In-Depth

IBM Supercharges pSeries

New Power4+ processors ratchet up performance by up to 65 percent

• By Stephen Swoyer
• 05/06/2003

Big Blue also tinkered with its capacity licensing schemes, announcing the availability of On/Off Capacity Upgrade on Demand (On/Off CUD) for its p690, p670 and p650 servers; a new memory on demand feature; and a free trial period—dubbed “Try and Buy”—in which customers can unlock and use excess capacity on their pSeries servers.

IBM’s top-of-the-line 32-way p690 and 16-way p670, along with its eight-way p650, a midrange server, and the p655, a high-performance computing system, have been refreshed with Power4+ chips running at 1.5 and 1.7 GHz. Big Blue first introduced the Power4+—running at a top speed of 1.45 GHz—in its p650 system in November 2002.

According to Jim McGaughan, director of IBM eServer strategy, IBM will charge roughly the same for the new Power4+-powered systems—“within about 2 percent or so”—as it did for the Power4-based configurations that they are now replacing. “Our customers are just going to love this. The price of all of this new technology is roughly equivalent to the price of the systems when we initially introduced them.”

Big Blue is packaging the new Power4+ chips in the multi-chip module (MCM) format that it first introduced with the Power4 architecture. MCM describes a configuration of four dual-core processors linked by a means of a high-speed interconnect. The Power4+-based p655 system Big Blue introduced in November—along with the low-end p630 that Big Blue also equipped with the Power4+ in February—exploited single chip modules, McGaughan explains.

Gordon Haff, a senior analyst with consultancy Illuminata Inc., says that the new, faster Power4+ processors should turn up the heat on competitors Sun Microsystems Inc. and Hewlett-Packard Co. “It is now accepted by nearly everybody that the Power4 is a faster processor than UltraSPARC III or PA-RISC, and the new Power4+ is even faster.”

HP and Sun both market systems that scale to 64 and 106 processors, respectively, but Haff speculates that IBM’s 32-way p690 is nevertheless competitive with both of these SMP behemoths. “There’s no question that IBM has a very fast system [with the p690], and that it is competitive with Sun and HP. Sun will claim that it can scale to 106 processors, but that’s not realistic in practice. So is a 32-way IBM system competitive with a 72- to 84-processor Sun system? For many applications, yes.”

Fleshing Out On Demand

McGaughan positions the array of new pSeries capacity licensing options as consistent with IBM’s broader e-business on demand strategy, which emphasizes a utility-based computing model along with autonomic capabilities that allow customers to respond more quickly to peaks in demand and system outages.

On/Off CUD, for example, gives customers a pay-after-use (temporary) facility to allocate resources and then de-allocate them if necessary. “That means that you can get inactive processors on your system so in the case of a p690 with 16 processors, you can have eight active and eight inactive processors. Those eight inactive processors—that’s your pool, that’s your resource that you can light up if you need it.”

On/Off CUD will be available for IBM’s low-end and midrange pSeries systems by late May. It won’t be available for the high-end p670 and p690 until September, however.

Try and Buy, McGaughan explains, provides an opportunity for customers to unlock a pSeries system’s excess capacity for up to 30 days. In this respect, he allows, a retailer could even choose to unlock excess capacity during the busy Christmas buying season—basically receiving the benefits of additional computing horsepower at no charge. Best of all, he points out, if a customer permanently activates at least two extra processors, the Try and Buy feature resets.

McGaughan stresses that IBM will not automatically shut down Try and Buy after 30 days; the feature resets only when an LPAR or system is rebooted. “This is not one of those Machiavellian things where we come in and pull the plug on the computer. [After 30 days] we tell the customer that it’s time to shut down that extra resource. The only thing that we do proactively for the customer is that if they reboot the machine, we take that idle capacity and return it to the pool.”

Similarly, Memory on Demand lets customers unlock additional memory for periods of increased demand. Once a customer unlocks additional memory—which is typically made available in 4 GB increments—it can’t be returned to the resource pool. Memory on Demand will be available in August.

Power5 on the Way

Since its introduction, Power4 has scaled from 1.1 GHz up to—at this point—1.7 GHz. McGaughan speculates that the architecture still has quite some way to go before it runs out of gas, but allowed that IBM is increasingly focusing its efforts on the next-generation Power5 architecture.

“Where we’re going is we’re heading very rapidly to Power5, which will be next year. I think it’s safe to assume that there’s still more performance in the Power4+, but major investments will be made in our next-generation architecture.”