Assure us that the bankruptcy of IBM i now imminent. They made the fastest CPUs of all and now they are surely exhausted by that and terminally sick.
HP Itanium will be the only rescue, right ?
IBM seems to finally be getting serious about Power-based blade servers. Today's launch of the Power Systems 701 and 702 blades, which use the new eight-core Power7 chips, offer about twice the performance of their predecessors, the JS23 and JS43 machines that were delivered using dual-core Power6+ chips last April. The use the …
Assure us that the bankruptcy of IBM i now imminent. They made the fastest CPUs of all and now they are surely exhausted by that and terminally sick.
HP Itanium will be the only rescue, right ?
Is there anywhere that shows real life performance numbers? price/watt/OPs comparison to a x64 chips would be nice.
You said this just last week:
" And I don't see any talk of Power8 blades on the IBM website - has IBM given up on Power blades? "
Well the whole world is still waiting for HP to announce Tukwila systems and IBM is on its second product announce of Power7.....or maybe the reality is the whole world does not care about Itanic.
The real question is has HP given up on rack based systems for Tukwila (they obviously have rack systems for Nehalem).......and you and I both know they have...except for a two socket box next year for telcos. HP-UX customers will be forced to buy blades unless they want to fork over millions for Superdome2.
Cheers from Glenshaw
<Yawn> Oh dear, the droolers really are out early today!
"....the whole world is still waiting for HP to announce Tukwila systems...." Do you even read what you right, let alone fact-check it? In one paragraph you say no hp Tukzilla systems have been announced, then in the next you insist "you and I both know" what those systems will be! Strange, are you using ESP to extract roadmap information from hp? Do you actually work for hp and want to diss your own products? Or are you an hp partner in breach of an hp NDA, maybe? Or could it be you're just wishfully talking from your nether regions? Personally, I'm going with the last option, given the rabidity of your response. Others may think it is just due to the voices in your head.
I'm guessing your only source of information (other than IBM FUD releases) is the vague CIO.com article from last October that said "....HP may introduce a modular, blade-like design for more of its Integrity systems...." - please note that's "may" and for "more of its Integrity sytstems", not all of them, and was on the back of hp having produced NonStop blades. Why that article? Beacuse it's the one I've already seen being used as FUD material by an IBM reseller. Until hp announce the final product, you're just guessing in a loud and childish manner.
PS: and I still don't see an IBM roadmap with Power8 blades on their webby; the Power7 blades announced have zero new features over the old Power6 ones, have even slower cores (3GHz vs 5GHz), and still don't have hot-plug drives; and they have killed the Cell blades. So, would you like your crow roast or fried?
We all know HP's plan is to have only blades and Superdome2. Well maybe a pesky 2 socket box for telco next year.
The problem with HP's tukwila blades is they only have 4 cores and half the cache per core vs. the prior generation. This forces them to have two sockets and why the chips are sooooo sloooow.
When you own 95% of the sales of Tukwila chip the speed bins are based on your bladed design.
I am hearing 4/27 will be date for the BL860/BL870/BL890 blade offerings
Here is the pricing for the systems
Woops those pesky Canadians again preannounced the products.
I'll be you a dinner at Hawksmoor HP does not announce any other hardware until August.
The BL860 and BL870 are decent except for the 4 cores and heat generated by 65nm old fab technology. The interconnect on the BL870 in the front looks like it will cause serious air flow problems.
The BL890 is an architectural albatross. Without the SX3000 chipset it cannot scale efficiently past 5 sockets. Curious why there will not be a 3wide (6 socket) blade offering.
And why is vPar/nPar not going to be supported anymore?
"We all know HP's plan..." Well, the truth is we don't all know, which rather sets the expectation that the rest of your post is going to be more unsubstantiated waffle and FUD.
"....The problem with HP's tukwila blades is they only have 4 cores...." Strangely enough, IBM said qaud-core was great when it was for P6. And dual-core Itaniums still beat quad-core P6 in many deals, which suggests having four cores which can't be kept spinning is really not much of an advantage compared to a dual-core system desinged to be efficient in all stages.
"...I am hearing 4/27 will be date for the BL860/BL870/BL890 blade offerings...." Hearing from where? Your anus? The BL860c and BL870c are current product numbers so it is more likely that hp will use new product numbering. I haven't a clue what those numbers will be but I would suspect either BL880a ("a" as first release of the model, replacing BL860c) and BL890a (replacing the BL870c). Or they could go BL960a and BL970a. Until the hp product announcement your rectum is just guessing.
".....Here is the pricing for the systems..." The link comes back as "page cannot be found". Did you capture a copy or can we assume it was again more farting?
"....I'll be you a dinner at Hawksmoor..." Even without your flatulence problems I wouldn't consider you as an option for a dinner date. I prefer my dates to have at least one foot grounded in reality.
".....The BL860 and BL870 are decent except for the 4 cores and heat generated by 65nm old fab technology. The interconnect on the BL870 in the front looks like it will cause serious air flow problems....." Hilarious! Am IBMer trying to say hp blades have cooling issues! I'm quite certain at this point that you have never touched a blade from either hp or IBM. Did you not see my point above about the way IBM have previously been forced to angle the industry standard memory modules in their narrow blades? That's because the only way they could fit them was at an angle to the mainboard. This means they do not have the spacing you will see between DIMMs on usual motherboards where the DIMMs are at 90 degrees to the motherboard. This overlap in the IBM design reduced the airflow between the memory sticks which is just one area the IBM blades ran hot. As I understand it, the only way round this would have been for IBM to use special low-profile memory sticks, which would have been much more expensive and made the IBM blades even more uncompetitive.
