HW News - SK Hynix Dispels BS Big Navi Rumors, Intel Memory Encryption vs. Vulnerabilities
Posted on March 3, 2020
As we prepare to fly out to Taiwan for factory tours, we've got another hardware news round-up to carry through the week: Our final donation count for the wildlife rescue charities is included, followed-up by SK Hynix's response to some Big Navi spec "leaks," JPR's GPU marketshare reporting, Biostar's H61 resurrection, AMD's chiplets aiding in cost reduction, and Intel promising Total Memory Encryption. We also talk about Plague Inc getting pulled due to the coronavirus scare.
GN’s Final Donation Count for Wildlife Charities
We discuss this one exclusively in the video embedded above! It's timestamped for your ease.
That “Big Navi” Rumor is BS
Earlier this week, as the internet is wont to do, it was lit aflame with more rumors of what form Big Navi would take. This time, a Twitter user -- who has since deleted the associated account -- shared a screenshot of an alleged spec sheet from SK Hynix outlining specs for an unnamed AMD GPU. You can probably imagine what happened next.
The story and screenshot made rounds online, being picked up by a number of outlets. The specs alleged the GPU would feature 5,120 shaders across 80 CUs, with as much as 24GB of HBM2e memory. The memory was claimed to operate over a 4096-bit memory bus with more than 2 TB/s bandwidth.
Those specs double what the current RX-5000 cards offer in terms of shaders and CUs. And while that itself isn’t necessarily unbelievable, the fact that a mainstream gaming card would come equipped with a 24GB VRAM buffer is. It’s complete overkill for mainstream gaming, and AMD’s margins on such a card would be completely lost trying to sell it to consumers. HBM2e is expensive to make and expensive to package, and while it might make sense for a workstation card, it doesn’t make a lot of sense for gaming over GDDR6 right now. Especially as much as 24GB.
Not to mention, the specs for SK Hynix’s HBM2e memory shown in the screenshot aren’t even accurate -- a point SK Hynix brought up itself with an official statement.
“Recently, there have been media reports about SK hynix’s memory HBM2E and AMD’s next-generation GPU, based on the allegations by a Twitter user CyberPunkCat, which are factually incorrect.
“SK hynix hereby announces that the company has not created or distributed such specifications as well as the document asserted to be leaked by an internal source. After a thorough investigation, we conclude that the screen capture of the allegedly internal document is fabricated.
“First of all, the specifications of our HBM2E shown in the document are incorrect,” says SK Hynix.
“Secondly, for those of you who are not familiar with Korean, the text in the document says ‘With Local Communities… 2020 Happiness,’ which is not related to HBM2E development or GPU. Moreover, we found out that this text actually appeared on the main page of our Korean Newsroom as a title for one of our content.”
JPR GPU Report: AMD up 22.6%, Nvidia down 1.9%
In the latest report from Jon Peddie Research, the firm states that quarterly GPU shipments for 4Q19 increased by 3.9% over 3Q19. More specifically, AMD’s GPU shipments last quarter were up by 22.6% Alternatively, Nvidia’s shipments were down by 1.9%. Keep in mind that, similar to Zen’s launch, increasing by 22.6% against almost nothing previously isn’t going to be enough to keep Radeon afloat, but it’ll get some cash injection to pay off debts and hopefully improve development on the drivers.
JPR notes that this is the third consecutive quarter of increasing GPU shipments. JPR also notes that a certain epidemic involving a certain pathogen in a certain country has disrupted supply chain operations for GPU makers, which has resulted in less growth. We can’t be more specific because our overlords will remove ads, but human malware is a real plague on the industry.
Coming as no surprise, Nvidia still commands the lion’s share of the market with 73% market share. AMD is a very distant second at 19%, with Intel trailing AMD at 18%.
Biostar Resurrects Sandy Bridge-Era H61 Chipset
Biostar has apparently decided to bring Intel’s H61 platform back to life with a new motherboard based on the Sandy Bridge/2011 era chipset. Known as the H61MHV2, it seems Biostar is trying to cater to users looking to keep older systems alive and forego the second-hand market.
The H61MHV2 is a micro-ATX board, albeit one with some oddities. For one, it comes with an extra PCIe slot, making the board a bit longer than your usual mATX fare. The DIMM slots are also situated at the top, horizontal to the CPU socket, and the chipset sits at a 45-degree angle to the CPU socket.
