Midrange Power7+ Servers: The More Oomph You Want, The More It Costs
April 1, 2013 Timothy Prickett Morgan
In last week’s issue of The Four Hundred, I walked you though the pricing for various sized configurations of the new Power 750+ and Power 760+ midrange servers using IBM‘s dual-chip module (DCM) variant of the Power7+ processor. That just looked at the system hardware costs. This week, as I have done in the past, I want to walk you through how the raw computing, operating system, and Software Maintenance costs compare across each processor SKU and how those costs stack up with prior Power 750 machines.
As has been the case since there was an AS/400, customers in the belly of the product line pay a considerably higher cost per unit of performance, and by a lot more than you might expect based on the configured system price. Decades ago, hardware was the expensive bit and while the OS/400 systems software was by no means inexpensive, considering it had a relational database built in, the software price seemed relatively reasonable. Of course, hardware prices have been driven down by Moore’s Law across processors, memory, and storage, and now hardware is the least of anyone’s worries. This is true of all systems at this point, by the way, not just the AS/400 and its progeny. And yet, here’s the funny bit, systems still account for well over $50 billion a year in revenues worldwide, twice the market for networking and external disk/flash storage individually.
There are lots of ways to dice and slice the costs of a Power Systems machine running IBM i, but the data presented in this story is, like the one I did for the entry machines a few weeks ago, designed to isolate the raw computing costs from chassis, memory, storage, and peripheral costs. The idea is to take each processor feature card, activate all of its cores, load up IBM i on each core, and see what that costs and then divide it by the Commercial Performance Workload (CPW) relative performance rating. This is, in effect, the price of raw computing on each processor feature. If you do that for the Power 750+ and the Power 760+, here are how the costs per CPW stack up for the first to processor cards in each chassis:
I put the configurations of the Power 740+ machines in the chart just to have a frame of reference.
The full set of data used to generate this chart can be found in this monster table, which I updated for the smaller Power7 and Power7+ machines to take out the Software Maintenance price hike that IBM announced on February 5 and subsequently retracted on March 11.
As you can see, the bang for the buck on the Power 740+ and Power 750+ is more or less the same, with a slight premium coming for the Power 750+ and the expansion capability in terms of I/O and memory absolutely warranting some sort of premium. Both machines are in the same P20 software tier, so software costs per core are the same and the performance per core is also roughly on par. Moving up to the Power 760+, which is essentially the same 5U chassis but with two six-core chips per socket instead of two four-core chips per socket as in the Power 750+, you jump up to that P30 software tier. And as you can see, that makes the cost per CPW go up by between 30 and 50 percent, depending on which machine you want to compare.
The share of the cost of the raw processor feature cards that is represented by the IBM i software licensed in each core is 80 percent or higher, as you can see here:
In effect, you can think of any Power Systems processor card as a base smartphone that the vendor is willing to give to you on the relative cheap because they want you on the service contract with that data plan. In this case, the Software Maintenance contract for one year costs the same or more than the hardware and the data plan represents 80 to 85 percent of the total cost of ownership. As I showed you a few weeks ago, on smaller Power 710+ and Power 720+ machines, the IBM i software on a fully activated processor feature card represented about 60 to 65 percent of the total cost of the card, with SWMA being between 25 and 30 percent. The Power 730+ and Power 740+ machines are virtually indistinguishable from the Power 750+ and Power 760+ in their ratios.
I have a hard time imagining there are a lot of IBM i shops that would go for a Power 760+ server over a Power 750+ machine. I can see more companies who might have otherwise had gone with a two-chassis Power 770+ setup backstepping to a Power 760+, although there is nothing in the way of software savings there because the Power 760+ and Power 770+ machines are both in the P30 tier. The Power 770+ has a lot more I/O scalability and twice the maximum main memory capacity, so you might be driven up there by your database and workloads. If you are thinking about choosing between a Power 730+, a Power 740+, and a Power 750+, the cost per unit of performance is not that different, but the Power 750+ can scale further in terms of CPU cores and main memory, so it may be a better box to make a foundation for a long-term investment where you can start out with one processor card and scale up to four as needed.
Next week, we’ll talk about how IBM is positioning the Power 750+ and Power 760+ and pitching them to customers.