It also meant they could not put hotplug disks on the front of their blades and get enough air through to the RAM, so IBM had to sacrifice hot-plug disks. As a user, we like onboard disk and SAN as they can give us advantages over just SAN disks alone, but we want them to be redundant and hot-swappable. IBM's designs don't meet this criteria, which has limited their appeal to us compared to hp blades.
"....The BL890 is an architectural albatross....." How? Please elaborate with some actual technical argument, otherwise readers will conclude you actually don't have any argument, just dribbling. Well, seeing as there is zero chance you have actually seen a BL890 it's pretty obvious that you are just dribbling.
But, if you want to compare architectures, why not explain why the "amazing" PS701 only has two mezz card slots and two onboard LAN ports, when the much older BL860c has three mezz card slots and four onboard LAN ports? Why does that matter? Well, if you want redundant SAN connections then you really want two fibre channel mezz cards, which means the PS701 then only has the two onboard LAN ports for all networking requirements. Not very good for IP-intensive work such as clustering (usually requires three+ LAN links, ideally four going through onboard and a separate mezz for redundancy) or webserving or virtualisation (two Gb ports is simply not enough for even weiner virtualised servers).
".... Without the SX3000 chipset it cannot scale efficiently past 5 sockets...." Wierd! Why is a replacement for a four-socket blade required to scale to five sockets? Where is the quint-socket Power blade? For that matter, where is the quint-socket any CPU IBM blade!?! If you're going to ruminate on complete irrelevancies you might as well enquire as to why the hp blades don't come in pink, it would be as technically lacking as your other statements.
"...And why is vPar/nPar not going to be supported anymore..." <Yawn> Once again, unless you can post a definitive hp statement to that effect, I'm going to assume it came from your nether regions. IBM have not announced what virtualisation tech will be available on Power8 with AIX7, but that doesn't mean I can immediately claim any future Power8 blade won't have a virtualisation capability.
"....Curious why there will not be a 3wide (6 socket) blade offering....." Here's where it gets really funny - why do they need a 6-socket blade!?!?! Where is the IBM equivalent if it is so vital? Get a clue! Better still, get someone else to get a clue for you, as I don't think you're capable of even finding a clue to start with.
I am very disappointed in you. HP must be keeping you in the dark. I think HP is keeping most customers in the dark though.
The URL above works for me. It is a spreadsheet with the new blade names and prices.
Just search for bl890c and ontario in google
AH384A HP Integrity BL890c i2 c7k server blade TQ $31,273.00
AH385A HP BL8x0c i2 Itanium 9310 2c Proc Kit TQ $3,400.00
AH386A HP BL8x0c i2 Itanium 9320 4c Proc Kit TQ $4,352.00
AH387A HP BL8x0c i2 Itanium 9340 4c Proc Kit TQ $7,854.00
AH388A HP BL8x0c i2 Itanium 9350 4c Proc Kit TQ $14,409.00
plus ddr3 memory
Sorry, Ms Kebabfart, but I don't usually Google on principle. The search didn't work on Yahoo so I slapped it into Google and promptly got a security warning! I think I'll just email our hp rep and see if he admits the price list did exist. Either way, all you have managed to ascertain is the model number (BL890c) and a few option codes, which hardly goes any way to confirm any of the wild suggestions you and the rest of the IBMers have made. Try again!
Can someone please call jlocke's Mommie? It seems that, with regard to technical comment, any time is long past his bedtime. I wasn't going to bother posting about the new Power7 blades, but jlocke' stupidity just painted a kingsize bullseye on the products.
"....HP Itanium will be the only rescue, right ?" Well, if you want to play the game, I suggest you at least get someone to help you with your posts. Like an adult, maybe. Pref one with some technical knowledge and experience in the industry, and not given to just regurgitating the IBM sales guides.
Anyway, since jlocke is so ready for a fight, would he like to explain why IBM still can't offer hot-plug disks on their so-called enterprise blades? At the moment, if the onboard disk fails on the IBM blades the whole blade has to be taken out of service and then out of the chassis to have the disk replaced. And since the internal disk slot seems to be a loner, there is no option to mirror that disk to make it resiliant. So, pointless as a boot disk, which means you have to add external storage just to get the mirrored boot disk that is just about essential for real enterprise roles. You have to go to the "snap-together" PS702 blade to get two disks in the blade, and they're still not hot-plug.
Compare that to the current hp Itanium blades, where even the single-width BL860c Itanium blade has two hot-plug boot disks and on a RAID controller, not just a "dumb" connector. The double-width BL870c has four! In fact, hot-pluggable boot disks have been on Itanium blades since their introduction, years ago. Why, after so many generations, can't IBM manage to put hot-plug disks on their blades? Could it be because they can't do so and fit enough memory on their narrow blades design? Are IBM still plugging the memory in at an angle to get it to fit?
And then we get to the CPU. What, only one option, and only 3GHz? What happened to all those IBM boasts of having 5GHz chips across all the IBM Power servers? Well, OK, admittedly IBM have failed to deliver a 5GHz Power7 CPU as promised, but at least the rack servers will have a 4.14GHz P7 offering when they're finally available to customers (down from the 5GHz P6). Seeing as power and cooling have always been the Achilles heal of the IBM BladeCenter chassis, I'm guessing that anything more than the 3GHz Power7 chip will cook itself in the H chassis, or suffer brown-outs due to IBM's poor power design. I suspect that IBM would have to upgrade both the blowers and PSUs (again) to get the 4.14GHz P7 chips into their BladeCenter designs, and that would be incompatible with the current range (again). Please do try and post a technical argument if you feel you can disprove that idea, not just the usual whining.
At this point I suspect the majority of the IBMers would also be quite happy if jlocke just dried up and blew away seeing as how he doesn't do them any favours!
Who cares about blades, It's a dead end anyway.
