Nvidia has released CUDA – its code that lets developers run their code on GPUs – to server vendors in order to get 64-bit ARM cores into the high performance computing (HPC) market.
The firm said today that ARM64 server processors, which are designed for microservers and web servers because of their energy efficiency, can now process HPC workloads when paired with GPU accelerators using the Nvidia CUDA 6.5 parallel programming framework, which supports 64-bit ARM processors.
“Nvidia’s GPUs provide ARM64 server vendors with the muscle to tackle HPC workloads, enabling them to build high-performance systems that maximise the ARM architecture’s power efficiency and system configurability,” the firm said.
The first GPU-accelerated ARM64 software development servers will be available in July from Cirrascale and E4 Computer Engineering, with production systems expected to ship later this year. The Eurotech Group also plans to ship production systems later this year.
Cirrascale’s system will be the RM1905D, a high density two-in-one 1U server with two Tesla K20 GPU accelerators, which the firm claims provides high performance and low total cost of ownership for private cloud, public cloud, HPC and enterprise applications.
E4′s EK003 is a production-ready, low-power 3U dual-motherboard server appliance with two Tesla K20 GPU accelerators designed for seismic, signal and image processing, video analytics, track analysis, web applications and Mapreduce processing.
Eurotech’s system is an “ultra-high density”, energy efficient and modular Aurora HPC server configuration, based on proprietary Brick Technology and featuring direct hot liquid cooling.
Featuring Applied Micro X-Gene ARM64 CPUs and Nvidia Tesla K20 GPU accelerators, the new ARM64 servers will provide customers with an expanded range of efficient, high-performance computing options to drive compute-intensive HPC and enterprise data centre workloads, Nvidia said.
Nvidia added, “Users will immediately be able to take advantage of hundreds of existing CUDA-accelerated scientific and engineering HPC applications by simply recompiling them to ARM64 systems.”
ARM said that it is working with Nvidia to “explore how we can unite GPU acceleration with novel technologies” and drive “new levels of scientific discovery and innovation”.
AMD is planning to bring its new Mantle API to Linux in the near future. Although Linux is not a big gaming platform at the moment, SteamOS could change all that starting next year.
AMD’s Richard Huddy says the decision was prompted by requests from developers who would like to see Mantle on Linux. However, he stopped short of specifying a launch date. Huddy confirmed that AMD plans to dedicate resources to bringing Mantle to Linux, but other than that we don’t have much to go on.
Mantle on SteamOS makes a lot of sense
Mantle is designed to cut CPU overhead and offer potentially significant performance improvements on certain hardware configurations. This basically means gamers can save a few pennies on their CPU and use them towards a better GCN-based graphics card.
However, aside from enthusiasts who build their own gaming rigs, the world of PC gaming is also getting a lot of attention from vendors specialising in out-of-the box gaming PCs and laptops. Many of them have already announced plans to jump the SteamOS bandwagon with Steam Machines of their own.
Should Mantle become available on Linux and SteamOS, it would give AMD a slight competitive edge, namely in the value department. In theory vendors should be able to select a relatively affordable APU and discrete GPU combo for their Steam boxes.
AMD already tends to provide good value in the CPU department. The prospect of using mainstream APUs backed by cheap discrete Radeons (or even Dual Graphics systems) sounds interesting.
It will take a while but the potential is there
Huddy told PC World that Mantle has some clear advantages over DirectX. Microsoft’s new DirectX 12 API has already been announced, but the first games to support it won’t arrive until late 2015.
“It (Mantle) could provide some advantages on Steam boxes,” said Huddy. “We are getting requests to deliver this high-performance layer.”
While DirectX 12 will be very relevant in the PC space, the same obviously cannot be said of Linux and SteamOS. Therefore Mantle on Linux makes a lot of sense. However, it all depends on AMD’s timetable.
Last month Valve announced Steam Machines would be pushed back to 2015. They were originally supposed to launch this summer and the first announcements were made months ago. The first designs were based on Intel and Nvidia silicon, but support for AMD hardware was added just a bit later.
When Valve announced the delay we argued that it could have a silver lining for AMD. It simply gives AMD more time to improve its drivers or add Mantle support, something Nvidia and Intel do not have to worry about.
It still remains to be seen whether Steam Machines can make a big dent on the gaming market. PC gaming is going through a renaissance, but the latest consoles are doing well, too (apart from the Wii U). The concept is very attractive on more than one level, but it is very difficult to make any predictions yet, since we are still about 15 months away from launch.
According to Jon Peddie Research (JPR), shipments of discrete graphics cards were down in the first quarter of the year. This is in line with seasonal trends, as the market cools down after the holiday season.
