HW News - RIP New RTX Models, AMD Supercomputer, Intel Selling Off NAND Business
Posted on October 31, 2020
This hardware news episode was filmed prior to the AMD news update, which we covered in full depth on the YouTube channel over here: AMD RX 6900 XT, 6800 XT, & 6800 Specs. We won't be going back over that announcement in this content, so you can check out the full RX 6000 details in that fully dedicated piece with additional AMD Q&A. In this one, we'll be talking RTX models being canceled or pushed back, AMD supercomputer pushes, Intel selling off part of its business, and more.
02:01 | Cyberpunk 2077 Delayed Until December 10
Luke Lafreniere from LMG said it best on twitter: “Cyberpunk 2077 delayed until 2077.” The good news is that, in a world where lots of products are rushed to market and result in ruinous repercussions for a brand’s image, delaying a launch -- although annoying -- could be the right move to ensure the product works.
CD Projekt Red went to twitter to tell everyone “we need to talk.”
In its tweet, the company said “we have important news to share with you,” then posted the following:
“Today, we’ve decided to move the release date of Cyberpunk 2077 by 21 days. The new release date is December 10th.” CD PR then apologized, and stated the following as reasons:
“The biggest challenge for us right now is shipping the game on current-gen, next-gen, and PC at the same time, which requires us to prepare and test 9 versions of it (Xbox One, Xbox One X, Xbox Series S and X, PS4 and Pro, PS5, PC, and Stadia)... while working from home. [...] We need to make sure everything works well and every version runs smoothly. We’re aware it might seem unrealistic when someone says that 21 days can make any difference in such a massive and complex game, but they really do.
“Some of you might also be wondering what these words mean in light of us saying ‘we achieved gold master some time ago.’ Passing certification, or ‘going gold,’ means the game is ready, can be completed, and has all content in it. But it doesn’t mean that we stop working on it and raising the quality bar.”
CD PR explained that this is when day 0 patches are often worked on, and noted that it miscalculated the time required for those day 0 patches. At this point, we’d be surprised if CD PR moves the dates again: The company is already missing out on significant sales potential during Black Friday and Cyber Monday sales periods, even without discounts, and is missing out on periods when new computers will be built using AMD and NVIDIA parts alike. The new consoles are set to launch on November 10th for Xbox and November 12th for PS5, so they’re missing those dates by a month.
So, with Christmas looming, CD PR will be under pressure to hit this date properly to ensure maximum revenue as people buy gifts.
Related to this, EVGA’s Jacob Freeman responded to a question from “Hans” on twitter, who asked “so, you’re saying there’s a chance we get 3080s before Cyberpunk?” Freeman replied, saying, “Chance, yes. But we’ll refrain from any estimates right now. Just too many variables unfortunately to predict it.”
Source: https://twitter.com/CyberpunkGame/status/1321128432370176002/photo/1
05:58 | Confirmed: Nvidia Cancels RTX 3080 20GB, 3070 16GB
GamersNexus has been able to independently confirm that NVIDIA, at least for now, has indeed cancelled the production of two would-be RTX SKUs slated for late 2020. The SKUs featured one based on the 8nm GA102 silicon (RTX 3080) and another based on the 8nm GA104 silicon (RTX 3070).
The GA102-based RTX 3080 variant would have doubled the memory buffer from 10GB to 20GB, and the GA104-based RTX 3070 variant would’ve done the same, moving from 8GB to 16GB. The SKUs would’ve been something of a response to AMD’s RX 6000-series/Navi 21, which are pushing memory capacity as a technical strong point.
However, with NVIDIA currently unable to keep pace with RTX 3080 and 3090 demand, and having already pushed back the RTX 3070 launch, Nvidia has instructed its AIB partners to abandon plans for any RTX 3080/3070 spin offs. Furthermore, GDDR6X memory supply from Micron is currently tight as it is.
It’s likely that some version of these cards will still come to fruition at some point, potentially coinciding with a 7nm Ampere refresh in 2021. The current rumor among AIB partners is that this may be used for retaking ranks against AMD if RX 6000 turns out well.
Source: GN
08:38 | RTX 3070 Delay: “We Wouldn’t Have Had A Single Card”
With the RTX 3070 launch, we started talking to board partners and OEMs about why NVIDIA initially delayed its launch window for the RTX 3070. The card was initially set to launch in mid-October, but got pushed back to the 27th and 29th (for partner models). In speaking with multiple board partners, GamersNexus was given the following anonymous comments:
“It’s good that they delayed. We wouldn’t have had a single card if they kept the original launch date.”
“Supply would have been worse than the 3080 and 3090 if [NVIDIA] kept the earlier October window.”
For other updates on supply, we asked the partners about off-record volume shipments. We’ve been told that supply is currently variable, pending NVIDIA’s allocation, and that they are doing anywhere between 400-500 units per week and upwards of 2000 for a good week, depending on the partner and how many units are stockpiled in a given production window. Our understanding is that companies with back orders open are still working through September back-orders, so if you aren’t on a list now, your best shot may still be sniping a restock on a retailer. Physical retail may be easier if you live near a computer shop, just because you won’t be competing with as many people. It’s a good time to make friends with your local floor staff.
