Sign up for our Technology Newsletter 
AMD’s Fiji GPU Goes High Bandwidth
New evidence coming from two LinkedIn profiles of AMD employees suggest that AMD’s upcoming Radeon R9 380X graphics card which is expected to be based on the Fiji GPU will actually use High-Bandwidth Memory.
Spotted by a member of 3D Center forums, the two LinkedIn profiles mention both the R9 380X by name as well as describe it as the world’s firts 300W 2.5D discrete GPU SoC using stacked die High-Bandwidth Memory and silicon interposer. While the source of the leak is quite strange, these are more reliable than just rumors.
The first in line is the profile of Ilana Shternshain, an ASIC Physical Design Engineer, which has been behind the Playstation 4 SoC, Radeon R9 290X and R9 380X, which is described as the “largest in ‘King of the hill’ line of products.”
The second LinkedIn profile is the one from AMD’s System Architect Manager, Linglan Zhang, which was involved in developing “the world’s first 300W 2.5D discrete GPU SOC using stacked die High Bandwidth Memory and silicon interposer.”
Earlier rumors suggest that AMD might launch the new graphics cards early this year as the company is under heavy pressure from Nvidia’s recently released, as well as the upcoming, Maxwell-based graphics cards.
Samsung Finally Starts 14nm FinFET
A company insider has spilled the beans in Korea, claiming that Samsung has started Apple A9 production in 14nm FinFET.
The A9 is the next generation SoC for Apple iPhone and iPad products and it is manufactured on the Samsung – GlobalFoundries 14nm FinFET manufacturing process. In the other news, Samsung’s Ki-nam, president of the company’s semiconductor business and head of System LSI business has confirmed that the company started production of 14-nanometre FinFET chips.
The report mentions Austin as a possible site for Apple products but we wonder if the GlobalFoundries Fab 8 in New York State could become one of the partners for the 14nm FinFET manufacturing. Samsung didn’t officially reveal the client for the 14nm FinFET, but Apple is the most obvious candidate, while we expect to see 14 / 16nm FinFET graphics chips from AMD and Nvidia but most likely in the latter half of 2015 at best.
Qualcomm is likely to announce new LTE modem based on 14nm FinFET and the flagship SoC Snapdragon 810 is a 20nm chip. Qualcomm is manufacturing its 810 chips as we speak to meet demand for flagship Android phones coming in Q1 2015. Flagship Samsung, HTC and LG phones among others are likely to use Snapdragon 810 as a replacement for this year’s Snapdragon 801, a high end chip that ended up in millions of high-end phones.
Samsung / GlobalFoundries14nm FinFET process is 15 percent smaller, 20 percent faster, and 35 percent more power efficient compared to 20nm processors. This definitely sounds exiting and will bring more performance into phones, tablets, GPUs and will significantly decrease power consumption. The move from 28nm is long overdue.
We believe that Qualcomm’s LTE modem might be the first chip to officially come with this manufacturing process and Apple will probably take most of the 14nm production for an update in its tablets and phones scheduled for 2015.
Amazon Web Services Goes Zocalo
Amazon Web Services (AWS) has announced two much-needed boosts to its fledgling Zocalo productivity platform, making the service mobile and allowing for file capacities of up to 5TB.
The service, which is designed to do what Drive does for Google and what Office 365 does for software rental, has gained mobile apps for the first time as Zocalo appears on the Google Play store and Apple App Store.
Amazon also mentions availability on the Kindle store, but we’re not sure about that bit. We assume it means the Amazon App Store for Fire tablet users.
The AWS blog says that the apps allow the user to “work offline, make comments, and securely share documents while you are in the air or on the go.”
A second announcement brings Zocalo into line with the AWS S3 storage on which it is built. Users will receive an update to their Zocalo sync client which will enable file capacities up to 5TB, the same maximum allowed by the Amazon S3 cloud.
To facilitate this, multi-part uploads will allow users to carry on an upload from where it was after a break, deliberate or accidental.