The sequential drop was 6.7 percent, which was still better than the overall desktop PC market, which slumped 9 percent. However, on a year-to-year basis add-in-board (AIB) shipments were down 0.8 percent. PC sales were down 1.1 percent.
Nvidia still controls two thirds of the market
Total AIB shipments in Q1 were just 14 million units. AMD and Nvidia both saw their shipments decrease 6.6 percent, so their market share did not change much.
Nvidia controls an estimated 65 percent of the market, up from 64.2 percent last year. AMD’s market share in Q1 was 35 percent, down from 35.6 percent a year ago.
The overall volume remains weak and in the long run things could get even worse, as on-die integrated graphics have already taken a big toll on sales of entry level discrete cards. As integrated GPUs become even faster, they are likely to cannibalize the low end market even further.
JPR points out that the AIB market peaked in 1999, with 114 million units shipped. Last year saw only 65 million units and the stagnant trend is likely to continue this year.
It’s not all bad news for AIBs
Although the slump in discrete GPU shipments is hurting AMD and NV hardware partners, JPR offers a rather encouraging outlook.
It points out that graphics cards are one of the most powerful, essential and exciting components in the PC market today. PC gaming is hardly dead, in fact it is going through what can only be described as a small renaissance. PCs will offer 4K/UHD gaming years ahead of consoles and the Steam Machine concept is looking good, too.
The compute market is another driver, as JPR points out:
“The technology is entering into major new markets like supercomputers, remote workstations, and simulators almost on a daily basis. It would be little exaggeration to say that the AIB resembles the 800-pound gorilla in the room.”
The AIB market is quite a bit less colourful and eventful than it was back in the day, but at least AIBs still have a lot on their hands and they are trying to tap new markets.
We spent a few weeks trying to find out what Nvidia has in mind for Tegra, as the company made quite a few sharp turns in its strategy. The first thing that happened is CUDA support, which is good for the security and defence markets, as it enables target recognition and similar tasks that can also be used for some peaceful technologies such as self-driving car.
Jetson is a cheap supercomputer base, but it is a very expensive microcontroller board. Apparently it is selling well, as it’s the fastest sub-10W supercomputer capable chip one can buy, and it costs $192. What Jen-Hsun Huang mentioned in the company’s financial Q1 2015 conference call is that Nvidia envisions “three growth drivers” for Tegra. They see Tegra in mobile devices, automotive and gaming.
Automotive is growing for Nvidia, but the company doesn’t really tell the world any meaningful numbers. Winning Tesla’s business means a few hundred thousand chips considering the fact that Tesla uses more than one Tegra chip per car. Tesla uses the old Tegra 2, something that the mobile devices market vaguely remembers. A total of 87 million cars were sold worldwide last year and there is a nice market opportunity for Nvidia there. Of course, the competition won’t stand still and let Nvidia conquer the automotive market unopposed, but we don’t see car manufactures changing Tegra for Qualcomm as quickly as this happened in mobile devices. Cars as platforms are built to last at least 5 years before any significant refresh and once you get a deal, you stay with the company for a while.
The other two catalysts that Huang mentioned, including mobile devices, might be a more troublesome component of Nvidia’s strategy. We simply cannot see Tegra K1 in any significant high-volume phone design in 2014. There are still some chances that Tegra K1 might end up in a few cool tablets, but it will be tough to land some top selling ones including the Nexus or Kindle Fire tablet refresh. In 2013 Qualcomm Snapdragon won both of these top selling tablets that are selling well. Gaming as a catalyst for Nvidia’s Tegra mobile strategy is a good playing card, but we are not sure how many Shield consoles you can sell. Nvidia has its own tablet, the Tegra Note 7, and probably a Tegra Note 8 in the works. However, these won’t outsell Google Nexus tablets anytime soon. Intel has big tablet plans to boost its market presence with 40 million units planned this year and Intel’s market development fund boosters are legendary and traditionally they work quite well with Taiwan, China, even US- and EU-based companies. Qualcomm and Mediatek have a strong presence in the tablet market, with Mediatek getting stronger every quarter especially in the lower end of the market. AMD wants a piece of Intel’s x86 tablet pie, too. It will be an interesting market to watch.