As for the BL870c it's a really nice product.. I really mean that Well engineered solid product, just to bad it is slow and expensive.
A 4 socket BL870c that does 114 on specint_rate2006 with 4 9150N's and 16GB RAM costs 40KUSD without OS.
Now for that price you can get
A 2 socket PS702 with 16 cores and 64 GB RAM which does 520 specint_rate2006.
Wait I mean you can get two PS702 for 38KUSD so who cares about hotswap disks... I'd boot from SAN anyway.
Now what is that.. a factor of 10 in price performance ?
Have a look at this benchmark btw.. so much for Nehalem-EX being cheaper than POWER.
"Who cares about blades, It's a dead end anyway...." Jesper, I'm shocked you could even post that! Both IBM and hp have stressed how important blades are to them, and their sentiments are echoed by the likes of Gartner and IDC, which both correctly predicted blades would be a massive growth area. To simply write them off like that comes accross a bit petulant, like a child sulkily saying "well who cares what anyone else thinks!"
Then you wander off into more irrellevant benchmarks. Please, show me the one busines that runs specint_rate2006 as a businees as opposed to real World apps. And then you don't have any figures for the new Tukzilla blades to compare to the carefully crafted PS701 or PS702 results, instead you compare with the old BL870c with the slower 9150N CPUs and DDR2 memory. The new Tukzilla blade is going to be much quicker than the BL870c just through using DDR3 memory, so your comaprison is quite pointless, even before we consider that the PS702 you mention had 16 cores and therefore double the software licence costs of the BL870c (not surprised you didn't factor that into your cost analysis).
As to all your posts which relie on vendor benchmarks, I shall reply with the usual advice - until you run a server in your environment, with your OS build, software stack, LAN and SAN, you cannot accurately predict performance, and any vendor benchmark should be treated at best as guidance. I bet you can't get IBM to guarantee every customer will get the same specint_rate2006 result in their environments and they got in IBM's labs. I know you won't as every time a vendor salesgrunt tries to push vendor benchmarks I simply ask them to guarantee I will get that exact performance, and so far it has proven a great way to shut them up!
I still don't like blades. I think they are only good for HPC like workloads.
As for using a benchmark on the BL870c, it is not like HP is shipping anything else.
Sure we can take the new bl860c i2 with Quad-Core Itanium 9350 it does 134 specint_rate2006 (128 base) with 2 sockets and 8 cores. And sure that is a good improvement compared to the old blade, wich did 114 on base.
That is how much you get from DDR3 and Quickpath, newer compilers and increase in Mhz and Turbo Frequency.. It's... 12% exactly the same as your clock frequency increase.
But still.. you are only packing 4 cores per chip and 4 cores that are significantly slower than the competition.
Still you have to compare that with the specint_rate2006 number of 520 for the 2 socket POWER7 blade. It is a factor of 4 times faster per socket..
And yes software will be more expensive.. or wait.. 520/16=32,5 specint_rate per core versus 134/8=16,75 specint_rate per core.. Ohh.. the Tukwila blade is still x2 slower per core...
That is why Oracle only charges you 0,5 licenses per Itanium core.. cause.. it is ... slow.
And your argumentation with benchmarks is flawed, Get real factor of 4 per socket. Itanium blades are slower and more expensive, it really is that simple.
"....I think they are only good for HPC like workloads....." Great for consolidation of existing UNIX systems too! And seeing as the word (subject to official hp confirmation) seems to be that hp will have the means to link their blades into larger SMP instances, it looks like they will be even more useful for replacing existing larger rack servers.
"...As for using a benchmark on the BL870c, it is not like HP is shipping anything else....." Well, IBM aren't shipping the PS70x blades either, yet. Please go try and get a delivery date, until then you are comparing IBM vapourware with existing hp kit.
And then you go off into specint fantasyland again. I'm losing count of the number of times I've told you that vendor benchmarks are completely irrellevant - did IBM do their usual trick and have a sub-$100k server using $4m of short-stroked storage for their benchmark? I'll tell you what, why don't you try and reproduce the IBM figures in your envirnoment (when IBM actually ship the PS70x blades that is). Then try with getting a whole chassis of blades to hit the same figure as your first run. You'll probably find the drop in available bandwidth as the blades share switch resources means the average performance drops to well below the best solo specint figure you can get, or that they brown-out or overheat as the IBM chassis just can't run all their blades at full load. Either way, you won't get the IBM lab figures, and they still won't show how your blades will behave with a real World app stack. I see you still don't supply the name of a business that runs specint as it's main app.
"....And yes software will be more expensive...." Yes, which means you can either do the job with hp kit and save money, or buy more hp kit for the same price and do more than the IBM solution. Come on, you IBMers have been FUDing Niagara for years on licensing costs, it's not like you can suddenly pretend it doesn't affect Power7.
".....Itanium blades are slower and more expensive, it really is that simple." And again - until you have benched both you can say all you like, but you're talking from your arse.
...thats what those hard disks will be doing, along with the RAM, chipsets, and the bladecenter connectors.
We ditched the ppc970 blades after we found they ran too hot frying their components, did to little, and the bladecenter support interfaces were simply crap. More IBM shovelware to sell to vulnerable golf-type managers who think that a blade server is still a cool idea.
You found one aspect where HP is better - disk hotplug. Maybe IBM customers don't demand that from a blade, I don't know.
But Power Consumption was always the weak point of - Itanic ! HPers themselves told me that their customers wanted PA Risc instead of Itanic because of that.
Power efficiency probably is the strong point of the SPARC multicores. But Power clearly is there to run existing, often single-threaded, apps to the max. That you cannot get without extreme power consumption. It does not matter whether it is Intel or IBM, and does not matter what architecture.
