Hardware stub

HW News - Supercomputer Cryptomining Malware, DDR5 & AMD, Ryzen 3 1200 AF

Posted on April 27, 2020

News is busy this week and features a story that we'd love to know more about: The Cedar Supercomputer accidentally running cryptomining software for 6 hours a day under the nose of researchers. We're also talking about DDR5 and AMD's future roadmap (all tentative), the Ryzen 3 1200 AF & 1300 AF CPUs coming to market, Unigine Community 2 SDK, TSMC earnings, and more.

Show notes continue after the video embed.

AMD Has DDR5 on Roadmap

This first one is short: GamersNexus has received an AMD roadmap from an insider whom we’ve previously verified. Internal roadmaps are constantly subject to change and aren’t public promises, so this is really more useful as something to look out for, not necessarily something to bet on. That’s especially true given the current economic and manufacturing uncertainty in the global climate.

All of those disclaimers out of the way, we ask that you not overly sensationalize this news and just keep it to the facts: The roadmap we received, which we are unable to share as it is likely marked to find leakers, indicates that 2022 is the year that AMD is intending to get DDR5 into its premium desktop lineup. The 2022 platform will retain PCIe4 and will likely be a Zen4 product, also intended to feature native USB4 support. The APUs for that year also presently have DDR5 on the roadmap and are listed as Zen3+ parts. The mobile roadmap has DDR5 LP5 on-board for 2022 in the premium and gaming lines.

We have more information than that, but since we know how wildly internal roadmaps can change, we’re going to exercise responsibility and withhold the rest until we get closer to those windows.

Source: GN

Ryzen 3 1200 AF & 1300 AF Surface at Retailers

Rumors have been going around for a while suggesting that the Ryzen 3 1200 would be getting the ‘AF’ treatment. That is, moving from GlobalFoundries’ 14nm node to its 12nm (12LP) node. AMD’s Ryzen 5 1600 was the first of such chips, and if the rumors are true, the Ryzen 3 1200 is next in line. 

In addition to an iterative node shrink, the AF part would also move the Ryzen 3 1200 to the Zen+ architecture, a slight optimization over vanilla Zen. The benefits of Zen+ mostly reside in slight IPC improvements per core, reduced power consumption, improved cache and memory latency, as well as an improved DDR4 memory controller. After all, Ryzen loves its memory, preferably the fast kind. 

Giving the rumors a bit of traction is German tech outlet, ComputerBase. Per ComputerBase, there are listings for the Ryzen 3 1200 AF at a German retailer. Its listing has a price of €54.73 ($60). If the listing is correct, AMD is apparently keeping the configuration the same as the original Ryzen 3 1200 (non-AF), meaning no change in clock speeds, cache, or TDP.

The Ryzen 5 1600 AF included the smaller Wraith Stealth cooler. We’d assume the same to be true for a Ryzen 3 1200 AF, especially as the original Ryzen 3 1200 shipped with the Wraith Stealth cooler anyway.      

Source: https://www.tomshardware.com/news/amd-ryzen-3-1200-af-12nm-zen

Layoffs Hit the Tor Project

As the current pandemic continues to play out, many organizations and companies are tightening the belt. Small businesses and nonprofits have become particularly vulnerable, ones just like the Tor Project. 

For those unfamiliar with the Tor project, it's responsible for pioneering onion routing through a decentralized network of servers and multiple layers of encryption. Tor’s flagship product, alongside the Tor network, includes the Tor browser. Some time ago, we put together a guide on maintaining privacy; the use of the Tor network and Tor browser make up a significant chunk of that guide.

The pandemic has apparently impacted the Tor Project severely, as Tor announced they had to lay off 13 staff members -- and remember, 13 isn’t an insignificant number when the resulting staff consists of 22 people. That’s roughly one-third of Tor’s staff. 

Over the years, the Tor project has increasingly grown to rely on donations from the private sector to fund its work. With the pandemic bringing nearly every aspect of the economy to a halt, it’s hard to imagine there being enough private donations for Tor to meet its goals. 

Despite the layoffs, Tor vows to forge ahead and continue maintaining the Tor network and making its technology available for all.    

Source: https://blog.torproject.org/covid19-impact-tor

Unigine Community 2 SDK

Unigine, the company most known for its GPU benchmarks, has announced that it's rolling out a free version of its Unigine SDK, known as Unigine 2 Community SDK. The free SDK features the Unigine 2 engine at its core, along with support for C++, C#, UnigineScript APIs. 

As a real-time 3D engine, Unigine has been used for games, but hasn’t gained the traction that say, Unity has -- and neither of them come anywhere close to Unreal Engine. Though, with Unigine opening up its platform to the public, that could change. However, the Unigine engine is far more prolific in the arenas of visualization and simulation. Of course, Unigine is also the foundation for a slew of benchmarks we all know, such as Heaven, Valley, and Superposition.   

