The Week in Review
Here's a collection of highlights, selected totally subjectively, from this week's HPC news stream as reported at insideHPC.com and HPCwire.
>>10 words and a link
Sun releases version 12 of its Studio parallel IDE;
New development release of the Globus Toolkit available for download;
LLNL adds 33 TFLOPS for Appro;
Diamond nuclei offer hope for room temp quantum computing;
Privately held cluster manufacturer Verari Systems landed $20M in new funding;
Platform released version 7.0.1 of its LSF workload management software;
NASA Ames buys IBM on ramp up to replace Columbia;
>>Cray's XT4 Delay: the drama! the intrigue!
There have been all kinds of shenanigans related to Cray's announcement on Monday of this week that issues with the “timing of volume parts availability” will push out acceptance of new XT4 systems. This dramatically changed Cray's 2007 financial outlook and caused them to warn that they might not achieve profitability this year.
There was much speculation about the source of the delay, with most of it focusing on the AMD processor. Cray has not yet confirmed which parts are causing delays. Ashlee Vance over at The Register is speculating that it's Barcelona's sister, Budapest, that's likely the source of the trouble (see Register articles at http://www.theregister.co.uk/2007/06/04/barcelona_delays_cray/ and http://www.theregister.co.uk/2007/06/05/amd_cray_dowjones/).
In what's turned into a Jerry Springer moment for the usually staid HPC industry, we even have a little public conflict and intrigue to go with the announcement.
Evidently a Cray spokesperson offered that a “four-core processor being made for the supercomputer by Advanced Micro Devices Inc.” is the cause of the delay, but this was later refuted by another spokesperson at the company who did not further elaborate on what parts are delaying shipment.
Then Dow Jones quoted an industry analyst (Citigroup analyst Glen Yeung) as saying that Barcelona will be delayed to the September/October timeframe from June/July, which is interesting, but potentially not directly relevant. AMD's stock has been falling since Friday on widespread speculation that its new white knight chips are not performing as well as anticipated in early tests (see, for example, http://money.cnn.com/news/newsfeeds/articles/djf500/200706051252DOWJONESDJONLINE000560_FORTUNE5.htm
AMD, for its part, has yet to comment publicly on Budapest, Barcelona, or any potential delays in its new chips.
I guess all this means that summer is officially here; when it gets hot in the South, people like to pass the time by sitting on the front porch and speculating about the news of the day. I'm proud to see this little bit of Southern heritage has made its way into our technology news community (and, yes, I'm part of the problem, too; but then, I actually live in the South).
>>Sun announces new blades
Sun this week announced its new Sun Blade 600 line. The new blades can be configured with Sun's own Ultrasparc T1, Intel Xeon, or AMD Opteron processors running Solaris, Windows, or Linux.
The range of supported chips makes Sun's offering fairly versatile and offers both 4 and 8 core processor options for customers. You can read the company's press release at http://biz.yahoo.com/prnews/070606/aqw164.html?.v=3.
>>Google acquires PeakStream
Google announced this week that it had acquired PeakStream, Inc. You'll recall that the company makes development software that turns single-threaded software into multi-threaded software ready to run on GPGPUs and multicore chips (we've covered PeakStream before, for example at http://insidehpc.com/2007/03/28/power-to-the-people/).
The Register argues that this is potentially a big deal for us, and I agree (see Ashlee Vance's opinion piece at http://www.theregister.co.uk/2007/06/06/google_peakstream_server/).
The problem is that the software industry at large (not just us) is heading for what most agree is a big problem: single-threaded software on multicore processors. Two companies had sprung up to deal with this challenge, RapidMind and PeakStream. It appears with this acquisition that PeakStream is out of the picture for everyone but Google. From The Register's article:
“Our most recent chats with relevant parties confirm that Google has no intentions of selling the PeakStream tools to chip, server or software makers. …In addition, we’re hearing that Google has little more than passing interest in crafting code to run on GPGPUs – the most immediate promise of an independent PeakStream's technology. Rather, Google ate up the PeakStream talent to develop better multi-threaded code capable of traversing x86 chips.”
This leaves us to hope that RapidMind, or some other startup, can carry the load for the entire industry with tools that matter.
John West summarizes the headlines in HPC every day at insideHPC.com, and writes on leadership and career issues for technology professionals at InfoWorld and on his own blog at onlytraitofaleader.com. You can contact him at email@example.com.