"....You found one aspect where HP is better..." Given hp's lead in installed blades marketshare I'd suggest us customers found a few other areas besides just hot-plug disks.
"....I don't know...." Yes, I'm beginning to think the boundaries to what you know encompass a very small area of either experience or technical knowledge.
I have noticed that the datasheets for the new IBM Power7 blades don't have any details on power consumption. The IBM EnergyEstimator also doesn't have any guidance on the new blades. Are we looking at the possibility that IBM have yet again produced a blade product only to find their chassis cannot supply enough power to run all the blades at load and a full set of switches? Are we going to see more IBM excuses about customers running most of the blades at 20% util for 80% of the time? Yeah, that should really work when we hit that end of year peak!
"....Power efficiency probably is the strong point of the SPARC multicores...." Ah, so now we see why you know so little about the hp or IBM kit - you're just another Sunshiner! No wonder blades are such a mystery to you given that Sun's blade offerings have always been comedic!
El reg readers really enjoys your posts.
LOL to you Matt!
Hot plug disks? Since when this really matters?
Do your blades boots from SAN or what?
About core frequency that happens to all vendors when you go into multi-core designs.
Intel did that from single core to dual cores, then quad, hexa, octo etc... (Sorry, Itanic stops at quad). Dude, there is no x86 running at 8GHz - 10GHz!
You are really funny Matty!
Performance is not related to clock frequency, I really thought you were smarter...
Now, you are showing to El Reg readers that you are just a clown!
Paris, cause she likes hot plugs too
"....Hot plug disks? Since when this really matters?..." Well, we found them useful enough that we more often buy hp blades than IBM ones.
"....Do your blades boots from SAN or what?..." Not all of them. In fact, we have some blades in regional offices without any SAN connections. How do you suggest we boot from SAN for those?
"....About core frequency that happens to all vendors when you go into multi-core designs...." But all you IBMers have been FUDing Itanium for years on the claim that Power6 had to be better just and only because it was 5GHz, and that Power7 would be even faster! Well, P7 is here, and it has a slower clock than P6, and the P7 blades will be further limited to 3GHz cores. I'm not surprised you're now backing away from your "frequency uber alles" blather.
"....Sorry, Itanic stops at quad...." Seeing as dual-core Itaniums have been beating quad-core Power in real World applications I can't say I'm too worried about how a quad-core Tukzilla will measure up against an octo-core P7, especially as the cores on the P7 are a step backwards to the P5 design. Until I have both to bench, I can happilly ignore your frothing and FUD.
"....You are really funny Matty!...." Gald you think so. Unfortuantely, I can only describe you as tragic.
"....Performance is not related to clock frequency...." Please tell the other IBM FUDers. Please write it on a Post-It note and stick it on the edge of your monitor, so the next time you are comparing Power to other chips and you feel the urge to squeal on about "frequency uber alles", you'll realise how stupid you sound and save yourself further embarassment.
I've just found the IBM Redbook for the new Power7 blades (http://www.redbooks.ibm.com/redpieces/abstracts/redp4655.html) and it descibes the memory sticks as "VLP", which I assume means "very low profile", probably to get them to sit at 90 degrees to the motherboard - looks like IBM had to go for specialised and very expensive memory rather than the standard DDR3 Tukzilla will be using. It also has some interesting diagrams for the PS70x blades, showing that IBM had to put the internal disk right at the rear of the blade so they could put the RAM at the front where it stood a chance of staying cool. With the PS700, where they have two internal disks, they had to sacrifice half the memory slots, and still could not make the disks hot-pluggable! Please also note you have to drop from 1066MHz to 800MHz when using the larger IBM memory sticks. More interestingly, it lets slip on page 46 that the 3GHz Power7 CPU in the PS70x blades only has a single memory controller to feed all eight cores - no bottleneck there then!
I'm going to stop there as I'm not getting paid to bash IBM kit, even if it is amusing to see how it makes jlocke and co spin like an Iraniam centrifuge! As I said before, at best it's all conjectuire until I actually get some kit to play wiht. Now, someone please take AC back to the nursery and put his toys back in his pram for him.
"Seeing as dual-core Itaniums have been beating quad-core Power in real World applications"
What quad-core Power? Power6 and power5 were dual core. I'd like to see where the dual core itanium beats a dual core power6 box. Do you have any evidence at all or any benchmarks you can point us to? All I can see is your babbling and hand waving. I think everyone here thinks that you are a joke.
"especially as the cores on the P7 are a step backwards to the P5 design."
Where do you get this nonsense from?
"IBM had to go for specialised and very expensive memory rather than the standard DDR3 Tukzilla will be using"
If it very specialised and high cost, then how come an equivalent Power7 blade with the same amount of RAM, double the amount of cores and 4x the performance is going to cost a lot less than your beloved upcoming Tukzilla baldes? Your posts are a joke.
"....What quad-core Power?...." Try Power5+ QCMs, which IBM flogged as "quad-core". Montecito versions of dual-core Itaniums walked all over them in our tests, let alone Montvale. Oh, and thanks for pointing out that IBM customers going from a current Power6 server to the new Power7 ones (when they eventually ship) will see their software licence costs quadruple. Now, how much of a point did the IBM FUD make of this with Niagara and Nehalem? A quick look through posts to these forums will show you how hard IBMers hammered that point, but suddenly it doesn't apply to Power, for some unexplained reason.
