From Computer Business Review, a sister publication
It is a long time since anyone has questioned the business acumen of Intel. The $16 billion chip supplier powers 80% of the world’s PCs, has enjoyed a compound annual growth rate of 33% for the past ten years, and has one of the strongest brand names in the industry. It has also been successful in executing an almost copybook PC business expansion plan, stealthily moving up the PC food chain from chip to motherboard supply, picking up an ever larger proportion of the available revenue stream. Leading PC suppliers, including Packard Bell, Gateway and Dell, all now base their systems on Intel motherboards rather than simply using Intel chips. But, Intel’s next planned assault – on the lucrative high-end server market – is, say detractors, in danger of being derailed. For once, they say, Intel’s focus on low-cost and high volume has led to some technological miscalculations which could prevent it from achieving its aims. Intel is already the dominant chip supplier in the network server segment. According to IDC, 63% of network servers sold are based on Intel. But at the enterprise server level, Intel’s penetration is slight. The majority of high-end server systems are based on RISC processors such as PowerPC from IBM, PA-RISC from Hewlett-Packard, Sparc from Sun and Alpha from Digital. Intel’s attempt to wrest control in the high-end server market comes in the form of its SHV (server high volume) ‘quad’ boards, which combine four Intel Pentium Pro processors, 512KB of cache per processor, up to 4 GB of memory and PCI/IO channels. Intel’s hope is that, using high speed interconnects and clustering technologies such as NUMA (non-uniform memory access) enterprise strength suppliers will harness large multiples of SHV boards to create systems which can rival the power of the mainframes. Furthermore, it says these systems will be based on PC pricing models, resulting in dramatic savings for users. But while a number of traditional high-end server vendors, including Data General, Sequent, NCR and Unisys, have expressed interest in using the technology, some are saying that Intel’s perpetual focus on costs means that the SHV boards are not suitable for powering large scale systems. Among the critics are some of its closest partners
Cutting corners
SHV simply isn’t correctly specified for mission critical computing, says Kevin Joyce, chief technology officer of Sequent. Because of its cost goals, Intel wouldn’t put in the right power supplies and connectors. As a result, he says, the SHV board cannot deliver the kind of reliability and security expected by enterprise users. Sequent’s solution to this problem has been to license the SHV design from Intel and then build its own variant. We have taken Intel’s 12-layer board design and done an 18-layer board for the datacenter. Joyce also disputes Intel’s PC pricing claims, saying that the price advantages provided by the SHV technology will be much smaller than many vendors anticipate. For a $1 million server system, 40% of the cost is in the disk, 25% in the memory, 25% is in the CPU and 10% is in the power supply. The only variable that changes with the SHV design, says Joyce, is the CPU. So although he believes that SHV will deliver improved price/performance, he expects base level prices to be much the same as those of existing server systems. This, he says, compounds Intel’s error of choosing cheap, PC-style components on the SHV board when it is clearly trying to be an enterprise-level product. The recognition of this fact by Intel, says Joyce, is one of the key reasons why the SHV boards are running around six months behind schedule. The boards, he says, should have already been delivered to its partners but Intel has had to re-examine the design in a bid to make the economics work. The market impact, he says, is that few enterprise suppliers will take the standard SHV boards, and Intel will miss its revenue targets. Compaq is doing its own quad (based on the Intel Pentium Pro). NCR is talking to us. Who else is going to take that board?
Commodity benefits
The answer, to date, is just two companies – Data General (DG) and Amdahl. Both have announced plans to build systems using the standard SHV building blocks, dismissing Sequent’s claims that the technology does not match the business requirements. Intel SHV servers are the new building block for the industry, says DG. They will push PC economics into large-scale computing and do for the datacenter what Intel has already done for the desktop. Ron Skates, CEO of DG, says that although implementing the commodity SHV technology will result in lower sales margins, the volume of sales will more than make up for the shortfall and encourage other vendors to follow in its path. Around 90% of [other server vendors] want to go the commodity approach, he says. Intel agrees: DG is a great example of standardize and scale, says Carr Biggerstaff, marketing manager for Intel’s emerging technologies group. Skates says that the company’s first SHV-based system will provide performance in line with an 8-way Hewlett-Packard T5000 machine for a cost of around $450,000 instead of $1 million. However, Intel’s problems with SHV mean DG’s delivery dates have slipped. Systems, which were due this year will now not ship until 1997. The real test of who is right and who is wrong will occur when systems are actually delivered. Certainly Sequent’s proprietary approach means it will have the advantage of being first. It says it will ship a 16-way NUMA-based system in December, followed up by a 32-way system in the first quarter of next year. DG will be 6 to 7 months behind us, says Joyce. In the meantime, uncertainties over the caliber of SHV is making it difficult for Intel to sign up other partners. But, according to Brian Richardson of the Meta Group, this is likely to prove a transient problem. Adding a $50 component to a board is a big deal for Intel but these vendors are helping to pull the company up the food chain and add the necessary capabilities for alternative mainframe systems. People will not take the boards initially, but they will take them next year when Intel has got the design right. Nevertheless, Intel will be mindful that its miscalculation has given the incumbent RISC chip vendors another year without serious competition.