Unigine 2 Community SDK is free of charge and royalties assuming one thing: individuals or studios using the SDK commercially must not exceed revenue or project funding of $100K/year. Unigine Community will be developed in parallel with the commercial editions -- Unigine 2 Engineering and Unigine 2 Sim -- and adhere to the same quarterly update cadence. 

Source: https://unigine.com/products/community/advantages/

TSMC: 1Q20 Earnings, Node Updates

TSMC reported its earnings for the first quarter of 2020, which ended in March. The world’s largest contract chip maker closed Q1 with a net profit of $3.89B on a quarterly revenue of $10.31B. That’s a ~90% increase YoY for TSMC and marks a 10-year quarterly profit growth.

However, the pandemic has been disruptive to TSMC’s operations, causing the foundry to be a little more consevative with its full year outlook. TSMC expects revenue growth to reach a “mid- to high-teens percentage” for the whole year of 2020. This is compared to TSMC’s previous revenue growth forecast of around 20%.  

Demand for TSMC’s advanced process technologies -- that is, anything 16nm and below -- was a key driver for the company’s quarterly growth. That coupled with the continued ramp for 5G and HPC demand. TSMC’s advanced technologies accounted for 55% of its wafer revenue in Q1. Notably, 7nm made up 35% of that, and in terms of revenue by node, TSMC’s N7 is at the top. 

TSMC notes that its N7 node is in its third year of ramp, while its N7+ is in its second year. N7+ will be TSMC’s first node to deploy EUV  for four critical layers. N7+ will also lay the groundwork for TSMC’s iterative N6 node, which is completely IP-compatible with N7/N7+. N6 will increase the EUV layer count by one and is already in risk production. TSMC expects to enter volume production with N6 by the end of this year.

Beyond N6, there’s N5, which is a full node evolution from N7 and will extend the use of EUV. N5 entered risk production last year, and TSMC has confirmed N5 is already in volume production with healthy yields. N5 is expected to deliver a rough 1.8x increase in density, as well as a ~15% performance uplift (speed) at iso-power versus N7. Alternatively, N5 could deliver a ~30% reduction in power at iso-speed compared to N7.   

TSMC also states it expects a “very fast and smooth ramp of N5 in the second half of this year driven by both mobile and HPC applications.” That timeline coincides with persistent reports that Apple will be among TSMC’s first customers to adopt its 5nm chips, in the form of the A14 Bionic SoC powering Apple’s 2020 iPhones, slated for announcement at Apple’s Fall iPhone event. 

TSMC also finally had a bit more to say about its N3 node, saying N3 is on track for risk production in 2021 and targeting volume production in 2022. TSMC has opted to remain on FinFET for N3 due to its maturity and costs. N3 is expected to deliver around a 70% gain in density compared to N5.        

Source: https://www.tsmc.com/uploadfile/ir/quarterly/2020/1OLf2/E/TSMC%201Q20%20transcript.pdf

https://www.reuters.com/article/us-tsmc-results/tsmc-first-quarter-profit-almost-doubles-on-chip-demand-but-sees-flat-growth-in-second-half-idUSKCN21Y0IN

AMD Announces Ryzen 3 3100/3100X & B550 

Following up to last week, AMD has indeed unveiled new Ryzen 3 parts for the desktop. AMD’s quad-core, non-APU R3 parts were notably MIA when Ryzen 3000 launched, and AMD has only gotten around to them just now. AMD has also finally taken the wraps of the highly anticipated B550 chipset for socket AM4. 

Over the course of three generations of Zen, AMD has found itself firmly in the premium CPU vendor position. When Ryzen 3000 launched, AMD did not offer a Zen2 SKU with anything less than six cores and a price tag of $200. Thanks to AMD, Intel’s products stack looks completely different these days, with Comet Lake-S i3 SKUs essentially becoming the i7 chips of yore. Intel has moved the i3 brand from dual-core to quad core, and is all but certain to extend hyper-threading to them as well. 

This would leave AMD vulnerable at the low-end, and we’re assuming AMD realizes this, thus we have the Ryzen 3 3100 and 3100X. By all accounts, AMD’s new R3 parts should be aggressively competitive, both in terms of performance and price. While we’ll have to fully test for the former, AMD has indisputably led the charge for the latter. In the wake of Ryzen 3000, Intel has mostly been relegated to competing in terms of price; Intel has historically enjoyed an advantage in pure gaming thanks to its single core/thread prowess, but that lead is threatened.

With the Ryzen 3 3100 and 3100X, AMD is seemingly setting the stage for another price war. The Ryzen 3 3100 is priced at $99, while the Ryzen 3 3100X is priced at $120. Both chips will pack Zen 2 cores, leverage SMT (a first for R3 chips), support DDR4-3200, come fully unlocked, and support PCIe 4.0. Any way you cut it, Intel’s looming i3 chips won’t have it easy. 

Lastly, AMD is finally rolling out the budget-oriented B550, noting that it has over 60 designs in development with its partners. Most notably, B550 will bring PCIe 4.0 to a much broader user base, as X570 boards have been somewhat price prohibitive.

