IBM's eight-core Power7 chip to clock in at 4.0GHz IBM looks set to join the seriously multi-core set with the Power7 chip. Internal documents seen by The Register show Power7 with eight cores per processor and also some very, very large IBM boxes based on the chip. The IBM documents have the eight-core Power7 being arranged in dual-chip modules. So, that's 16-cores per module …
you must be the guy who stole my briefcase in the Westchester Airport restroom! If I ever find you, I'm going to run your hands through a paper shredder - you'd better have some good voice-to-text software or your hack days are done. (BTW, you might want to take a look our IBM ViaVoice products, it's award winning technology). Also, when I do catch you, you'd better have my f*ing iPod on you - I just got my play lists organized and I want it back!
"And then IBM still has the Quasar project lurking in the background, where it's combing Power and Cell chips. Stand back, friends. Stand back. ®"
Crikey, Jumping Jehosophats, Ashlee, that must be a Colossus of a Virtually IntelAIgent Machine which would be QuITe a Perfect Fit for their VTThrust. .... http://www.theinquirer.net/gb/inquirer/news/2008/07/10/virtual-teleportation-arrives.
I wonder what all those Sunny West Coast hippy type software gurus would counter that with....... or would they just hang on for the Roller Coaster Ride/Helter Skelter Trip, which would be a Shame/Abject Abdication of Potential Opportunity.
Here I thought Lawrence Livermore's Lab concoction of some 1200 Power PC powered crazy Mac's operating in tandem and parallel was faster than the some of the best overprice Japanese monster machines they can produce today , this baby would not even break out in a sweat running all of JPL's heaviest floating point orbital routines to the six millionth decimal place !
What a beast , indeed !
Since IBM have been bashing the UltraSPARC T1/T2/T2+ at every event I've been too its quite funny to see them make an almost exact Power-based copy.
What scares me, is if those 8 cores (16 on the MCM) are as hungry as the Power4/5/6 range, you'll blow the mains if you put two in a rack without its own nuclear power station. IBM really need to start thinking about efficiency so it will also be interesting to see if AIX actually scales with it.
The problem with IBM, as shown by their recent shameless WPAR copy of Solaris Zones after slamming that technology is that they don't understand these issues at the customer end as they are so focussed on (big-)blue-sky.
Try not to break the posties back when the leccy bill comes in for these monsters!
...go, vista haters, go, its not too late to be the first to make the point about vista running slowly...oh, wait, yes it is
it wasn't that funny the first time, it sure as hell hasn't gotten any funnier. and no matter how many times the comments are made, it isn't actualy going to achieve ANYTHING, let alone make vista run faster
and on the subject of la la land, whose gonna give me a ride to the aftershow? i hope i have enough change, to make my brain rearrange
All the other vendors have been killing Sun for going to a many multicore, multi-threaded model. Sun did it in a few years back and when you think about how long it took to have the vision, design and implementation work, you're almost talking 10 years ago. During all of this time, IBM has been slamming Sun for the multi-core, multi-threaded processors i.e. UST1, T2, T2+, Niagara 3, Rock. All of a sudden, IBM is going to do the same in 2010. Whoopie....
Reminds me of how Solaris Zones/Containers and DTrace were insignificant until IBM came out with WPARs and Probevue. Maybe Sun would get more credit for their innovations if their marketing were as good as IBM's. Can someone fire Anil Gadre already???
Actually AC, it is funny. Still. This long after Vista is released, and speed is still an issue? Says something important methinks. Particularly given the state of hardware way back when the project kicked of, and the state of hardware way back when it was first supposed to be released. Perhaps MS are wishing it had been even further delayed so hardware had even more of a chance to catch up.
