HW News - 4nm TSMC Node, RTX 3080 Leaks, Sony PS5 Design, & Intel Lakefield
Posted on June 15, 2020
While the week started off rather slowly, the news crescendoed towards the end of the week, capped by Sony’s Future of Gaming event where we finally caught a glimpse of the elusive PlayStation 5 console. Also interesting is the fate of Kaby Lake-G, held in limbo while Intel and AMD decide who should deliver driver support.
Some lesser stories include news on TSMC -- on fronts both manufacturing and geopolitical. There’s finally also a speculative execution attack that doesn’t come with Intel’s name attached to it. We also have Intel’s Lakefield CPUs, which may be Intel’s most interesting CPU line in years. There’s also news of a particular ISP throttling entire neighborhoods to deal with heavy internet traffic.
This past week at GN, we revisited AMD’s Ryzen 7 1700 for 2020, as well as getting back to case reviews with Cooler Master’s TD500 Mesh case. We also detailed our experiences, to date, with Thermaltake’s marketing and engineering.
Follow below for the video embed and article.
00:55 | GN’s Thoughts on the RTX 3080/3090 Rumors
We’re sure that some of you have now seen the alleged early mock-ups of NVIDIA’s next-generation video cards. We won’t really be talking about it in too much depth, but did want to offer some of our thoughts from an industry perspective.
In our Patreon Discord when the images were first leaked, we had someone ask whether the design might be final since they figured that a tool would be needed to have made the product that was leaked. We answered it there already, but the short version is that no, you wouldn’t need tooling to make a plastic shroud. Typically, companies will make early mock-ups with 3D printing, then later move to a CNC with one of the many variants of plastic. Other options include making resin molds, depending on what the factory intends to do with the model once done. Tools are used for things like case panels, while shrouds would be done with molds. For what it’s worth, we’ve made one-off molds for our own products, like the GN toolkits, where we might spend a couple grand just to try something out. One of the factories we’re working with on an unannounced new product only charged us $400 for the mold, so the point is that doing mock-ups like this isn’t prohibitively expensive for a company like NVIDIA. It’s likely that a lot of these exist, so we wouldn’t really say that it’s guaranteed any of the leaks will ever show up in the wild. It’s definitely cool to see some of the early concepts, though, and opens the window to something we don’t normally see.
05:54 | PS5 Design Unveiled
Sony, which has been locked in a waiting game with Microsoft, has finally revealed what the PS5 will look like, and as usual, the internet has some thoughts. Online commentary of the design has mostly included references to routers, humidifiers, and euphemisms of anatomical componentry.
At its Future of Gaming event, Sony made a slew of announcements, mostly in regards to titles coming from its first-party studios, but also offered a glimpse of the PS5 hardware. There’s the PS5 itself, obviously, but Sony also showed off the new DualSense Controller and Pulse 3D wireless headset, all sporting a white coat of paint to match the aesthetic of Sony’s new console.
For the PS5 itself, as usual, it’s an acquired taste. It’s definitely more bold and aggressive than Microsoft’s mini-ITX inspired design with the Xbox Series X. The PS5 has some interesting angles and curves, with blue lights to add a soft glowing effect. It appears to have vents along the perimeter of the interior, partially obscured by the white hood. We’re concerned primarily about ventilation and airflow on the cooler, but our past leaked information has indicated that the console has spent disproportionately more on cooling than previous models, so we’ll test it once it’s in.
As with past models, the PS5 can be oriented vertically or horizontally. The PS5 will come in two models, at least initially; a “fatter” model sporting an optical drive, and a slimmer version sans optical drive.
There was no word on price, but Sony is still promising a holiday 2020 shipping date. Sony also announced that an enhanced and expanded edition of Grand Theft Auto V will come to the console in 2021. After all, nothing says welcome to 2021 like a game from 2013. We look forward to the 10-year anniversary, remastered remaster in 2023.
Source: https://www.cnet.com/news/sony-revealed-ps5-console-heres-what-we-know/
09:03 | Intel & AMD Can’t Decide Who Supports Bastard Silicon
Kaby Lake-G, the once unfathomable collaboration between archrivals Intel and AMD, seems to be caught in the middle of a spat between its parents. The issue, it seems, is one of driver support. Specifically, which company should provide it.