Since Tegra K1 doesn’t have an on board LTE it is a hard sell for phones in 2014. Top four phones in 2014 don’t have it as Samsung Galaxy 5, HTC One M8, LG G3 chose Qualcomm and Apple is using its own chips. Other top brands including Sony, Motorola are using Qualcomm for their high end phones. Even Chinese Xiaomi chose Snapdragon for its Mi3 phone, but there is a slim chance that there will be Tegra version too. China doesn’t really care about LTE, at least not yet. Nvidia might have a chance in the mainstream phone market, but its chances are not good. Mediatek is getting really strong in this market and Qualcomm has some great solution for this market as well. The Tegra 4i Gray chip has three design wins so far, Wiko Wax, LG G2 Mini in South America and security focused Blackphone. That’s simply not enough, not even close. At this point it seems increasingly likely that the Tegra 4i will not even have a successor.
Nvidia hopes that having two chips, one 32-bit based on Cortex A15 cores and one based on Denver 64-bit cores, might work. We will have to wait and see as the Tegra K1 has currently shipped only in the Jetson TK1 kit and the reason behind is probably the super high margin Nvidia can make with a $192 supercomputer board. We expect to see Tegra TK1 based products in June time, around Google IO. An 8-inch Tegra TK1 tablet would not surprise us, either. However, it’s not easy to be optimistic, as we simply don’t see a lot of potential design wins this year.
A part of Nvidia’s Financial Q1 2015 conference call Q&A session included some questions about micro servers, whether or not the 64-bit Tegra K1 can make it into the GRID market.
Nvidia’s CEO Jen-Hsun Huang was a straight shooter saying that Nvidia is “seeing a lot of interest in putting something like Tegra in micro servers,” but he added a caveat: “one step at a time, one step at a time.”
Jen-Hsun addressed the importance of the software component, or software stack for this market. He points out that the software stack Nvidia is building for GRID will eventually used on top of Tegra.
Denver could have what it takes
The Denver 64-bit architecture used in the Tegra K1 64-bit might have a shot in the micro server market as it will offer a lot of compute power and the CPU core should end up faster than the Cortex A57. Back at GTC 2014 we saw a demo at the GE booth where a single Jetson board hooked up to a camera over 10Gbit network port managed to trace a multiple targets at once.
The significance of such capabilities is that they can could easily find applications in the security market, drones, self-driving cars and all this can be done with close to 10W of power. Of course, this can be done with existing chips, but the CUDA powered Tegra K1 needs significantly less power to pull it off, which results in a much smaller footprint than say a 100W system that would handle the same task.
Eventually Nvidia could make a move in the traditional server market. The way things are going some 64-bit Cortex A57 servers might hit the market in early 2015 and Denver 64-bit might be the only custom based 64-bit ARM core ready at that time. Qualcomm’s 64-bit Krait is expected in the first half of 2015 and it might give Nvidia, AMD and other players a run for their money.
The server market is more than just chips and John Byrne, Chief Sales Officer at AMD, covered a few interesting points in a brief discussion with Fudzilla last month.
Does an ARM server push make sense for Nvidia?
In order to be successful in server market you need software, hardware, a great field application engineer network, customers that want to work with you and this is exactly why Calxeda failed. Nvidia has an advantage over Intel as it can do ARM based micro server, but then again, everyone else can. AMD is taking the ARM server market quite seriously and it has a lot more experience in the field than Nvidia.
Should Nvidia choose to proceed with an ARM server push, it will ultimately be a matter of research and development money that company can afford to put behind this risky strategy. Nvidia is making some progress with Tesla and Grid into the server market, but it will take some great products to convince Dell, HP and the rest of the market go your way.
In theory, Nvidia could leverage its compute software stack, but that’s only relevant in a handful of niches. AMD is targeting a much wider micro server market, while Nvidia could go after niche systems that would benefit from its compute tech. However, whether or not this niche is worth the investment remains to be seen. Both Nvidia and AMD can offer unique compute capabilities that could differentiate their ARM parts from the rest of the field (CUDA, Open CL), rendering their ARM server parts more competitive in a number of market segments.
AMD launched the first Kaveri parts a couple of months ago, but the rollout has been limited. The company is currently selling just two desktop SKUs in retail. There are no mobile parts and there are no 45W desktop parts, either.
However, the first mobile Kaveri parts could be just around the corner and this would hardly be news were it not for the surprisingly low TDP. Computerbase.de came across a curious HP leak which points to a new 19W part with some rather interesting specs.
Meet the A10-7300 low-power Kaveri
The part in question is the A10-7300, a quad-core rated at just 19W. We still don’t know anything about its GPU though, but the rest of the spec looks rather promising. The A10-7300 runs at 2GHz, but it can hit 3.2GHz on turbo. That compares well to older mobile APUs. The same HP laptop is available with the A10-5750M, which is a 2.5GHz part capable of hitting 3.5GHz with Turbo Core, but it’s a 35W part.