Source: GN
11:02 | Intel Sells Off NAND Business
What started out as a rumor earlier this week quickly morphed into fact: Intel has sold off its NAND and storage business to SK Hynix. The acquisition will see SK Hynix pay Intel $9 billion for “the Intel NAND memory and storage business, which includes the NAND SSD business, the NAND component and wafer business, and the Da4lian2 NAND memory manufacturing facility in China,” according to a joint press release by both companies.
Interestingly, Intel will still retain its Optane technology and related IP, and it will still (presumably) fall under Intel’s Non-volatile Memory Solutions Group (NSG). This suggests that Intel isn’t getting out of the storage business -- it’s getting out of the NAND flash business. Intel will still develop products based on Optane and 3D XPoint.
According to Intel and SK Hynix, the acquisition timeline will play out across two phases, and will not fully close until March of 2025. Both companies expect all regulatory hurdles to be cleared by late 2021, by which SK Hynix will pay Intel $7B and take possession of Intel’s SSD business, all related IP, and Intel’s NAND fab in Da4lian2, Liao2ning2 China. However, SK Hynix will not bring on any NAND flash employees or acquire any NAND flash-related IP, despite acquiring the fab.
Instead, Intel will continue to develop NAND for the next four years, until March 2025, where phase 2 begins. During this phase, SK Hynix will pay Intel the remaining $2B and take possession of Intel’s NAND IP and the employees that operate Intel’s NAND fab in China. The exaggerated timeline for this acquisition is a bit odd, and neither company has offered an explanation for it as of this writing. We’re sure to know more about it in the future, as the deal progresses.
Source: https://www.wsj.com/articles/intel-nears-deal-to-sell-nand-memory-unit-to-sk-hynix-11603129781
https://news.skhynix.com/sk-hynix-to-acquire-intel-nand-memory-business/
13:40 | Google DOJ Lawsuit, Google’s Response
While not strictly hardware or enthusiast related news, this one is still relevant: the biggest news item this past week, other than the 3070 and AMD stuff, was the landmark antitrust lawsuit slapped against Google by the Department of Justice. Although this is just one suit against one company, it’s likely the opening salvo in the long brewing fight against Big Tech -- companies like Amazon, Facebook, and Google, who all wield obscene market power and influence.
In a press release -- subtly titled “Justice Department Sues Monopolist Google For Violating Antitrust Laws” -- the Department of Justice and 11 state Attorneys General have filed an antitrust lawsuit against Google. The lawsuit alleges the following:
- Entering into exclusivity agreements that forbid preinstallation of any competing search service.
- Entering into tying and other arrangements that force preinstallation of its search applications in prime locations on mobile devices and make them undeletable, regardless of consumer preference.
- Entering into long-term agreements with Apple that require Google to be the default – and de facto exclusive – general search engine on Apple’s popular Safari browser and other Apple search tools.
- Generally using monopoly profits to buy preferential treatment for its search engine on devices, web browsers, and other search access points, creating a continuous and self-reinforcing cycle of monopolization.
That last one is a particular doozy, as it loosely refers to Google’s lucrative deals with Apple and Mozilla, in which Google literally pays billions of dollars a year to be the default search engine in web browsers like Safari and Firefox. Google, of course, maintains that these deals are perfectly legal, and even Mozilla -- who often masquerades as the anti-Google -- offered a kinda-sorta defense of its own.
“In this new lawsuit, the DOJ referenced Google’s search agreement with Mozilla as one example of Google’s monopolization of the search engine market in the United States. Small and independent companies such as Mozilla thrive by innovating, disrupting and providing users with industry leading features and services in areas like search. The ultimate outcomes of an antitrust lawsuit should not cause collateral damage to the very organizations – like Mozilla – best positioned to drive competition and protect the interests of consumers on the web,” says Mozilla.
So, Mozilla’s take is more or less “Hey, feel free to shoot the baker, just don’t slap the cake off of our plate.”
As this case is just now beginning -- and will likely take years to resolve -- there’s not much in the way of specific to discuss. However, there’s a few points worth mentioning. First, while the DOJ is making some broad, sweeping allegations against Google, it’ll have to pick its battles based on what it can make stick. Secondly, no amount of hand waving by Google, who has already called the lawsuit “deeply flawed,” is going to make this go away. Third, Google will undoubtedly be under the scrutinous gaze and lens of lawmakers for a long time to come.
Lastly, while there’s some speculation about how prepared the DOJ is to take on Google and how its case may differ from past precedents, the DOJ doesn’t just trot out these cases everyday. And if history is any indication, as with AT&T in the 1970s and Microsoft in the 1990s, when the DOJ swings, it usually doesn’t miss.
Source: https://www.blog.google/outreach-initiatives/public-policy/response-doj/
17:41 | AMD Notches Another Supercomputer Win
AMD’s supercomputing deals just keep coming. This time, AMD is set to outfit the Australian Pawsey Supercomputing Centre with a new supercomputer built out of both AMD CPU and GPU silicon.
