By Joe Clabby, Clabby Analytics
In May of this year, Clabby Analytics published a report called “The Power Systems Renaissance” that described why we believe that IBM’s Power Systems have turned the corner after several quarters of declining revenue. In Q1, 2015, IBM Power Systems revenue increased by a percentage point over the same quarter in 2014. Revenue increased again in Q2 and Q3 earnings announcements are just days away.
The growth of Power Systems was led by strong sales in scale-out systems – and although IBM did not say this, I suspect Linux scale-out systems accounted for much of that growth. As I said in the above-mentioned report, IBM had been asleep at the wheel when part of the market started to migrate from Unix to Linux and was not properly prepared with the right systems architectures (low-end, scale-out, Linux-oriented Power Systems) or the right pricing/deployment models to compete with the dominant x86-based Linux servers. What the past two quarters have shown me is that IBM has gotten its act together – and Power Systems-based Linux servers are now starting to contribute more heavily to overall Power Systems revenue. (Note: IBM’s last earnings call also mentioned that high-end Power Systems are also experiencing a return to growth – and that the AIX Unix business is still solid).
A real indicator that IBM’s Power Systems organization now “gets it” when it comes to Linux is the company’s latest Power Systems new hardware, consumption and service support model announcement. A big focus for IBM’s Power System organization in 2015/2016 will be to better serve hybrid cloud environments and hyper scale data centers with new cloud and cluster servers, as well as with new deployment models.
On the hardware side, IBM just announced a new “LC” line of Power Systems servers. Note the “L” designator in each model – these are Linux-focused scale-out offerings. The LC line features three new configurations: 1) the IBM Power Systems S812LC (designed for entry and small Hadoop workloads); 2) the IBM Power Systems S822LC for commercial computing (which is particularly well suited for processing data in a cloud; and, 3) the IBM Power Systems S882LC for high performance computing.
As for deployment in cloud, big data and business critical scenarios, IBM is offering several models designed to fit specific business needs,including on-premises, hybrid cloud, custom and converged infrastructure deployments. As for cost, IBM is now providing a self-service mechanism for ordering Power Systems enabling customers to buy them on-line with simple pricing.
For custom configurations, customers can connect with IBM Business Partners to create a solution specific to their need. Service and support options with “ultra-flexible” terms are also available. Giving customers a lot more control over picking and choosing the configurations and pricing models that they want mimics x86 configuration/pricing models – and is a huge step in the right direction for the Power Systems group.
A Closer Look at the New Linux Server Configurations
As we have mentioned in several Clabby Analytics reports, we see Power Systems as particularly strong at performing computational and analytics tasks. With the ability to process significantly more threads per clock cycle than competing x86 servers, Power System users can get results (and business advantages) more quickly than those using x86. But processors are no longer the primary consideration when it comes to achieving fast performance. Designers have implemented new system busses, faster input/output channels, and, in some cases have introduced different types of processors in order to achieve accelerated results. The Power Systems organization has been particularly aggressive in building advanced, highly-integrated systems designs that take advantage of many of these advances.
As for the new servers, the IBM Power Systems S812LC has been tuned and optimized as a server for entry and small Hadoop and Spark workloads. It offers 16X the memory capacity of servers using Intel’s Xeon E3-based, and twice the performance of more modern 1PXeon E5 servers. This translates in some cases to 2.3X better performance per dollar spent for Spark workloads. As for density, 94% more Spark workloads can run in the same rack space as a competing Intel Xeon E5-2690 v3 system. The performance differential translates into 1.94 better performance per systems (when comparing a 10 core Power System S812LC to a 24 core HP DL380). This model is a 1-socket, 2U machine with up to 10 cores (up to 3.3 GHz), 1 TB of main memory, 115 GB/sec memory bandwidth, room for 14 LFF (HDD/SSD) drives with access to up to 84 TB of storage. It also has 4 PCIe slots, 2 of which are CAPI (specialized Flash to CPU) enabled. And there is a default of 3 years, 9X5 warrantee support and 100% customer replaceable parts.
IBM’s Power System S822LC is a more powerful 2-socket server, with up to 20 cores (up to 3.3 GHz), a TB of main memory, 230GB/second memory bandwidth (twice as fast as the S812LC model), 2 integrated NVIDIA GPUs for accelerated parallel processing, 5 PCIe slots (2 CAPI enabled, 1B add-in), and the same 3 year, 9X5 warrantee default as the single socket solution. The S822LC has been tuned for high performance computing with accelerators produced by the OpenPOWER Foundation for compute, fabric and storage acceleration. And the CAPI interface also accelerates the performance of other attached devices. In addition to tuning for high performance computing, the S822LC has also been tuned for commercial computing, particularly for data processing in the cloud – and is a flexible commercial computing server for managed service providers who want to build their own configurations. Of note, this commercial computing server offer 2.7X more transactions per second per core when using PostgreSQL as compared with a Xeon E5-2699 v3. Further, it offers 40% better price/performance for PostgreSQL vs. Xeon E5-2699 v3-based servers. And its pricing is very attractive for on-premises and in-cloud applications.
A Word about Open Standards and OpenPOWER
In October, 2014, we wrote a report the described the genesis and operation of the OpenPOWER Foundation launched by IBM, NVIDIA, Google and others. We would be remiss to overlook the contribution that open standards and the OpenPOWER foundation are making to the success of the new Power Systems lines. IBM is now delivering PowerVM NovaLink which enables direct OpenStack (an open cloud standard) connection to PowerVM hosts. And the company is also delivering PowerVC Dynamic Resource Optimizer to actively balance workloads and resources based on business policies. Further, the OpenPOWER ecosystem has evolved greatly since our last report, with over 100 Foundation members developing and delivering systems, storage, network and accelerator solutions to market – and enabling users of POWER processor-based systems to get insights faster and more rapid return-on-investment than ever before. Expect IBM’s open ecosystem to spur even more growth in Power Systems in the future – as well to drive even more innovative designs.
We guessed right at the beginning of this year when we reported that Power Systems had returned to profitability. Given instability in various parts of the world, we expect IBM’s Power Systems revenue road to have some bumps in the road ahead. But overall we are highly encouraged with the direction that the company has taken Power Systems over the past two years (continued emphasis on AIX Unix and much stronger emphasis on Linux on POWER). We are now seeing even more impressive POWER/Linux system designs emerging from IBM – and expect many more to come as IBM continues to optimize its scale-out Power Systems for specific markets. And we are especially encouraged that IBM Power Systems have entered the self-service age, allowing customers more flexibility in web-based configurations and pricing. We continue to like what we are seeing from IBM’s Power Systems group – and we’re even more excited to see what IBM has been able to do with its OpenPower partners in building a viable, vibrant, POWER-based open systems community.