HW News: GDDR6 Mass Production, MSFT Telemetry, & AMD Newhires
Posted on January 29, 2018
This week's hardware news recap teases some of our upcoming content pieces, including a potential test on Dragonball FighterZ, along with pending-publication interviews of key Spectre & Meltdown researchers. In addition to that, as usual, we discuss major hardware news for the past few days. The headline item is the most notable, and pertains to Samsung's GDDR6 memory entering mass production, nearing readiness for deployment in future products. This will almost certainly include GPU products, alongside the expected mobile device deployments. We also talk AMD's new-hires and RTG restructure, its retiring of the implicit primitive discard accelerator for Vega, and SilverStone's new low-profile air cooler.
Show notes are below the embedded video.
Samsung GDDR6 Enters Mass Production
Samsung recently revealed they’ve entered mass production with the industry’s first GDDR6 memory, aimed at advanced graphics, in addition to other applications such as AI, automotive, VR (virtual reality), AR (augmented reality), and network systems. Indeed, for gamers, the most enticing application is in next-generation graphics cards.
Samsung’s first GDDR6 ICs will be available in 16 Gb densities, offering 18Gbps per pin transfer rate and up to 72 GB/s of bandwidth per chip. Samsung’s GDDR6 speeds are more than double than that of GDDR5’s 8Gbps, and higher than GDDR5X memory at 11Gbps. The new memory is built on Samsung’s 10-nanometer “class” technology—class referring to anything between 10 and 19nm. We can assume that Samsung isn’t at 10nm with GDDR6, otherwise they’d unambiguously acknowledge the achievement. Still, according to Samsung, 10nm class technology brings about a 30% productivity increase—presumably compared to 20nm—to manufacturing. That productivity increase probably translates to more chips per wafer.
Additionally, Samsung’s new GDDR6 will operate at 1.35V, down from the 1.55V required for GDDR5/GDDR5X. This means higher efficiency, less heat waste, and higher performing modules. Samsung doesn’t elaborate on this accomplishment, and only credits their low-power circuit design.
Samsung has currently made no mention of when its GDDR6 will ship globally/commercially, but Samsung has touted for some time now that they would have GDDR6 memory in 2018. With Samsung seemingly on schedule, we can reasonably assume Samsung’s clients are planning products with the new memory solution.
AMD Hires New RTG SVP & Engineering SVP
AMD announced its hiring of two senior members of RTG, filling-in for the role left behind by former RTG Head Raja Koduri. The newhires are Mike Rayfield, SVP and GM of AMD RTG, and David Wang, SVP of Engineering for RTG. David Wang’s roles will pertain entirely to architecture and hardware development of future AMD GPUs, while Rayfield will handle overall direction, strategy, and business development.
Mike Rayfield has been in the industry since about 1983. His work experience includes 16 years as Director of Sales at Texas Instruments, 2 years at Cisco for Business Development, 7 years at NVIDIA’s mobile division as a General Manager, and 5 years at Micron as a General Manager of the mobile unit.
David Wang’s history dates back to the beginning of consumer graphics, including history as a design manager at SGI, the company from which Jensen Huang later defected to form nVidia. Wang also worked his way up from Engineering Manager at ATI, to Engineering Director, to Senior Director, and stayed at ATi and AMD from 2000 to 2012, later leaving for Synaptics in 2012 through present.
Source: http://www.amd.com/en-us/press-releases/Pages/graphics-industry-leaders-2018jan23.aspx
Secondary source: https://venturebeat.com/2018/01/23/amd-hires-former-nvidia-tegra-exec-and-synaptics-leader-to-run-its-graphics-chip-division/
AMD Kills Implicit Primitive Shader Function
According to a post by Marc Sauter on 3DCenter.org, AMD intends to kill its implicit primitive shader driver support for the RX Vega series graphics cards, which users have clamored over for performance increases since launch. As of now, it looks like AMD is implementing a driver path that will allow developers to explicitly write code to take advantage of the primitive shader function; unfortunately, because AMD’s Vega marketshare is insignificant, and because poor yields, high production cost, and mining demand have made Vega unavailable to gamers, it would be unlikely a developer would allocate time to such programming. Maybe we’ll see it in a DOOM patch, but that’d be it.
To bring everyone up to speed, the primitive shader function is supposed to more intelligently discard primitives that are unnecessary, done by combining the vertex and geometry shader functions. AMD claimed that vertex shader code could increase in speed by 2x with this function.
Old Mike Mantor video: https://www.youtube.com/watch?v=XC5Dy_b-kE8
Official source: https://www.forum-3dcenter.org/vbulletin/showthread.php?p=11610696
Microcenter’s Pricing
Microcenter notified us that they are working to inform all sales associates on the new discount policies, as this is something with which we've experienced frustrations. Official policy below:
- Video Cards are now one per household
- No Price Matches
- To qualify for the coupon you already have to have $700 in revenue attaches to the processor
Microsoft Telemetry
Some major notes on Microsoft's upcoming Windows 10 update, all regarding telemetry and data collection:
-”Windows Diagnostic Data Viewer” utility to be added in next big Windows 10 release
-Windows Insiders will get early access
-Microsoft describes this as part of their commitment “to be fully transparent on the diagnostic data collected from your Windows devices, how it is used, and to provide you with increased control over that data.”
-Remember, earlier Microsoft had to release a special telemetry-free version of Win 10 for China
-Will be able to view all data stored in the cloud about your specific device, including:
>Common Data, like the Operating System’s name, the Version, Device ID, Device Class, Diagnostic level selection and so on.
>Device Connectivity and Configuration such as device properties and capabilities, preferences and settings, peripherals, and device network information.
>Product and Service Performance data that show device health, performance and reliability data, movie consumption functionality on the device and device file queries. It’s important to note that this functionality is not intended to capture user viewing or, listening habits.
>Product and Service Usage data includes details about the usage of the device, operating system, applications, and services.
>Software Setup and Inventory such as installed applications and install history, device update information.
SilverStone AR11 LP Cooler
-LGA115x low profile cooler, supports 95W TDP or greater depending on case airflow
-92mm fan (15mm thick), 1200-3000RPM, 55.76CFM @ 44.5dBA max
-97mm x 94mm x 47mm (47mm + 15mm = 62mm tall)
-4 x 6mm thick copper heat pipes, aluminum fins
-Price TBA
http://www.silverstonetek.com/product.php?area=en&pid=785
https://www.intel.com/content/dam/www/public/us/en/documents/guides/4th-gen-core-lga1150-socket-guide.pdf (has images of socket LGA1150 for B roll)
News: Steve Burke, Patrick Lathan, & Eric Hamilton
Video: Andrew Coleman