Battlefield in AMD-Intel server wars takes shape

NOW WE'VE SEEN recent roadmaps from both AMD and Intel outlining their server strategies next year, we're able to have a stab at working out where the battles will be.

This is complicated by the way Intel positions its Xeon processors, which are all based on the Pentium 4 architecture with extra tweaks.

And although AMD will certainly use its Opteron server processors to attack Intel's Itanium where it hurts, it's clear the first battlefield will be against those Xeons. Unholy hell will break out next year, we anticipate.

One of the fronts will be performance, with Intel still positioning its Xeon servers on frequency/cache, while AMD clearly thinks frequency is not that important.

The Itanium comes under attack from AMD in the second quarter of next year, and here both firms discount clock speeds. Also, here, software applications will become ever more important.

Dual Systems
The information we have shows that in the first quarter of next year, AMD will have the following microprocessors available: The Thoroughbred based MP 2600+, the 2400+, 2200+ and the 2000+. At some time in the first quarter, we will see Sledgehammer (Opteron) systems with clock frequencies currently projected at 1.4GHz and 1.6GHz. Remember, these frequency figures for the Sledgehammer may well change before launch date - that's the nature of a CPU roadmap.

During this period, Intel will have Pentium 4 Xeons at speeds of 3GHz, 3.06GHz and below, aimed at the performance, volume and value two way marketplace.

In Intel's nomenclature, a "value two processor" system has a typical price of $2,300 to $2,800, while AMD's dual processor systems will only include the MP 2400+, the 2200+ and the 2000+ during this period.

Volume and performance two way systems in Intel's counting, for typical systems cost between $3,000 and $5,500.

Here, AMD has a category called $3,000 plus (and less than $6,000), in which it will position the MP 2600+ and the Sledgehammer 1.4GHz and 1.6GHz dual processing chips.

These early Opteron chips will have 1MB of cache, it appears, competing against Intel Xeon MP chips with 512K of level two cache. Intel is able to produce its "Gallatin" chips with up to 2MB of level three cache. Will it suddenly start being able to produce big L3 cache Xeons in the mainstream line? We'd consider that doubtful.

AMD is clearly undercutting Intel's prices in the dual systems arena - typically by $1,500 or so. That could be achieved by it making bill of material (BOM) savings at every level of a two way server's production, including motherboards designed for the processor.

And what about performance? If AMD introduces a Sledgehammer Opteron at 1.6GHz, will that be a better performer than a Xeon at 3.06GHz with a 533MHz bus - also slated for introduction in Q1 of next year.

This, we anticipate, will not be an easy comparison to make. Benchmark comparisons are very different for server platforms - even dual processor server platforms, than for desktop chips.

Certainly, clock speeds on their own will not be an adequate way to compare peformance - we expect to see a flurry of SPEC benchmarks appearing with claims and counterclaims flying like arrows at the Battle of Hastings in 1066. Will AMD or Intel be ye King Harold who gets a sharp pointed stick in his eye as a result?

Four Way/Eight Way Systems
Current AMD roadmaps show Sledgehammer multiprocessing systems appearing in the second quarter of next year, with the speeds rising to 2GHz and 1.8GHz.

The top end multiprocessing systems will cost $25,000 and more. A lot? Perhaps not, because by this time AMD may well have secured the application support it needs for the 64-bit element of the Opterons, and this will pitch these systems against the Itanium 2 and later in the year Intel's Madison and Deerfield 64-bit chips.

And also, let it not be forgotten, against the four way Xeon Gallatin systems.

The Itanium 2 comes at clock speeds of 1GHz with 1.5MB of level three cache, but by Q2 Intel will have projected the Madison version of the Itanium, likely to be a 1.5GHz 64 bit chip with 4MB/6MB of cache.

A typical eight way Intel Itanium or Xeon MP system will cost between $38,000 and $44,000 during next year. A typical four way server using the Itanium 2, or the Madison or the Xeon MP will cost between $14,000 and $26,000.

It's clear from these figures that that is where AMD is pitching its high end Opterons and it wants to fight both the Itanium and the high end Xeons.

Again, the highest end Xeon MP predicted thus far for 2003 is one over 2.50GHz, with 2MB of level three cache. That will have to fight against a 2GHz+ Sledgehammer Opteron with 1MB of cache.

AMD is undercutting the Intel server roadmap at every level of the game, including the Itanium processor during 2003

Can AMD succeed in the server marketplace?
We believe that whatever happens in the long term, AMD will not make a signficant dent in Intel's server market share during 2003, except at the dual server and workstation level.

Whether AMD gets support from HP, IBM, or even Dell, large corporations will not wish to transfer wholesale to eight way Sledgehammer systems until they've thoroughly evaluated them.

They will also wish to evaluate the 64-bit application software intended for the high end Opterons, and this whole process will extend into 2004.

The big advantage Intel has at this end of the market is that it has the infrastructure and the money in the bank to persist, while AMD's financial structure cannot be compared with Chipzilla's.

The introduction of these server processors from AMD will have a salutary effect for big business, however. We anticipate a decrease in the price of average selling prices for Intel Xeons, coupled with more rapid introduction of future technology from the chip giant.

Irrespective of the financial status of AMD and Intel, we know from speaking to vendors at every level of the marketplace that there's broad support for alternatives. The third party firms are not Intel nor AMD fanboys, but nearly every one we've spoken to believe that the survival of the smaller player is good for competition and good for the development of better technology.

The biggest question mark for us is how these introductions will affect Intel's Itanium thrust. Having spent billions and billions of dollars on development over 10 years, now Intel will have to seriously consider the future direction of this platform, on which many have already cast doubt over those years.

Performance at this 64-bit level will also be critical to the future of the Intel Itanium platform, we believe.

Next year promises to be one of the most interesting fights in the microprocessor marketplace. We'll cover where we think the battles will be at the desktop level in a future article. µ

See Also
Intel server roadmaps 2003
AMD server roadmaps 2003