Feeds

back to article IBM's Power7 pitch deconstructed

Some of IBM's Power7 machines have been shipping for several weeks, and the high-end Power 770 and 780 boxes start shipping this coming week. Now, the sales pitching and smooth talking by IBM and its local business partners will begin. What, exactly, will that sales pitch be? That all depends on what gear you have installed, …

COMMENTS

This topic is closed for new posts.

IBM Marketing

I get the impression that IBM is really scared of Intel and the EX/Westmere's that are coming.

I'm not a dyed-in-the-wool or blind x86 religious bigot, but I will point out that those Power7's ain't exactly cheap.

At least from my perspective, it would take a LOT to convert everything (as if that were really possible for us with as many Windows desktops as we have running under VMWare) to AIX or to the Power7 version of Linux. That seems to be something that IBM always brushes over. It certainly would not be a walk in the park to do so.

I find the Power7's interesting, but I think in two weeks to a month and a half, the Power7 will have lost its luster to some extent.

0
0

Well...

Nahh.. it will still be the top dog, sure Nehalem EX will be barging at it's heels but only barking.

Lets try to look at a specint_rate2006 for the current Nehalem clocked at 2.93 GHz, best result is:

203 with 8 cores and 2 sockets.

http://www.spec.org/cpu2006/results/res2009q2/cpu2006-20090511-07354.html

That is 25,4 specint_rate per core.

Best POWER7 result is 1462 for 32 cores

http://www.spec.org/cpu2006/results/res2010q1/cpu2006-20100208-09580.html

That is 45,7 specint_rate per core. The more humanely priced POWER7 will still do 33.5 specint_rate per core.

Do you think a Nehalem EX clocked at 2.26 GHz will make that gap bigger or smaller ?

What Nehalem EX it will do is to bring x86 server on equal footing with regards sockets with POWER7. And that will be the good thing for x86.

Cause right now you might think that a power7 server is expensive. But try to buy a decent x86 server that gives you the same scalability.

It's more or less the same price, if you use a lower clocked power7 power 550 server versus for example a x3950 or a DL785.

Sure monsters like the power 780 are much more expensive, but that is kind of out of reach for x86 servers. Both with RAS and scalability.

// Jesper

0
0
Silver badge
Linux

Systemic Alien Probes for Realities that are Spun Perceptions/Phantom Tales with Telling Trails*

"First and foremost, IBM wants to remind everyone that it has dominant Unix market share ... to get business partners psyched up with a chart like this: .... Here's another comparison IBM cooked up, which kills the ProLiant and Integrity birds with one comparison stone: ..... "

The perception presented by IBM with those little nuggets, Timothy, is that they are in industrial scale decline as a decisive commanding force ...... faced as they are with the SMARTer Virtual Operating Systems with Advanced and/or Artificial Intelligence and/or Information.

* To either Follow Blindly [Sacrificial Lamb Mode] or Dynamically Modify [Global Operating Device MindSet] for Reality is just Beta AI ProgramMING Projects.

0
0
Silver badge

Shock horror probe!

"Ultimately, what customers need to look at to do comparisons is raw data on the systems themselves, then they need to examine benchmarks and other performance data, gather up pricing information, and finally try to weave together a picture that has more depth and breadth than the IBM sales pitch."

You mean IBM marketing's sales pitch might not mean much in the RealWorld(tm) that I live in? And here I was hoping that I could just coast along, allowing the marketers of the world to make my technical decisions for me ...

1
0
Headmaster

Be carefull comparing the SAP ..

benchmarks you are using here.

There is a big difference in the numbers of SD users that a machine benchmarked with SAP ERP 6.0, Enhancement Pack 4 (Unicode) can support and then SAP ERP 6.0 (2005).

First the response time requirements have been cut in half and furthermore you now have to run the benchmark with UNICODE. This means that server 'juice' needed to run a certain amout of users has pretty much exploded. I would say a good rule of thump is a factor of 1.5.

An example is the DL380c

SAP ERP 6.0, Enhancement Pack 4 (Unicode)

versus

SAP ERP 6.0 (2005)

http://download.sap.com/download.epd?context=D572BCCD2B192563572F68E04554632AFF2ECD46FA2B579C45FC0469C0F1379C99C40453761A2066

http://download.sap.com/download.epd?context=40E2D9D5E00EEF7C0DCD4FC7250EF29D3829644CE4C28A5A1C573FBF6F0F9FAE

Now that is a factor of 1.5 times fewer users, for the exactly same machine.

