Entry Power S812 Gets A New – But Still Short – Lease On Life
March 18, 2019 Timothy Prickett Morgan
Despite the fact that Moore’s Law increases in performance in CPUs have been slowing for years, for many customers, the growth in the throughput performance of processors as more cores and threads are added to a Power9 chip have outstripped the capacity growth requirements for many IBM i shops. For many of these customers, a single core Power7, Power7+, or even Power8 processor did the trick just fine, and is better suited to their needs than an entry Power9 machine with just one core running IBM i.
We would argue – and have argued many times – that what IBM needs to do is make the Power chip cores and the IBM i and Linux licenses that run on them cheaper so more customers will consolidate X86 Linux workloads onto Linux partitions on Power and, wherever possible, port X86 Windows Server workloads to Linux on Power and pull these in, too. This would mop up all of that extra capacity, and provide a more integrated, hybrid system than is possible over the network, and give Power Systems a nice jolt in the arm, too.
We were a bit concerned that when the Power9 processors launched in entry systems a year ago, there was not a Power S912 system aimed specifically at these very entry IBM i shops with modest processing needs. The single socket Power S914 has four, six, or eight cores and a performance range of 52,500 CPWs on the four cores to 78,500 on those six cores to 112,500 across those eight cores. This seems like a lot of performance for customers that may only need somewhere between 5,000 CPWs and 10,000 CPWs of oomph to run their transaction processing and batch workloads. (A year ago, we went into the performance and price/performance of the entry Power9 lineup at this link.) On the Power S814, the four-core Power8 chip is rated at around 39,700 CPWs running at 3 GHz, so the single core is just a tad under 10,000 CPWs, and that means the performance of the single core in the Power S812 “Mini” configuration that was announced last year on Valentine’s Day as part of the 30th anniversary for the AS/400 is about the same and is also considerably lower than the 13,125 CPWs per core of the Power9 chip used in the Power S914. As long as IBM was selling the Power S812 and Power S814, therefore, it didn’t matter that the Power9 was overkill. But then, last October, Big Blue told the IBM i community that it was pulling the plug on the Power S812 and Power S814 on May 31 of this year. That means the price of an entry system and its performance would be much higher than many IBM i shops would need, which would presumably slow sales down or drive customers into the second-hand systems market for vintage gear if they did need to add capacity or upgrade from a smaller machine.
On March 12, in announcement letter 019-021, the Power S812 Mini has gotten a temporary stay of execution and will now not be removed from the IBM sales catalog until November 29, 2019. (This machine is sold under the 8284-21A product number in that catalog, if you need to look for it.) The Power S814 (8286-41A) is still for sale until May 31, so if you want to get one of these machines, you can still do it for the next couple of months. The Power S812 Mini and the single-core Power S814 both have a 64 GB cap on main memory, which is very restricting to performance, but this is IBM’s way of charging more to customers who need richer memory configurations and still modest numbers of Power8 cores.
IBM doesn’t extend the life of a product beyond its plans unless there is a reason for it, and clearly the reseller channel and the customer base wants an entry machine of modest performance that has a reasonable price, and clearly many of these customers think that a Power9-based Power S914 machine with just one core activated to run IBM i is overkill and they don’t want to pay the premium even if they probably will get significantly better batch run times and probably better response time on transaction processing compared to a Power8-based system.
So now entry IBM i shops running their applications on vintage iron have to make a choice, and they have to make it fast or the market is going to make it for them. They can go out and quickly get a Power S814, they can get a Power S812 Mini before the end of the year, or they can get a Power S914 in the next year or two before the Power10 chips come out, probably in 2021 in entry machines and probably with around 2X the CPW throughput and maybe 10 percent to 20 percent more single-threaded performance, if history is any guide. So you might have a base eight-core system with 125,000 CPWs of performance and a single core at somewhere around 16,000 CPWs.
Here is the point. There is no reason to cancel the Power S812 Mini at all. IBM should just sell it as long as customers want it. Just put the processors and motherboards in a barn somewhere, with maybe 50,000 of them all ready to go, and offer them at aggressive pricing on the hardware and the IBM i software and maybe even take the memory cap off so it can run a few Linux cores and multiple partitions and use it as a way to start pulling workloads off those external Linux and Windows Server platforms so customers get on track to needing a really hefty Power S924 or Power E950 (which should be running IBM i but only supports AIX and Linux) or even a Power E980 someday.
We talk all the time about modernizing applications and databases, but we don’t talk about the large number of customers running on old hardware that runs out of date or nearly so operating systems and databases. Getting customers current on hardware is the first step in getting current on the systems software stack and therefore modernizing applications. Every software vendor in the IBM i community should be encouraging customers to get modern on hardware because they will end up with a larger and more vibrant customer base and therefore more money to do even more innovative products and help customers do interesting things. If they can’t make that jump from vintage (Power6+ and earlier, Power5+ and earlier, you draw the line) to new (Power9) or new-ish (Power8) hardware, then they can’t modernize apps. And so they are effectively not really part of the IBM i base, but really part of the AS/400 and System/3X base. The part no one sees and that doesn’t spend money on hardware and software and does not invest in the future, but just keeps doing what they been doing for a long, long time.
Keeping a machine like the Power S812 Mini alive is the first step in getting customers to bridge out to Power10 and actually needing it when they get there in a few years. That’s the game. So let’s play, and let’s play to win. For everybody. That’s going to take some rethinking and refactoring of the Power9 line, which I am noodling now.