Zocalo was launched in July as the fight for enterprise storage productivity hots up. The service can be trialled for 30 days free of charge, offering 200GB each for up to 50 users.
Rival services from companies including the aforementioned Microsoft and Google, as well as Dropbox and Box, coupled with aggressive price cuts across the sector, have led to burgeoning wars for the hearts and minds of IT managers as Microsoft’s Office monopoly begins to wane.
Amazon Intel Zeon Inside
Amazon has become the latest vendor to commission a customized Xeon chip from Intel to meet its exact compute requirements, in this case powering new high-performance C4 virtual machine instances on the AWS cloud computing platform.
Amazon announced at the firm’s AWS re:Invent conference in Las Vegas that the latest generation of compute-optimized Amazon Elastic Compute Cloud (EC2) virtual machine instances offer up to 36 virtual CPUs and 60GB of memory.
“These instances are designed to deliver the highest level of processor performance on EC2. If you’ve got the workload, we’ve got the instance,” said AWS chief evangelist Jeff Barr, detailing the new instances on the AWS blog.
The instances are powered by a custom version of Intel’s latest Xeon E5 v3 processor family, identified by Amazon as the Xeon E5-2666 v3. This runs at a base speed of 2.9GHz, and can achieve clock speeds as high as 3.5GHz with Turbo boost.
Amazon is not the first company to commission a customized processor from Intel. Earlier this year, Oracle unveiled new Sun Server X4-4 and Sun Server X4-8 systems with a custom Xeon E7 v2 processor.
The processor is capable of dynamically switching core count, clock frequency and power consumption without the need for a system level reboot, in order to deliver an elastic compute capability that adapts to the demands of the workload.
However, these are just the vendors that have gone public; Intel claims it is delivering over 35 customized versions of the Intel Xeon E5 v3 processor family to various customers.
This is an area the chipmaker seems to be keen on pursuing, especially with companies like cloud service providers that purchase a great many chips.
“We’re really excited to be working with Amazon. Amazon’s platform is the landing zone for a lot of new software development and it’s really exciting to partner with those guys on a SKU that really meets their needs,” said Dave Hill, senior systems engineer in Intel’s Datacenter Group.
Also at AWS re:Invent, Amazon announced the Amazon EC2 Container Service, adding support for Docker on its cloud platform.
Currently available as a preview, the EC2 Container Service is designed to make it easy to run and manage distributed applications on AWS using containers.
Customers will be able to start, stop and manage thousands of containers in seconds, scaling from one container to hundreds of thousands across a managed cluster of Amazon EC2 instances, the firm said.
Google Continues A.I. Expansion
Google Inc is growing its artificial intelligence area, hiring more than half a dozen leading academics and experts in the field and announcing a partnership with Oxford University to “accelerate” its efforts.
Google will make a “substantial contribution” to establish a research partnership with Oxford’s computer science and engineering departments, the company said on Thursday regarding its work to develop the intelligence of machines and software, often to emulate human-like intelligence.
Google did not provide any financial details about the partnership, saying only in a post on its blog that it will include a program of student internships and a series of joint lectures and workshops “to share knowledge and expertise.”
Google, which is based in Mountain View, California, is building up its artificial intelligence capabilities as it strives to maintain its dominance in the Internet search market and to develop new products such as robotics and self-driving cars. In January Google acquired artificial intelligence company Deep Mind for $400 million according to media reports.
The new hires will be joining Google’s Deep Mind team, including three artificial intelligence experts whose work has focused on improving computer visual recognition systems. Among that team is Oxford Professor Andrew Zisserman, a three-time winner of the Marr Prize for computer vision.
The four founders of Dark Blue Labs will also be joining Google where they will be will be leading efforts to help machines “better understand what users are saying to them.”
Google said that three of the professors will hold joint appointments at Oxford, continuing to work part time at the university.
Will The Chip Industry Take Fall?
Microchip Technology has managed to scare Wall Street by warning of an industry downturn. This follows rumours that a number of US semiconductor makers with global operations are reducing demand for chips in regions ranging from Asia to Europe.