// jesper

1
0
IT Angle

"maximum power usage for data center site planning"

Try to get anyone (including IBM STG) to commit to any other power usage figure, I dare you!

1
0
Linux

What is your problem ?

http://www-947.ibm.com/systems/support/tools/estimator/energy/index.html

// Jesper

0
0
FAIL

Lord of the Flies

So, IBM now has the fastest Unix chipset? Wonderful.

Now, let's do a measurement around Price-performance and compare to Intel Nehalem.

The good old days of scaling up on a Unix server for maximum performance are now gone due to intelligent software algorithms that allow for an application to scale out across multiple x86 compute nodes.

1
2
FAIL

Apples and Lemons

...have different performance characteristics. IBM is doing a dirty trick by assuming that x86 and Itanium don't have virtualization software and therefore a low rate of utilization.

Nobody is using VMWare, Solaris Containers and HP IVM, or what ?

I bet four-core x86 machines with VMWare or Parallels will do very well.

This comparison is like comparing one tape drive without compression to a new tape drive with compression enabled. IBM is playing a ridiculous trick, and you smell it, when you hear "97% reduction in space".

What was the price of a Gallon Of Original IBM Snake Oil again ?

0
0
Go

Power matters

Having recently been on a tour of one of our datacenters, I've gained a better appreciation for why power is a concern beyond the green issues.

A large datacenter has a huge and expensive infrastructure built in to manage power: generators, inverters, battery banks, UPS systems... all of it designed to be able to deliver 'x' amount of power.

If you want to put more equipment on the floor than that system can support, you're going to need more than another extension cord to the local power co. You're going to have to upgrade all that expensive gear. You HAVE to assume worst case (i.e. max power load on your servers), unless you're willing to gamble on never exceeding the rated capacity and having the whole electrical system implode. In fact, you probably should never exceed 80% of your rated capacity if you like being able to sleep at night.

With ever-expanding workloads, the best hope for many datacenters (ours included) is to be able to reduce the power load per unit compute power; through more efficient computers and/or virtualization.

0
0
Boffin

Utilization

Presuming the consolidated workload has a reasonably random dispersions of peak loading per consolidated task, having a higher utilization on the machine doing the consolidation would make sense.

On a small box, more overhead is needed when the box gets hit with a spike in workload if one is to maintain some performance metric. If one consolidates 100 small box workloads onto a larger box, it is unlikely that all 100 processes would spike simultaneously-- so the average workload on the large box can be higher and still maintain some metric such as response time due to a spike in work on one (or a few) of the 100 consolidated workloads... of course each data centre is different, and this doesn't apply to all of them, it depends on what the work is.

0
0

Do they give Trumpets to everyone?

Another exciting day to hear the trumpet blasters. First off, HP calls the Integrity 'Superdome' (http://h20341.www2.hp.com/integrity/w1/en/high-end/integrity-high-end-servers.html) so someone can just SHUT UP about IBM being condescending with that.

Complain about the benchmarks if you like, but they are fair ways to see if two things perform the same.

I LOVE to hear about folks comparing chips to systems. Some people are comparing IBM 750 systems (32 cores lists at $190KUS with 192 GB of memory) to a handful of chips that can be assembled onto a Supermicro motherboard (street price ~$10K). Those are not the same thing at all. The Supermicro cannot hold 192GB of memory, the Supermicro doesn't have the uptime and partition-ability of the 750. It would be amusing to see what kind of benchmark scores the Supermicro would have, they would be less than the x64 machines above. You cannot even get eight 8Gb SAN cards and four 10Gb Ethernet cards on any Supermicro that I have ever seen.

With all of that, IBM clearly is still charging more. I just haven't seen a credible report with apples-to-apples performance. This article certainly isn't pretending to do that, either.

People that compare VMWare with PowerVM just don't understand what PowerVM does. PowerVM is firmware-assisted for both Ethernet and Fibrechannel operations. You implement NPIV so that the WWN follows the partition when you move it about. You implement HIE to the VIO servers so that twin VIO servers share the low-cost 10Gb Ethernet ports. You implement internal-VLANs so that firmware can route traffic once it is inside the server to whatever-host needs them. There is nothing like this (yet) in VMWare. Multi-CPU clients are highly firmware assisted in PowerVM. They are high-maintenance issues for VMWare. Perhaps VMWare will get there, but VMWare must invent everything as CISC software.

