HW News - Ryzen 9 16-Core, AMD Supercomputer, & Intel's Shortage Coming to an End
Posted on May 14, 2019
Computex is just a few weeks away. Mark calendars for May 28 to June 1 (and surrounding dates) -- we're expecting AMD Ryzen 3000 discussion, Navi unveils or teases, and X570 motherboards in the CPU category, with potential Intel news on 10nm for mobile and notebook devices. This week's news cycle was still busy pre-show, though, including discussion of an impending end to Intel's CPU shortage, the AMD supercomputer collaboration with Cray, NVIDIA's move away from binned GPUs, and more.
As always, the show notes are below the embedded video.
Livestream Results & Computex
We've been quiet on the website while doing livestreams the past weekend, including an overclock of the Ryzen 7 2700X to 5550MHz, nearly 5.6GHz, and overclocking the KINGPIN 2080 Ti to about 2600MHz. You can find our recaps of each event here:
Rumor: 16-Core Ryzen 9 CPU Leaked
Rumors and speculation have been running rampant for some time regarding 16-core Ryzen 3000 parts, as Dr. Lisa Su has hinted that AMD could go beyond the disclosed 8-core Ryzen 7 SKUs with its new chiplet design.
Well known leaker, a Twitter user by the name of TUM_APISAK, is laying claim to a 16-core engineering sample out in the wild, purportedly sporting a base clock of 3.3 GHz and a boost clock of 4.2 GHz, and that it will run on the X570 chipset. The chip would likely assume the Ryzen 9 mantle. There’s also talk of 12-core Ryzen 7 SKUs.
TUM_APISAK is allegedly a reliable source, in so far as leaks and rumors are concerned. Still, take the rumor with the usual prescribed amount of salt.
People should stop expecting 5GHz all-core stock frequencies, though.
Source: https://twitter.com/TUM_APISAK/status/1126466183740321798
https://www.tomshardware.com/news/amd-ryzen-3000-series-16-core-cpu-specs,39304.html
NVIDIA to Stop Dividing Turing Into A and Non-A
Nvidia has been dividing the Turing silicon into A and non-A variants for its RTX 2070, 2080, and 2080 Ti cards; however, soon there will only be one silicon, at least for the 2070 and 2080 cards.
Igor Wallossek with Tom’s Hardware Germany recently disclosed via retail channel sources that Nvidia aims to stop binning the Turing A-dies, and further cease production on the non A-dies. If you recall, when we dissected the EVGA RTX 2070 XC Ultra, we compared the TU106-400-A1 and TU106-400A-A1 dies.
The separate Turing dies allowed Nvidia to cherry pick the best yields, and offer them in the form of A-dies for its partners’ flagship, overclocked cards. Meanwhile, Nvidia could offer the lower quality non-A dies -- which Nvidia forbade overclocking on -- for lower priced cards down the stack. Users could of course overclock these cards themselves, but AIB partners were not allowed to pre overclock them. It’s also highly unlikely a non-A die would overclock as well as its binned, A-die counterpart.
Via Tom’s Hardware’s report, at the end of May, the RTX 2080 and 2070 cards will use only one variant of Turing: the TU104-410 and TU106-410, respectively. This implies that TSMC’s 12nm "FFN" process that the GPUs are built on is mature enough for optimum yields, and Nvidia no longer needs to segregate the silicon.
This is theoretically good news, as the new silicon should level the playing field in terms of price, relative to the GPU users are buying. It should also mean we can expect higher quality Turing silicon in the future. We’ll see if Nvidia plans to do the same with the RTX 2080 Ti and the TU102 die.
Tom’s Hardware via Igor Wallossek: https://www.tomshardware.com/news/nvidia-binning-turing-die-geforce-rtx-2080-rtx-2070,39250.html
Intel Will Debut 7nm Process by 2021 to Fight TSMC
Also in Intel news, the company recently revealed at its 2019 investor meeting that it will ship its first 10nm CPU in volume, Ice Lake, this June. This also aligns with Intel’s plans to have Ice Lake-based devices on shelves for Holiday 2019. Intel also anticipates a steady cadence of 10nm products throughout 2019 and 2020: 10nm AgileX FPGAs, a 10nm GPU, Tiger Lake, and the 10nm Snow Ridge SoC aimed at 5G.