And yes it does achieve something. It gives us a fresh chance to question just what we're getting for our money. Given the enormous advances in hardware since (for arguments sake, say) the 80486, can we really say we've seen the same with desktop software? Is Vista as many times more productive than win 3.11? Or is it time to ask some hard questions about just what we're getting for our money? For my part, I've asked and answered those questions, and the answer wasn't Vista. FWIW, my current choice doesn't meet that standard either, but it is pushing the boundaries in that direction.
"Since IBM have been bashing the UltraSPARC T1/T2/T2+ at every event I've been too its quite funny to see them make an almost exact Power-based copy."
You clearly haven't the foggiest notion about the architecture of either. The PowerProcessor is a rip-snorting, single-thread monster and sticking 8 of those on a chip is not going to change that; just the economics. No doubt it will come at the price of using a serious amount of electricity. It's vastly better suited to applications which require fast single-thread performance than is the T1/T2. For those that care, that includes things like high transactional databases where it is essential to keep code-paths short where exclusive access is required. If that's not done, then your multi-core machines will end up burning up a higher and higher proportion of their CPU waiting for spin locks.
The T1/T2 range is fine for low-power, high-throughput workloads where there are lots of largely independent threads. The T1./T2 hardware threading effectively treats the core resources as a system in its own right and despatches different parts of different hardware threads onto the core resources. A sort of multi-tasking within the core (for instance, the T1 has 4 threads per core and one integer unit whilst the T2 has 8 hardware threads and two integer processors per core). Now this is very efficient as resources that would otherwise be unused due to thread stalls for things like memory access can be used. However, it comes at the cost of seriously slower single-thread speed, especially so when the underlying core resources are over-committed (which, incidentally, means that CPU utilisation as reported by the operating system is seriously non-linear as the OS sees the hardware thread, effectively a virtual CPU, as the resource and not the core utilisation).
The T1/T2 is not a competitor to Power. Either a shop has committed itself to AIX (in which case Power is a done deal). The real competition to T1/T2 comes from the x86 side where sheer volume (and better single thread speed) is the real challenge from the latter.
The Power competes with Itanium and the SUN/Fujitsu SPARC64 IV (and the coming Rock processor), not the T1/T2. Frankly the SPARC64 IV is not in the same performance league at the moment. SUN will have to up their game on Rock, and they are going to have a hard time with selling volumes. Incidentally, Power, Itanium & SPARC64 IV all support a form of hardware multi-threading, but it's of a fundamentally different type to that of the T1/T2 which has the feature as its heart and soul. It's innovative, and clever, but it has its limitations and I think it will get steam-rollered by the x86 bandwagon.
If this is to be believed, then this is the REAL "shock and awe". Am I the only one sitting here, mouth open in amazement (and, quite frankly, disbelief) at those numbers?
A processor that can handle 256 BILLION float-point operations per second?
A 1.30PB/s interconnect to link the systems? What kind of interconnect can even come close to that? Even if that 1.30PB/s is a total of the combined 38,900 systems, that would still be 35.04GB/s per system. That's almost 3 times the maximum DDR3 bandwidth (12800MB/s), and about 4.38 times the maximum PCI Express bandwidth (8GB/s).
As other commenters have said, I hope they have their own power station because between powering these systems and powering the A/C needed to cool these systems, they'll definitely need it.
Just some rough numbers because I'm bored... If each processor uses 50W, that would be 1.945MW for the processors. 26PB storage at 1.5TB/disk (we won't even consider RAID, controllers, etc) = 17,333 disks at 8W each = 139KW. So just considering the processors and non-RAID storage, that's 2MW. I think we can all agree that's a massive lowball number since it doesn't include other system components, RAID, controllers, AC-to-DC loss, etc... And then think of the A/C needed to pump out 2MW worth of heat in the summer.
Wow. Just, wow.
@George
That was actually clever. Next time I see you, I'll buy you a loaf of bread. I don't know why, but I hope it will make my feelings clear. Ignoring that, I don't spend ALL my time reading The Articles on The Register, it would normally be only once a day, so 12 hours is a perfect average. Really makes you think, eh? Of course, its all skewed due to The Reg Hack Delay. But there's no need to go into that now.