Lastly, power. It is clear that many datacenters have their backs against the wall on AC. They would prefer to stretch-out overtaxed infrastructure by purchasing devices that use less power, even if they will only last 24 months. It just isn't a good choice in the long term.

0
0
Stop

Just a Marketing FUD .. No real Vale

Interesting IBM ..

What was costing me 2 Million dollars (M9000-256 core ) will cost me 0.25 Million dollars (32 core IBM) . or you want to sell a 32 Core machine for the 2 Million Dollars ? - If so , where is the Innovation ? (Proof - Show me a customer you haven't sold upgrade with in 6 months of original system purchase on P6)

So you have managed to shrink the computing power required by a factor of 1/8 , Now How will you grow your revenues ? I still have to make my numbers, Pay my employees , give Bonus and Increment , wonder How do i manage , if my revenue is going to shrink by 1/8 by this innovation ?

I really Wish we don't do the same mistakes done by the financial companies, Artificially inflated numbers without any real value - Why not create something useful to the world , which is long lasting than creating unrealistic benchmarks and sell them at ridiculous prices to un-suspecting customer. (I hate this discount ranging from 40-95% , 95% discount .. go to hell ).

But IBM - Look at it this way , you want to replace a 256 Core M9000 with a Small 32 Core machine - Use your engineering Brain , if this is really true , why would any customer buy a M9000 any more ? It is like buying a 386 machine today . HP/SUN - this is end of world for you guys if IBM's claims are true .

I believe your Innovation is just worth , replacing a similar size server with 97% discount from List price and without forced support bundle.

Prove me wrong ..

0
0

What ?

"What was costing me 2 Million dollars (M9000-256 core ) ? "

Damn do you get 80% discount or something.

Shrinking the computing power required by a factor of 1/8 ?

Ehh.. math..

// Jesper

0
0
IT Angle

And when will AIX 7.x arrive to allow us to use all off this new goodness?

So this new CPU will leap tall buildings with a single stride, accelerate faster than a cheeta, have more functions than an iphone but when will the magic sauce i.e. AIX 7.x be released that will set the goodness free? If do not think that most large corps will buy a P7 with Aix 6 to upgarde to Aix 7 in six months - the large part of the solution is still not released.

We have all seen that movie before!!!

0
0
Flame

AIX version 7 ?

AIX version 7 ?

Damn, we haven't even moved from AIX 5.3 to AIX 6.1 yet. And AIX 5.3 is still sold, so an end of Service haven't even been announced yet.

Besiden POWER7 runs both AIX 5.3 and 6.1. And all benchmarks seems to have been done on 6.1 so what are you waiting for ? If you buy 6.1 you are getting 1 year of swma anyway. So it's a free upgrade. Not that I wouldn't let the OS battlehardne a bit first.

AIX 6.1 will at least be under SWMA support up to end of 2013.

Seems like a bit of a rhetorical problem.

// jesper

0
0
Stop

AIX 7

Hmm, AIX 5.3 and 6.1 are still been sold, therefore the P6 plaform are still been sold so I will get it at a discount? Oops forgot - I wil get a massive discount anyway.

The problem is not the support, the problem is that all the real performance enhancements available with the P6 was only enabled with AIX6. Same with P7 - 4 threads per processor for example - till then 2 threads... helps to read the redbooks! Its like driving a Ferrari with low octane - ie. stupid.

Till then - the P7 is just marketing Hooha - ok if you are at a price point (e.g. your three year inclusive support contract is up and the support cost will balloon massively) but if you do not need to move.... stay put. Its hard enough to get the *&%* app guys to move anyway - why would I go through the pain twice, platform move and then AIX move. I'd rather eat worms!!

0
0

WARNING!

I came across some fine print last week: quad-SMT (amongst others) will not work using AIX 5.3. Was in the systems mag, iirc.

And we really should get off our lazy behinds and move to 6.1TL4, I do not expect there to be a 5.3 TL12, so the last two years are already counting.

0
0
Linux

Yeah right.

Ehh.... Sorry but you aren't right.

