The Linley Group has released a 6-page analysis of IBM POWER8 vs Intel Xeon
and it should have IBM laughing all the way to the bank. The analysis was done by Tom Halfhill and has an original publication date of December 29, 2014 although the report has only just started to be circulated.
Having disposed of its x86 business,
IBM has bet big to the tune of billions of dollars on the technology in its POWER processor line.
This edition of the Microprocessor Report suggests that IBM has got it right in being able to attack Intel. That said, careful reading of the report shows that this is not a slam dunk for IBM but for now, it does have a distinct advantage over Intel.
One of the key benefits of POWER8 over Intel Xeon is the number of threads per processor core. While Intel continues to deliver 2 threads per core, IBM is delivering a whopping 8 threads per core. For complex workloads such as data analytics which are very dependent on the number of threads, IBM clearly has the advantage.
It is not just the thread count that goes in IBM's favour. IBM has twice the amount of L2 Cache memory per core 512kb vs 256kb and delivers almost twice as much L3 Cache than Intel. Unlike Intel who doesn't support L4 Cache, IBM provides 64MB by using its own buffer chips. Processor frequency is also a win for IBM because they are able to run POWER at a much higher frequency but this comes at a cost. The power required to support much higher clock speeds can be expensive, something that Intel admitted when it stopped chasing clock speeds a few years back.
As systems scale up, IBM POWER8 delivers a significant benefit over Intel Xeon. While Intel is limited to 2 sockets for Coherent SMP, IBM is able to use 12 sockets. This is a major bonus for IBM where it has customers who want very systems capable of handling extremely complex workloads.
Perhaps the biggest surprise in this report is the list price of the processors. Out of the two different versions of processors from both companies that were compared, IBM POWER8 with 8 cores is less than 50% the cost of an Intel Xeon E5-2697v3 with 12 cores. Meanwhile the IBM POWER8 with 12 cores is around 70% the price of an Intel Xeon E5-2699v3 with 18 cores. In each case, the price includes the 4 buffer chips that deliver the L4 Cache.
While IBM will take as many of the positives out of this report as it can, and there are plenty to take, it must also look at how it addresses the criticism. The main criticism is around the power required to support the processors. The basic processor comparison shows Intel running at 145W Thermal Design Power (TDP) compared to the POWER8 at 190W. When you add in the buffer chips at 20W each, this adds up to another 80W required for a POWER8 system.
It is not just the power required to run the system but the ensuing impact on cooling and system design that is not in IBM's favour. Although there is no attempt in the report to estimate the cost of the additional power over any period of time or to calculate the power per workload it does say that customers should spend the savings on buying IBM on a bigger fan and offsetting the higher electric bills.
A more serious criticism and one that IBM is showing signs of solving is one of market penetration. Halfhill points to the development of the OpenPOWER Alliance and how well IBM is doing in pulling in new partners. In the last year IBM has persuaded more than 90 companies to join the OpenPOWER Alliance. These companies cover the whole gamut of system design from processor fabricators through to system design companies.
Despite this, there has been little sign of a third party server market growing up around POWER8. Halfhill rightly suggests that this could be IBM's Achilles heel. Intel has a well developed partner world building motherboards and servers based on its processors. This means that it can rely on those partners to deliver the sales of systems to customers while it concentrates on processor design. IBM, by comparison, is stuck building the market at the moment while still having to fund the future development of the next generation of POWER9 processors.
Working in IBM's favour, however, is Watson. IBM has deployed Watson into its SoftLayer cloud business and that has taken POWER8 systems with it. In addition to offering Watson on POWER8, SoftLayer is also offering POWER8 as a bare bones solution running any of the three main Linux distributions. This is already beginning to filter through in terms of sales and that takes us to Halfhill's final criticism of IBM's strategy.
Halfhill believes that customers will be put off porting their systems and IBM partners who are still selling x86 are likely to relegate POWER8 to niche rather than mainstream applications. This should change as IBM rolls out its education programme for partners. When POWER8 launched, IBM deliberately targeted little-endian Linux and now has RedHat, SUSE and Unbuntu running on the platform.
Customers currently running Linux are able to move straight to POWER8 with many applications not even requiring to be recompiled. This means that there is no reason to see POWER8 as a niche solution when it is clearly able to compete head-to-head with Intel in the Linux space.
What will please IBM's management most about this report is not the fact that POWER8 is able to thrash Intel Xeon but Halfhill's final paragraph:
Even if OpenPower systems never break out of this niche, their high performance can support the high margins that sustain a profitable business. And those margins should remain high as long as Intel keeps its prices and margins high. OpenPower can generate extra revenue that helps IBM defray the costs of designing new Power processors for its own systems and for OEMs. The cost of offering Power processors as merchant products is relatively small, and any incremental profit is good for IBM’s server business.
One of the reasons for IBM getting rid of x86 to concentrate on its POWER and mainframe strategy was to bring back higher margins for the company. Halfhill believes that POWER8 is able to do that and in a month where IBM was again forced on to the backfoot with less than stellar financial results, there is still a glimmer of light at the end of the tunnel.