The board is also keeping with 2011 tradition, being devoid of any moderin I/O such as USB 3.0 (or any of its confusing generational successors) and even PCIe 3.0 isn’t fully supported unless an Ivy Bridge chip is being used. Also, SATA II ports are on tap, a far cry from today's SATA III ports -- which have shown their age by today’s standards. Biostar has more notably improved the NIC to a 1Gbps option, up from its original 100Mbps solution.
Biostar hasn’t revealed pricing at time of writing/filming. However, if the price is right, it could be a nice option for anyone still maintaining older Sandy Bridge or Ivy Bridge machines.
Source: https://www.biostar.com.tw/app/en/mb/introduction.php?S_ID=972
AMD’s Chiplets Reportedly Reduce Costs by 25%-50%
With Zen and its accompanying chip families (Ryzen, Threadripper, Epyc), AMD doubled down on MCM designs over single, monolithic processor dies. This has given AMD a lot of flexibility up and down its product stack, as we’ve seen AMD continually up the core count in successive generations. AMD’s chiplet design has also been more economical and predictable in manufacturing. Monolithic dies can be very expensive, especially as transistor density increases and chips get more complex. Bigger dies also constitute more manufacturing problems on a per wafer basis, increasing the chances of less than desirable yields. While any new process needs time to mature, and AMD has had its fair share of growing pains transitioning to TSMC’s N7 node, AMD’s aggressive pricing has been boosted by lower manufacturing costs thanks to its chiplet design.
At the International Solid-State Circuits Conference (ISSCC), there were apparently talks of these savings (as reported by Chinese website MyDrivers) where AMD illustrated its 7nm CCX-based chips net anywhere between 25% to 50% savings over “hypothetical” 7nm monolithic designs. AMD states that making its 16-core Ryzen 3950X using a monolithic die would cost twice as much, while its 8-core variants would cost at least 25% more with a monolithic design.
Source: https://www.notebookcheck.net/AMD-says-Chiplet-design-can-cuts-cost-by-more-than-half.454676.0.html
Intel Is Still Promising -- And Not Delivering -- Total Memory Encryption
Intel has been promising full memory encryption within its CPUs as far back as at least May 2019 -- nearly a year ago. It was talked about long before then. At its recent Security Day, which is a bit ironic given all the security vulnerabilities Intel is beset with, Intel wanted to remind everyone that it still has plans for full memory encryption. Some day.
Intel’s TME (Total Memory Encryption) or MKTME (Multi-Key Total Memory Encryption) got something of a fleeting mention at the event, with nothing more said about where it rests on Intel’s roadmap. Intel’s TME and MKTME are equal parts about bringing parity with AMD’s SME (Secure Memory Encryption) and addressing weaknesses and limitations in Intel’s SGX enclaves.
Intel’s full memory encryption would work in similar fashion to AMD’s SME. TME or MKTME would be able to fully encrypt memory, whether it be at rest or in transit. Single, ephemeral keys would be generated at the kernel level with a new 128-bit key generated with every reboot. MKTME is an extension of TME that would add support for multiple keys to be in use.
While full memory encryption isn’t an answer to side-channel/speculative execution attacks, it’s an important first step. It also does have inherent mitigations for certain types of data leaks. It’ll also be important as non-volatile memory sees more use, and if non-volatile memory ever replaces RAM.
China Pulls Plague Inc. Game Over, uh, “Human Malware” Concerns
Plague Inc., a “plague simulator” game, has been pulled from China’s Apple app store. The game was reportedly yanked due to “illegal” content, according to developers.
The game has apparently surged in popularity following the outbreak of a certain virus that we won’t name here. But it starts with a C and rhymes with demonetized. Not really, but thanks to YouTube, we’ve resorted to just calling it “human malware.” Anyways, moving on.
“We have some very sad news to share with our China based players. We've just been informed that Plague Inc. “includes content that is illegal in China as determined by the Cyberspace Administration of China” and has been removed from the China App Store. This situation is completely out of our control,” says Ndemic, the developer of Plague Inc.
Ndemic notes that its removal may have been prompted over concerns and sensitivities regarding the current epidemic in China. However, it’s worth mentioning that China has wildly oppressive censorship motives. Confusingly, the ban seems to only apply to thi iOS version of the game; it can still be had in China via Steam for PC.
Source: https://kotaku.com/china-pulls-plague-inc-from-app-store-1841955831
Editorial: Eric Hamilton
Host: Steve Burke
Video: Josh Svoboda, Keegan Gallick