At the last minute, as it was rolling out the low-end and high-end of its Power7-based server lineup on Tuesday, IBM rushed out a new TPC-C online transaction processing benchmark. No surprises there. And there are no surprises that Big Blue has pushed the OLTP benchmark, showing it could put a box into the field that can do …
The world had grown up a little and stopped posting pointless TPC-C cluster scores. Since the benchmark can scale perfectly the score you get is simply a factor of how much money your marketing dept. has to throw at the problem. For years all the top scores were funded by MS as they had more money that everyone else.
The one box score should have been quite impressive. And these SSD solutions are probably more practical than the approach the vendors used to take of sticking well over 100 baby disk arrays on the back of these boxes coz it was cheaper (just for the benchmark) than using a real storage solution.
No.1 in single system, No.1 in clustered system
That was the reason -- to beat Oracle's clustered configuration with cluster. I guess they could beat or come close to result of the Sun cluster with single 780 too (it's a bit faster than 595 with same number of cores), but what for? Let's not mix clustered results with non-clustered ones.
They are now number 1 on TPC-C with clustered results and number 1 with single system results (with P6 595). I guess IBM does not need to audit 795 performance. Not until someone beats their single-system 595 record result. They are on top and nobody expects 795 to be slower that 595... :-)
SPARC is dead
Larry has been spending millions on ads talking about the TPC-C benchmark from last October.
Check out how Power7 compares
3 Systems vs. 12 systems
10.36M transactions vs. 7.65M transactions
$1.38/transaction vs. $2.36/transaction
224 flash drives vs. 4,800 flash drives
$11.5M Full DB2 license vs. $7.85M 3 year only license (like anyone buys 3 year...IBM should create one of those BS 3 year term things as the $/trans would be even lower)
I wonder what the next ad will be from Oracle.
35% less transactions
71% more expensive
Oracle SPANC is not dead...please keep paying our maintenance fees.....Larry needs a new America's cup yacht. All he wants to do is sell you Exadata and has yet to publish any benchmark.
The Difference Between God and Larry Ellison: *God Doesn't Think He's Larry
I can add to your comparison, that IBM response times are not too great.
Congrats IBM, to first place. According to IBM, they have always had the first place. Even when Oracle had the record, IBM claimed that IBM still had the fastest TPC-C result in the world, because "IBM had faster cores"!!!! So, sure IBM used fewer cores, but at the top spot there was an Oracle system. Not IBM. I wonder how many IBM customers they fooled they still had the record, because "IBM used fewer cores"? (If you have faster cores, you can use fewer cores). I dont think Oracle will claim they still have the record, just because they "uses more cores"? IBM just twists everything.
I would like to see a comparison of all next gen CPUs, such as SPARC T3 vs POWER7. That can be interesting. Oracle will release a 16.384 thread Solaris machine with 64TB RAM, maybe T3 cpus. I would like to see comparison with POWER7.
For POWER6 vs Niagara T2, Niagara T2 won in TPC-C.
For POWER7 vs Niagara T2, POWER7 wins in TPC-C.
For POWER7 vs Niagara T3, I wonder who will win. I wouldnt be surprised if T3 wins in TPC-C.
IBM have the fastest system, the TPC-C was originally supposed to be a single node benchmark. Oracle twisted this with their cluster. Oracle claimed XX Millions transactions before they released their clustered benchmark and only reached to 7.x million (and at a horrible price for a cluster (if you include software)).
The claims that Oracle had the fastest TPC-C system was and is:FALSE, FAKE or FUD - call it what you want. The Power 6 595 is still the fastest single node database server out there.
This benchmark is just made to show what you can do with a cluster DB2 database. They only used three servers.
Will be very interesting to see if IBM publishes a benchmark based on the new 795 server later this year.
Niagara T3 is still under development, SUN/Oracle has not been able to deliver on their roadmap for years unfortunately.... let's see if they do this time. IBM will deliver a POWER7+ in a few years time and the POWER8 will be out in about three years (probably).
First kebb, now POWER has the fastest benchmarks both in the clustered and non clustered category for TPC-C. Period, so stop your ramblings.
