Something EPYC is missing from mainstream servers

HPE Cloudline CL3150 server

The major server vendors have all now officially announced a refresh of their portfolios, most timing this to coincide with the official launch of Intel’s Xeon Scalable processors this week.

HPE detailed its Gen 10 wave of servers last month, but Dell EMC, Lenovo and Cisco all saved much of the detail until Intel’s official announcement regarding its new server processors, based on the Skylake architecture.

This is no real surprise, as we’ve got used to the vendors tying their refresh cycles to the availability of new processor platforms.

But wait, where are the systems based on AMD’s EPYC processor platform, a chip designed to take on Intel’s Xeon processors? At the EPYC launch in Austin, Texas last month, AMD disclosed a number of vendors supporting their new chips, which included Dell EMC, HPE and Lenovo.

I asked Lenovo where its AMD-based systems are, and the firm was somewhat less than forthcoming. One product manager stated that he had no knowledge of any EPYC models, while the official line from the firm’s press office is that their servers are Intel-based.

So, I checked back to AMD’s announcement of the EPYC server chips, and sure enough, there is an endorsement from Lenovo:

“The AMD EPYC processors present unique opportunities for our customers to lower Total Cost of Ownership via an unprecedented balance of cores, memory bandwidth, and I/O. We are excited to collaborate with AMD and several global Hyperscale customers to develop and deploy single socket and dual socket EPYC-based servers,” said Paul Ju, vice president and general manager, Lenovo Global Hyperscale Business.

It would seem that if Lenovo is using AMD EPYC chips in systems, these are being supplied by a different business unit that targets the hyperscale market, while its enterprise servers are entirely Intel-based.

Lenovo is not the only vendor taking this path; HPE’s sole EPYC server disclosed so far is the Cloudline CL3150 (pictured), an ultra-dense rack-mount system targeting software-designed storage clusters, again for the hyperscale and service provider markets.

Many of the other brands lined up by AMD are well-known white-box vendors such as Supermicro, Tyan, Inventec and Wistron, which also sell into the hyperscale market.

Meanwhile, Dell EMC gave a supporting presentation at AMD’s event in Austin, yet its 14G announcement makes no mention of EPYC-based systems, nor are any listed on its site.

The situation thus seems to be that the Tier 1 vendors are all steering away from offering EPYC-based systems to compete directly with Intel’s Xeon platform, except in the hyperscale arena, where factors such as up-front cost and performance per watt override other considerations.

Is this because the big vendors not want to risk the wrath of Intel by putting up Xeon and EPYC systems against each other, or is it that their analysis of AMD’s chips has convinced all of them that it is ideally suited to a different role than that of Intel’s?

The reality is that it could be either reason, or both. Or to put it another way, vendors such as HPE enthusiastically backed AMD’s earlier Opteron server chips, only to see the company fail to keep up with Intel’s rapid pace of development, and so the sizable fraction of the server market that AMD managed to grab for itself slowly declined and customers drifted back to Intel-based systems as their servers reached end of life and were replaced.

Lenovo, HPE and Dell EMC are likely taking the approach that they won’t rush to offer AMD-based enterprise servers immediately, and are waiting to see if there is customer demand for them. With server shipments to enterprises reportedly declining because of the growth in cloud services, this may never materialise.

Then again, perhaps AMD was wise to develop a chip that is ideally suited to high-density servers for the service provider market: this is the area that is showing the strongest growth at the moment, after all. With up to 32 cores, 8 memory channels and 128 PCIe lanes, AMD can claim with some justification that a single-socket EPYC server can do the duty of a traditional two-socket system, at much lower cost.

UPDATE: I asked Dell for details of its EPYC-based servers, and this is the response I received:

Yesterday Dell EMC launched the first wave of its 14th Generation of PowerEdge servers aligned to Intel processors, PowerEdge servers based on the AMD EPYC processors will be available in the second half of 2017.