Kaby Lake-G eventually took the form of Hades Canyon, pairing Kaby Lake-G processors with AMD’s Radeon Vega M graphics. At the time, we praised Hades Canyon for its incredible processing power and had a lot of fun overclocking it with custom cooling. Fast forward a few years, and now Intel and AMD’s collaboration is without proper driver support, as well as without indication on when and if that will change. Intel defaulted support to AMD some time ago, and as Tom’s Hardware reports, AMD hasn’t offered an updated driver package either, outside of the main Adrenaline 2020 suite. This will prevent Kaby Lake-G systems from moving to the Windows 10 version 2004 update.
Tom’s Hardware reached out to both Intel and AMD, but only received an official comment from Intel. "We are working to bring back Radeon graphics driver support to Intel NUC 8 Extreme Mini PCs (previously codenamed ‘Hades Canyon’),” Intel told Tom’s Hardware.
Source: https://www.tomshardware.com/news/amd-pulls-driver-support-for-intel-kaby-lake-g
10:50 | SLS Side-Channel Attack Affects Armv8-A Chips
Were we to personify Intel, we’d assume it’s smiling right now. The latest speculative execution flaws don’t affect its chips this time, but in a rare occurrence, affect ARM. ARM is now dealing with a new side-channel attack preying on chips based on the Armv8-A architecture.
Arm actually claims the new flaw, known as Straight Line Speculation (SLS), is a derivative of Spectre. Where Spectre affects all chip manufacturers (mostly Intel, but also AMD and Arm), SLS is strictly aimed at Arm. SLS was discovered through research done at the Google SafeSide Project, and has been given the CVE-2020-13844 entry.
According to Arm, SLS occurs when a processor speculatively executes instructions beyond what should be a change in control flow. When an Arm-based chip speculatively executes the next instructions linearly in memory past a point of change in control flow, it could open up changes in the cache subsystem, offering a chance for an attacker to leak data. So far as we understand it, anyway.
While Arm says the security risk is low and such an attack would be hard to pull off in practice, it is offering mitigations nonetheless. So far, Arm has contributed patches to open-source tooling and communities such as GCC, LLVM, Trusted Firmware-A, Op-Tee, FreeBSD, and OpenBSD. Arm notes that, in most cases, it doesn’t expect a performance impact from the mitigations.
Source: https://www.zdnet.com/article/arm-cpus-impacted-by-rare-side-channel-attack/
Arm SLS whitepaper: https://developer.arm.com/support/arm-security-updates/speculative-processor-vulnerability/downloads/straight-line-speculation
13:04 | TSMC Preparing N4 4nm Node
TSMC has recently announced that it will bridge its N5 and N3 process nodes with a 4nm N4 node. This announcement seemingly came from left field, because while TSMC has detailed its 5nm and 3nm processes, 4nm has never been mentioned.
TSMC Chairman Mark Liu told EE Times that “N4 is an evolution from N5. We’re already in business negotiations with customers on N4.” It was assumed TSMC was going to skip 4nm on the road to 3nm, but it appears that’s not the case.
On the surface, it seems as if the N4 node will serve as an iterative upgrade to TSMC’s N5P. TSMC’s N5P is itself an enhanced intra-node upgrade from N5, similar to what TSMC has been doing at 7nm (see: N7/N7P/N7+). It seems as if N4 will add another rung in the ladder between N5P and N3, assuming TSMC doesn’t take 5nm beyond N5P. We’ll see.
TSMC’s N5 hit risk production last year, and ramped in April 2020. Apple and AMD are already rumored to be among TSMC’s premier customers at N5. Meanwhile, N5P is supposed to arrive in 2022, while N4 looks good for 2023.
Source: https://www.eetimes.com/tsmc-discloses-secret-4nm-node/#
14:31 | TSMC Prepared to Recover From Huawei Order Loss
In more TSMC news, it seems as if TSMC is prepared for the loss of Huawei. Now, that’s not to say TSMC wants to lose Huawei, or HiSilicon; however, Huawei’s US blacklisting and the recently amended export rules from the US Department of Commerce may force TSMC to drop Huawei whether it wants to or not.