Now this benchmark is IMHO terrible, and I cannot understand why it was even made. It doesn't make POWER look good IMHO, only in the eyes of the unenlightened, cause they beet an equal pointless benchmark made by Oracle. That is my take on this.
Now The T3 will get it's butt kicked by POWER7. The T3 boxes will still be more expensive, both in procurement and in price/performance. IMHO Adding even more slow threads will only give you so much.
Now your who won is absolutely bull. The only Oracle Niagara T2 TPC-C benchmark made is a clustered TPC-C benchmark. For single system image the POWER6 power 595 benchmark is still king.
Now for a 64 TB machine with 16.384 threads.. yeah.. righ.. sure oracle might make a cluster and slap a single serial number on it and call it a machine. But hey that still makes i a cluster.
"The claims that Oracle had the fastest TPC-C system was and is:FALSE, FAKE or FUD - call it what you want. The Power 6 595 is still the fastest single node database server out there."
No one denies that IBM had the fastest single node database server. The question was, who could deliver most TPC-C performance. Who had the record. Until recently it was Oracle who could deliver most TPC-C performance. Call it FUD if you wish, but it is true that Oracle had the top spot. You dont have to believe me on this, but you can check it up. I dont get it how IBM can claim they still had the TPC-C record, because IBM "had faster cores"??? Just sick reasoning. It reminds of when Jesper Frimann explained to me, that even though you need four POWER6 at 5GHz to match two Nehalem at 2.93GHz in TPC-C, the POWER6 is the faster cpu. Because "the POWER6 core is faster". Just sick reasoning. You need to be an IBMer to show such a flawed logic. If one core is faster, it doesnt make the entire CPU faster. Nor does a faster ALU or faster register, does not make the CPU faster. Just weird and flawed logic.
No one denies IBM has the record. What is your problem? I have not denied it. I accept the official POWER7 benchmarks. You, on the other hand, did not accept Niagara benchmarks, you rejected them all because they where "crafted by Sun", "cherry picked by Sun", etc. I dont have problem admitting the POWER7 is fastest right now. You OTOH, could not admit Niagara was fastest on some benchmarks. You could not admit that Nehalem was faster than POWER6, even when the benchmarks showed it.
POWER6 was slower than Niagara T2 on some workloads. I expect Niagara T3 to be faster than POWER7 in a similar vein. The problem is the turmoil with Sun. Development has not been prioritized, lots of people have been axed. When develpment of T3 continues without disturbances again, I would not be surprised if T3 utterly crushes POWER7 and makes it look just as pathetic as Niagara T2 at 1.4GHz made the POWER6 at 5GHz look like.
For the 16.384 thread machine, most probably it will no be a cluster. But a single server. Now it looks insane with 16.384 threads. But remember, back then Niagara T2 looked insane, but it delivered. I expect T3 to deliver. Sun has always gone first, and others have followed. Like when IBM talked about few high clocked cores is the way forward and downtalked many lower clocked cores as Niagara has, and now with POWER7 the way forward is many lower clocked cores. Ridiculous. But I dont expect you to see that.
Keb. There is a big difference between being fast and
"just sick reasoning. It reminds of when Jesper Frimann explained to me, that even though you need four POWER6 at 5GHz to match two Nehalem at 2.93GHz in TPC-C, the POWER6 is the faster cpu. Because "the POWER6 core is faster". Just sick reasoning."
That is because You have never understood what I said, either deliberately or due to lack of ... well.. understanding. The problem is that people use the terms fast as it was a one dimensional label. It isn't.
On the other hand then terms like "total chip throughput" is a relatively well defined metric. And here you are quite right, T2(+) is "faster" than POWER6.
"Like when IBM talked about few high clocked cores is the way forward and downtalked many lower clocked cores as Niagara has, and now with POWER7 the way forward is many lower clocked cores. Ridiculous. But I dont expect you to see that."
Eh ? A POWER7 core is faster than a POWER6 core (delivers more single threaded throughput) A POWER7 core also delivers more throughput than a POWER6 core and finally a POWER7 CHIP also delivers more throughput than a POWER6 chip. So what is your problem ? IBM managed to do all three things. Increase speed and increase throughput and throughput.