AMD Ryzen Pro to bring workstation-class performance to desktops

AND Ryzen Pro graphic

AMD is aiming at the enterprise desktop market with an upcoming Ryzen Pro processor family based on its Zen core design, taking on Intel’s Core chips by offering more cores and greater security through encrypted memory support.

With Ryzen Pro, AMD is aiming at the corporate market, in contrast to its consumer-focused Ryzen chips launched earlier this year. The firm claimed that this is an important market for it, with over 350 major corporate customers, and said that it expects to double the number of enterprise-class products using its chips by the end of the year.

“We have a lot of momentum in the commercial space of the market, and now with the technology of Ryzen Pro, we believe we can take that to the next level,” said John Hampton, director of commercial business development for AMD.

Many people will be surprised to learn that desktop PCs are still around. The reality is that while laptops may have largely displaced desktop systems, they still have value for businesses in office environments and especially for smaller businesses.

Ryzen Pro is clearly being positioned against Intel’s Core processors, with AMD dividing up its portfolio into the Ryzen 7 Pro, Ryzen 5 Pro and Ryzen 3 Pro segments against the Core i7, Core i5 and Core i3, with clock speeds of up to 3.7GHz with frequency boost.

AMD Ryzen Pro vs Intel Core

As with the EPYC server processor that AMD recently launched, the firm is trying to compete by offering more cores in each segment. Thus Ryzen 7 Pro has 8 cores against the 4 of the Core i7, the Ryzen 5 Pro has up to 6 cores against the 4 of the Core i5, and Ryzen 3 Pro has 4 cores against the 2 of Core i3.

“Ryzen Pro for us will be a no-compromise solution versus Intel vPro,” said Hampton, claiming that AMD’s new platform has “more cores and more threads at every price point”.

AMD claimed that benchmark figures show that a Ryzen 7 Pro delivers up to 62 percent more multi-threaded performance than Core i7.

“We’re designing for the future. We believe that more and more commercial workloads will demand more core, more threads, more power, and so on,” Hampton added.

The target markets for each segment are power users running content creation and scientific applications for Ryzen 7 Pro, while Ryzen 5 Pro is the mainstream platform for advanced productivity, and Ryzen 3 Pro is for office productivity and entry-level tasks.

AMD Ryzen Pro SKUs

While the firm claims ‘workstation class’ performance for Ryzen Pro, the chips are aimed at business desktops and so not intended to compete with Intel’s Xeon chips in the actual workstation market. The upcoming Ryzen “Threadripper” may address this market. It will be comprised of two silicon dies rather than the single one of the Ryzen Pro, making for a processor with twice the number of memory channels and up to 16 CPU cores.

And despite AMD touting Ryzen Pro for content creation and scientific applications, it is a purely CPU-based chip, not an APU that integrates GPU cores like many of AMD’s existing processor lines. It is expected that the Ryzen Mobile chips for laptops will be APUs and combine Zen cores with AMD’s Vega GPU core. Meanwhile, Ryzen Pro systems will have to rely on discrete GPUs, either on the motherboard or in a PCIe slot.

But AMD is not merely counting on performance to differentiate from Intel’s chips. Like the EPYC line, Ryzen Pro features a built-in AMD Secure Processor that provides a hardware-based root of trust against malware threats, plus the transparent memory encryption technology that can be used to secure areas of memory used to store sensitive data, or the entire memory of the system.

As with EPYC, there is a small performance overhead when memory encryption is enabled, but this is in the order of a one or two per cent reduction in performance, AMD claims.

The AMD Secure Processor provides a secure boot feature by validating the system firmware to ensure it has not been tampered with before it will allow the processor cores to come out of their reset state and begin the normal boot process.

This is designed to prevent malware attacks that come in under the level of the operating system and attack the firmware, making them hard to detect and remediate.

AMD Ryzen Pro graphic

Ryzen Pro also features AMD’s SenseMI technologies, a set of adaptive features to a tweak the processor for optimum performance, as seen in the consumer Ryzen chips. These include Precision Boost to adjust the clock speed in 25MHz increments, and Pure Power, which is the flip side of this and optimises the power consumption for the workload.