TSMC Chairman Mark Liu said while TSMC hopes not to lose Huawei’s business, that if it does, “we will replace it in a very short time.” Liu also went on to say that “It’s difficult for me to predict how fast it could be filled immediately.”
According to Liu, TSMC is still assessing how the new rules will affect the company, and what implication they will have for its business. Per the amended export rules, any foreign chipmaker using American IP, technology, or equipment will have to apply for a license before supplying Huawei. TSMC has stated that it may apply for a license or exemption.
Around the same time the US Department of Commerce announced the new export rules, TSMC announced its plans to build a new $12B megafab in Arizona.
16:04 | Intel Launches Hybrid Lakefield CPUs
Intel finally released its Lakefield CPUs, which could prove to be its most interesting CPU line in years. Intel bills Lakefield as a hybrid x86 platform, and it uses Intel’s newest Foveros packaging technology. Foveros allows Intels to mix-and-match various IP “blocks” with different memory and I/O elements within a 3D package. One of the highlights of Foveros is pairing “big” cores with “little” cores. Foveros is also similar to AMD’s chiplet approach, in that Intel is mixing different dies, such as a 10nm package die and a 22nm I/O die for Lakefield.
This is evident in the first two products that are using Foveros, the Intel Core i3-L13G4 and i5-L16G7. Both of these SKUs will pair one big 10nm Sunny Cove core with four little Tremont cores. Intel hasn’t detailed the new chips heavily, but we know that these are Intel’s first chips shipping with package-on-package (PoP) memory and a standby SoC power as little as 2.5 mW.
The chips will come with 4MB of cache, and Gen11 graphics with up to 64 EUs. These chips will also have a package area of 12x12x1 -- or roughly the size of a dime. Intel states that this is possible by stacking two logic dies and two layers of DRAM, which eliminates the need for external memory. Lakefield packs support for LPDDR4X-4266, although it’s unclear how much memory the chips will support.
Initially, Lakefield will power two designs shipping this year: The Samsung Galaxy Book S and the Lenovo ThinkPad X1 Fold. It’ll be interesting to see what’s under the hood with Lakefield and how Intel is making it work. Sunny Cove and Tremont obviously use different architectures and support different instruction sets. Intel notes that there’s hardware-guided OS scheduling to funnel the right jobs to the right cores, but we’d like to know more about how the cores work individually.
20:12 | Cox Is Throttling Entire Neighborhoods To Mitigate Heavy Data Usage
Cox Communications is apparently slowing internet speeds for entire neighborhoods to mitigate what it calls “excessive usage.” What’s more, Ars Technica reports that Cox has imposed these kinds of slowdowns across neighborhoods before, although Cox won’t say how many users need to be “excessive” in their internet use to result in a neighborhood-wide throttle.
Ars Technica details the story of Mike, a Cox customer from Florida. Mike was apparently warned about his excessive use before Cox slowed the upload speeds on his gigabit plan from 35Mbps to 10Mbps, for him and the rest of the neighborhood. According to Mike, who spoke with Ars, his internet usage was between 8TB to 12TB a month. Furthermore, according to Mike, this wasn’t unusual data consumption for him, but Cox only recently flagged his account in May.
Ars reports that Mike currently pays $150/mo for Cox’s gigabit plan, which includes an additional $50/mo for “unlimited data” that allows him to go beyond Cox’s arbitrary 1TB data cap. According to Ars, Mike’s usage consists of scheduled device backups and "data sharing via various (encrypted) information-sharing protocols," between the non-peak hours of 1AM and 8AM. Again, Mike states that this has been his normal internet use for the last four years.
The fact that Cox only recently flagged his account and issued slowdowns no doubt suggest that Cox is struggling under spiking internet traffic, thanks to a certain pandemic. However, Cox is reluctant to admit that, as all are ISPs. In fact, Cox recently told Ars that its network was “performing very well overall.”
Mike initially started to receive phone calls about his usage, and when he called Cox to discuss it, he was told to lower his monthly usage or have his account terminated.
