IPF Marketshare and Product Cycles
One of the major criticisms of the Itanium architecture is that to date, the vast majority of sales have been from HP, which co-designed the architecture and the first few MPUs. Historically, around 70% of Itanium system revenue has been from HP, so it is a fair criticism that while Itanium is a vendor neutral architecture, there clearly is a king of the hill. However, in the last quarter, HP has been selling an even higher percentage of Itanium systems; as a result, many companies, journalists and analysts have claimed that Itanium is in fact, a single vendor architecture. Unfortunately, many keen eyed observers seemed to have missed a simple explanation for this situation: product cycles.
HP upgraded their product line early this year with the sx2000 chipset, which is used in the larger multi-node systems, which scale from 8 to 64 sockets. Consequently, many customers opted for these newer systems which offered a reasonable performance boost over the prior generation (and could easily be upgraded to use Montecito). HP also released Integrity Nonstop systems (a legacy of Tandem) last year, and probably just started picking up many customers in the beginning of 2006. While hardly commonplace, Nonstop systems are ubiquitous and are in the same general class as IBM mainframes: unquestionably the best tool for certain tasks, and priced accordingly. However, most of the other Itanium system vendors were waiting for Montecito to roll out their new systems; after all, when qualification and testing costs millions of dollars, why bother doing it twice in 18 months? As a result of the product cycles from OEM partners, HP took nearly 90% of Itanium system revenue in the first quarter of 2006. Is this a sign that Itanium is just an HP product? Not really. Most likely it is just an anomaly caused by HP’s strong product release cycle in mid to late 2005, and early 2006, relative to other Itanium vendors. Of course, only time will tell going forward…so watch the market share numbers.
Montecito represents a rather significant advance for the entire Itanium community. Previously, Itanium systems generally lagged competing pSeries based servers from IBM, often by rather significant margins in OLTP performance benchmarks, such as TPC-C. The performance gains from going to a 2-way CMP design, along with 2-way multithreading should put Itanium systems comfortably ahead of their x86 counterparts, miles ahead of SPARC, and will go a long way towards catching up with IBM’s pSeries systems. For data warehousing, this advance should put Itanium safely ahead of all competitors, most other RISC systems perform poorly on more analytical workloads. Intel offered plenty of benchmarks for the launch, including a 4S/8C rx6600 TPC-C submission from HP, which scored 344K tpmc, compared to 421K tpmc for a 4S/8C p570. While this is far from parity, the rx6600 would score better if it were using HP-UX and Oracle instead of Windows and SQL Server 2005. Over the next six months, quite a few new results for TPC-C, SAP SD, TPC-H and other benchmarks should crop up, which will clarify the exact position of Montecito based systems, relative to competitors.
While performance relative to current competitors is unquestionably important, a larger driver for system sales may in fact be performance relative to older generations of hardware. The easiest pickings for Itanium vendors (and likely the vast majority of sales) are users looking to migrate from moribund architectures; there is OpenVMS on Alpha or VAX, HP-UX on PA-RISC, GCOS, etc. For these customers to migrate not only must the required applications all be available on Itanium, they will also want to see a substantial performance boost, since new purchases are usually on a 5-7 year cycle. Unfortunately, for some of these customers, the performance advantage was not compelling enough with prior generation products. Montecito systems are another story entirely; the performance improvements will be there, and more over, there won’t be anything substantially better until Tukwila arrives in 2008, so there is little value to waiting.
As Itanium partners release new systems in the coming months, this should drive a significant number of sales and lift the fortunes of Itanium. However, at the same time, it will be a test of the architecture. It is quite clear that Itanium must make significant inroads to the high-end market to achieve the goals set out by the community. Montecito is a good step along that path, as it will offer roughly twice the performance of its predecessor at lower power and the same cost. However a stable, consistent and competitive roadmap will be required, along with reliable execution for Itanium in the future.
I’d like to thank the following people:
- Nick Knupffer for his feedback on my article
- Kirk Skaugen for the interview
- Pat Gelsinger for chatting with me when he’s trying to prep for his talk
- Ruben Simpliciano and Diana Swanson for arranging the interview
Also, I’d like to point out that there was a very interesting demo from Secure64, which I didn’t have much time to address. Hopefully we’ll have a chance to discuss that in a future article.