All results are on AIX/iOS 6.1 or LINUX:

http://www.spec.org/cgi-bin/osgresults?conf=cpu2006&op=fetch&field=CPU&pattern=POWER7

http://www.spec.org/cgi-bin/osgresults?conf=jbb2005&op=fetch&field=CPU&pattern=POWER7

http://www.oracle.com/apps_benchmark/doc/E-Bus-R12-Pay_ORA_Med_IBM_p7-6-core-Single-Audited.pdf

http://download.sap.com/download.epd?context=40E2D9D5E00EEF7CC663DDC83061BA6F9242BF63884801E0C4E97204702C1F27

Operating system, central server: AIX 6.1

So your argument is that I won't get the AIX 6.1 performance documented here unless I go to AIX 7 ?

Damn that is a statement that is close to FUD. Sure when AIX version 7 is battlehardened going from Version 6 to Version 7 perhaps will get you 5 percent better throughput or perhaps a tiny bit more.

But get real. You are talking about a few percent. The real deal is, that for example a box like the power 750 is almost 6 times faster (SPECJbb2005) / 4 times specint_rate than the power 550 it replaces and cheaper.

So please keep that SUN sales pitch to yourself :)

// Jesper

0
0

I hate it when IBM does this to me:

5.3 TL12 will be released on 23rd of April...

0
0

Yes.

Ok 5.3 TL12, that is one more "TL/MP" than 4.3.2/4.3.3. Damn, who would have thought that.

Yes, we need to get moving to 6.1. Also new VIOS version 2.1.3.

// Jesper

0
0
Silver badge
Happy

"Leakage"

I'm thinking IBM are more worried that migrations to x64 mean a lot more going to hp than IBM's x64. On the other hand, I am eager to get some Power7 kit to play with, see what it can do in the real World. It should give a nice boost over some of the older Power6 kit we have.

0
2
Linux

jup.

Yes, much the same here.

Although I think some of the POWER7 models are a bit starved on Memory.

// Jesper

0
0

FUD or not FUD?

I think IBM marketing is best. According to wikipedia article on "FUD", IBM was the first company to embrace FUD in a wide scale. Here we have some more IBM examples:

I really like when IBM claim they can consolidate many computers onto a single IBM machine. It is like their Mainframes: IBM can apparently consolidate 1500 x86 server to a Mainframe. What IBM does not tell you (you have to read carefully) is that each x86 server is utilized at a few percent and the Mainframe has close to 100% utilization. With the same weird reasoning, I can utilize a couple of idling x86 servers on my iPhone. Ergo "iPhone can virtualize several x86 servers!!!". True statement or untrue? FUD?

Also, IBM does not tell you that IBM Mainframe cpus are dog slow. 16 Nehalem-EX cpus can easily match one z10 Mainframe with 64 cpus when we talk about raw cpu performance. So how can 1500 x86 be utilized on some 10ish crappy Mainframe cpus? What happens if the x86 servers start to do some cpu intensive work?

I would not be surprised if the x86 servers that IBM wants to consolidated onto a Mainframe, are 1GHz Pentium 3 or even older. IBM loves to compare new h/w to old - giving the impression of superiority.

.

Regarding this POWER7 consolidation, it is the same. Apparently IBM can consolidate many machines onto a POWER7 box. But IBM does the same old trick again

1) Compare to very old machines with bad performance

2) Assume the old machines have low utilization whereas the IBM box has high utilization.

I wonder how IBM would react to this advertisement from Intel?

"Now we can consolidate 125 POWER boxes onto a 8 socket Nehalem-EX machine! Save money and power! x86 is much faster than POWER" - and Intel dont tell anyone that the POWER boxes are all old idling POWER5, and the Nehalem-EX box is fully utilized. Do you think IBMers would cry out loud and call this FUD? Or would they accept this as a fact, gladly? Would this statement be FUD or not?

.

Earlier I complained on a Sun comparison of an Sun box to an inferior HP box. I asked Sun to withdraw the benchmark. I wrote

"How can you compare HP 4 dual core 1.6GHz vs SUN 4 quad 2.53GHz? Even a dog would understand that it is not fair. This is something IBM could do, but SUN didnt. I thought? But maybe I am wrong?"

http://blogs.sun.com/mandalika/entry/peoplesoft_na_payroll_240k_ee

And in the article on this site here, I am proven correct. To compare new h/w to old h/w is typical for IBM, to try to give the impression of superiority. Not fair play in my point of view. But maybe IBM think this is fair? IBMers would never try to fight a guy in their same size, they only fight smaller guys, and only if the smaller guy is tied up. Fair play does not exist.

0
1
This topic is closed for new posts.