Intel will also be debuting its 7nm process in 2021, competing with TSMC’s 5nm process. Intel’s first 7nm product will be an Intel Xe Graphics Architecture based GPU, following its first discrete GPU in 2020. Intel’s 7nm process will be the first time the company uses EUV, and Intel expects a 2x improvement in terms of density over 10nm. Intel will also lean heavily into its EMIB and Foveros technology at 7nm.
Intel also plans to focus on intra-node optimizations, i.e., more “+” steps within the node. Intel expects this will help avoid taking on too much risk with design goals, à la 10nm, while also improving scaling between generations. Intel also expects to deliver one Moore's Law gain at the beginning of a node, and one at the final revision of a node.
Source: https://newsroom.intel.com/news/2019-intel-investor-meeting/
AMD & Cray Collaborate on World’s First Exascale SuperComputer
The world’s current fastest supercomputer, Summit, is set to be dethroned in 2021 with a joint venture between AMD and Cray.
Frontier, as the system is called, is expected to be delivered to ORNL (Oak Ridge National Laboratory) in 2021, and will be the world’s first exascale supercomputer targeting 1.5 exaflops of compute performance -- roughly 5 times faster than Summit’s 200 petaflops.
The exact details haven’t been confirmed yet, but Frontier will leverage AMD EPYC CPUs, Radeon Instinct GPUs, AMD’s Infinity Fabric, and Cray Slingshot interconnect technology. Each node will feature one AMD EPYC CPU, and four Radeon Instinct GPUs.
“We are excited to work with the team at AMD to deliver the Frontier system to Oak Ridge National Laboratory,” said Steve Scott, senior vice president and CTO at Cray. “Cray’s Shasta supercomputers are designed to support leading edge processor technologies and high-performance storage, all tightly interconnected by Cray’s new Slingshot network. The combination of Cray and AMD technology in the Frontier system will dramatically enhance performance at scale for AI, analytics, and simulation, enabling DOE to further push the boundaries of scientific discovery.”
Frontier will be deployed to study weather, sub-atomic structures, genomics, physics, and more.
Intel Expects Rebound from CPU Shortage in 2H19
Supply chain vendors for notebooks are signaling that Intel’s CPU shortage will ease in June, as Intel expects to increase its shipments of entry-level processors. Intel has been aggressively focusing on higher-margin products, as well as its top customers, during the CPU drought.
Digitimes reports that Intel has informed notebook partners that it will begin shipping entry-level chips in June, and while there will still be some shortages, the shortages will greatly narrow. Notebook shipments are expected to rebound in the second quarter, after a sluggish first quarter, primarily riding on the hype for Intel and Nvidia’s newest mobile chips.
Intel’s CPU shortage has driven many vendors and OEMs to put in orders with AMD. However, according to Digitime’s sources, vendors such as Dell, HP, and Lenovo are expected to be putting in more orders with Intel instead of AMD.
Intel CEO Bob Swan recently acknowledged Intel’s CPU troubles in an earning’s call, stating that the CPU shortage would not be fully rectified unil Q3. Swan also vowed “never again to be a constraint” in Intel’s customers’ growth.
Source: https://www.digitimes.com/news/a20190508PD211.html
AMD’s EPYC Persuades Dell to Triple EPYC-Based Server Platforms
Last year, Dell EMC CTO John Roese didn’t seem to be overly impressed with AMD’s EPYC platform.
“AMD is doing some interesting things, and by adding them to the portfolio we pick up a few extra areas, but let's be very clear: there is a huge, dominant player in compute semiconductors, and then there is a challenger which is doing some very good innovative work called AMD, but the gap between them is quite large in terms of market share and use-cases. So our portfolio is not going to change in any meaningful way,” Roese said in an interview last year. “ Don't expect it to be a duopoly any time soon.”
Now, in very contrasting statement, Dell’s Dominique Vanhamme states that the company will essentially triple its EPYC server offerings, and support the upcoming 7nm EPYC Rome. "Out of, let's say, 50 or so platforms that we have today," said Vanhamme to Itpro.co.uk “three of them are AMD - we'll probably triple that by the end of this year."
Aside from the promises 7nm brings, Vanhamme cites a high demand for EPYC from customers, particularly in the general purpose market. Still, AMD’s representation in Dell’s server portfolio will pale in comparison to Intel’s ~50 SKUs, highlighting just how much of a stranglehold Intel has on the server market.
Still, challenges aside, AMD’s EPYC adoption continues steadily as the company has been picking its shots wisely with cloud service providers.
Editorial: Eric Hamilton
Host: Steve Burke
Video: Josh Svoboda, Andrew Coleman