HW News - AMD Closes, then Opens "Open Source" Code, 6.5GB/s SSD, & Unpatchable Vulnerability
Posted on May 17, 2020
It’s been another interesting week in the realm of hardware and technology. The week started off slowly, but ended with a deluge of interesting stories, mostly as it relates to US semiconductor manufacturing. In addition to Intel and Samsung in talks with the Department of Defense, it looks as if TSMC will be adding a second fab to its US roster.
We also have news on AMD’s open-source GPUOpen, and its apparently not so open-source Radeon Rays solution. Sometimes. There’s also news on the recently unveiled Unreal Engine 5 and how Epic CEO Tim Sweeney feels about the SSD storage solutions in the PlayStation 5.
Elsewhere at GN, we recently covered Nvidia’s GTC 2020 keynote where Ampere was formally announced -- check out both the article and video. We’ve also been extensively overclocking the Ryzen 3 3100, as well as the AMD Ryzen 3 3100 Infinity Fabric clock (FCLK).
GamersNexus Wireframe Mousemat Launched
We launched our new desk Mouse Mat! Check it out here.
Rumor: Intel, AMD, Nvidia, Case Makers Bail on Computex 2020
Computex, along with other trade shows like CES, are among the most important for our industry and audience. Unfortunately, the global pandemic has left most of these shows cancelled at worst, and rescheduled and pared down at best.
Such is the case with Computex this year, as it’s been rescheduled for the end of September with a less broad focus, as well as going from five days to just three. That said, up until recently it was still expected the big players like AMD, Intel, and Nvidia would still have some sort of presence. Now, according to Digitimes, that’s up in the air.
Citing sources familiar with the matter, Digitimes reports that these chipmakers are unsure about their Computex plans. What’s more, is that motherboard and graphics card vendors such as Gigabyte and MSI are also likely to bail on the show. AMD has been promising Zen 3 and RDNA 2 for 2020 for some time, and it’s been expected that, at least for its CPUs, AMD would take the wraps off at Computex this year.
If AMD and other chipmakers are indeed going to be absent from this year’s show, we’d expect them to pivot to an online event, such as a live stream. Digitimes’ report notes that other vendors and OEMs are already planning such online events to show off their products that were slated for Computex 2020.
Source: https://www.digitimes.com/news/a20200508PD216.html
AMD Goes Back on Open Source Promise
AMD’s Scott Herkelman posted on reddit recently about Epic’s Unreal Engine advancements, and in so doing, Herkelman remarked, quote:
“What is exciting about the new consoles launching is that for those game developers who build games across PC and consoles, it will push them to incorporate leading next-gen techniques to all audiences. [...] The other point that we, here at AMD, have been planning for is the timing with the console launches, to ensure that no hardware vendor-specific “proprietary” ray tracing technique or other GPU features slows down and bifurcates the industry to adopting next-gen features. With this console momentum and Microsoft’s DXR for PCs, I’m hopeful we can push towards an open ecosystem for all gaming and gamers.”
Shortly after recommitting to openness, AMD shattered another one of its glass walls after years of casting stones at its competitors over closed-source development practices. There may be a legitimate reason for this new change, but it’s hard to forget the interminable nagging of NVIDIA for their closed-source graphics solutions. AMD’s Radeon Rays solution, hosted on AMD’s GPUOpen website that specifically calls for non-proprietary graphics standards, has now gone closed-source. Radeon Rays version 4.0 has dropped the OpenCL and open source checkmarks from its marketing bullet list, but has added Dx12, Vulkan, and optimization for BVHs. We reached-out to AMD yesterday for comment and are still awaiting an answer as to why this change was made, and why Radeon Rays 4.0 remains hosted on GPUOpen, which has marketed its entire existence on open source code.
We should note a few key things here: First off, we don’t begrudge a company for closed-source code. You’re potentially talking millions of dollars of investment, if not more, and not every single software package needs to be given away for free. If AMD wants to go closed-source, we’re completely fine with that from a business standpoint; it could even be because of relationships with other companies. However, AMD has built its brand on top of smearing others for closed source software development, just like AMD smeared Intel for limited chipset survivability, and that’s not a good look. It’s another mark in the books about AMD’s marketing department taking out its engineering department.
Source: https://gpuopen.com/radeon-rays/
AMD GPUOpen: New FidelityFX Effects For RDNA
All that stated, AMD does get credit for breathing new life into the rest of its open-source GPUOpen suite, both with a revamped website and a collection of new technologies optimized for RDNA-based GPUs. GPUOpen was launched in 2016 as an open-source alternative to Nvidia’s GameWorks, but AMD is overhauling it for developers with four new FidelityFX effects.