“I tried to explain that my usage is not out of the ordinary for me. My day-time bandwidth usage is paltry (most of my bandwidth consumption is scheduled from 1am-8am), and that Cox should have been upgrading their infrastructure instead of oversubscribing nodes and pocketing the record revenue. I was told if I did not make a substantial decrease in my upload data usage, my service would be terminated,” Mike told Ars.
We’d like to acknowledge that Mike’s response is especially on point here, particularly the part about not upgrading infrastructure while oversubscribing nodes and pocketing the profit. That’s been the buffet of choice for American ISPs for years.
In a later email, Cox stated that Mike and his neighborhood would be getting neutered upload speeds. “During these unprecedented times, many people are working and schooling from home, and maintaining connectivity is important. We are working to provide a positive Internet experience for everyone, so we've adjusted our Gigablast upload speeds in your neighborhood from 35Mbps to 10Mbps, now through July 15, 2020. Your download speeds have not changed,” reads the email.
This is normally the part where I’d remind everyone about the gutting of net neutrality, and then subsequently look forward to some nonsensical comment about how ISPs don’t need regulation, free market, etc. However, as Ars points out, the net neutrality framework had in-built exemptions (known as reasonable network management) that likely would have applied to the current circumstances ISPs are dealing with right now (see: pandemic internet traffic).
Cox didn’t answer Ars’ questions as to exactly why Cox was struggling with heavy internet traffic.
24:51 | Jim Keller Has Left Intel
In a bit of unexpected news this week, Jim Keller has resigned from Intel, effective June 11, 2020. Keller’s resignation is apparently due to personal reasons. With his resignation,
Keller ends a two year stint at Intel, after being hired as Intel’s Senior Vice President of Silicon Engineering Group.
Keller is a well known processor engineer with past tenures at Apple, AMD, and Tesla, just to name a few. Intel’s press briefing notes that while Keller has resigned, he will act in a consulting capacity for six months to help ease the transition. Dr. Ian Cutress confirmed that while Keller is consulting for the next six months, he’ll be doing so exclusively for Intel.
Intel also noted that some of its leadership will be changing. From Intel’s press release:
- Sundari Mitra, the former CEO and founder of NetSpeed Systems and the current leader of Intel’s Configurable Intellectual Property and Chassis Group, will lead a newly created IP Engineering Group focused on developing best-in-class IP.
- Gene Scuteri, an accomplished engineering leader in the semiconductor industry, will head the Xeon and Networking Engineering Group.
- Daaman Hejmadi will return to leading the Client Engineering Group focused on system-on-chip (SoC) execution and designing next-generation client, device and chipset products. Hejmadi has over two decades of experience leading teams delivering advanced SoCs both inside and outside of Intel.
- Navid Shahriari, an experienced Intel leader, will continue to lead the Manufacturing and Product Engineering Group, which is focused on delivering comprehensive pre-production test suites and component debug capabilities to enable high-quality, high-volume manufacturing.
Computex 2020 Has Been Called Off
As the global pandemic continues to claim tradeshows, Computex 2020 was one of the last remaining shows holding out hope that it could go on, which seemed a bit over-optimistic.
Initially, the Taiwan External Trade Development Council (TAITRA) and Taipei Computer Association (TCA), the organizers behind Computex, had planned on rescheduling Computex for September of this year. The rescheduled Computex would’ve been much smaller in scale, ranging from September 28 through September 30, and focusing on “5G and communication, smart solutions, gaming, and InnoVEX (startups).”
Now, Computex 2020 has been shelved until 2021. However, TAITRA is offering a series of online services in lieu of a proper show. There was already the #ComputexOnlineTalks and Computex Online Sourcing Meetings that took place, but still coming up are the #InnoVEXOnlineDemo and Computex Online 2D Exhibition. The latter two events will focus on InnoVEX startups and exhibitors’ products and services, respectively.
The #InnoVEXOnlineDemo will be held from the Taiwan Trade Shows YouTube channel on June 29. The Computex Online 2D Exhibition will be a new online platform for exhibitors launching September 28.
Source: https://www.computextaipei.com.tw/en_US/news/info.html?id=DEA9521F075CFA8A
Editorial: Eric Hamilton
Host, Additional Reporting: Steve Burke
Video: Keegan Gallick, Andrew Coleman