by Tom Yager

AMD’s Opteron launch

analysis
Apr 22, 20034 mins

AMD's Opteron launch is far more important than the low-key participants in the event were willing to convey. Microsoft and IBM were presented as headliners, while workhorses SuSE and Newisys got pushed into the background. Opteron has a brilliant future that starts right now. It's too bad AMD telegraphed the message that customers should wait for IBM's servers and Microsoft's OS. I'll say what AMD couldn't: Do

AMD’s Opteron launch is far more important than the low-key

participants in the event were willing to convey. Microsoft and

IBM were presented as headliners, while workhorses

SuSE and

Newisys got pushed into the background. Opteron has a brilliant

future that starts right now. It’s too bad AMD telegraphed the message

that customers should wait for IBM’s servers and Microsoft’s OS. I’ll say what AMD couldn’t: Don’t wait. Buy from one of the

gutsy third-tier hardware OEMs that had the nerve to back this architecture

from the jump. You can run validated 64-bit SuSE or Mandrake Linux

on those systems right now. Or install any 32-bit Linux. If you grab one

based on a kernel after

2.4.19, you can pull down updates from kernel.org

(they’re tagged “x86-64,” named for the licensable standard rooted in

AMD’s design) and do an in-place recompile. You can recompile your

commands and utilities when you get around to it, but with rare

exceptions for small projects, all of Linux is 64-bit clean.

If you want to run Windows, run 32-bit

Windows until Microsoft’s marketing debt to Intel and HP

is paid up (my wager: after HP’s Itanium 2 product line

delivery in mid-2003). If it’s smart, Microsoft will deliver Opteron

support as a Service Pack or optional free download instead of

a separate Windows product.

I don’t blame AMD for failing to give its hard-working, but

less visible partners more play at the launch event.

IBM and Microsoft are the kind of partners that can impose onerous

terms on their deals. I give Microsoft’s

Brian Valentine and IBM’s DB2 team

a lot of credit for truly championing the AMD64 architecture from

the beginning.

Their influence wasn’t enough to motivate both vendors to

throw in wholeheartedly behind Opteron, as I believe they

shoud have. Opteron would have taken off like a rocket

today if IBM and Microsoft had given it the unqualified support that I

know people inside both companies feel AMD64 deserves.

Instead, Brian took the stage at AMD’s launch event and said

that Windows Server 2003 will ship “64-bit enabled.” That’s a

veiled and, I think, borderline inappropriate reference to the software’s support for Itanium, the sole 64-bit Windows platform.

I like Brian, but he should have stuck to

Opteron-related news while he stood on AMD’s stage. One fascinating factoid: DB2 is about 10,000,000 lines of C++

code. IBM’s developers were able to port all of that code

to Opteron in two days. The IBM spokesman said that DB2 has

never ported with a simple recompile. It’s always had to be

hand-tuned for each new platform.

A Computer Associates exec had similar remarks about the ease

of porting to Opteron.

Its Ingres DBMS port to Opteron took two days to perform and validate. A

CA exec said that Ingres’s performance on Opteron “was so unbelievable that we

had to double-check the results.”

Damn.

The headline-making news from the event will undoubtedly be

IBM’s commitment to build Opteron-based eSeries servers. If AMD and

IBM were engaged today, IBM handed its new mate a gold-plated ring.

IBM’s Mark Shearer gave an almost grudging

welcome to Opteron, narrowly positioning it

as a niche processor for scientific and technical applications. Baloney.

Brian Valentine said as much. IBM’s own DB2 engineers said as much.

This is a fast, affordable general-purpose commercial CPU.

IBM won’t reverse gears on the eSeries marriage to Intel

Xeon CPUs. From AMD’s presented SPEC and TPC numbers,

Opteron kicks Xeon’s butt and gets within slapping distance of

Power4+ and Itanium 2 on integer computing and throughput.

Opteron can sling sci/tech workloads, but Alpha, Power4+ and Itanium 2 all have faster floating point, so they’re actually better suited to

sci/tech than first-generation Opteron. That is, if you ignore their

higher purchase cost and the fact that Alpha is (sadly) end-of-lifed. Don’t buy IBM’s spin and relegate Opteron to the narrow market that

Itanium is stuck in.

I am disappointed with some aspects of the Opteron launch,

but really jazzed about Opteron.

I’ve had a dual-processor Newisys 1U rack server running in my lab for a few weeks

now. Even though it’s not final hardware–the motherboard still

has a couple of jumper wires on it–it is fast and stable running

64-bit SuSE Enterprise 8.0 and the 32-bit version of Windows Server 2003.

Windows has a run-time switch that activates AMD’s very sexy

HyperTransport-based NUMA (non-uniform memory architecture) circuitry, which boosts

memory throughput and blows away the x86’s 4 GB memory ceiling.

Now that AMD’s non-disclosure period has expired, I’ll be able to report on my hands-on experience with Opteron. So stay tuned.

——–