Geared Down, Low Cost Power IBM i Box Rumored
February 6, 2017 Timothy Prickett Morgan
We have been talking for a long time about the impedance mismatch between some IBM i shops in terms of what processing, memory, storage, and I/O capacity they need and what IBM delivers to them with each new generation of Power Systems iron. For many shops, even the smallest machine that Big Blue currently delivers is too much of a box.
This is a testament to the efficiency of the IBM i platform and its integrated relational database management system in that it takes very little iron to get useful work. It also shows that for many customers, transaction processing is not a big part of their overall workload, even if it is absolutely critical to the business mission. As I have said many times before, IBM should be encouraging for more workloads to be moved onto Power Systems iron from Windows and Linux servers based on X86 processors. But there is another tack to take. IBM could make smaller and less expensive versions of the entry Power Systems – something like the “Invader” AS/400 systems from days gone by that are small and suited for these customers and their database workloads and just get as many people as possible onto newer iron.
If the rumor mill is accurate, then Big Blue is in fact working on a new entry Power Systems machine that fits the bill. It is not clear at this time if IBM is creating a whole new system, perhaps based on a variant of a geared-down Power8 processor, or just repackaging and repricing an existing Power Systems machine. A lot depends on how many Power8 chips IBM has in stock, how many existing entry Power Systems machines it has in the barn, and what demand it thinks it might have for a less capacious, less costly IBM i and AIX box.
Yup, I said AIX there. From what I hear, IBM thinks that a segment of the market that wants a cheaper and less powerful (yes, it does happen) Power Systems box wants to run AIX. So this is apparently not going to be a machine marketed to IBM i shops, like the past several generations of single-socket, four-core machines have been.
How Low Can You Go?
The advent of a lower-end Power System machine, or machines if there is more than one, brings to mind two key questions: How low can the performance be geared down and how much lower can the price be dropped?
What I remember during the initial Power8 launch is that IBM did not even want to commit to doing the four-core version of the Power S814 system, which can be geared down to have as little as a single core loaded up with the IBM i operating system. When the entry Power8 machines rolled out in April 2014, nearly three years ago it is hard to imagine, IBM was telling business partners it might create a four-core machine, as it had done with the Power7 and Power7+ systems before it, but that it should not count on it. The plan called for IBM to ship a memory-crimped variant of the six-core Power S814 machine, with a maximum of four cores that could be activated, and the plan was to ship it in the second half of 2014. It was unexpectedly launched in June of that year, ahead of schedule, and that was probably because customers, resellers, and yours truly were grousing that there needed to be a low-cost machine in the P05 software tier to appeal to customers who had modest computing needs but who wanted to get current.
We think there is similar pressure on IBM to do something to make sales to entry IBM i and AIX shops, and to do so well ahead of the Power9 systems launch later this year. With the Power8-based S814 that came out in June 2014, a four-core machine was rated at about 39,800 units of aggregate performance across the motors based on the Commercial Performance Workload (CPW) benchmark test. Those Power8 cores ran at just a hair over 3 GHz. That was a 40 percent bump in performance compared to the four-core Power 720 using Power7+ chips running at 3.6 GHz, which was rated at 28,400 CPWs. We don’t think IBM will offer more performance in whatever box it is working on, but rather less, and it might impose a relatively skinny memory footprint on the box, too.
The peculiar thing is that IBM is already offering a pretty sweet deal to customers buying entry machines with four or six Power8 cores, as we pointed out back in November. The deal was secretly launched to business partners back in September, and then IBM finally revealed that it was in fact the one behind the deal. This Double Up Core and Memory promotion allows for customers to get four cores on the entry Power S814 machine for the price of two cores, and to get double the memory capacity at 64 GB at the cost of 32 GB. This amounted to about a 10 percent discount on the overall system hardware cost.
With the new entry machine, IBM could put out a box with only two cores being accessible, and it might chop the memory capacity quite a bit. If a four-core machine was capped at 64 GB, then it stands to reason that a two-core machine would top out with a cap of 32 GB. IBM could dial back the cores and not dial back the memory, which I think would be wise. It might also switch to the Power S812 chassis for this new machine, which is a smaller system with fewer peripheral expansion options, particularly if there were a lot of them sitting around unsold. (I think IBM had higher hopes for this single-socket Power S812 system as a competitor to X86 boxes way back three years ago.) If the price can be made low enough, IBM can clear out inventory and make some money that it was not going to make with future machines with 24-core Power9 chips in them. For these customers, one or two cores running at around 3 GHz is more than enough, and 48 cores is overkill in a big way.
A lot of the software stack, including logical partitioning, might be overkill, too. So IBM could just fire up the machine with one logical partition, plunk IBM i or AIX on it, and get rid of PowerVM (or at least the way it is exposed to administrators and licensed to customers) and other extraneous stuff, like the Hardware Management Console. This way, IBM could not just cut the cost of processing, but the overall system. You don’t need an HMC to use Windows or Linux, after all.
IBM has a real opportunity to make some incremental sales with this rumored machine, but it is likely that IBM won’t do a lot of engineering on this system. It will be a geared down Power S812 or Power S814 is my guess, with some price cuts.
I would like IBM to do something more dramatic with what I am calling the Power Systems Mini, which would be based on the future Power9 processor. I want IBM to take a chunk of the 24-core chip and where there are mostly dud cores, preserve the few good ones and then gear them down to like 2 GHz or maybe 2.5 GHz. Slow them down and make them much cooler – and I mean temperature, not hipness there. Build a small motherboard with the processor and some main memory and only Ethernet ports and a few flash modules. Cram this into the smallest form factor possible, and seal the box and paint it light blue. And put all the necessary software onto the box and cap it at maybe five concurrently users, or ten perhaps. And sell it for like $7,500, all in.
With many tens of thousands of customers running OS/400, i5/OS, and IBM i on vintage iron, the Power Systems Mini – maybe we call it IBMini? – could be appealing to many tens of thousands of customers who just need a really modest machine. Or, you could go really crazy and make it a coprocessor for an X86 server. . . . I am going to noodle this a bit more. Hmmm. Where did I put that soapbox?
In any event, I will keep my ear to the ground and see if I can’t find out more about this future entry Power Systems machine I am hearing about and that could come to market in February or March. Stay tuned, and if you are buying a Power S812 or Power S814 right now, it is probably a good time to reach out to your business partner, who might know more than I have been able to find out at this moment.