Where do you get the 16 threads/core number? The presentation on the crypto units mentions 8 strands/core.
Sun Microsystems, muzzled by Oracle's impending $5.6bn takeover, somewhat surprisingly showed up at the 21st annual Hot Chips conference sponsored by the IEEE and hosted at Stanford University - the birthplace of Sun - where Sun's chip techies talked about the future Rainbow Falls Sparc T processors and their integrated …
If the Power7 share the same characteristic as the slow Power6: a small cache that can never fit in an entire server workload, with many client data sets. That is the reason the Power6 is so slow on server-client workloads, because the cache is too small. Frankly, no CPU has large enough cache to fit in thousands of clients data into it's cache.
Therefore you have to use a design which is not dependent on a fast cache to achieve speed: the SUN Niagara - to reach high performance on client-server workloads. And that is the reason a SUN T3 will smoke a Power7, unless the Power7 has drastically changed architecture (to not be dependent on a cache).
In fact, I suspect even a SUN T2 will smoke the Power7 because today you need four Power6+ at 4.7GHz to match one Niagara T2 at 1.4 GHz . Will the Power7 be four times as fast a Power6, so it can match one Niagara T2? No? Maybe even a SUN T1 will smoke the Power7? Hmmm... Surely, nothing will catch up a Niagara T3. It is a killer.
.."Geronimo..", IIRC. Was it Zork? Oh, forget it. I feel like a pig with 'flu. Feel like shi*te (G/F "You smell like it, Kulta*!").
Don't access your computer unless you've had a jab. Virii come from allsorts of places - and I aint talking about Bassett's liquorice ones. TTFN.
* Means gold, or darling. But I spotted her spade outside, and as I'm insured, and there's no gold in Oulu, I'm a bit twitchy. Hope she has no idea about boiling rabbits....
Dude....you are clueless. Stop drinking the koolaide and do real life comparisons of systems. The T2 can preform fairly well for java apps if they are thread heavy, but put any kind of batch, single thread workload or database on the system and they will grind to a stand still. KISS keep is simple architecture only works for simple applications.
We bought some of these boxes to replace UltraSPARCIII boxes and they looked pretty good, but it was an easy compare. Then we brought in p6 for the database servers as we would not pay the Oracle .75/core on an 8 core chip. We were amazed with the performance and true virtualization.
We are now standardized on Nehalem for 2 socket and Power6 for "data" applications/ workloads.
If you need to run single thread apps then go to a PC or SPARC64. I don't believe a word of what you're saying. High end applications need throughput and POWER does not provide much of that. T2/T3 provides throughput at the low-mid end and SPARC64 provides throughput at the high end and single thread perf at the low end. All of this at half to a quarter of the cost of IBM.
".....I like the hp part in the link..." Well, that's probably because the Oracle infrastructure is largley running off hp kit. When they used to brag about their Collaboration Suite they always went on about how they replaced umpteen dozen Exchange servers with three hp Itanium servers - guess SPARC just couldn't do the job!
I'm seriously beginning to suspect Kebabbert is actually an IBM marketing plant, created to make Sunsiners look stupid and encouraging people to say nice things about Power. I mean, can anyone really be that obnoxious, naive and technically inept?
Actually, I have a lot of friends that work for Oracle and the majority of the DataCenter still runs on Sun.
To say "replaced umpteen dozen Exchange servers with three hp Itanium servers - guess SPARC just couldn't do the job!" is really pretty funny and disingenuous. Exchange is a Microsoft product and Microsoft only puts Exchange on Windows. They replaced PC's with Itanium, not SPARC as you would have people beileve... Not really much of an accomplishment considering that the old PC's they used to run Exchange on likely did not have domaining, which the Itanium servers do. Try again Bryant.
"I mean, can anyone really be that obnoxious, naive and technically inept?"
LOL. No comment.
You have stated several things, many of them turned out to be FUD. You are a FUDer, and a liar. How in earth can you state that the slow Power6+ is faster than a Niagara? If you claim something false, you are lying. Right? Ergo, you are a liar.
And besides, I wonder who is technically inept? As I said, I have a double Masters, one in comp sci in algo theory and discrete math, and one in math. Ive also tried to explain to you that a cache will never be able to fit in a server work load, to no avail. You STILL believe that the Power6 is able to fit in thousands of clients dataset + OS + kernel + what not into a 24MB cache?
I suggest you just think over it. Does it sound reasonable? Does it sound logical? Yes or No? Does the bear shit in the forest? Yes or No?
We have some T2 boxes (no T3 boxes) in our shop and they work fine. However, when we were looking at a EMR (EPIC) we were going to go with AIX or Solaris. So when we invited IBM for their dog and pony show, they went over the p570 and the p595 boxes. When it was Sun's turn for their dog and pony show, they went over the M8000/M9000 boxes with us. They never mention any T3 boxes which I believe is that even Sun knows the T3 doesn't play in all area's and especially at the high end.
So, the T2/T3 our nice boxes, but they compete with the low to mid range while the Power6 and SPARC 64 compete from the mid range to high end. So when we were deciding on the M9000 or the p595, it actually boiled down to IBM is on solid ground while Sun was not (this was decided around January 09).
I can't believe I'm siding with Matt on this one.