As AMD notes, the FidelityFX family was announced last year with its first open-source effect, Contrast Adaptive Sharpening (CAS), that made its way to games like Rage 2 and Shadow of The Tomb Raider. Now, FidelityFX has been expanded to include Stochastic Screen Space Reflections (SSSR), Combined Adaptive Compute Ambient Occlusion (CACAO), Luminance Preserving Mapper (LPM), and Single Pass Downsampler (SPD).
Stochastic Screen Space Reflections is AMD’s take on Screen Space Reflections (SSR), which is an old approach to cost-saving graphics that reduce GPU load by only calculating reflections based on the view frustum. Specifically, the FidelityFX implementation will look to shore-up shortcomings with traditional SSR with a different algorithm that can add varying degrees of roughness to support glossy surfaces. SSSR also comes with an RDNA-optimized denoiser, combining results from several frames to construct a noise-free image.
Next is FidelityFX CACAO, or Combined Adaptive Compute Ambient Occlusion. This is an RDNA-optimized version of ambient occlusion. CACAO, based on Intel’s Adaptive Screen Space Ambient Occlusion, has been optimized to support a range of hardware and sampling quality options.
FidelityFX Luminance Preserving Mapper (LPM) is an open-source library for bringing over HDR and wide gamut tone and gamut mapping into games. There’s also support for packed FP16 when used with Vulcan.
Lastly, there’s FidelityFX Single Pass Downsampler (SPD). This can be used for downsampling the resolution of assets using MIP levels. SPD is capable of a single pass for all MIP levels, rather than a pass for every MIP level, meaning the GPU doesn’t have to be synchronized after each step. This makes SPD suitable for async compute and should speed up performance.
Source: https://gpuopen.com/introducing-our-new-fidelityfx-effects/
US Wants Chipmakers To Bring Production Stateside
The current administration is in talks with chipmakers like Intel and TSMC about bolstering domestic chip production. The talks between the US government and chipmakers come as global geopolitics heat up with China, thanks to hot seat issues such as trade and the current pandemic.
The US government has become increasingly concerned, or paranoid, about reliance on chip manufacturing and supply chains in China. To that end, representatives with Intel and TSMC have been in contact with the Department of Defense about increasing domestic chip production, according to Reuters.
“Intel is well positioned to work with the U.S. government to operate a U.S.-owned commercial foundry and supply a broad range of secure microelectronics”, said Intel.
In unrelated news, Intel has posted 28 security advisories, including new vulnerabilities, since January of 2020.
TSMC has long straddled the fence on its position about moving any manufacturing to the US, saying “We are actively evaluating all the suitable locations, including in the U.S., but there is no concrete plan yet.” Furthermore, according to a separate report from Digitimes, TSMC continues to weigh its options for locations outside of Taiwan, with the US being one such possible location.
However, new reports have seemingly contradicted what Digitimes is saying. According to WSJ, TSMC has imminent plans to build a 5nm foundry in Arizona, with construction expected to begin in 2021. TSMC expects production to begin in 2024.
The new fab will get a $12B investment, spread over nine years and will create an estimated 1,600 jobs. The fab will eventually be capable of producing 20,000 wafers per month. This will mark TSMC’s second manufacturing site in the US, with the other being located in Camas, Washington. TSMC also holds design centers in Austin, Texas and San Jose, California.
The US government is also apparently looking into ways to help Samsung expand its foundry operations in the states. Samsung currently has a plant in Austin, Texas.
https://www.digitimes.com/news/a20200512PD202.html
Unpatchable Thunderspy Attack
News has surfaced of a new hardware-based exploit affecting Thunderbolt ports, aptly named Thunderspy. The news comes to light thanks to the work of researcher Björn Ruytenberg.
The exploit affects any PC with a Thunderbolt port and can apparently bypass passwords and encryption to capture data. While the attack is serious, it should be noted that this attack requires the attacker to have physical access to the hardware, which should theoretically limit the attack surface.
Thunderspy is a collection of attacks that essentially break any security measures put in place by Thunderbolt 1, 2, and 3 and allow DMA (Direct Memory Access) attacks.
“These vulnerabilities lead to nine practical exploitation scenarios. In an evil maid threat model and varying Security Levels, we demonstrate the ability to create arbitrary Thunderbolt device identities, clone user-authorized Thunderbolt devices, and finally obtain PCIe connectivity to perform DMA attacks. In addition, we show unauthenticated overriding of Security Level configurations, including the ability to disable Thunderbolt security entirely, and restoring Thunderbolt connectivity if the system is restricted to exclusively passing through USB and/or DisplayPort. We conclude with demonstrating the ability to permanently disable Thunderbolt security and block all future firmware updates” says Ruytenberg in his report.