And as for your wet dreams about Great SPARC Machines.. well there is no SUN anymore, there is only Oracle. It isn't even Oracle who makes the machines it's Fujitsu. They don't even fabric the CPU's anymore, that is outsourced to TSMC.
// Jesper says BLEH.
Has no merit what so ever, he failed to read what was actually going on.
"That is because You have never understood what I said, either deliberately or due to lack of ... well.. understanding. The problem is that people use the terms fast as it was a one dimensional label. It isn't.".... It doesnt matter what you call it. You have numerous times said (just as IBM) because one core is faster, then the whole cpu must be faster. You said that POWER6 is faster in than Nehalem in TPC-C just because one POWER6 core is faster. You have also talked about how slow cores Niagara has, and how fast a POWER6 core is.
So what does it matter how slow one core is, or how slow the ALU is? The only thing that matters is how fast the CPU is, when we talk about the fastest CPU. I explained this to you many times, but you kept rambling about "faster cores", "higher license costs if you use many cores", etc. When you claim that POWER6 is faster than Niagara, then we must compare CPU vs CPU, I said. But you kept talking about how fast POWER6 core is. Just like the Nehalem vs POWER6 benchmark. Just sick reasoning. You need FOUR power6 cpus to match TWO nehalem - and still the power6 is faster because the core is faster. Great thinking from IBMers.
"On the other hand then terms like "total chip throughput" is a relatively well defined metric. And here you are quite right, T2(+) is "faster" than POWER6."...... Wow! Stop the presses! Jesper admits that T2 is faster than POWER6 in any aspect. I didnt expect that. Usually you just deny and reject all benchmarks that show T2 being faster in any aspect. The benchmarks are "crafted by Sun", "cherry picked by Sun", etc. What has happened, Jesper? When T3 comes out, and crushes POWER7 just like T2 crushed POWER6, will you deny and reject all those new benchmarks, just like you usually do?
FYI, when you talk about speed. If a T2 has much higher throughput than POWER6 and can finish the job much faster it means T2 is.... faster, yes? T2 finish the job in much shorter time, because higher throughput. To most people, it means T2 is faster. But not to you IBM FUDers I suspect.
"Eh ? A POWER7 core is faster than a POWER6 core (delivers more single threaded throughput) A POWER7 core also delivers more throughput than a POWER6 core and finally a POWER7 CHIP also delivers more throughput than a POWER6 chip. So what is your problem ? IBM managed to do all three things. Increase speed and increase throughput and throughput."...Yes, but IBM has always bashed Niagaras approach with many and slower clocked cores. The only way forward is few and very high clocked CPUs IBM said. Back then, it was radical to have 8 cores. No one used 8 cores, except Sun. Now everone does it. Including IBM. I have heard about a new IBM cpu with 16 cores and 64 threads - does it not sound similar to Niagara cpus? You know, the very same Niagara which has such a baaaaad approach, according to IBM? To copy someone else, is the sincerest form of flatter.
If IBM had such a good approach, where are the cpus with 1-2 cores clocked at ridiculously high speeds? Where are the single core cpus clocked at 8-10GHz? No one does that, not even IBM. It is a dead end. Sun leads the way, again. T3 will have 128 threads, which sounds sick today. But tomorrow everyone will have it. Oracle will release SPARC machines with 16.384 threads which sounds sick to today, which is usual when we talk about Sun tech. But tomorrow everyone will have it. Including IBM.
I just dont get it. When IBM bashes something, IBMers says it is the worst in the world, it is slow, the fast benchmarks are lies, they are "crafted and cherry picked". When IBM does exactly the same thing a few years later, it it the best in the world. Jesus. Sun released the container full of servers, Black Box(?) years ago. Have IBM also released their copy yet? Sun had a technique where you run different cpus: SPARC, niagara, x86 and one Solaris to rule them all. Recently IBM has released the same thing for a Mainframe, it will rule POWER7 blades, x86 blades etc - but they all run their separate OS. This IBM Mainframe technique is truly "revolutionary and unique and new" yes? Hilarious. Just look what Sun does today, and expect IBM to copy that a couple of years later. Just like Microsoft copies Apple.
