Original URL: http://www.theregister.co.uk/2006/08/05/survey_server/
More Rev F delays, IBM's HTX embrace and Sun's mystery 'Eunuch'
Sorry, we mean 'Munich'
Thanks to our sharp readers, The Register manages to chuck a broad net over the server industry. Every now and then, we like to give that net a hurl and see what comes back and this week we're happy to report a banner haul. If server happenings/gossip are your thing, buckle up. We're going into fast and furious mode.
Waiting for the Dough
Many of you know that IBM, Sun Microsystems and Cray are locked in a fierce competition to receive a huge portion of federal pork. DARPA has been evaluating these vendors' proposals for an advanced type of supercomputer that would put the US a full computing generation ahead of other countries. Most of those close to the competition expect DARPA to pick two winners and to divide hundreds of millions of dollars between them.
The contest winners were meant to be announced in July, but here we are in August. What gives? Well, a DARPA spokeswoman said the agency will now announce the winner sometime "this Summer." Little whispers in our phone indicate that DARPA has told the vendors to expect a late August announcement.
Waiting for Rev F
The word on the street is that AMD will announce the Rev F versions of Opteron on Aug. 15. The chips can't arrive quick enough for AMD, which has seen Intel come out of nowhere to usurp the two-socket server performance crown with Woodcrest.
As it stands, however, the Rev F chips will be more delayed than previously expected. AMD already pushed back its official "launch" announcement from July to August. Now we hear that HP doesn't expect to slot Rev F chips into boxes until "late September or the first week of October."
"AMD is privately saying that Rev F is a holding place for quad-core, since it will go into the same socket," one source tells us. "The best Intel can offer up at the moment in the four-way market is Tulsa, which I understand is going to shoot back up to 130W or so. Intel won't have anything viable until Tigerton, which is expected in the second half of '07.
"There's going to be a lot of VMware sold between now and then, and it's landing increasingly on four-ways."
IBM goes HTX
That last comment may help explain why IBM finally decided to bed down with AMD rather than just flirt.
When IBM last week announced its five, new Opteron-based systems, it was shy about providing detailed system specifications. Well, those details have since trickled out. Big Blue has a handy page here where you can get an overview of the boxes.
Of particular note is the appearance of HTX slots in the x3455, x3655 and x3755. Unless we're mistaken that makes IBM the first Tier 1 to latch onto HTX.
A number of third parties have started building accelerators and networking cards that plug right into the HTX slots, which are found only on AMD-based systems. The HTX slots let third parties make use of Hypertransport's strong performance and give AMD a large potential edge over Intel, which has a closed architecture.
HP is expected to head to HTX country as well, while Sun plans to ignore the slots for now.
Thanks to Greg L for flagging that up.
Lets turn to our readers for some more color on IBM's Opteron move.
"I suspect a lot of companies have called IBM on the carpet about delivering decent opteron servers to the market," writes a reader who runs IT for a Latvian bank. "Up until this year, my shop was a pure IBM shop. IBM could not deliver what I needed. This year HP and Sun get my business. I doubt I am the only one to do so. HP beat IBM on the performance, price, heat and power usage criteria I had set. HP sucks on the remote management aspects in comparison to IBM, however. I would not be surprised if IBM realises this mistake now. How much of their business has walked away to the Sun and HP opteron servers?"
"There is a lot more background to IBM's past Opteron support," writes another reader. "The original systems (325,326 and 326m) are all rebadged motherboards from MSI with the 326m only having been released because of the lead free standards that have been put in place.
"This means that the only part of IBM's product set that was designed fully in house was the blade card.
"I can't say where the new systems have been sourced from, but the specs indicate a lot more commitment from IBM than people have seen in the past."
One last bit of weirdness on the IBM Opteron front is this appearance of the mystery x3105 on this site. So far it's just a name. Anyone have some help here?
Sun's mystery boxes
Speaking of server weirdness, reader Dale Ghent has tracked down two mysterious Sun systems called the X2100 M2 and the X2200 M2. The boxes are apparently code-named Leo and Taurus and meant to ship with a complementary workstation dubbed Munich. Sun doesn't want you to know about this it would seem. They've blanked out the boxes from the Sun web site and seem to have convinced a Phillip Fayers to remove his blog post about the systems. [Update: Not the case, says Phil.]
Ah, but Google's cache can be so kind.
"Leo sounds like a motherboard update to Sun's current cheapest server, presumably still a single socket box. The M2 moniker could just be coincidence (refering to a Mark 2 x2100) but the new AMD socket was to be named M2 at one point. Now it will be called AM2. The AM2 socket will be used for single socket AMD x86 systems and use DDR2 memory so Sun might want to offset the increased RAM cost by offering cheaper Athlon CPUs.
"The Taurus is based on one of the nVidia mcp55 chipsets (there are 4 available according to NforcersHQ) and it may be called the Sun Fire x2200 M2 (another machine on the warranty list). These chipsets have started appearing in commercial motherboards with single CPU sockets but the bug report indicates a system with 4 CPU cores. Now either that's very early silicon of a quad core Opteron (unlikely) or Sun are using the mcp55 in a 2 socket system. The system appears to be using the built in ethernet controllers provided by the mcp55 chipset (2 x 1GB/s). This system could be an update for the x4100/x4200 or possibly a new machine targetting a different market. Given the reference to new servers in the Infoworld article about the Rev F Opteron launch and the M2 tag I'm pretty sure that Taurus is based on the new 1,209 pin Socket F.
"Some of the nVidia mcp55 chipsets support multiple graphics cards with nVidia SLI technology (which gives you faster graphics by allowing more than one card to work together on driving a single display). So this nVidia system sounds a lot like the 4x4 initiative which AMD demonstrated recently. 4x4 provides desktop systems with 4 CPU cores (2 x duals) and up to 4 graphics cards, the platform seems to be targetted at gamers but it sounds like it would make a decent platform for a workstation as well.
"I wouldn't be at all suprised if it turns out that Leo, Taurus and Munich are all based on the same nVidia chipset as it would reduce development costs for Sun. It would be a return to Sun's roots if Taurus and Munich share the same motherboard, again it reduces development, manufacturing and support costs by having common components."
Seeing Red Hat
We'll close this gossip session on a software note.
A Red Hat VP caused quite the stir this week when he started talking about how not data center ready Xen is. As Red Hat tells it, only a clown would shove Xen onto production boxes at this point. The company, however, believes all the kinks in Xen will be worked out by the time it ships RHEL 5.0 at the end of the year. How convenient.
In the meantime, you will find Xen inside of SLES 10. Novell really, really wants to make sure you know that and the company's CTO has dished out a globule on the matter.
With a company like Amazon.com going Xen crazy with the 2.0 version of the open source code, we have to wonder how strong Red Hat's "it's not quite baked" story really is. ®