With Ryzen Pro aimed at business buyers, other important features are reliability, stability, and manageability.

On the reliability front, AMD claims that the Pro products are sourced from silicon wafers with the highest yields to ensure fewer defective parts. Meanwhile, AMD offers 18 months stability for drivers and other low-level software, plus 24 month availability of processor SKUs for vendors.

On the manageability side, AMD supports the DASH standard developed by the DMTF (Distributed Management Task Force). The advantage of DASH is that it is cross-platform, and will work with both Intel and AMD platforms.

Availability of Ryzen Pro is set for the official launch on 29 August, when AMD will disclose the vendors which are offering systems based on the new chips, and information of the pricing versus Intel’s Core chips.


HPE Gen 10 servers offer built-in security and custom Intel Xeon SKUs


HPE Gen 10 servers

HPE is aiming to widen its appeal to enterprise customers with its Gen 10 wave of servers based on Intel’s latest Xeon processors, adding security into the hardware to combat security threats, more predictable performance in turbo mode, and new payment options for mid-size firms.

Announced at HPE Discover in Las Vegas, the 10th generation of HPE’s x86 server systems is based on Intel’s Xeon Processor Scalable family, announced by the chipmaker last month. These will be rolled out across the breadth of its portfolio, according to HPE, including the ProLiant, Synergy, Apollo, Superdome and Cloudline families.

HPE said that the new systems have been designed to address three key areas of concern for organisations; agility, security and economic control.

“When we were trying to think a few years back about building our next generation of compute, we thought, how can we have the best possible agility on-prem? How can we have security that is unmatched in the industry? What can we do so that customers can decide how to manage best control their economies of scale, the way to spend their money, and how they scale in the future?” said Alain Andreoli, senior vice president for HPE’s Data Center Infrastructure Group.

HPE iLO management chipHPE’s new “silicon root of trust” security capability is built into a custom chip fitted to the system board and works with the integrated lights-out (iLO) management controller in each HPE server. This enables it to protect the integrity of the system firmware right from its point of manufacture all the way through to operational deployment, what HPE dubs the Secure Compute Lifecycle.

Uniquely, according to HPE chief technology officer Mark Potter, the silicon root of trust can also recover the system back to a previous authenticated state, rather than just preventing it from booting up if it detects that the firmware does not match the correct digital fingerprint.

“We feel it is very important to keep the system up and running, so we built an automatic recovery capability right into our secure system,” he said.

The new iLO also includes the ability to automate maintenance upgrades for servers with support for scheduling and rollback, simplifying operations management in data centres where HPE systems are being operated at any kind of scale.

Addressing the agility question is the latest version 3.1 of HPE’s OneView unified infrastructure management tool, which provides HPE systems with new composable storage capabilities and support to help customers to deploy key applications and infrastructure software including the Mesosphere containers platform.

“We have made OneView now template driven so that these key applications can be moved and migrated and run automatically, and we have also integrated Mesosphere as one of the templates in OneView,” said Andreoli.

HPE also disclosed it is using custom SKUs of Intel’s Skylake-based Xeon Processor Scalable family, in order to deliver greater performance for customers when using turbo mode. This is implemented as three features collectively called Intelligent System Tuning, to be available on selected ProLiant Gen 10 servers.

“We’ve worked very closely on Intel on how do we take better advantage of the ability to run turbo mode, which is based on workload and environment, how do we speed the processor up, and so we uniquely have the ability to run in turbo mode with these special SKUs,” said Potter.

Intel Xeon Processor ScalableCore Boosting offers the ability to dynamically increase the turbo frequency, while Jitter Smoothing makes for more consistent performance when switching core speeds. Workload Matching provides templates to optimise the processor for specific workloads.

“One of the drawbacks of turbo mode is this notion that you’re going to go into turbo mode and change frequency, and every time you make these big changes in clock frequency because of the dynamic of the workload, you can create delays or latency. So we have a patent on how to smooth this out and keep a consistent deterministic behaviour for the application, but run it at the fastest level of performance,” Potter explained.