As for the GPU, we can only speculate at this point. When it comes to low-power APUs, AMD usually tries to keep the core count high and saves energy by reducing the clocks. For example, the A10-5545M, a 19W Richland quad-core, ships with 384 shaders clocked at 450MHz to 554MHz (the CPU is clocked at 1.7GHz/2.7GHz). The A10-5747M is a closer match, as it’s clocked at 2.1GHz/2.9GHz. It features 384 Radeon cores clocked at 544 to 626MHz, but it’s a 25W part.
We would be very surprised to see anything less than 384 GCN cores on quad-core Kaveri ULV parts. On dual-core designs with a single module the number should be 192, but AMD could surprise us.
Does it stand a chance against Haswell refresh parts?
AMD hasn’t had much luck in the mainstream mobile market for years and Kaveri isn’t about to change that overnight. However, the TDPs look quite a bit better and we’re expecting AMD’s integrated GPU to be as competitive as ever. After all, this is a 28nm part, we expected significant improvements.
It’s still not enough to make a dent on Intel’s market share, but Kaveri is starting to look a lot better than Richland. While it won’t be able to take on many Haswell SKUs, it should offer relatively good overall performance and value, with superior GPU performance as its main selling point.
AMD’s iGPU performance isn’t only a concern for Intel, as it has long-term implications on the discrete market. The attach rate is slowly going down as Intel and AMD pack ever more powerful GPUs into their mainstream parts. This is bad news for Nvidia, which has dominated the mobile discrete landscape for years. The need for low-end discrete GPUs in mainstream notebooks is disappearing fast.
Red Hat has announced that it bought storage system provider Inktank.
Inktank is the company behind Ceph, the cloud based objects and block storage software package used in a number of Openstack cloud configurations.
Ceph will continue to be marketed alongside Red Hat’s own GlusterFS in a deal worth $175m, which the company does not believe will adversely affect its financial forecasts for the year.
In a statement, Brian Stevens, EVP and CTO of Red Hat said, “We’re thrilled to welcome Inktank to the Red Hat family. They have built an incredibly vibrant community that will continue to be nurtured as we work together to make open the de facto choice for software-defined storage. Inktank has done a brilliant job assembling a strong ecosystem around Ceph and we look forward to expanding on this success together.”
As part of the deal Ceph’s Monitoring and Diagnostics tool Calamari will also become open source, allowing users to add their own modules and functionality.
Inktank founder Sage Weil used his blog to assure users that the two storage systems will be treated with equal respect. “Red Hat intends to administer the Ceph trademark in a manner that protects the ecosystem as a whole and creates a level playing field where everyone is held to the same standards of use.”
Red Hat made the announcement fresh from Red Hat Summit in New York, where the company reaffirmed that it is the Linux distribution of choice at the CERN supercollider in Switzerland.
The Inktank deal is set to close later this month.
AMD has not had much luck in tablets, that’s no secret and Intel hasn’t exactly done a great job, either. For years mass-market tablets were powered solely by ARM-based chips, but last year Intel upped the ante with Bay Trail-T, arguably the first truly competitive x86 SoC in the tablet space.
This year it’s AMD’s turn. Mullins offers a big improvement over Temash, it delivers a lot more performance and a few new features that make it a lot more attractive than its predecessor. Performance is not an issue, either.
However, having a good product simply isn’t enough.
Why Mullins could succeed where other APUs have failed?
Mullins offers vastly superior efficiency compared to Temash. AMD started using the SDP metric last year, in response to Intel’s decision to use SDP for some of its mobile/tablet parts. However, AMD still uses TDP, too.
Mullins parts feature an SDP of 2.8W. AMD previously stated that Mullins would end up with an SDP of ~2W, roughly on a par with Bay Trail-T parts. The actual TDP is of course somewhat higher. Temash parts feature an SDP of 3W to 4W, but the TDP is about 8W. Mullins offers a huge improvement, with an SDP of 2.8W and TDP ranging from 3.95W to 4.5W. Thanks to STAMP and other efficiency tweaks, Mullins can deliver quite a bit more performance than Temash in the same thermal envelope, and then some.
For example, the Temash based A6-1450 packs four Jaguar CPU cores clocked at 1GHz and it can hit 1.4GHz on Turbo. The GPU is clocked at 300MHz, but it clock up to 400MHz on Turbo. However, the Mullins-based A10 Micro-6700T can hit a max CPU clock speed of 2.2GHz, while the GPU can reach 500MHz. This is not only much higher than what Temash was capable of, it is also higher than what we saw on mainstream Kabini APUs with a TDP of 15W, yet the A10 Micro-6700T is a 4.5W part.
