming the center of a bidding war between global chipmakers, including Samsung and TSMC, with additional interest from board partner Foxconn.
Early reports have suggested that Samsung was only interested in acquiring a minority stake -- something to the tune of 3 to 5 percent, according to an anonymous industry official (via Korean Times). According to the report, the investment would be similar to the equity investment Samsung has made in other companies, such as ASML.
The same report out of Korean Times also suggested that Arm would “be acquired by a consortium led by multiple parties from the semiconductor industry given the complex nature of Arm's shareholding structure." However, Samsung has refuted this statement, saying that it is “groundless.”
Additionally, according to another report from Nikkei Asian Review, both TSMC and Foxconn have been approached. Sources told the outlet that financial data and projections were given to both companies regarding an investment in Arm. However, while it appears both TSMC and Foxconn may be interested, the two are monitoring how talks with Nvidia progress.
"TSMC and Foxconn are closely monitoring how the talks between SoftBank and Nvidia progress. The two are still interested but are balking at a full acquisition,” sources told Nikkei Asian Review. This statement seems to echo that of Samsung’s alleged interest in buying a smaller stake, rather than trying to buy the company outright.
Samsung update: https://hothardware.com/news/samsung-joins-chip-consortium-counter-nvidia-bid-arm
11:23 | Intel Data Breach: 20GB of IP Hits the Internet
The hits just keep coming for Intel, it seems. Intel is still dealing with the fallout coming off the heels of its 7nm delay, and now, it has to contend with what looks like a major IP leak. We won’t be directly linking the leak documents, as we’re uncertain if they are secure to download. We’d exercise caution playing with dumps like this, as it can be a security vulnerability.
The leaked documents were published by Till Kottmann, a Swiss software engineer. Kottman has a history of publishing leaks from various tech companies, according to ZDNet. Kottman claims to have received the leaked files from a hacker who purportedly gained access to them earlier this year, with Kottman indicating on twitter that it was closer to May. According to the hacker, the files were scraped from an unsecured server hosted on Akamai’s CDN. Furthermore, Kottman claims this is the first of many Intel leaks to come.
The leaked files contain information on Intel’s various Lake platforms, such as Tiger Lake and Elkhart Lake, and going back as far as Kaby Lake. While the files don’t seem to contain anything intrinsic to CPU design, the information still seems to be very sensitive, and no doubt valuable. According to Kottman, the leak contains the following:
- Intel ME Bringup guides + (flash) tooling + samples for various platforms
- Kaby Lake (Purley Platform) BIOS Reference Code and Sample Code + Initialization code (some of it as exported git repos with full history)
- Intel CEFDK (Consumer Electronics Firmware Development Kit (Bootloader stuff)) SOURCES
- Silicon / FSP source code packages for various platforms
- Various Intel Development and Debugging Tools
- Simics Simulation for Rocket Lake S and potentially other platforms
- Various roadmaps and other documents
- Binaries for Camera drivers Intel made for SpaceX
- Schematics, Docs, Tools + Firmware for the unreleased Tiger Lake platform
- (very horrible) Kabylake FDK training videos
- Intel Trace Hub + decoder files for various Intel ME versions
- Elkhart Lake Silicon Reference and Platform Sample Code
- Some Verilog stuff for various Xeon Platforms, unsure what it is exactly.
- Debug BIOS/TXE builds for various Platforms
- Bootguard SDK (encrypted zip)
- Intel Snowridge / Snowfish Process Simulator ADK
- Various schematics
- Intel Marketing Material Templates (InDesign)
Intel issued a statement confirming the leak, but refuting Kottman’s claim that it was hacked. Instead, Intel maintains that the information originated from its Resource and Design Center. This would be indicative of a partner leak, which GN has received from partners covertly in the past -- most recently, it was from AMD’s similar resource center. That’s much less severe than a server hack, if so.
"We are investigating this situation. The information appears to come from the Intel Resource and Design Center, which hosts information for use by our customers, partners and other external parties who have registered for access. We believe an individual with access downloaded and shared this data,” says Intel.
15:39 | AMD x86 Market Share Reaches Historic Highs
The last couple weeks, we’ve discussed AMD’s meteoric rise on the stock markets and record earnings over the last quarter. AMD’s upward ascension continues, as the company is riding historic highs regarding its market capitalization, stock price, and x86 market share.
AMD’s stock price has been trending upwards all year, but within the last few weeks, it entered new territory, as it was higher than Intel’s for the first time in years. On August 5, AMD’s share price peaked at $85.31, and broke the $100B market cap barrier for the first time. However, at time of writing, AMD’s stock has tumbled down a bit to $83.26, with a market cap of ~$98B. Such is the stock market.
Separately, AMD has also notched its highest x86 market share since 2013, as new market share data has been published by Mercury Research. AMD has also secured its highest ever notebook market share, and continues to claw away server market share from Intel.