This capability could be important for industry sectors such as financial trading, where predictable deterministic performance is important.

“Every transaction has to behave the same, so jitter smoothing is key for that segment,” he added.

The Gen 10 systems also feature Scalable Persistent Memory, an extension of HPE’s existing persistent memory technology that combines DRAM with flash chips to make a non-volatile DIMM (NVDIMM). HPE gave away few details other than saying this will now be available at “terabyte scale” for all ProLiant systems.

Meanwhile, HPE is offering new purchasing options that bring the benefits of flexible purchasing to a wider range of customers. Flexible Capacity, which has been available for several years, offers enterprises a pay-as-you-go option based on the compute capacity they use.

The new Capacity Care Services are available for Gen 10 systems and aimed at mid-size customers, offering them usage tracking reports with a quarterly consultation to assess and manage their compute capacity as necessary. This will help customers save money by eliminating overprovisioning of capacity, HPE said.


Some thoughts on Samsung’s Galaxy S8 and its DeX docking cradle

Samsung’s newest Galaxy smartphone range was launched to great fanfare last week, but came with an unexpected surprise. No, not an exploding battery, but a docking cradle that turns it into a desktop computer.

The latest line-up of Samsung’s flagship handset comprises two models, the Galaxy S8 and Galaxy S8+, that are everything you would expect from the leading smartphone vendor. They offer a choice of display size – 5.8in and 6.2in – multi-core processors, support for the latest high-speed networks, and the Android 7.0 Nougat operating system.

Owners of the new devices can also pick up an unusual optional extra for their Galaxy S8 or Galaxy S8+, in the shape of the DeX. This looks rather like a fancy black ashtray, but turns out to be a docking cradle that comes with an Ethernet port, HDMI video output and full-size USB 2.0 and USB Type C ports.

As well as charging the phone, the DeX lets you connect up to a desktop monitor, keyboard and mouse, and presumably also to a LAN using the Ethernet port, effectively turning your phone into a desktop computer. When plugged in like this, the user interface switches to a desktop-style user interface to make use of the larger display.

Samsung Galaxy S8 and DeX desktop cradle

This is an interesting concept, and one that has been mooted before. The comparison many in the tech industry are making is to the Continuum feature of Windows 10, which is intended to offer a similar large-screen experience for users of phones running Windows 10, when connected to a monitor. But the Windows Phone platform is essentially going nowhere, and many mobile watchers have pretty much written it off as dead.

However, it is also a feature that Ubuntu firm Canonical touted for its proposed Edge “superphone” back in 2013. This was to have been a high-spec smartphone running Ubuntu Linux, which would switch between a mobile user interface and a full Linux desktop shell, depending on whether it was docked or not. Sadly, the Edge never met its crowdfunding target on Indiegogo, and so did not go into production.

Samsung’s implementation of the concept could potentially attract interest from professional users because of the fact that key apps such as Microsoft’s Office Mobile suite are available for Android, and someone whose role mostly involves document work in Microsoft Office could conceivably run this on a Galaxy S8 or Galaxy S8+ connected to a monitor, in place of a desktop computer.

Another potential use case is not as a PC per se, but as a thin client for accessing virtual desktop (VDI) sessions. Here, the user would have access to a full-blown Windows desktop running full Windows apps, with the docked Galaxy S8 or Galaxy S8+ serving as a terminal. Both Citrix and VMware offer Android versions of their VDI client software, which Dell Wyse used to deliver a pocket-sized thin client in the shape of the Cloud Connect device several years back.

Citrix also showed off its receiver client running on a smartphone connected to a monitor screen as far back as 2010, although this required a handset with an HDMI output to function, and few had this.

Although Samsung seems to have made a splash with DeX at the Galaxy S8 launch, is there actually much call for this usage model? While the ability to turn your smartphone into a desktop client system is a neat trick, who would actually use such a system?