And all this false IBM marketing: one Mainframe can consolidate 1500 x86 servers - if they all idle. Jesus. According to wikipedia, IBM was the first company to employ FUD. The FUD continues.
"....eans T2 is faster. But not to you IBM FUDers I suspect."
Now wipe your eyes and have a nice cup of tea and a piece of cake, my boy. Sorry you haven't been attending one of my classes here where I work, in my "Designing Solutions on UNIX", I use a 2-3 hours just explaning what the different vendors use of terms and what they mean with regards to threads,CPU's, Processors etc etc. U've been mixing it up since day one.
As for T3 crushing anything. Well it'll be the tables of Solaris sysadmins, when they have to have 500 inch monitors to be able to see all the threads. T3 is just more threads, 16 threads on 8 cores on one chip. That is 128 threads per CHIP.
A POWER 750 does 2410483 specJBB2005 on 4 sockets and 32 cores.
A Ora T5440 does 841380 specJBB2005 on 4 sockets and 32 cores.
So as for crushing.. what is that x2 the throughput ? Or just x 1.5 ?
So you are basically saying that a T3 based T5440 with 4 sockets and 32 cores will do around 3.5M-5M sepcJBB2005 ? So doubling the number of threads will give you a factor of 4-6 in throughput. You should really cut down on the psychedelic drugs.
"Ramblings about IBM is Microsoft and SUN is apple"
Listen. IBM isn't really copying anything from Niagara. Your whole premisses for your ramblins arewrong. Niagara and POWER7 are like night and day. The idea behind Niagara is many simpler cores with many threads for great efficiency. It's a good concept for the types of workloads it was designed for.
POWER7 is totally different it's 8 really really fat cores, each core has 12 execution units.
Sure IBM uses 4 threads on a POWER7 core and Niagara uses 8 (soon to be 16) but just cause they are
using >1 thread doesn't make it the same. Niagara uses statically round robin scheduled fine grained
multithreading. It's terrific when used for many light independent threads, and very efficient, for
that type of workload. But it sacrifices single threaded performance, and you need to have a lot of
threads that want to execute for it to be efficient.
POWER7 uses a SMT, with quite a few bells and whistles for example the OS can fold together
threads(virtual processors) , hence you actually don't have to have 4 threads executing at the same time, you can also have one that takes up the whole processor, the processor adapts to the workload that runs on it. So at 1-8 threads per chip it's a single threaded beast, and then as the number of threads increases up to 32 it becomes more and more of a throughput CHIP. Going from perhaps 800 units of work in throughput at 8 threads to 1440 units of work at 32 threads.
For the Niagara T2+ CHIP running the same workload it might do something like 85 units of work at 8 threads and 425 at 64 threads. With the T3 this might change to 100 units of work at 8 threads to something like 900 at 64 threads.
This little calculation is based on the 97 SPECint_rate2006 for one T2+ chip and 330 Specintrate2006 for one POWER7 chip, and from various papers that suggest that going from 1 thread per core to 8 threads give you something like a factor of 5 in throughput. And not a factor of 8.
This is a huge difference, and if you cannot see that then well... your loss.
"And all this false IBM marketing: one Mainframe can consolidate 1500 x86 servers - if they all idle. "
Problem is with servers is that most of them are idle... With a click of a mouse I have access to utilization data of thousands of thousands of servers... And guess what stand alone Wintel servers, do have lousy utilization. Often between 1-5%. Now how you then choose to consolidate and where is your own business. The premisses for their calculations are IMHO valid. I've have personally run consolidation projects where we consolidated hundreds of Wintel servers running Oracle DB's onto a single POWER server. HUGE savings.. HUGE saving.. And the fun part is when you talk with Wintel sysadmins, when they talk about utilization it's peak utilization... so 'My server is 50% utilized' normally mean that at peak times it uses 50%.
- +Analysis Microsoft: We're making ONE TRUE WINDOWS to rule us all
- Apple: We'll unleash OS X Yosemite beta on the MASSES on 24 July
- Pics It's Google HQ - the British one: Reg man snaps covert shots INSIDE London offices
- White? Male? You work in tech? Let us guess ... Twitter? We KNEW it!
- Apple fanbois SCREAM as update BRICKS their Macbook Airs