‘Contra revenue’ is Intel’s biggest competitive advantage
AMD has been showing off its Discovery tablet platform for a while, but so far AMD-based tablets have been scarcer than hen’s teeth. Now that AMD finally has a truly competitive part that can take on Bay Trail-T, it would be logical to expect more design wins.
However, thanks to Intel’s ‘contra revenue’ scheme this won’t be easy. Intel insists it’s not doing anything wrong and it doesn’t like it when someone describes its tablet push as a massive subsidy programme. Ultimately, that’s what it really is. Just because Mullins could succeed doesn’t mean it actually will.
AMD is not thrilled by the prospect of more Intel subsidies and market development programmes. The company has been dealing with similar Intel shenanigans for almost two decades and it knows it cannot compete on a level playing field. AMD cannot afford to burn hundreds of millions of dollars per quarter to gain a few dozen tablet design wins. Therefore AMD is targeting a somewhat different market, mid-range $299 tablets. Intel is trying to grab everything from $99 to $299 with its tablet SoCs, while Haswell and Broadwell should take care of the higher end of the market.
Intel hopes to ship 40 million tablet parts this year. We don’t know what AMD has in mind, but it is probably not even close to 40 million. It will be tricky, but this time around AMD appears to have a truly competitive product. In addition, not even Intel can afford to keep spending $1bn per year on its tablet push, so we should see its contra revenue taper off moving forward.
Even so, it might give Intel a huge competitive advantage. Intel is on track to quadruple its tablet shipments this year. If it manages to double them next year it will end up with 80+ million units, which seems like a relatively conservative estimate at this point. Intel is stealing design wins from the likes of Mediatek, Rockchip, Nvidia and so on. AMD will have to steal them back from Intel, which sounds a bit more difficult, even with competitive products.
He is one of the drivers behind AMD’s transformation, with the ultimate goal of turning the chipmaker into a new organization that is not so heavily dependent on the PC market. John confirmed that the company is on the road to achieve a huge milestone in its transition plans, generating approximately 50 percent of its revenue from the non-PC market by the end of 2015.
The time for the talk could not been better, as the market reacted positively to AMD’s Q1 earnings and at press time the stock was at $4.14, up $0.45 or 12.06 percent which is a huge jump for a tech stock. Keep in mind that many tech stocks have been bearish over the last four weeks, with several massive selloffs, especially in software and internet companies.
AMD fighting back in CPU space
We covered numerous topics from desktops, notebooks and tablets strategy all the way to the server, semi-custom APUs and of course the graphics market.
John said that leadership in the graphics sector is critical in AMD’s strategy, none more so than in the PC space where AMD wants to use their performance APU’s to compete with Intel’s Core i3 and Core i5 processors in the lucrative mainstream market. This is what AMD wants to address with Kaveri and to some extent with Kabini APUs.
AMD has high hopes for its upcoming server parts where they just launched their first ARM 64-bit product for the dense server space, where AMD expects to be a leader. On the other side of the spectrum the frugal AM1 platform launched a few weeks ago and it is getting very positive reviews. The first Kaveri parts have been on sale for a while, although we would like to see more desktop SKUs, not to mention mobile Kaveri APUs, including ULV variants.
Semi-custom APUs are blurring the line between AMD’s traditional product classes, but sales appear to be good, with more than 12 million Xbox One and PlayStation 4 consoles in the wild.
Phenomenal discrete GPU sales
Byrne is quietly confident when it comes to the GPU market, having just seen very strong sales in the performance and enthusiast high end segments of the market. The surge was driven by competitive products, great games and bundles, even with the cryptocurrency craze which was more or less a fluke for AMD.
The company remains committed to the GPU market, and expects to bring the successful R9 / R7 architecture further down into the mainstream price points in 2014, with similar traction. This means AMD will continue the fight against Nvidia in desktop and notebook GPU markets, while at the same time taking on Intel on desktop and notebook side with new APUs.
AMD thinks that the mix of great gaming performance, HSA, Mantle, Open CL, compute performance and some cool technologies like facial recognition can boost its position in the GPU market. This is just one part of the magic potion that is really starting to work for AMD, but it’s good to know that when it comes to graphics and gaming, AMD will stay committed to these markets in 2014 and beyond.
Enthusiasts need not worry. Although the company is reinventing itself and pursuing non-PC revenue streams, AMD will still be there to cater to their needs.
AMD posted some rather encouraging Q1 numbers last night, but slow PC sales are still hurting the company, along with the rest of the sector.
When asked about the PC market slump, AMD CEO Rory Read confirmed that the PC market was down sequentially 7 percent. This was a bit better than the company predicted, as the original forecast was that the PC market would decline 7 to 10 percent.