Just to make it clear to you IBMers that obviously don't actually work with enterprise servers, just having an "uber fast" CPU is not going to solve all your problems. For the last ten-plus years the industry has been more concerned with implementing high-speed connections between CPUs and memory, disk, SAN and LAN, because CPU speeds were well ahead of what the rest of the system could supply. Whilst IBM like to concentrate on a very tiny set of benchmarks that highlight pure core throughput, the reality is this is curently not the issue with real World applications. The IBM TPC record was a good exampe where IBM had to use a rediculous database and storage setup ($4m of SAN!?!?!) to keep the Power chips spinning. Even IBM have been forced to admit the doubling of CPU core speed in P5 to P6 only produced a 10% gain in performance at the coalface, which just goes to show how badly IBM did on optimising the rest of the system.
Instead, hp have concentrated for years on making the whole system (OS, CPUs, interconnects, memory, etc) work better rather than just selling CPUs that spend most of their time idling. Now, if that wasn't true, hp would never have sold a single Integrity server if Power made the pSeries just so gosh-darn fast as you lot like to claim. The facts say otherwise, indeed the recent Gratner figures as discussed here on the Reg forums show hp are picking up more of the high-end deals, not IBM. Seeing as there is no way that massive chip on your shoulder will let you ever consider any benchmark I can post to prove that Integrity can beat pSeries, I'll just let the benchmark of the market show how wrong you are.
"....Where do you get this nonsense from?...." Please try and deny that P6 was a diferent architecture with in-order execution, whereas P7 is a return to out-of-order execution, as last seen in P5. Please also try and deny that P7 pics look just like lots of P5 cores squeezed onto a die. Why "please"? Because I need a good laugh.
".....If it very specialised and high cost, then how come an equivalent Power7 blade with the same amount of RAM, double the amount of cores and 4x the performance is going to cost a lot less than your beloved upcoming Tukzilla baldes?...." I'll tell you what - when you actually have two quotes for the same solution from hp and IBM to meet the same real World case, then we'll see. Until then, any pricing is just sales waffle as neither product is available to us customers today. It wouldn't be the first time a product was announced at one price and there was a price hike by the time it actually hit the streets.
".....Your posts are a joke." <Insert the sound of laughing here. At you.>
"Seeing as dual-core Itaniums have been beating quad-core Power in real World applications I can't say I'm too worried about how a quad-core Tukzilla will measure up against an octo-core P7, especially as the cores on the P7 are a step backwards to the P5 design. Until I have both to bench, I can happilly ignore your frothing and FUD."
Ok you don't really have that much knowledge about POWER do you ? Or .. well I think you are playing tricks.
The only Quad Core POWER that has been shipping is the POWER5+. And it's actually not a real Quad core but two dual cores in an MCM. Just like the very first Quad core Xeons.
So you are basically taking POWER5+ from 2005, and comparing it to a unspecified dual-core itanium, Must be Montvale which is ok. Montvale is from 2007.
You don't really specify how you measure faster.
Now can Montvale beat a QCM POWER5+ on any commercial workload ?
On single threaded performance I'd say Jup. No problem. The POWER5+ core relies on SMT to get throughput. So if you took the lowest clocked POWER5+ QCM which runs at 1.5GHz and matched it to a 1.66GHz (9140) Montvale I'd say that the 1.5GHz POWER5+ would get beaten by as much as 60-70%.
Itanium have always been about single threaded throughput and it's still good at it.
But if we then look at the per socket performance. No way. Sure there are people who are terrible at setting systems up. So sure you can configure your server so badly that it runs like hell.
But for something like SAP for example, then my sizing data puts a Montvale 9140 at around 1500 SAPS per core/3000 SAPS per chip. And a POWER5+ QCM@1.8 GHz at around 5600 SAPS/QCm or 1400 SAPS per core.
Again Itanium is faster per core but slower per QCM module.
All SAP ECC 6.0 numbers.
So what you said isn't really technically wrong... But you are comparing a current version of HP's Itanium line to the generation before the generation of current shipping POWER products, and again you are just using 'our own real life performance' as a
BLEH. I meet guys like you in our Solaris depardement and our HPUX department all the time. Clinging on to their jobs, desperately trying to shoot down the business case for a move off their platform to POWER, with strange arguments just like you are using. I've heard so much sh*t these last few years.
"Yes but Real life performance ... , What RL performance ? You are using a bloody M5000 for what your colleges across the room are using a power 570 for. And they are sharing it with 8 other projects."
"Oh.. no IVM is not something we use.. it is much better to use npar and vpars.. We know that and... " your bloody conservativeness is putting us out of business.
The facts are that it's very hard for us outsourcer like us to stay competitive, in the Unix area if we don't use POWER. That is just the way things are, and the ones you should be barking at is HP and Oracle. Just like the rest of us is screaming Make better products, so that there can be competition. Competition is good.
Like for example when we lost a big chunk of a customers workload here some time ago. Oh no lets not offer them a migration project off SD to some POWER7 hardware, with overcommitment. No the technical Unix people on the account says that we'll loose if we do that, and they know the bla bla bla.. We lost.. Big time.. to .. EDS!!! of all. And what are they going to do.. migrate it all to x86 blades. BLEH.
And nice.. reading manuals and redbooks trying to find weaknesses. "IBM had to go for specialised and very expensive memory ". Who cares as long as the blades have superior price performance compared to the competition..
So you are pointing at other crying FUD, but remember when you point at someone there are three fingers pointing back at yourself.
Yes, desperate is the word I would use to describe the whole Power7 launch. Terrified of Nehalem and Tukzilla, IBM have announced more vapourware that won't be available to customers for months (high-end probably not until 2011?), and then won't be able to perform to its best until a new version of AIX arrives (maybe late 2010, more likely 2011?) and all the applications are available for that version (at best 2011). As regards your reply, I'd use the words presumption and avoidance.