Most mobile professionals currently use a laptop, and plug that into a desktop display when they are in the office. The laptop has a decent sized screen that you can take out on the road with you, while with the Samsung DeX, you leave your big screen and keyboard behind in the office when you go roaming.

Samsung DeX desktop cradle

While it is conceivable that there may be some mobile worker roles in which a big screen is needed only in the office, and a smartphone is sufficient while out on the road, these would seem to be a bit of a niche. Few people would suggest that a smartphone with its small screen and lack of a physical keyboard would be suitable for intensive work – they tend to be used for checking emails or looking up information.

Then there is the fact that the suggested price of the DeX docking cradle – $149.99 in the US – is not much lower than many existing thin client terminals from established vendors in this market such as HP and Dell Wyse, so anyone choosing to equip their staff with a Galaxy device and a DeX to use as a VDI client would not really be saving much.

In addition, the lifecycle of a smartphone tends to be much shorter than that of a corporate device like a thin client terminal. Users tend to upgrade their phones every couple of years, whereas thin clients are often good for up to seven years of use.

Another potential pitfall is that you may invest in a bunch of Galaxy S8 or Galaxy S8+ handsets and DeX cradles, only to find that Samsung may no longer support DeX with any successor generation of Galaxy devices.

While Samsung is to be applauded for exploring a novel use of smartphones with the DeX hardware, it may find that the main market for this will be found among consumers or tech enthusiasts rather than business customers.

Windows Server on ARM: what does it mean?

Qualcomm Centriq ARM server

The demonstration of a version of Windows Server running on ARM-based servers came as a shock to many, especially as this is something that Microsoft has expressly ruled out in the past. But look closely and there is no suggestion as yet that this will lead to commercial availability of such products.

This first public demonstration of Windows Server running on ARM-based systems came at the Open Compute Project (OCP) Summit 2017 in Santa Clara. It was conducted by chipmaker Qualcomm, using its Centriq 2400 platform that boasts up to 48 cores based on ARM’s 64-bit ARMv8 architecture.

Microsoft is also working with at least one other chipmaker, Cavium, which trumpeted its own involvement in a statement saying it was “collaborating with Microsoft on evaluating and enabling a variety of cloud workloads running on Cavium’s flagship ThunderX2 ARMv8-A Data Center processor for the Microsoft Azure cloud platform”.

The key phrase here is “for the Microsoft Azure cloud platform”. This version of Windows Server, and the systems it is running on, seem to be part of an evaluation by Microsoft to see how well ARM-based servers can run some of its cloud services operated from its network of data centres.

ARM has been taking aim at the server market for at least the past five years, at least as far back as the launch of its 64-bit ARMv8 architecture. The proposition is that ARM cores are less complex and consume less energy than rival architectures, such as Intel’s x86 and IBM’s Power processors.

[For more on this, see my article on IDG Connect: No ARM in a bit of server market competition]

However, expert opinion has so far been that the economics of this would only really make sense for hyper-scale environments – typically meaning the large cloud service and internet companies such as Google, Facebook, AWS, and Microsoft, which operate tens of thousands or even millions of server nodes. These are the companies for whose requirements the OCP was started in the first place.

In a post on Microsoft’s Azure blog, Distinguished Engineer Leendert van Doorn confirmed that the ARM servers are currently for Microsoft’s internal use only:

“We have been working closely with multiple ARM server suppliers, including Qualcomm and Cavium, to optimize their silicon for our use. We have been running evaluations side by side with our production workloads and what we see is quite compelling.”

What this may mean is that Microsoft could be planning to migrate some of its cloud services over to ARM-based infrastructure at some point in the future. How worried should Intel be about this move?

The reality is that x86 systems are not going to go away, for the simple reason that the virtual machine workloads that Microsoft customers have hosted on Azure require an x86 server to run on: pretty much every enterprise in the world is run on x86 servers, and these customers expect any public cloud infrastructure-as-a-service (IaaS) to do the same for compatibility reasons.