For the numbers themselves, Mercury research reports that AMD has captured 18.3% of the overall x86 market. Again, that’s the highest share AMD has had since 2013. To break it down by segment, AMD currently holds 19.2% of the x86 desktop market, an increase of 2.1 percentage points yearly and 0.6 quarterly. Moving onto notebooks, AMD is seeing record market share, with a reported share of 19.9%, a 5.8 point gain yearly. When it comes to servers, AMD CEO Lisa Su recently said the company finally achieved double-digit share, but stopped just short of citing any numbers. AMD bases its server share numbers off of IDC’s research, which reports only on 1P and 2P server market share. Mercury Research reports not only on 1P and 2P, but 4P and 8P configured x86 servers as well.
As such, there’s some discrepancy between AMD’s claims and Mercury’s numbers. Mercury Research reports that AMD has captured 5.8% of the x86 server market. Intel should be better represented when counting 4P and 8P servers.
18:05 | Bill English and Frances Allen Pass Away
This past week, we lost not one, but two iconic names in computing. Both Bill English and Frances Allen have passed away, each well known for their seminal contributions to home computing and computer programming, respectively.
Bill English is credited along with Douglas Engelbart (who died in 2013) for the creation of the computer mouse. English built a prototype based on Englebart’s concept in 1963. The first prototype was built from a block of wood, with one switch and metal discs for wheels that recorded movement. We’ve actually personally seen this mouse at the Computer History Museum in Mountain View, California, which you all should visit if ever in that area. This early version of the venerable rodent was shown during the 1968 “Mother of All Demos” convention, where it was shown alongside early versions of word processing, hyperlinks, and video calls.
English later moved onto the famous Xerox Palo Alto Research Center, where he began to refine the mouse by including a track ball rather than metal discs. This version of the mouse would serve as an early progenitor for computer mice popularized by Apple and Microsoft in the coming decades of home computing.
Frances Allen, on the other hand, was a leading computer scientist, and the first woman to receive the A.M. Turing Award and the first female IBM Fellow. Frances Allen is well known for contributions in software compiler research, specifically as it relates to parallel computing and compiler efficiency. According to IBM, where she spent some 45 years, her work on parallel optimization and inter-procedural analysis still leads the way.
Frances Allen’s work and research produced the Stretch-Harvest compiler, the Cobol compiler, and the Parallel Fortran. She also led the way into research into parallel processing, which would become the basis for GPUs and supercomputers, among other things.
Bill English passed away at the age of 91, while Frances Allen was 88.
Ex-Intel Engineer Francois Pidnoel On Current State of Intel
The former Principal Engineer at Intel, Francois Pidnoel, is back with another deep dive into Intel’s troubles. Pidnoel recently shared some insight into what he believed led Apple away from Intel, and now he’s offering his thoughts on the current state of Intel’s manufacturing woes -- and how he thinks they can be fixed.
Pidnoel offered up his take via his YouTube channel, but PCWorld’s -- and friend of GN -- Gordon Mah Ung parsed the video and dissected some excellent points we want to highlight. First and foremost, Pidnoel echoes others’ criticisms that Intel is currently being led by non-technical people, and that the chipmaker promotes a culture of “MBAs” over those with a technical background. The somewhat opposite is true for AMD currently, whose CEO is an electrical engineer with extensive semiconductor and silicon design experience.
Secondly, Pidnoel asserts that AVX512 was a mistake. This would no doubt please the curmudgeon Linus Torvalds, who recently dubbed AVX512 a power virus. AVX512 is a wider vector instruction set aimed at certain workloads. However, as Pidnoel puts it, AVX512 “is designed for a race of throughput that is lost to the GPU already.”
Pidnoel states that the inclusion of AVX512 leads to larger, and more complex CPU dies, as well as having inherent power costs. That echoes Torvald’s sentiments almost exactly, as Torvalds stated he’d rather see FP math used in the GPU, and see Intel’s transistor budget aimed at other things than AVX512, and also complained about AVX512 taking away top frequency for Intel CPUs. Intel CPUs usually downclock when executing AVX512 instructions.
Finally, Pidnoel also criticized Intel for wasting die and core space on its Xeon line, saying Intel should specialize its Xeons for niche applications, rather than selling chips that waste as much as 10% of the chip space. Pidnoel highlighted applications that don’t need AVX512 as a prime example to be more conservartve. The former Intel engineer also touched on Intel losing focus on its core CPU business and issues with Intel’s hyperthreading performance compared to Ryzen. The video, as well as PCWorld’s recap, are well worth a look. Both are linked below.
Editorial: Eric Hamilton
Host, Additional Reporting: Steve Burke
Video: Keegan Gallick, Andrew Coleman