Rory pointed out that AMD can grow in the PC market as there is a lot of ground that can be taken from the competition. The commercial market did better than expected and Rory claims that AMD’s diversification strategy is taking off. AMD is trying to win market share in desktop and commercial segments, hence AMD sees an opportunity to grown PC revenue in the coming quarters. Rory also expects that tablets will continue to cannibalize the PC market. This is not going to change soon.
Kaveri and Kabini will definitely help this effort as both are solid parts priced quite aggressively. Kabini is also available in AMD’s new AM1 platform and we believe it is an interesting concept with plenty of mass market potential. Desktop and Notebook ASPs are flat which is something that the financial community really appreciated. It would not be so unusual that average selling prices were down since the global PC market was down.
Kaveri did well in the desktop high-end market in Q1 2014 and there will be some interesting announcements in the mobile market in Q2 2014 and beyond.
GlobalFoundries should be rolling out 20nm chips later this year and we hope that some AMD 20nm products might actually launch this year. The foundry failed to conquer the world with its 28nm process, but after some delays it got sorted out the problems and managed to ship some high-volume parts based on this process.
GlobalFoundries is manufacturing AMD’s new Kaveri APUs, while TSMC is making the Jaguar-based 28nm parts. We are not sure who is making the new server parts such as Seattle or Berlin, both 28nm designs. It is expected that GlobalFoundries should commence volume production of some 20nm parts later this year and the company has big plans for a faster transition to 14nm.
GlobalFoundries cozying up to Samsung
It is no secret that Intel leads the way in new process transitions and that Intel plans to ship 14nm parts at the time TSMC and GlobalFoundries are struggling to ship their first 20nm parts.
GlobalFoundries has now announced that it will start a strategic collaboration with none other than Samsung for its 14nm transition. It is easy to see that these two big players need each other in order to fight against bigger competitors like Intel and TSMC. GlobalFoundries and Samsung don’t have much overlap, either.
This joint venture will result in faster time-to-market for 14nm FinFET-based products. We see at least two advantages. According to Ana Hunter, Vice President of Product Management at GlobalFoundries, the process design kits are available today and the the foundry should be ready to manufacture 14nm FinFET products by the end of 2014. This sounds a bit optimistic, as we heard these bold announcements before, especially as both companies didn’t really start shipping 20nm parts yet, at least not in high volume high performance parts. It should be noted that Samsung joined the 28nm club quite late and shipped its first 28nm SoC just a year ago, in the Galaxy S4.
Sawn Han, Vice president of foundry marketing at Samsung Electronics, calls this partnership a ‘game changer’ as it will enable 14nm production by a total of four foundries in the world, three from Samsung and one from GlobalFoundries. Samsung will offer 14nm FinFET from S2 Fab in Austin Texas, S3 Fab in Hwa Seong in South Korea and S1 Fab in Gi Heung South Korea. GlobalFoundries is preparing its Fab 8 in Saratoga, New York State, for the 14nm push.
14nm FinFET crucial for next-gen SoC designs
The companies say 14nm FinFET technology features a smaller contact gate pitch for higher logic packing density and smaller SRAM bitcells to meet the increasing demand for memory content in advanced SoCs, while still leveraging the proven interconnect scheme from 20nm to offer the benefits of FinFET technology with reduced risk and the fastest time-to-market.
The 14nm LPE should deliver 20 percent more performance compared to 20nm parts and the power required should sink 35 percent versus 20nm LPE parts. Compared to 20nm parts it will save 15 percent of die space as well making it possible to cram more components into the same die size.
We have yet to see the first mobile 20nm parts in actual products. Qualcomm announced its first Snapdragons based on the new process a few weeks ago, but they won’t be ready for months. You can expect that a SoC manufactured on 14nm could end up 40 to 50 percent faster than its 28nm predecessor and that the power requirement could go down by 50 to 70 percent at best.
The total market for mobility, wireless and computer network storage market is expected to hit around $20 billion by 2017. Of course, everyone wants the piece of that action. The joint venture will offer both 14nm LPE (Low Power Enhanced) and 14nm LPP (Laser-Produced Plasma) process.
All we need now are the design wins from high-volume customers and if we were to bet we would place our money on Samsung, namely its Exynos processors. We would be positively surprised to see 14nm SoC in mobile phones and tablets in 2015, but it is a possibility. Keep in mind that we are still waiting to see the first 20nm SoCs and GPUs in action.
Canonical has announced its latest milestone server release, Ubuntu 14.04 LTS.
The company, which is better known for its open source Ubuntu Linux desktop operating system, has been supplying a server flavor of Ubuntu since 2006 that is being used by Netflix and Snapchat.
