Big Blue has been talking about the Power7-based "Blue Waters" supercomputer nodes for so long that you might think they're already available. But although IBM gave us a glimpse of the Power 775 machines way back in November 2009, they actually won't start shipping commercially until next month – August 26, to be exact. The …
"...a full rack weighs 7,502 pounds..."
Not gonna be sitting on a raised-tile floor, then...
Yes they are, but the floor needs re-inforcing.
All of the cables and hoses go under the floor, and are arranged so that once in place nobody actually has to go under the racks to work on them. The weight of each rack is spread using steel spreader bars so that the point weight will not overload individual tiles, and the upright posts are being augmented and have diagonal bracing.
It's not just the static weight that is a problem. If these racks are moved full, the rolling resistance is such as to make any raised floor give in, so it is necessary to move the racks partly empty, and fill them in situ.
You would not believe the planning effort going in to the initial deliveries. The ship group includes a battery powered tractor to move the racks!
And it is not just Blue Waters getting these. There are other customers that will be getting deliveries before the end of the year.
Hopefully not giving any secrets away, but anon. anyway.
Now THAT is some kind of kit!
I can't want to read more about it!
Can it run??? You know what.
Lattice Gauge Computations?
I hope it does.
Re: Lattice Gauge Computations?
haha - Havn't you guys been beavering away on something called Q for a few years? Now is not the time for computer envy....
In other news I got one to run Crysis but it crashed.... through the floor into the downstairs flat :-)
half the number of cores and double the RAM, and we'd have a new prime candidate for our 'shared'/cloud like environment.
// Jesper beer cause it weekend.
PS: TPM a type seems to have sneaked into the article "for a total of 246 cores, on a single, massive motherboard", should have been 256 :)=
For a total of 256 cores
I think every IT specialist should be able to multiply 4 CPUs by 8 cores each by 8 MCM modules per box...
It is a nice round number, isn't it? what will the next be? 512, 1024?
We have 1000 on order
Can't wait to start getting them in.
oh that's right I am just kidding...would be cool to get a few if we had the right workload and money
It's worth the money
Just to watch the end of solitaire :)
Why AIX first?
While the hardware looks good I'm curious about the delay on Linux support. How many AIX systems are in the top 500?
Spoiler: It's 21 (4.20 %)
Is AIX first?
Blue Waters will be Linux, not AIX.
The training that is being offered in Poughkeepsie to software installers is mostly Linux to start with, not AIX, and many of the systems available for test run Linux rather than AIX.
It is clear that AIX is normally bid first, but that is because the RAS features of AIX are better than those on Linux, and Loadleveler support integrates better with AIX and the hardware than Linux (Loadleveler is the IBMjob scheduler IBM normally bid).
Linux has some advantages. Because of AIX 7.1 has a hard limit of 256 logical CPUs, this means that a single drawer (8 way multithreaded on 256 cores) it is necessary to divide each drawer into at least 8 LPARs in order to be able to have SMT turned on. As I understand it, Linux can support the 2048 logical CPUs that a single drawer has, so can be configured as a single system image.
re:Is AIX first?
I'm afraid the quote from AC of 21 July2011, 06.06GMT is quite misleading.
AIX 7.1 has a current public support limit of 256 physical cores, and 1024 threads.
This is the same as later technology levels of AIX 6 and AIX 5.3.
People have been playing with systems this size since at least 2009 on AIX.
It doesn't take long to look at the public IBM APAR database and find stuff to do with it:
Reported component name:AIX 5.3
Reported component ID:5765G0300
@AC 23rd July 2011 11:51 GMT
Just because the APAR says that support for 1024 CPUs has been added to curt/tprof (part of the command set) does not necessarily mean that the AIX kernel supports it at present. The AIX 7.1 release notes (and I believe one of the later 6.1 TL release notes) talk about raising the scheduling limit to 256 CPUs, and from experience of AIX 5.3 on Power6 575's, it is logical CPUs that are important (575s have 32 processors which actually appear in the system as 64 logical processors - the AIX 5.3 scheduling limit as far as I am aware - when you have SMT turned on).
I speak with knowledge of an in-flight Power7 775 install, and it is clearly the case that there are 8 LPARs (or Octants as they are called) per drawer in the design. I may have put 2+2 together and got 5, but I don't think so.
I believe that for the systems that have had large numbers of processors (and I think it was only the 595 that could have the required number of processors for AIX 5.3) you had to run with SMT off if configuring all processors in a single LPAR.
For Power7 systems that may contain more processors, there is a minimum level of AIX 6.1 with a TL to support up to 256 CPUs. If this is real CPUs, then this would equate to 2048 threads. As I understand it (I have not got a multi CPU Power7 system to play with yet) you either have SMT on or off, so with 8-way SMT would only allow you to use 128 processors in a single system image if 1024 was the limit on threads.
IBM is getting very poor in documenting how things work internally, so I think we are all just reading between the lines with the larger systems at the moment.
- Product round-up Ten excellent FREE PC apps to brighten your Windows
- Review Tough Banana Pi: a Raspberry Pi for colour-blind diehards
- Product round-up Ten Mac freeware apps for your new Apple baby
- Analysis Pity the poor Windows developer: The tools for desktop development are in disarray
- Chromecast video on UK, Euro TVs hertz so badly it makes us judder – but Google 'won't fix'