"....So you are basically taking POWER5+ from 2005, and comparing it to a unspecified dual-core itanium, Must be Montvale which is ok. Montvale is from 2007....." Presume again. The 2.2GHz Power5+ systems in question didn't arrive until 2006, and went up againt Monetcito Itaniums in an hp Integrity Superdome hardware partition that year. The Superdome solution (tested with RHEL and hp-ux) was faster and cheaper than the IBM solution with an Oracle/SAP stack and an EMC array. We let IBM and their reseller do the tuning for the IBM solution over a two week period. The hp solution was faster out-of-the-box. When we looked at the upgrade to Montvale we went back and tested it against a revised IBM solution using P6 CPUs - IBM still lost.
"....All SAP ECC 6.0 numbers....." And here we go with the avoidance. More vendor benchmarks to try and imply every business runs just like an IBM labs benchmark session.
""....IVM is not something we use...."" As for IVM, I happen to like the product and championed its use in our own consolidation work, so if your own "experts" advise against it that's not my problem. It doesn't mean we don't have Integrity servers using npars and vpars because in some situations they are the better option. I suggest you do a lot more reading up on the hp Partitioning Continuum before condemning the use of vpars or npars. Then again, maybe not as I really don't want an outsourcer getting better at their job, thanks!
"....And what are they going to do.. migrate it all to x86 blades....." Where do you think a large portion of our SPARC-Slowaris went? I would have loved to port it all to hp-ux, but it was a much better business case to move it to RHEL on ProLiant. We have also replaced some older Power and PA-RISC servers with Lintel for the same reason - it is much cheaper, especially on hp blades! I'm pretty sure if your outsourcing salemen had had an xseries and Linux solution ready to put up against EDS's offering then you might have stood a better chance, but - as with most IBM-centric salesteams - yours seems to have thought AIX-on-pSeries and nothing else. Which reminds me of the same attitude from the old SPARC-Slowaris salesgrunts that used to try selling to us, and look what happened to them.
".....Who cares as long as the blades have superior price performance compared to the competition...." It's easy to postulate three reasons customers will care. Firstly, specialised memory will be more expensive (just through economies of scale), so when customers look to upgrade beyond the subsidised memory prices in the standard bundles they will suddenly find the RAM sticks are pricier. Secondly, specialised RAM will lag in development behind standard RAM, which means IBM will either have to pay more to keep abreast (making the RAM more expensive again) or lag in offering RAM upgrades compared to competitors. Thirdly, specialised RAM reduces the choice of suppliers, meaning IBM cannot take advantage of the number of RAM manufacturers as a server vendor using standard RAM sticks, so what happens if your limited choice of RAM suppliers (probably just one) has a problem delivering to demand or a problem in fabrication? All three points will impact the price/performance comparison, which is why us customers will care.
"....And nice.. reading manuals and redbooks trying to find weaknesses...." Why are you so upset that a customer should want to read the publicly available documentation that explains the product? Isn't that the point of the Redbooks? Or do you think us customers should just shut up and take your word as gospel? Not likley! Think of it this way - I'm an amateur FUDist at worst, but the technical marketting people from hp, Soreacle, Fudgeitso and even CISCO will be going over all the IBM documentation with a fine toothcomb, looking for anything that can be pointed to as a possible issue (or FUDed, if you prefer) in the new P7 designs. Given the resources they have I'm sure they will do a darn sight better job than I can, so if you find my amateur attempts (sparked by jlocke and co's stupidity, not mailce), then I think you're going to really cry like a baby when those experts get to work! And don't act so innocent, IBM do exactly the same with competitor kit.
/Shakng my head in disbelief but SP&L.
No matter how many times you keep saying the same mantras over and over again, it won't make them true. Wake up.. smell the new world order. You repeat the same and the same and the same and the same and the same things again and again. It's like you are trying to convince yourself against your better judgement.
"You need AIX version 7 that is not here to get good performance out of POWER7"
Power7 doesn't need AIX version 7 to run fast. Sure AIX 5.3 is not going to give you the performance, that AIX 6.1 has, but it's not like that is a secret. But with the kick butt benchmarks it does right now with AIX 6.1, where it is absolutely cleaning house. Then I am really looking forward to AIX version 7. If you are right then .. MAN is it going to run fast with AIX version 7... Maaannn..
"Benchmarks are irrelevant "
No matter how much you keep disbelieving benchmarks. The facts are that this is what sizing data used to sell solution are made upon. This is how it's done almost everywhere. But I guess where you work is different.. I can just imagine.....
HR: Hello is this IT ?
IT: Yes it is IT, what can we do for you?
HR: I would like to have 4000 Extra SAPS allocated for our HR system.
IT: Ok, it will take 3 months, then we will be finished.
HR: WHAT ? 3 Months.. but we need it in a week.
IT: Yes, but we have to get the vendors in and do benchmarks, we don't thrust sizing tools, and benchmarks.
HR: You, gotta be kidding.
IT: No, 3 Month minus one day to benchmark, and then one day to setup the system on the winning platform.
HR: Well if you say so. <CLICK>
And what is fun is that when benchmarks suit your purpose you have no problem quoting them.
But wait this is even better, you really like to link to HP's case study website:
Which surely is much much better than benchmarks. Yeaahh.. right. In Denmark we have an expression that goes like this:
You shouldn't throw with stones when you live in a glass house.
*CACKLE* Damn if you weren't so desperate you would be funny. Lets see what you found.
1) Single Memory controller.
That is actually kind of bull, as the blades has better per core/per GHz throughput than the POWER 780, who has both enabled.
2) That the Memory modules were angled for better space utilization. and is special and will be more expensive...
Yeaaaahhh.. right. Again.. the POWER7 blades are much cheaper than Itanium ones so what's the problem.
3) Lack of hotswap disks.
Now that is a valid point, if you want to boot from external disks. But hardly something worth paying many times the price performance for. But then again most people don't boot from internal disks today. Or they use partition migration and move the workloads to another blade when they want to service the blade.