Ubuntu 14.04 Long Term Support (LTS) claims to be the most interoperable Openstack implementation, designed to run across multiple environments using Icehouse, the latest iteration of Openstack.
Canonical product manager Mark Baker told The INQUIRER, “The days of denying Ubuntu are over, and the cloud is where we can make a difference.”
Although Canonical regular issues incremental releases of Ubuntu, LTS releases such as this one represent landmarks for the operating system, which only come about ever two years. LTS releases are also supported for a full five years.
New in this Ubuntu 14.04 LTS release are Juju and Maas orchestration and automation tools and support for hyperscale ARM 64-bit computing such as the server setup recently announced by AMD.
Baker continued, “We’re not an enterprise vendor in the traditional sense. We’ve got a pretty good idea of how to do it by now. Openstack is gaining a more formal status as enterprise evolves to adopt cloud based solutions, and we are making a commitment to support it.
“Openstack Iceberg is also considered LTS and as such will be supported for five years.”
Scalability is another key factor. Baker said, “We look at performance. For the majority of our customers it’s about efficiency – how rapidly we can scale up and scale in, and that’s something Ubuntu does incredibly well.”
Ubuntu 14.04 LTS will be available to download from Thursday.
The Red Hat Summit kicked off in San Francisco on Tuesday, and continued today with a raft of announcements.
Red Hat launched a new fork of Red Hat Enterprise Linux (RHEL) with the title “Atomic Host”. The new version is stripped down to enable lightweight deployment of software containers. Although the mainline edition also support software containers, this lightweight version improves portability.
This is part of a wider Red Hat initiative, Project Atomic, which also sees virtualisation platform Docker updated as part of the ongoing partnership between the two organisations.
Red Hat also announced a release candidate (RC) for Red Hat Enterprise Linux 7. The beta version has already been downloaded 10,000 times. The Atomic Host fork is included in the RC.
Topping all that is the news that Red Hat’s latest stable release, RHEL 6.5 has been deployed at the Organisation for European Nuclear Research – better known as CERN.
The European laboratory, which houses the Large Hadron Collider (LHC) and was birthplace of the World Wide Web has rolled out the latest versions of Red Hat Enterprise Linux, Red Hat Enterprise Virtualisation and Red Hat Technical Account Management. Although Red Hat has a long history with CERN, this has been a major rollout for the facility.
The logging server of the LHC is one of the areas covered by the rollout, as are the financial and human resources databases.
The infrastructure comprises a series of dual socket servers, virtualised on Dell Poweredge M610 servers with up to 256GB RAM per server and full redundancy to prevent the loss of mission critical data.
Niko Neufeld, deputy project leader at the Large Hadron Collider, said, “Our LHCb experiment requires a powerful, very reliable and highly available IT environment for controlling and monitoring our 70 million CHF detectors. Red Hat Enterprise Virtualization is at the core of our virtualized infrastructure and complies with our stringent requirements.”
Other news from the conference includes the launch of Openshift Marketplace, allowing customers to try solutions for cloud applications, and the release of Red Hat Jboss Fuse 6.1 and Red Hat Jboss A-MQ 6.1, which are standards based integration and messaging products designed to manage everything from cloud computing to the Internet of Things.
Based on the firm’s Kabini system on chip (SoC), the APU is named the “AM1 Platform”, combining most system functions into one chip, with the motherboard and APU together costing around between $39 and $59.
Launched at the beginning of March and released today in North America, AMD’s AM1 Platform is aimed at markets where entry-level PCs are competing against other low-cost devices.
“We’re seeing that the market for these lower-cost PCs is increasing,” said AMD desktop product marketing manager Adam Kozak. “We’re also seeing other devices out there trying to fill that gap, but there’s really a big difference between what these devices can do versus what a Windows PC can do.”
The AM1 Platform combines an Athlon or Sempron processor with a motherboard based on the FS1b upgradable socket design. These motherboards have no chipset, as all functions are integrated into the APU, and only require additional memory modules to make a working system.
The AM1 SoC has up to four Jaguar CPU cores and an AMD Graphics Core Next (GCN) GPU, an on-chip memory controller supporting up to 16GB of DDR3-1600 RAM, plus all the typical system input and output functions, including SATA ports for storage, USB 2.0 and USB 3.0 ports, as well as VGA and HDMI graphics outputs.
AMD’s Jaguar core is best known for powering both Microsoft’s Xbox One and Sony’s Playstation 4 (PS4) games consoles. The AM1 Platform supports Windows XP, Windows 7 and Windows 8.1 in 32-bit or 64-bit architectures.