Microchip Chief Executive Steve Sanghi warned that the correction will spread more broadly across the industry in the near future. Microchip expects to report sales of $546.2 million for its fiscal second quarter ending in September. The company had earlier forecast revenue in a range of $560 million to $575.9 million. Semiconductor companies’ shares are volatile at the best of times and news like this is the sort of thing that investors do not want to hear.
Trading in Intel, whiich is due to report third quarter results tomorrow, was 2.6 times the usual volume. Micron, which makes dynamic random access memory, or DRAM, was the third-most traded name in the options market. All this seems to suggest that the market is a bit spooked and much will depend on what Chipzilla tells the world tomorrow as to whether it goes into a nosedive.
nVidia Finally Goes 20nm
For much of the year we were under the impression that the second generation Maxwell will end up as a 20nm chip.
First-generation Maxwell ended up being branded as Geforce GTX 750 and GTX 750 TI and the second generation Maxwell launched a few days ago as the GTX 980 and Geforce GTX 970, with both cards based on the 28nm GM204 GPU.
This is actually quite good news as it turns out that Nvidia managed to optimize power and performance of the chip and make it one of the most efficient chips manufactured in 28nm.
Nvidia 20nm chips coming in 2015
Still, people keep asking about the transition to 20nm and it turns out that the first 20nm chip from Nvidia in 20nm will be a mobile SoC.
The first Nvidia 20nm chip will be a mobile part, most likely Erista a successor of Parker (Tegra K1).
Our sources didn’t mention the exact codename, but it turns out that Nvidia wants to launch a mobile chip first and then it plans to expand into 20nm with graphics.
Unfortunately we don’t have any specifics to report.
AMD 20nm SoC in 2015
AMD is doing the same thing as its first 20nm chip, codenamed Nolan, is an entry level APU targeting tablet and detachable markets.
There is a strong possibility that Apple and Qualcomm simply bought a lot of 20nm capacity for their mobile modem chips and what was left was simply too expensive to make economic sense for big GPUs.
20nm will drive the voltage down while it will allow higher clocks, more transistors per square millimeter and it will overall enable better chips.
Just remember Nvidia world’s first quad-core Tegra 3 in 40nm was rather hot and making a quad core in 28nm enabled higher performance and significantly better battery life. The same was true of other mobile chips of the era.
We expect similar leap from going down to 20nm in 2015 and Erista might be the first chip to make it to 20nm. A Maxwell derived architecture 20nm will deliver even more efficiency. Needless to say AMD plans to launch 20nm GPUs next year as well.
It looks like Nvidia’s 16nm FinFET Parker processor, based on the Denver CPU architecture and Maxwell graphics won’t appear before 2016.
Intel Sampling Xeon D 14nm
Intel has announced that it is sampling its Xeon D 14nm processor family, a system on chip (SoC) optimized to deliver Intel Xeon processor performance for hyperscale workloads.
Announcing the news on stage during a keynote at IDF in San Francisco, Intel SVP and GM of the Data Centre Group, Diane Bryant, said that the Intel Xeon processor D, which initially was announced in June, will be based on 14nm process technology and be aimed at mid-range communications.
“We’re pleased to announce that we’re sampling the third generation of the high density [data center system on a chip] product line, but this one is actually based on the Xeon processor, called Xeon D,” Bryant announced. “It’s 14nm and the power levels go down to as low as 15 Watts, so very high density and high performance.”
Intel believes that its Xeon D will serve the needs of high density, optimized servers as that market develops, and for networking it will serve mid-range routers as well as other network appliances, while it will also serve entry and mid-range storage. So, Intel claimed, you will get all of the benefits of Xeon-class reliability and performance, but you will also get a very small footprint and high integration of SoC capability.
This first generation Xeon D chip will also showcase high levels of I/O integrations, including 10Gb Ethernet, and will scale Intel Xeon processor performance, features and reliability to lower power design points, according to Intel.
The Intel Xeon processor D product family will also include data centre processor features such as error correcting code (ECC).