And then some strange quotes from someone that POWER blades are hot.
Let me tell you young man, that blade systems are hot. Damn hot, that be either HP or IBM. You don't put blade lots of blade chasis in a single rack unless you have a cooling system that can deal with the heat flux. That is why the nice people at HP and IBM make really nice manuals that you should read about how the air flow and cooling capacity.
Jesper, you really need to take a chill pill!
".....Wake up.. smell the new world order....." Yeah, I think you'll find that's called x64, not Power. You've overdosed on the IBM marketting and need to really sit down and think before saying such laughable nonsense as Power7 being "the new world order".
"....Power7 doesn't need AIX version 7 to run fast....." Yes, but you won't get the best out of Power7 without AIX7.x. I see you're back to your avoidance issue again.
".....No matter how much you keep disbelieving benchmarks....." Because I've seen vendors fail to deliver what the shiney marketting brochures said before. If you are so happy to believe in benchs then that's your funeral, just don't expect the rest of us to be so gullible. I also see you still don't post the name of a company that does nothing but run specint_rate2006 as a business, so you're still avoiding that one as well.
"...I can just imagine..." The problem is you actually can't. Because for me to say yes to such a request I would have to have confidence that we could actually deliver that additional performance on schedule and at minimal disruption and cost. Which is why we insist on POCs before purchase, so we can actually get a real World view of performance, and then can confidently predict how to scale up as required. We're not so stupid as to simply say "Yes, we can do that 'cos the salesbod from IBM, who knows SFA about our enviornment or business, says so." If, however, you are happy to do so then this goes a long way to explaining why you're being beaten in deals by EDS.
"....And what is fun is that when benchmarks suit your purpose you have no problem quoting them...." You'll usually find I only quote benchmarks to show up others that base all their arguments on benchmarks because they have no hands-on experience. But please try and pretend that an IBM server will deliver the same performance in real World use as it does in a carefully crafted IBM labs benchmark session. You can't and neither do IBM, as they do not guarantee that customers will get the same figures.
You then get very uptight about the three bits of example FUD I mentioned - the angled RAM on the old IBM blades; the single memory controller; and the lack of hot-pluggable disks. As I said before, you should be worried because that took five minutes work by an amateur, and the vendors will have their own "competitve analysis" going to their marketting teams soon.
But the real icing on the cake is yet to come! "....That is why the nice people at HP and IBM make really nice manuals that you should read....." But when I posted comments after reading the IBM Redbook for the PS70x blades you got all uppitty! Seems like I just can't win, I'm damned if I do and damned if I don't. Or is it that just in your eyes I can't win as long as I think for myself?
"Jesper, you really need to take a chill pill!"
No, I don't do drugs. But some homebrewed beer later on helps with the allergies, will be needed.
".....Wake up.. smell the new world order....." Yeah, I think you'll find that's called x64, not Power. You've overdosed on the IBM marketting and need to really sit down and think before saying such laughable nonsense as Power7 being "the new world order".
No I am not overdosed on IBM Marketing, they are just as big dorks as SUN and HP marketing.. perhaps even a bit dorkier.
As for the x86 new world order. Well I've been hearing that for the last 10 years. It still hasn't happned. There have never been sold as much UNIX capacity as there is now. It's kind of like the "Mainframe is Dead", in my last job we broke the companies record for buying mainframes 2 years ago, in the mids of the recession.
"Yes, but you won't get the best out of Power7 without AIX7.x."
Sure it won't. But that doesn't change that what you are getting right now on AIX 6.1 is more than good. And your argument is stupid, don't buy this product cause it'll get better with time. I don't think I need to say more.
"Benchmarks and sizing"
Matt. Sizing isn't an exact science. I've been doing it for 10 years+. The problem with sizing is that you have to know what you are doing to get a pretty decent result. And normally no platform will deliver what a benchmark promises. But you can't run vendor benchmarks all the time, or perhaps your company has. Fine with me, if it makes you happy then it makes you happy.
I've done a shitload of benchmarks, as a consultant. I've done EMC benchmarks in Cork to showed the 'flight envelope of DMX 2000", I've benchmarked some of the very first x86-64 machines versus SGI versus p595. I have even benchmarked code on some of the whiteboxes Merced beta machines from Intel.
And sure as an IT manager/CIO you get a CMA report that you can use to say "I did everything I could it isn't my fault". But when that is said, I've never been in a situation where we/I didn't hit the expected result +/-10-15%. Which was based upon what.. a sizing
But we are others that doesn't have that luxury. We don't have 3 months for benchmarking, so we actually have to sit down and try to understand how things work, this is how it is with most companies. And besides 80% of all sizing is upgrades anyway, hence you have a source system and a target system which makes things much easier. But for real new workload sizing, that hasn't become easier. Today some of the processor architectures have what the IT press have called accelerators. If your application can utilize those then you get the 'benchmark' performance, if it can't you don't.
Does POWER7 have accelerators that you need to be aware of ?
Well according to you, No. Cause POWER7 is just a shrink of POWER5 Quote "Please also try and deny that P7 pics look just like lots of P5 cores squeezed onto a die. Why "please"? Because I need a good laugh."
Well sure it has, I can understand why you need to do benchmarks, cause you don't know what you are talking about. POWER7 have several, what the IT press calls, "accelerators", it has a Vector execution unit that is now fused into the FPU unit. In POWER6 it was a separate unit. Furthermore POWER7 has a DPU unit. A unit that do decimal floating point math, which will make your financial software go like a rocket if the software uses it.
Now the same goes for x86, and for the Venus SPARC processor.