AMD said that it is going after Intel’s Bay Trail with the AM1 Platform, and expects to see it in small form factor desktop PCs such as netbooks and media-streaming boxes.
“We see it being used for basic computing, some light productivity and basic gaming, and really going after the Windows 8.1 environment with its four cores, which we’ll be able to offer for less,” Kozak added.
AMD benchmarked the AM1 Platform against an Intel Pentium J2850 with PC Mark 8 v2 and claimed it produced double the performance of the Intel processor. See the table below.
The FS1b upgradable socket means that users will be able to upgrade the system at a later date, while in Bay Trail and other low-cost platforms the processor is mounted directly to the motherboard.
AMD lifted the lid on its Kabini APU for tablets and mainstream laptops last May. AMD’s A series branded Kabini chips are quad-core processors, with the 15W A4-5000 and 25W A6-5200 clocked at 1.5GHz and 2GHz, respectively.
It appears that AMD’s professional graphics push is finally starting to pay off.
AMD’s graphics business is chugging along nicely, thanks to the success of Hawaii-based high-end cards, solid sales of rebranded mainstream cards, plenty of positive Mantle buzz and of course the cryptocurrency mining craze, which is winding down.
However, AMD traditionally lags behind Nvidia in two particular market segments – mobile graphics and professional graphics. Nvidia still has a comfortable lead in both segments and its position in mobile is as strong as ever, as it scored the vast majority of Haswell design wins in 2013. However, AMD is fighting back in the professional market and it is slowly gaining ground.
Mac Pro buckets boost FirePro sales
Last year AMD told us at the sidelines of its Hawaii launch event that is has high hopes for its professional GPU line-up moving forward.
This was not exactly news. At the time it was clear that AMD GPUs would end up in Cupertino’s latest Mac Pro series. The question was how much AMD stands to gain, both in terms of market share and revenue.
Although we are not fans of Apple’s marketing hype and hysteria associated with its consumerish fanboys, we have to admit that we have a soft spot new Mac Pro buckets. The bucket form factor is truly innovative and as usual the Mac Pro has the brains to match its looks. Basically it’s Apple going back to its roots.
Late last year it was reported that AMD would boost its market share in the professional segment to 30 percent this year, up from about 20 percent last year. For years Nvidia outsold AMD by a ratio of four to one in the professional space. The green team still has a huge lead, but AMD appears to be closing the gap.
It is hard to overstate the effect of professional graphics on Nvidia’s bottom line. The highly successful Quadro series always was and still is Nvidia’s cash cow. AMD is fighting back with competitive pricing and good hardware. In addition, the first Hawaii-based professional cards are rolling out as we speak. AMD’s new FirePro W9100, its first professional product based on Hawaii silicon, was announced a couple of weeks ago.
Can AMD keep it up?
2014 will be a good year for AMD’s professional graphics business, but it still remains to be seen whether the winning streak will continue. Apple does not care about loyalty, it’s not exactly a monogamous hardware partner. Apple has a habit of shifting between Nvidia and AMD graphics in the consumer space, so we would not rule out Nvidia in the long run. It might be back in future Mac Pro designs, but AMD has a few things working in its favour.
One of them is Adobe’s love of Open CL, which makes AMD’s professional offerings a bit more popular than Nvidia products in some circles. Adobe CC loves Open CL and AMD has been collaborating with Adobe for years to improve it. Support now extends to SpeedGrade CC, After Effects CC, Premiere, Adobe Media Encoder CC and other Adobe products.
Pricing is another important factor, as AMD has a tradition of undercutting Nvidia in the professional segment. When you happen to control 20 percent of the market in a duopoly, competitive pricing is a must.
Also, changing vendors in the professional arena is a bit trickier than swapping out a consumer graphics card or mobile GPU in a Macbook. This is perhaps AMD’s biggest advantage at the moment. Maintaining such design wins is quite a bit easier than winning them. AMD learned this lesson the hard way. Nvidia did not have to, at least not yet.
According to Seeking Alpha, demand for Mac Pro buckets is “crazy-high” and delivery times range from five to six weeks. Seeking Alpha goes on to conclude that AMD could make about $800,000,000 off a two-year Mac Pro design win, provided Apple sells 500,000 units over the next two years. At the moment it appears that Apple should have no trouble shipping half a million units, and then some.
If AMD manages to hold onto the Mac Pro deal, it stands to make a pretty penny over the next couple of years. However, if it also manages to seize more design wins in Apple consumer products, namely iMacs and Macbooks, AMD could make a small fortune on Cupertino deals alone.
Bear in mind that AMD’s revenue last year was $5.3 billion, so $800 million over the course of two years is a huge deal – even without consumer products in iMacs and Macbooks.