While the supercomputer itself is a bit of a mystery, it’ll be built on HPE’s Cray EX platform. We went over some details for HPE’s Cray Ex architecture previously for the upcoming Crossroads supercomputer, so check that out for a primer. The unnamed new supercomputer will also replace Pawsey’s current systems: Magnus and Galaxy.
While lots of details are being kept close to the chest on this one, it seems the new supercomputer will use future AMD Epyc (see: Milan) CPUs and Radeon Instinct GPUs. Additionally, the machine will leverage over 200,000 CPU cores and over 750 GPUs, plus 548TB of memory. Compute power is set to top out at 50 petaFLOPs.
Construction of the system will begin next year, with two cabinets going online towards the end of 2021, with the rest of the cabinets coming online in 2022.
Source: https://www.theregister.com/2020/10/19/new_pawsey_supercomputer/
18:48 | Intel 3Q2020 Earnings
Intel has reported its revenue for Q3 2020, and while Intel has been notching record quarters throughout the year, Q3 marks a decline in revenue and profits for the chipmaker. Notably, Intel is also feeling the sting of the Covid-19 pandemic and the strain and uncertainty it has put on the various markets that Intel deals in.
For the third quarter, Intel is reporting a total revenue of $18.3 billion, which surpassed Intel’s previous expectations, but marks a 4% decline YoY. It also marks a significant dip QoQ, compared to last quarter’s $19.7B in revenue. Intel’s gross margins are also noticeably softer, coming in at 53.3%, compared to 58.9% YoY.
As usual, Intel’s Client Computing Group (CCG) and Data Center Group (DCG) are the stars of the show, even if revenues from these business units are a bit flat for CCG and declining for DCG. Intel’s CCG brought in $9.8B for Q3, which is 1% up YoY, driven by demand for notebooks for learning and working from home. Intel’s DCG brought in $5.9B, marking a 8% decline YoY and 17% QoQ. According to Intel, this is due to the current pandemic affecting the economy, as well as weaker demand for Intel’s enterprise and government products. Intel also notes that ASPs are trending down across several product segments.
Outside of revenues, Intel is reporting that its investment in 10nm manufacturing is paying off, as the company’s 10nm fab in Arizona is now fully operational. As such, Intel expects to ship 30% more 10nm volume by the end of 2020 than previously expected. However, despite the increased 10nm capacity, Intel made it clear that its upcoming Ice Lake-SP parts would not ship until early 2021, leaving Intel to continue to lean on 14nm server parts for the remainder of 2020.
Regarding 7nm, Intel is still keeping its plans for third party foundry use close to the chest. Intel’s 7nm parts are still delayed to 2022 at best, if not more like 2023. Intel is expected to make some important manufacturing calls regarding 7nm at some point in 2021 -- e.g., what parts it will manufacture itself (if possible) and what parts will be manufactured by the likes of TSMC.
Additionally, Intel announced that it has taped out and powered on its Intel Xe HPG DG2 GPU, which should be Intel’s first discrete add-in GPU card.
22:36 | UMC Agrees to $60M Settlement for IP Infringement
According to a report from TrendForce, United Microelectronics Corp (UMC) has agreed to pay up to $60 million in fines in response to the lawsuit brought against it by the Department of Justice back in 2018. The crux of the lawsuit alleged that UMC funneled Micron’s DRAM IP and manufacturing secrets to Fujian Jinhua Integrated Circuit Company (JHICC).
JHICC is a Chinese established firm that was created to bolster domestic semiconductor production in China. JHICC struck a deal with UMC to help it develop a DRAM manufacturing process. Somewhere along the way, Micron accused UMC of poaching its engineers and persuading them to bring along IP and manufacturing secrets with them. Micron eventually sued both UMC and JHICC for patent infringement.
Fast forward a couple of years, and UMC has decided to settle with the DOJ; it’ll still have to settle with Micron separately, as will JIHCC.
UMC released a statement (per Tom’s Hardware):
“As previously disclosed, United Microelectronics Corp. has been responding to charges brought by the U.S. Department of Justice based on, among other things, an allegation of conspiracy to engage in theft of trade secrets of Micron Technology by the Company, Fujian Jinhua Integrated Circuit Co., and certain individuals.
The Company has been negotiating and discussing potential resolution of these charges with DOJ, and the Company anticipates reaching a resolution with DOJ in the foreseeable future. It is anticipated that a court hearing will occur in the near future to address the anticipated resolution of these charges.
In connection with this anticipated hearing, UMC and DOJ have each submitted a sentencing memorandum to the court regarding the terms of the proposed resolution, which would include a proposed plea to a lesser charge and a proposed $60 million fine.
As such, the Company can now reasonably estimate a probable loss and has recorded an aggregate accrual of $60 million with respect to these matters. As the discussions are continuing and any final resolution will require court approval, there can be no assurance that the Company's efforts to reach a final resolution will be successful, including as to the amount of a fine, or the timing and terms of such a resolution."
Source: https://www.trendforce.com/presscenter/news/20201022-10524.html