@Tim
An interesting argument. The way you describe it, I wouldn't say it's funny, as much as I would say it is sad and demoralising. As far as achieving something goes though, I disagree. Making a "wisecrack" in a comment on an IT news site gives you a fresh chance to question what you were getting for your money? What prevented you from doing so before? Did you have to read a comment by someone else stating Vista was slow before you saw the performance problems on your own machine? Or did it not occur to you to stop and think when you first started up the machine after installing Vista, "Hmm, XP loaded a lot faster, maybe this wasn't an upgrade at all, it was a trick"? I imagine there is a word or phrase to describe that kind of mentality, but I don't know what it is offhand. Anyway, that aside, "fresh"? Really, "fresh"? "Fresh" is when something is relatively new, straight out of the bag. If you take your anti-Vista attitude out of the bag every day, at every given opportunity, it becomes worn out. Perhaps even - dare I sat it - stale?
Regardless, we soldier on, striding quietly but forcefully into the night, and evermore shall this era be known as the turning of the otter.
SPARC went multicore because it was way easier to cut and paste cores and add some front-logic than redesigning the entire core for faster executions.
Even Power4 (released in 2001, multicore before SPARC) is probably faster than Suns Fastest T2. Unless you run java or some exotic legacy SPARC-code no-one bothers to port. It should be your last choice. I bet even Intel Atom runs faster.
Why haven't Sun released any spec int/float numbers for T2, because it sucks big time. Just check http://www.spec.org
Nothing tells me Rock will be anything else. Just even more slow cores. I rather doubt it will come out at all. It is just disinformation to assure current SPARC customers the architecture is alive and kicking. SUN has lots of intel/amd boxes they can sell you instead.
How long 'til we see one in our Salvage bins?
Can't wait to morph it to something that will work for a poor child somewhere!
Gift From God Computer Foundation.org sent out, in 2007, over 1400 systems to school aged kids on the dark side of the Digital Divide.
Units with no COA from Microsoft get Ubuntu, Xubuntu...
The anti-Vista jokes were funny and the pro-Vsta posts were funny for their bad timing.
I just want to know what happened to all the "Where is Osama Bin Laden now that they have the technology" jokes?
Or is this one going into the great Microschism in the NSA?
Tetris anyone?
Whilst I agree with you that Vista sucks, I'll think you'll find that Webster issues Apple-baiting posts of a non-technical nature in line with the technical ability of the average iBoner, whereas I offer technical reasons and observations (with glee) as to just why Sun are going down the pan. Would you care to supply even a slightly technical argument as to why you believe (a) the Niagara's single-threaded capability compared to just about any other CPU in modern servers is not cr*p, and (b) why you think Sun developed multi-cored CPUs? We promise not to laugh too much at your reply.....
"Not sure why you feel that XP should be faster coz it's 7 years old.
Normally you look for progress in new versions of software. The new version should be faster than the version it replaces. It might be more resource hungry but if it ain't faster it has no right to exist."
So XP was faster than 95, 95 was faster than 3.11, 3.11 was faster than 3 and....
With greater functionality and more complex GUI comes greater background workload. Windows 3.11 would fly on a modern PC "assuming you could get drivers working for modern hardware" but would still look ugly and dated.
Yes Vista has its problems and is slower than its predecessor, and yes these things will no doubt only be fixed just as 7 hits the shelves but, this is the same situation as faced by 95 and every MS desktop OS since.
They copied the Sun partitioning, probevue, and whatnot because they can.
They added dual serviceprocessors because they can.
And whenever someone comes up with a point they think IBM is lacking, IBM will order their people to go out and make sure the bat they get hit with is nerfed.
IBM wont let others be able to say "X is _so_ good and look at how IBM is lacking that".
And if it is not a question about $ VS $ then I'll generally take the IBM technology when I can.