Again, Microsoft confirms this in the blog:

“One of the biggest hurdles to enable ARM servers is the software. Rather than trying to port every one of our many software components, we looked at where ARM servers are applicable and where they provide value to us. We found that they provide the most value for our cloud services, specifically our internal cloud applications such as search and indexing, storage, databases, big data and machine learning.”

“To enable these cloud services, we’ve ported a version of Windows Server, for our internal use only, to run on ARM architecture. We have ported language runtime systems and middleware components, and we have ported and evaluated applications, often running these workloads side-by-side with production workloads.”

So, don’t expect to see ARM-based Windows Servers anywhere except in hyper-scale cloud data centres for now. Of course, where Microsoft leads, others may follow, but the huge installed base of Intel-based servers out there means that the average company is not going to be buying ARM servers anytime soon.

Raspberry Pi Zero gets wireless to celebrate the Pi’s fifth birthday

The Raspberry Pi is officially five years old today, and the folks behind the low-cost single-board computer are celebrating with the release of a new model, which brings wireless connectivity to the smallest form factor Pi model on the market.

Known as the Raspberry Pi Zero W, the new model keeps the same dimensions as the existing Raspberry Pi Zero design, but adds in the wireless functionality (802.11 b/g/n WiFi and Bluetooth 4.1) that was introduced with the Raspberry Pi 3 Model B a year ago.

Raspberry Pi Zero W

Also unveiled today is a new injection-moulded case designed to snugly fit either the new Pi Zero or the existing model, and which comes which a choice of three interchangeable lids; one with a cut-out for the camera module accessory (the ribbon cable for this is also included), another with a cut-out exposing the 40-pin expansion connector, and the third one with no cut-out.

Pi Zero W case and lids

Pricing for the new Raspberry Pi Zero W is expected to be £9.60 inc VAT in the UK, while the new Pi Zero case is expected to cost somewhere in the region of £5. Both are expected to be available from the usual Pi outlets such as The Pi Hut, Pimoroni and Adafruit.

With wireless capability, the Raspberry Pi Zero W fixes what was the only major drawback of the Pi Zero: no network connectivity. True, you could plug in some kind of USB network adapter, but this would be cumbersome, especially as the Pi Zero has only a micro-USB port and needs an adapter to take standard USB devices.

Pi Zero and Pi Zero W

Top: original Pi Zero
Bottom: Pi Zero W

However, the lack of wireless connectivity has not stopped the Pi Zero from being used for a variety of projects, including build-it-yourself digital camera, when combined with the camera module.

With the addition of built-in WiFi and Bluetooth, the Raspberry Pi Zero W looks certain to find its way into numerous new devices, especially projects aimed at the Internet of Things (IoT), given the device’s small size. The wireless support means the new Pi Zero can connect to other devices or to the internet via a WiFi connection.

In fact, given the small size of the device – 65mm long by 30mm wide – we were at a loss to work out where the antenna is on the Raspberry Pi Zero W. Locating the actual wireless chip is relatively easy, as it is a small silvery package, as seen on the Raspberry Pi 3.

However, the Raspberry Pi 3 also had a small surface-mounted antenna, of which there is no sign on the Pi Zero W. It turns out that the new model has an antenna actually built into the circuit board, located between the miniature HDMI and USB connectors (see image below).

Pi Zero W antenna

This piece of high-tech wizardry comes from a Swedish firm called ProAnt, which specialises in antenna design, and is alluded to by text on the rear of the new Pi that says “antenna technology licensed from ProAnt”.

Here is the hardware specifications for the new Raspberry Pi Zero W, as detailed by the folks at Raspberry Pi.

– 1GHz, Single-core CPU

– 512MB RAM

– Mini HDMI and USB On-The-Go ports

– Micro USB power

– HAT-compatible 40-pin header

– Composite video and reset headers

– CSI camera connector

– 802.11n wireless LAN

– Bluetooth 4.0

These specifications are essentially the same as for the original Pi Zero, save for the addition of wireless, meaning that the new model has a less powerful processor than the Pi 3, but still more powerful than the original Raspberry Pi launched five years ago.