“With high levels of I/O integration and energy efficiency, we expect the Intel Xeon processor D product family to deliver very competitive TCO to our customers,” Bryant said. “The Intel Xeon processor D product family will also be targeted toward hyperscale storage for cloud and mid-range communications market.”
Bryant said that the product is not yet available, but it is being sampled, and the firm will release more details later this year.
This announcement comes just days after Intel launched its Xeon E5 v2 processor family for servers and workstations.
Vendors Testing New Xeon Processors
Comments Off on Vendors Testing New Xeon Processors
Intel is cooking up a hot batch of Xeon processors for servers and workstations, and system vendors have already designed systems that are ready and raring to go as soon as the chips become available.
Boston is one of the companies doing just that, and we know this because it gave us an exclusive peek into its labs to show off what these upgraded systems will look like. While we can’t share any details about the new chips involved yet, we can preview the systems they will appear in, which are awaiting shipment as soon as Intel gives the nod.
Based on chassis designs from Supermicro, with which Boston has a close relationship, the systems comprise custom-built solutions for specific user requirements.
On the workstation side, Boston is readying a mid-range and a high-end system with the new Intel Xeon chips, both based on two-socket Xeon E5-2600v3 rather than the single socket E5-1600v3 versions.
There’s also the mid-range Venom 2301-12T, which comes in a mid-tower chassis and ships with an Nvidia Quadro K4000 card for graphics acceleration. It comes with 64GB of memory and a 240GB SSD as a boot device, plus two 1TB Sata drives configured as a Raid array for data storage.
For extra performance, Boston has also prepared the Venom 2401-12T, which will ship with faster Xeon processors, 128GB of memory and an Nvidia Quadro K6000 graphics card. This also has a 240GB SSD as a boot drive, with two 2TB drives configured as a Raid array for data storage.
Interestingly, Intel’s new Xeon E5-2600v3 processors are designed to work with 2133MHz DDR4 memory instead of the more usual DDR3 RAM, and as you can see in the picture below, DDR4 DIMM modules have slightly longer connectors towards the middle.
For servers, Boston has prepared a 1U rack-mount “pizza box” system, the Boston Value 360p. This is a two-socket server with twin 10Gbps Ethernet ports, support for 64GB of memory and 12Gbps SAS Raid. It can also be configured with NVM Express (NVMe) SSDs connected to the PCI Express bus rather than a standard drive interface.
Boston also previewed a multi-node rack server, the Quattro 12128-6, which is made up of four separate two-socket servers inside a 2U chassis. Each node has up to 64GB of memory, with 12Gbps SAS Raid storage plus a pair of 400GB SSDs.
Intel Discusses SoFIA
Intel has shed more light on its upcoming SoFIA SoC, which stands for Smart of Feature Phone with Intel Architecture. For SEO purposes we’ll just call it Sophie.
The SoFIA project is being spearheaded by Intel’s Singapore office and CNET had a chance to catch up with recent developments. The chip was announced back in December, but details were sketchy and to some extent they still are.
SoFIA is going after MediaTek and Qualcomm in the mainstream and entry level segments. It features a dual-core Silvermont processor on a 28nm die. This is where it gets interesting, as Intel simply doesn’t do 28nm. SoFIA will be manufactured by TSMC instead, but Intel is planning to build SoFIA chips in-house in the future. The company has already made it clear that it is moving to 14nm, so there is a good chance the next generation chip will be a 14nm design churned out by Intel fabs.
SoFIA has an integrated 3G modem, but a 4G version is coming later. Bluetooth and WiFi are on other chips. Lack of 4G connectivity might not be a big deal for potential SoFIA customers, as the chip is supposed to go after entry level smartphone designs. Intel told CNET that SoFIA could pave the way to $50 smartphones in Asian markets and realistically nobody expects a $50 phone to feature 4G connectivity.
It is hard to say anything about SoFIA’s real-world performance. Quad-core tablet processors based on the Silvermont architecture can still hold their ground against many high-end ARM SoCs, so a dual-core version optimised for smartphones should have no trouble keeping up with lesser ARM parts.