After all, who needs a usable interface when you can run batchfiles.
AC Power requirements may be hefty but we could all easily compensate for that by not running Vista. (Which, as the readership is reminded, also requires it's own, preferrably nuclear power station - windmills won't cut it for this particular Quixotian incarnation of the Bloated Ogre)
AC will of course balk and complain it is not funny, and spend the rest of the day writing contrived prose to point out the fallacies in my post, all the while grumpily sucking on their stolen eye candy. I know I know... we all need that kiddie-drool-interface with the big buttons and toolbarf 'cos otherwise our humungous flatscreens will be so empty and desolate!
Not to mention no AC can do without the kind of priceless assistance with our daily tasks that only Redmond dares to offer: "The Vista Ridicule Wizard has noticed you are trying to make a fool of yourself. Would you like the Vista Ridicule Wizard to help you with that?"
AC because the AC keeps my head cool - our Power servers don't really need it, just the Pentiums, and they won't be missed anyways.
Paris because she knows an a day keeps the doctor away...
that IBM were already shipping a mix of power/x86. Oh wait, wasn't that RoadRunner?
It's no news that IBM was working on Blue Waters and that it would be terrifying, and I wouldn't be surprised if the actual POWER7 project started out just for this system. IBM cannot afford to lose #1 spot on the top500 for PR, Marketing and investor reasons.
These massive projects, typical of IBM, should be seen as such (though they are really cool machines), so that they can inspire everyone aiming at spots between #10 and #200 to give their hardware budget to IBM.
Vista so horney it love you long time.
But it will still lag like crap while the indexer that you can't murder runs and it oh-so-helpfully precaches the "top 10 programs" on that system you use regularly for 50 or 60 programs a day, (the joys of being a sysadmin!)
Over a petabyte of interconnect? Sever ha-jillion floating point operations a second? You'll probably need them, to hash all yoru personal data up for "anonymous statistics" sent to mircoso-, I mean the pigopolists.
And don't forget if you try to do anything useful in the multimedia department it'll downgrade your experience for your own protection.
There, that'll give you vista lovers something to chew on. (OMG INACCURACIES AND BAD SPELLING ELL OH ELL)
Typed on the only vista box I own, which I must maintain in order to help the sutpid who downgraded to vista.
Vista: it has two purposes: to provide a reference for me to walk customers through issues for phone support...and to RDP into my XP box so I can get real work done.
Don't know how many people remember Elite on the 32K BBC B, but it was a revelation when it first came out. Realtime hidden line removal on an 8-bit micro running at 2MHz with not a GPU in sight! When I first saw it I was amazed, as I had been playing around on the BEEB in assembler to do 3-D wireframe, and I could only get simple objects (cubes mainly and other regular objects) without hidden line removal running at 2-3 frames a second. But I think that the main problem was that I was using the OS linedrawing primatives, whereas Elite used a quick and very dirty algorithm.
I would love to know how they did the hidden line stuff so fast on a limited system. They wern't even using colour switching to hide the drawing (Elite ran split screen, with the top 3/4 running effectivly in mode 4 [actually a hybrid mode 3] 1 bit per pixel == 2 colours, and the bottom running in mode 5 2 bits per pixel == 4 colours) except when you were using a 6502 second processor, when it ran in mode 1 all the time. Really used the available hardware to it's best.
This Power7 monster IBM is proposing sounds like you will need serious communication skills to get the best from it. Make the p4 and p5 stuff I am working on at the moment look a bit lame.
5 years ago IBM created a hybrid biological-electronic chip based on the human genome, that gave quantum-computing-like performance figures. Unfortunately the genome they sampled was female, so its used the last 10^10^10^10^10^10 flops to try to decide what colour it wants to be.
Ba boom tsh!
