HW News - Apple Could Move on from AMD, Huawei 7nm CPU, 3995WX Threadripper CPU, & Atari VCS
Posted on July 14, 2020
Another big week in hardware and technology news. It looks like the perpetually delayed Atari VCS is finally set to ship later this fall, hopefully. We also have news of Huawei’s Kunpeng 920 CPU, talk about Volta and Turing encoders shipping with the GTX 1650, news of Apple also ditching AMD, and more.
At GN, we recently went over the basics of overclocking the Ryzen 5 3600 XT, Infinity Fabric, and memory. We also have extensive review coverage of AMD’s new XT series of CPUs: The Ryzen 3900X, the Ryzen 7 3800XT, and the Ryzen 5 3600XT. We further introduced our new GN PC Component shirt, and you can get one here. The GN store continues to be the best way to support our work.
News follows below, with the article and video embed.
02:44 | Threadripper Pro 3995WX
If anyone remembers our report from September of 2019, we broke early news about the TRX40 and an sWRX80 platform. The TRX40 motherboards came to fruition with new Threadripper, so our sources were on the money about the existence of those boards and their replacement of the other SP3 sockets. After a few months more of waiting, our source privately said, “sorry, you don’t get them all,” and indicated that they thought the WRX80 platform might not actually come to fruition, despite early presence on AMD roadmaps.
We were fine with that, because we did note in the news stories that it may not happen. That said, it looks like things have turned around: New rumors of an alleged AMD Threadripper Pro 3995X CPU have been posted by Chiphell. The image could be faked, mind you, but the current rumor is that it’s an 8-channel solution. If that’s true, it’d align with our September 2019 exclusive leak that sWRX80 would be an 8-channel Threadripper platform.
Source: https://twitter.com/9550pro/status/1280706446515036160
04:51 | Atari VCS Set To Ship In October. Maybe.
The Atari VCS, which is built in the spirit of the Atari 2600, is set to ship to customers by no later than December 24th, as shipping begins this October. The Atari VCS has been on a long and controversial road to get to this point, and you can check out a fairly comprehensive timeline over at Reddit.
The Atari VCS was originally teased at E3 2017, back when it was called the Atari Box, and well, things just seemed to go off the rails after that. Atari opened up a crowdfunding campaign via IndieGoGo, but pushed back pre orders and shipping dates multiple times. In the meantime, Atari was also dipping its feet into other ventures, such as bringing Rollercoaster Tycoon to the Nintendo Switch and building its own Atari Chain cryptocurrency, among other things.
The console itself also underwent several hardware revisions, further delaying production and eventual release. The console was given a bump in memory going up from 4GB to 8GB. Then, a decision was made to upgrade the CPU, going from a Bristol Ridge-based APU to one of the Raven Ridge variety. Details around the CPU are still vague, other than it’s a 14nm APU featuring two Zen cores and Radeon Vega graphics. Possibly something semi-custom, though there likely won’t be enough volume for that, nor is it within the $3M budget afforded through the crowdfunding campaign. We’ll see, though.
On top of the CPU and RAM upgrade, the Atari VCS will also come with 32GB of storage, a custom Linux distro (called Atari World), Atari’s “Modern Controller,” as well as the classic Atari joystick. The whole setup will cost $390.
Source: https://www.pcgamer.com/atari-vcs-console-shipping-date/
09:00 | Huawei’s 7nm Kunpeng 920 CPU Tested, Kind Of
Last year, Huawei announced it was entering the server market with its own line of processors, known as the Kunpeng 920-series. This line of chips is designed in house by Huawei (HiSilicon) and is based on the Armv8 architecture and aimed at the server market. The Kunpeng 920 is based on Huawei's own TaiShan v110 microarchitecture and is the building block of its TaiShan servers.
The Kunpeng 920 line scales up to 64 single-threaded cores, and as far as we can tell, those cores are custom (or semi-custom) Cortex A-72 derivatives, as that’s what the TaiShan v110 architecture is based on. The Kunpeng 920 chips are built on one of TSMC’s 7nm nodes, though exactly which one isn’t clear.