Thunderspy affects all PCs shipped between 2011 and 2019 and is essentially unpatchable via software or firmware mitigations. According to Ruytenberg, it will require a silicon redesign. Intel notes via a blog post that Kernel Direct Memory Access (DMA) has been in place since 2019 to mitigate Thunderbolt-based attacks.
However, Ruytenberg warns that Kernel Direct Memory Access protection isn’t a silver bullet for Thunderspy, and that Thunderspy stands to threaten future connection standards such as USB 4 and Thunderbolt 4.
Source: https://blogs.intel.com/technology/2020/05/more-information-on-thunderspy/#gs.5sypvk
Unreal Engine 5
Epic made waves this week by taking the wraps off its Unreal Engine 5. Epic showcased a tech demo running on PlayStation 5 hardware, a devkit no doubt. The reveal -- and subsequent headlines -- were replete with sensationalism about how our PCs are now being put to shame and included the liberal use of “next-gen”. Which, we suppose, is about right on time.
That aside, the tech demo, titled “Lumen in the Land of Nanite,” highlights a few things that Unreal Engine 5 will bring with it. The demo gets its namesake from the core technology in UE5: Nanite virtualized geometry. The idea behind Nanite is that developers can port film-quality assets into UE5 with automatic scalability in terms of LODs depending on the scene.
“Nanite virtualized micropolygon geometry frees artists to create as much geometric detail as the eye can see. Nanite virtualized geometry means that film-quality source art comprising hundreds of millions or billions of polygons can be imported directly into Unreal Engine—anything from ZBrush sculpts to photogrammetry scans to CAD data—and it just works,” says Epic. Hey, we’ve heard that somewhere before. Maybe Jensen is still haunting the studio.
Then there’s the Lumen System, which Epic says is a “fully dynamic global illumination solution that immediately reacts to scene and light changes.” Epic frames this as a way around the static lighting limitations found in current game engines. Epic claims that developers will no longer need to develop light maps by hand or wait for “lightmap bakes to finish and to author light map UVs.”
Additionally, the idea of streaming assets from storage to system memory is apparently being realized thanks to SSDs finally making their way into consoles. -- which isn’t a new idea, mind you. It’s just one that’s been shelved thanks to spinning storage. With SSDs, increased memory, and the improved I/O of new consoles, games should be able to (theoretically) load in assets as the scene is rendered. This obviously has big implications for loading screens and pop-in effects.
On the business side, Epic is transitioning UE5 to a new model where royalties are waived until a game reaches $1M in gross revenue.
Source: https://www.unrealengine.com/en-US/blog/a-first-look-at-unreal-engine-5
Epic CEO Excited for PS5 Storage Implications for PC
Epic Games CEO Tim Sweeney noted in the company’s Unreal Engine 5 press briefing that he anticipates knock-on impact to the industry from the Sony PS5 storage solution change. We’ve already known that the PS5 and Xbox are both moving to solid-state storage, but Sweeney seems to indicate it’s more of an architectural advancement than a strict jump to SSDs.
Sweeney seemed to imply that the new SSDs would be sufficiently fast to render game worlds “tens of gigabytes in size” on-the-fly, highlighting that it’s specifically the PS5’s storage architecture that outpaces what PCs are currently capable of. Sweeney specifically stated that it’s “ahead of anything you can buy on PC for any amount of money right now. It’s going to help drive future PCs,” further elaborating that SSDs need to catch up to the PS5’s storage solution.
Note that Sweeney, despite being a CEO of a massive company, is extremely technically competent and likely knows what he’s talking about. This isn’t just a fluff piece. He’s always been a PC guy first.
We’re looking forward to technical analysis as the console nears launch.
Source: https://www.unrealengine.com/en-US/blog/a-first-look-at-unreal-engine-5
Samsung 6.5GB/s Read PCIe Gen4 SSD
Samsung just announced its new PCIe Gen4 SSD, the PM9A3 SSD, the highlight of which is among chart-topping sequential read performance at 6500MB/s. Some of the other stats aren’t as impressive, particularly the random write speed at 180,000 IOPS. This drive seems to be optimized specifically for high sequential reads, which would make it useful in data center or other targeted use cases where that’s the primary need. The SSD also writes at 3500MB/s sequentially, putting it among top performers, but barring it from being the top performer. Random read IOPS is reported as 900,000.
Samsung intends to make a full range of the PM9A3, starting at 960GB and reaching almost 8TB, with the max at 7.68TB after over provisioning.
Source: https://www.sammobile.com/news/samsung-unveils-new-ssd-pcie-gen-4-e1-s-form-facator/
Editorial: Eric Hamilton
Host, Additional Reporting: Steve Burke
Video: Andrew Coleman, Keegan Gallick