The Ryzen 3 3100 and 3100X are expected to be available May 21, 2020. Meanwhile, B550 motherboards should arrive from AMD’s ODM partners on June 16, 2020.            

Source: https://www.amd.com/en/press-releases/2020-04-21-amd-expands-3rd-gen-amd-ryzen-desktop-processor-family-unleashing

Seagate and WD Acknowledge SMR Concerns

In a quick update to last week’s SMR story, both Western Digital and Seagate have responded publicly. 

Western Digital initially defended its position on SMR, mentioning that it doesn’t always discuss “what’s under the hood,” like drive-managed SMR (DMSMR) and host-managed SMR (HMSMR), preferring to default to specific use-cases instead. Then it seemingly walked those comments back a bit, with a decidedly more humble tone.

WD went on to disclose all of its client HDDs that use any form of SMR; these include drives in its WD Red, WD Red Pro, WD Blue, WD Black, and WD Purple lines. WD went on to state it would update marketing materials, add information about SMR, and include benchmarks and ideal use cases.

Seagate gave comments to ArsTechnica regarding the use of SMR in some of its HDDs. “Seagate confirms that we do not utilize Shingled Magnetic Recording technology (SMR) in any IronWolf or IronWolf Pro drives—purpose-built for NAS solutions. Seagate always recommends to use the right drive for the right application,” Seagate said to ArsTechnica. 

Seagate was less forthright about its use of SMR in client desktop drives, though. Seagate seems to maintain that SMR is appropriate for consumer drives, and while Seagate’s Exos and Archive HDDs are identified as using SMR, it still seems content to market its Barracuda drives with undocumented SMR use.     

GN reached out to Toshiba directly for clarification on its use and disclosure of SMR. However, we have not heard back as of this writing. If we receive a comment from Toshiba, we will include it in next week’s HW News. 

Source: https://blog.westerndigital.com/wd-red-nas-drives/

Seagate: https://arstechnica.com/information-technology/2020/04/seagate-says-network-attached-storage-and-smr-dont-mix/

LGA1200 Retains LGA115x Cooler Compatibility

Last December, a drawing surfaced detailing Intel’s upcoming LGA1200 socket, the upcoming successor to Intel’s perennial LGA115x socket design. LGA1200 will necessitate a couple of changes, the most obvious being new motherboards and chipset silicon -- those will come via the 400-series platform.

This is due to the LGA1200 socket commanding 1,200 pins, exactly 49 more than the previous LGA1151 socket. Additionally, Intel is also adding an extra two cores for its halo Comet Lake-S SKUs. We have some early/preliminary details on those last two points that GN confirmed independently, and are worth checking out. When we looked at the drawing last December, it looked as if the dimensions for the LGA1200 socket would be identical to that of LGA115x, meaning that, at least in theory, LGA115x compatible coolers would work with LGA1200. However, that was just speculation at the time.

Now, though, Noctua has confirmed that LGA1200 will in fact retain LGA115x cooler compatibility. 

“The heatsink mounting on Intel's new LGA1200 platform (code name Comet Lake S) is identical to all LGA115x sockets (LGA1150, LGA1151, LGA1155, LGA1156). Therefore, all Noctua CPU coolers that support LGA115x also support LGA1200 and don't require any mounting updates,” says Noctua via its product page. 

So, this is good news for any users that are in possession of an LGA115x cooler and may be interested in adopting Comet Lake-S, assuming the cooler is up to the job. 

 Source: https://noctua.at/en/sip_productfaqnew/productfaq/view/id/236

Cedar Supercomputer Hit With Hidden Cryptomining Process

GN received a tip from a viewer who goes by Dr. Goose, presumably friends with our other informant, Mr. Fahrenheit, although with a Phd in waterfowl and avian species. The tip was about Cedar, a petaFLOP scale supercomputer that resides at Simon Fraser University in Canada, being taken offline due to some security mishaps.

An investigation was launched by Compute Canada, who deploy and maintain many of Canada’s academic supercomputers, after reports that Cedar was performing unusually slow. The team investigating (including members from both Compute Canada Federation and Simon Fraser University) found that there were hidden processes running between midnight and 6:00 am (PDT) that were leading to significant system load.

Upon discovery, Cedar was taken offline from Wednesday, April 15 to midnight on Sunday, April 19. While offline, Cedar was technically still running and queuing jobs. The investigation uncovered a cleverly hidden cryptocurrency mining process that was running from 00:00-06:00 PDT daily, poaching compute cycles. It’s believed the mining process was injected into Cedar through a compromised user account and key. 

It’s noted that the mining process obscured itself from Cedar’s standard monitoring tools, and automatically removed evidence of its activities daily at 06:00 PDT. Therefore, the executable was not seen in the file system. Compute Canada notes that there was no instance of personal or private data being compromised.    

Editorial: Eric Hamilton
Host, Additional Reporting: Steve Burke
Video: Keegan Gallick