And btw POWER5 have neither of those, nor did it have a recovery unit like POWER6 has which is now worked into the individual execution units in POWER7. Now from the looks of it, haven't really seen any hard confirmation, POWER7 can issue two instruction groups, yes instruction groups just like on Itanium, per clock cycle just like POWER6 compared to one group on POWER5.
Statements like the above only gets you put in the same box as Kebbabert.
And I must admit that you scares me that you can't accept a request on your system for additional capacity without doing a vendor benchmark. But hey looks like your IT depardement is populated by BOFH's. Which is nice++, only problem is that depardements like that usually gets outsourced, at some point in time cause Management wants to get rid of them. And that usually has a tragic ending, seen that way to many times.
"As I said before, you should be worried because that took five minutes work by an amateur, and..."
Yes you are right it took an amateur 5 minutes to find those, hence the validity of the claims.
And as for other vendors finding FUD, you know the problem with FUD is that if it is desperate enough it backfires. And that goes for every vendor IBM,HP,Fujitsu or whatever.
"...Statements like the above only gets you put in the same box as Kebbabert..."
Oh, is it so? And how about your statements that "despite you need four POWER6 to match two Intel (ordinary) Nehalem, the POWER6 is faster"? Or, when you claimed that because Niagara T2+ had an Hz upgrade from 1.4GHz to 1.6GHz it could be considered a new generation CPU, playing in the same league as Nehalem-EX, POWER7, Niagara T3, Venus, etc? I claimed that POWER6 and Niagara are in the same generation, and POWER7 and T3 are next gen - you said: No.
These above are just some of the weird comments you have spewed out. I have many many more weird comments from you. I, OTOH, always asks for proofs, and links to benchmarks and white papers when I claim something - you OTOH always rejected my benchmarks as ("cherrypicked by Sun", "Carefully crafted by Sun", and what not). I have always accepted your benchmarks. The best thing is, when recently, you claimed that "I have changed and now supplies benchmarks and white papers to my claims" - that is pure bull shit from you. That is only an attempt from you to make look like I am an FUDer coming with unsubstantiated claims.
I have talked lots about mathematicians being taught to supply proofs, and to ask for proofs. Always. As a mathematician I do that. You are just FUDing again, but, now about me. I also asked for proofs from you: prove it, show that I am not able to backup my claims. You have never supplied such a proof. FUDing again. Just like the rest of IBMers.
So, these remarks put you in box of your own.
Cool it Keb.
As for what is fair game to compare with regards to systems.
IMHO If you are shipping (and it is not a legacy product), or have announced priced/benchmarks etc of a product/system Then it is fair game to compare against it.
And what do I mean with a legacy product, well IMHO it wouldn't be fair to compare Nehalem-EX, Westmere -EP, POWER7 based products to for example the SUN Netra T2000, which is still a product you can buy from Oracle. Then there is the T5120, that wouldn't really be fair either. You could argue that on price/performance it might be fair but otherwise no.
But it is IMHO fair game to compare with T2+ products, and your insistence on comparing with a product (T3) which isn't here and to which we only have very little information is not serious.
And you have NO WHAT SO EVER, problem with comparing to T2+ products when it suits you, like the SUN T5440 cluster TPC-C benchmark. sooo..
Now, comparing against legacy products and mismatched systems, that is what the FUD machines of the Big Server Vendors do. Cherry pick, trying to find weaknesses etc etc. That is why you should always check the context of a comparison on a vendor site. And this goes for all vendors, that is why I constantly say to you that when you pick case studies, or link to comparisons on vendor sites, that this is not the way to do it.
I have no problems with vendor sites that list their own benchmarks, no problem, but when they start to compare and glorify their own products I say no. And cause you have in the past mainly picked from sources like, I use the terms cherrypicked by SUN or the like. If you had done the same from an IBM or HP site I would have used just the same terms.
Don't blame it on me that you can't tell marketing from
But comparing against T2+ systems against current x86 systems and POWER and Itanium is totally fair. You on the other hand with your hand waving about T3, and that it's another generation than the T2+, and that it is only fair to compare equal generations .. bla bla..
T2+, SPARC64 VII, Nehalem-EX, Westmere-EP, Itanium Montvale, Tukwila POWER7/POWER6 are shipping products. And fair Game.
Sure when we start to get benchmarks on T3 then it's also fair enough to compare with that. But we haven't yet. And with the whole ROCK history of SUN then I only think it's fair that we get some semi hard facts on the table before you start to go into a religious frenzy of worship.
I mean it's just like hearing the people who really like Itanium, who are already preaching Poulson.. I mean come on, Tukwila isn't even shipping yet.
And for your whole compare "POWER6 to Nelahem EP and claiming that POWER6 CHIP is faster than Nehalem EP." I'd like to have a link to where I've said that.
It seems to me you often forget that things are different seen in different contexts.
For example which CPU is the fastest Database platform ?
Well it sure depends on the database, Neither T2+ or POWER will run SQL server. So it will be a match between Itanium and x86.
Which CPU is the fastest ?
It depends. Is that seen in the context of a fully loaded Enterprise class system ?
Or is it in a cheap whitebox 1-2 socket server ?
Or Or ...
For me it is quite simple, I need both single threaded throughput, Core throughput and scalability, cause that is what I we need where I work.
And that is the basis for my look upon the world...
"And how about your statements that "despite you need four POWER6 to match two Intel (ordinary) Nehalem, the POWER6 is faster"
The Power6 is generally faster than Nehalem - per core, which is what most software vendors base their charging on. If you are talking per chip, then a 4 core Nehalem is faster than a 2 core Power6 chip.
And don't get sucked into generation this and generation that. If the chips are available in the general timeframe, then they should be compared against each other. I have no idea when T3 will be shipping - if it is within this year, then it will be compared against the POWER7 later in the year, if not, then it might have to go against the POWER7+ next year.