While the Kunpeng 920 line is aimed at servers and the data center, there’s at least one SKU that appears destined for consumer PC use. Per Tom’s Hardware, a Chinese YouTube channel got its hands on an 8C/8T Kunpeng 920 chip, paired with Huawei’s D920S10 motherboard, both of which came in a third party system. The video can be found here.
As Tom’s Hardware notes, the primary focus of the video is one of software support. Being that the Kunpeng CPUs are Arm-based, any x86 software won’t run natively and will require a compatibility layer. That problem is further compounded by the operating system, which appears to be some forked Linux distro and only supports 64-bit. The user was able to get a 4K resolution at 60Hz, all while using the Yeston RX550 -- and you all know how much we love Yeston.
That said, the user did manage to get a couple of modern benchmarks going. There was a blender BMW render test that completed in almost 12 minutes; not great. The PC was also able to stream some 4K content, but suffered due to poor encoder performance. The user also noted that during the 4K streaming, the CPU usage was “not very high.”
There were no gaming benchmarks to speak of, and again, the video seemed to highlight software and OS compatibility issues, and how usable the PC was out of the box. These kinds of tests do offer a glimpse of the massive challenge China has ahead of it, though, as it relates to weaning itself off of American chip technology in an effort to produce domestic CPUs.
If you haven’t already, check out our deep dive into the ZhaoXin ZX-C+ 4701 CPU, which is a home-grown x86 Chinese CPU.
13:35 | MSI CEO Passes Away
Unfortunately, we learned this week that MSI President and CEO, Charles Chiang, has passed away, aged 56. Chiang had been with MSI for 20 years, before taking up the mantle of CEO and President. Previously, he headed up the company’s desktop computing segment, being responsible for the MSI’s foray into gaming focused motherboards and graphics cards.
As CEO, Chiang spoke rather candidly on topics such as Intel’s CPU shortages and the US-China trade war. While the details of his death are being investigated, MSI gave the following statement to Tom’s Hardware:
"Earlier today, MSI GM and CEO Charles Chiang passed away. Having been a part of the company for more than 20 years, he made outstanding contributions and was admired by his colleagues. Mr. Chiang was a respected leader in the MSI family, and helped pave the way for the brand’s success. We are all deeply saddened by the news, and are mourning the loss of Mr. Chiang. He will be deeply missed by the entire team."
Digitimes: https://www.digitimes.com/news/a20200708PD205.html
Source: https://www.tomshardware.com/news/msi-ceo-charles-chiang-dead-56
15:27 | GTX 1650 GDDR6 Uses Both Volta And Turing Encoders
If Nvidia’s product wasn’t confusing enough -- especially as it relates to the forever refreshed GTX 1650 -- it seems it’s about to get a bit more convoluted. First, the old news. Nvidia’s GTX 1650 line has historically used the older Volta NVENC encoder, which has been a point of contention among some users. Even despite multiple refreshes, like moving to GDDR6, the GTX 1650 line retained the Volta encoder.
However, with its most recent refresh -- TU106-125 rumors notwithstanding -- the newer TU106-based GTX 1650 cards seem to use the Turing NVENC encoder. Or, at least some of them do. Nvidia’s product page lists the GTX 1650 GDDR6 as coming with either “Volta or Turing” in regards to the encoder. It appears the die type will dictate what encoder the card uses.
Upon closer inspection, down in the fine print (where we often find these things), Nvidia notes read: “TU117 GPUs have Volta encoder, TU106/TU116 GPUs have Turing encoder. Check with manufacturer for exact model.”
So, if you’re in the market for a GTX 1650 with GDDR6, it may be worth trying to identify the die type, if the encoder variant is important to you.
Source: https://twitter.com/momomo_us/status/1280131139382865920
Source: https://www.nvidia.com/en-us/geforce/graphics-cards/gtx-1650/
19:07 | Intel’s Gemini Lake Reaches EOL
Another Lake bites the dust. This time, Intel’s Gemini Lake is being taken out to pasture. Gemini Lake, which was based on Intel’s Goldmont Plus architecture, rolled out in 2017. Gemini Lake ranged from two to four cores (with no hyperthreading), and mostly targeted low power devices. The EOL notice also comes as Intel just launched Gemini Lake Refresh last year, which is meant to hold Intel over as its new Tremont Atom architecture gets off the ground.