Anonymous, because my co-workers read these post
Yeah, I think they're both fudsters, but I'm not really sure of the point behind your post. Sun showed M8/9000 boxes so you are leaning towards IBM? Huh? Just because the T-Series does not cover all areas? Use T-Series for low to midrange and M-Series for mid to highend... You'll save a ton on hw and service costs. IBM has nothing at the low-end, which I'm sure you have some need for.
"And besides, I wonder who is technically inept? As I said, I have a double Masters, one in comp sci in algo theory and discrete math, and one in math. Ive also tried to explain to you that a cache will never be able to fit in a server work load, to no avail. You STILL believe that the Power6 is able to fit in thousands of clients dataset + OS + kernel + what not into a 24MB cache?"
There are lots of engineers over at http://www.realworldtech.com/forums/index.cfm?action=list&roomid=2 You should dazzle them with your intelligence. There's a long thread about Power7.
Anonymous Coward, you call me "a FUDer as well"?
"...However, when we were looking at a EMR (EPIC) we were going to go with AIX or Solaris..." I have never claimed that the Niagara is fastest on ALL work loads. I claim that the Niagara is faster on some specific server workloads, for instance Siebel 8.0. SUN has explained that Niagara sucks on single threaded workloads, everyone knows this.
"...They never mention any T3 boxes which I believe is that even Sun knows the T3 doesn't play in all area's and especially at the high end...." If the things you said were true, then you would have known that no T3 machines does exist yet. SUN offers no T3 machines. I wonder if the T3 chip is finalized yet? Maybe not? Maybe SUN are doing the last polishing on the chip?
Somehow... I suspect you have no clue what you talk about. When you write that the T2 and T3 are fine, but are killed by IBM Power boxes. And you call ME a FUDer? Maybe I should start to post lots of testimonies like "I work at a large bank/company/etc and I love the IBM Power boxes, but sadly, when we tried the Niagara boxes on SIEBEL it turned out that one SUN T5440 is twice as fast as three IBM P570 servers, so now we are migrating to Niagara. Can't understand I am siding with Kebabbert on this one". Who is the FUDer? You or me? The good thing is that the T5440 is twice as fast as three P570 on SIEBEL benchmarks, according to Oracle white papers. I dont have to FUD nor lie.
Yes, I AM smart. I mailed MENSA about my test results on their web page (they have a small, informal test), and they told me that they have members that achieved lower scores than me. If I am a member of MENSA, then I am smarter than like 95% of the world's population. If I fail the true MENSA test, I will fail just barely, which means I am smarter than like 90% of the world's population. Still, I am smart. This is fact. (Not surprising though, as IQ tests favours mathematicians. It is like a musically gifted person takes a music test - he is likelly to score high there, as well as a mathematician is likely to score high on IQ tests)
That sounds stupid to me. Build a supercomputer on Power6??? Do you know why it sounds stupid to me? Because one MAIN problem with supercomputers are the cooling. And you know that a high clocked CPU as the Power6 uses lots of power. Maybe 400watt? 500watt? To use such CPUs goes against all supercomputer design. Therefore I suspect you know not what you talk about. Are you FUDing?
My post was to basically tell Kebabbert he was also a FUDer. He always mentions the T2/T3 are better than the P6 570 or the P6 595. Which I'm sure for some, yes, but other apps, not even. Even Sun matches the p570 and p595 with the M8000/9000 because they really compete with each other. The T2/T3 are for low to mid range, but not the high end range.
I'm sorry if I confused you, but the T2 had nothing to do with our EMR decision since it wasn't an option but rather the M8000/9000 were are Sun options. We chose IBM over Sun, not because the p595 was superior to the M9000 (they both have their pros and cons and they both would have worked fine). It boiled down to upper management going with IBM since they were worried about SUN (which proved to be a smart move on their part).
As a UNIX admin, I always tell my co-workers here to go with Xeon processors for the low to mid range if the application will work with Linux. Of course, sometimes its not supported (or not many clients running linux) so we choose the T2 boxes, or the M3000-M5000 for Solaris or the p520/550 for AIX (again, depending on the app).
Of course I know that several supercomputers are based on Power CPUs. What am I? An ignorant business person or a double M Sc?
Maybe what YOU dont know is that the power usage is a major problem for super computers, and the cooling. If you study, e.g. IBM Blue Gene which is ranked no 5 on top500, you will see that it uses PowerPC CPU at ~700MHz. One of the world's fastest computers uses 700MHz CPUs. Now, why is that? Take a wild guess.
To build a supercomputer on Power6 (which maybe consumes ~500W) would go against all super computer design principles. Therefore I strongly question that claim. It is preferable to use low power CPUs.
If you call me a FUDer, then you have to prove it. I always mention that the T2 wins over Power6+ on certain server workloads. I have never claimed that the T2 is faster than the Power6 in every aspect.
Now, prove that I am FUDing, or you are a liar. Just like some random morons claiming stupid things like that a "Niagara suffers from a small cache" - how can that be true if the Niagara smokes a Power6 on Siebel 8.0 benches? It is not true, therefore that claim is a lie. Ergo, he is a liar and FUDer.
Now, you prove that I am lying. If you can not, then you are a liar. Why is that several IBMers is a FUDer and liar?
Your Sun kool aide cup runith over. Yes the T2 can do very well in some workloads. The problem is Nehalem has passed it up in the workloads it does well at and it has never been able to compete with Power for the workloads Power does well at.
Word of advice if Sun ever gives you a free trip to Jonestown Guyana....think about switching your kool aide supplier.
Biting the hand that feeds IT © 1998–2021