...i don't actually like vista, i didn't say i did, its just boring that almost every comment thread on here ends up turning into a dig at vista. either that, or someone claiming "let me be the first to welcome our new XY overlords", or another argument between apple fans and haters etc.
personally, given the choice, i'd still be using win 2k pro, as i did for the first five or so years after xp was released. there was an operating system, no fucking tellytubby hills and garish blue/orange tosh in my face.
"but can it beat Kasparov? " .
Almost certainly. But if you are interested in a pure-skill no-luck game where moderately skilled humans can still beat a computer you might want to have a look at the board game known as go (Japanese), or wéiqí (Chinese). See http://en.wikipedia.org/wiki/Go_(board_game) . That said, It took me a year and a half of playing this game before I could beat a computer go program on a Pentium 4 more often than not. I'd certainly be interested to see how a massively parallel go program using simulated annealing on this machine would play against someone a few handicap stones stronger than me.
Now as a UNIX admin (Solaris, Linux, AIX) I'm glad to see Sun's new SUN SPARC64 VII chip. We need to replace some of our older Sun boxes and I'm excited to see this new chip.
However, if you look at the benchmark, Sun used twice as much memory when IBM ran the benchmark. So I would suspect IBM will load the same amount of memory and we shall see which one does better. Also, note that the P6 595 server had 32 procs (max out) and the M9000 had 64 procs (max out). I would have to say the P6 5.0GHz is superior to the new SPARC64 VII chip since it takes twice as many SUN procs to meet the same performance level. That's why we are replacing our Sun's with AIX boxes running Oracle since Oracle license's per core.
I wonder how the Rock will do compared to the P7??
Sun is already in trouble. Our testing showed a 4x advantage per CPU with IBM Power6 over Sun's latest CPU's, under our application and database load. This meant that when evaluating new server purchases, Sun could have GIVEN us the database servers, but the difference in Oracle licensing costs would have still made it more expensive than purchasing IBM servers. So I'm in the process of configuring five new Power6 based p570/MMA servers...
As boring as Vista comments may be - I'm glad so many people are dinging Vista for bad performance everywhere they can. While anyone who is even a junior level IT person knows Vista sucks donkey balls, a lot of people who don't know much about computers are still stunned when they find out vista is crap. They say things like - isn't Vista newer than XP - assuming it will perform better. Then their are MS Fanboys who have replaced their 10 year old PC that was running XP and hasn't been reformatted in 5 years with a new Core 2 system with a couple of gigs of memory running Vista and they think it is the Vista that is making it run OK - not knowing that that new machine with XP would run faster.
Why do I want people to ding Vista - not because I hate MS - on the contrary I've been supporting MS products for over 20 years now. What I hate is Ballmer ignoring the complaints of IT professionals, IT journalists, the public in general and just saying Vista runs fine - it is all in your imagination - look how many vista licenses I've sold (ignoring the fact they originally forced vendors to sell PCs with Vista only until the hardware manufacturers rebelled, ignoring the fact people are buying Vista licenses but using them to upgrade to XP, ignoring the fact they have their first class action against a bad OS, ignoring the fact they had a petition from over 200,000 people saying they preferred XP over Vista etc etc.)
""I agree with "It's so cool because IBM did it! " ... Especially the last sentence!" ....
By Anonymous Coward Posted Tuesday 15th July 2008 10:04 GMT
I also agree, AC, that have Failed to Market and Capitalise on both their Current Real Potential and Future Virtual Strength ....... but that is easily reversed/fixed with a New Creative Marketing Red Team evangelising /eulogising Assets and Tapping into ITs Latent [as in Sensitive Proprietary Commercial Information] AIResearch and dDevelopment .
Interesting article, but is Power7 really going to be 8 core or more like 2 x 4-core? as per the QCM / dual processor cards/books that IBM stuffed out for 595 machines and will probably do for the Power6+?
Good reflection and reaction to your article Ash by TPM...
http://www.itjungle.com/tfh/tfh072108-story02.html
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