Most of Gemini Lake CPUs were sold under the Pentium Silver and Celeron families, both for desktop and mobile applications. Under the EOL transition, customers will have either until October 23 or January 22 to place final orders, depending on the SKU. Last orders will ship April 2nd or July 9th, 2021.
Source: PCN(s): 117659, 117657, 117653 -- https://qdms.intel.com/Portal/SearchPCNDataBase.aspx
20:17 | Apple Is Saying Goodbye to AMD, Too
When Apple recently announced the move away from Intel in favor of its own Arm-based Apple Silicon, it wasn’t exactly clear what that meant for the future of AMD GPUs in Macs. Well, let there be no more confusion: Apple will be developing its own GPUs for Macs, much like it’s been doing with its A-series SoCs for iOS devices.
In a developer video titled “Bring your Metal App to Apple Silicon Macs,” Gokhan Avkarogullari, Apple’s director of GPU software, discusses porting non-native Metal-based apps over to Apple’s new platform.
In the video, Avkarogullari details Apple’s TBDR (Tile Based Deferred Rendering) architecture that it will be using for its GPUs, which will be supported through Apple’s Metal API. The video also showcased Dirt Rally running on an Apple Silicon based Mac, with Rosetta translating the x86 binaries. This is seemingly to assure developers and users that Apple is committed to broad x86 support, and not just Photoshop, as PCGamer quipped.
To that end, however, Apple is still encouraging developers to build natively for the platform. The video outlines a path for developers. For x86 code, Apple assures it will run out of the box thanks to its Rosetta translation layer, which is also supposed to extend certain features from the Metal API to non-native code. Obviously, translation has some performance costs, so Apple says this should only be the first step.
The next step is for developers to recompile, using Apple's MacOS SDK. Then, developers will want to optimize for Apple GPUs and Metal, of which Apple has an entirely different video about. Apple seems to think all of this is worth it, as it’s convinced it can unlock better power and efficiency with its own in-house designed GPUs.
We won’t have to wait terribly long to see what Apple is cooking up, as it will bring its first Arm-based Macs to market towards the end of the year. Those will almost certainly be the smaller MacBooks, as Apple’s workstation/Xeon replacement likely isn’t ready for prime time.
Source: https://www.pcgamer.com/apple-ditch-amd-gpus-and-intel-cpus/
Source: https://developer.apple.com/videos/play/wwdc2020/10631/
24:11 | Nvidia Dethrones Intel as Most Valuable Silicon Company
Intel’s position as the highest-valued chipmaker has been usurped by NVIDIA, but to be fair, NVIDIA isn’t exactly a “chipmaker” -- more of a chip designer, sort of like AMD.
As Reuters reports, Nvidia’s stock has been steadily rising, moving up 2.3% earlier this week to a record $404. This puts Nvidia’s market cap at $248B, which is about $2B above that of Intel, who is sitting at $246B at the time of this writing.
Reuters further notes that while Intel’s stock has dropped 3% this year, Nvidia’s has raised by 68%, which analysts seem to think is due to the datacenter and the surge of remote work in the last several months. The report suggests that Nvidia has also made better plays at diversifying its business, while Intel has notoriously struggled with manufacturing and supply chain woes.
However, Nvidia’s sales still pale in comparison to Intel’s. “Analysts on average see Nvidia’s revenue rising 34% in its current fiscal year to $14.6 billion, while analysts expect Intel’s 2020 revenue to increase 2.5% to $73.8 billion, according to Refinitiv,” says Reuters.
26:46 | Super Mario Bros Sells for $114,000
Heritage Auctions emailed us this week to inform us that it had brokered the sale of an original, new-in-box Super Mario Bros. 1985 game cartridge. The cartridge went through 28 attempted bids before landing on the winner, who paid $114,000 at public auction. Heritage told the media that this is the most money ever paid for a video game at public auction, and indicated that the previous high-record holder was 1987’s Mega Man, also sealed, sold for $75,000 in 2019.
The Super Mario Bros cartridge in question still had its hangtab, so it looks like it’s new-old stock that never sold at its original retail location. This probably sat in some former store owner’s basement for a long time.
We look forward to a future where the first Steam accounts are sold for millions of dollars.
Editorial, Reporting: Eric Hamilton
Host: Steve Burke
Video: Keegan Gallick