so...real cores are better then virtuals....

[charged3800z24]

Sorry, i am board and just about done eating lunch. I really am not trying to start something here;). Just noticed thise and thought I would show it.

Stole this from the news section showing Intels new 6core (12thread) chip.
the MC from AMD's lower clocked chip shows that in a multi threaded loaded app, real cores are btter. Now, real cores can usually mean more cost, power etc. But if done right, should be close.

Intel 2.4ghz
http://www.abload.de/img/w3wlt8.jpg

AMD 1.6ghz
http://img15.imageshack.us/img15/2255/wprimep.jpg

for those confused.... here is 2ghz

http://img168.imageshack.us/img168/7618/wprime2.jpg

I am sure everyone has seen his thread on here ~~~~>http://www.xtremesystems.org/forums/showthread.php?t=233565

[FlanK3r]

good comparsion!Think, AMD Thuban will very competive with 8 threads Nehalem i7 900 :)

[ELItheICEman]

AMD image link is broken - I'm curious to see how it looks though.

[Helloworld_98]

only 800MHz? that must have some ridiculous IPC to compete with a 2.4GHz Nehalem.

[Dr_Swizz]

How does the intel chip perform with SMT disabled?

[Jakalwarrior]

only 800MHz? that must have some ridiculous IPC to compete with a 2.4GHz Nehalem.

That's speed step in play. The chip is a 1.6ghz.

[Andi64]

12c/24t vs. 24c/24t? If that's the case... they are not the same market segment IMHO.

Two Nehalem-EX 8c/16t is the real enemy of these Opterons.

[EniGmA1987]

I seriously doubt a couple Magny-Cours could be that fast at only 800MHz, but just to be certain on the test results could you turn off Cool-n-Quiet or whatever to have the CPUs run at full speed always? It would at least for a better screenshot so people arent confused.

[charged3800z24]

I seriously doubt a couple Magny-Cours could be that fast at only 800MHz, but just to be certain on the test results could you turn off Cool-n-Quiet or whatever to have the CPUs run at full speed always? It would at least for a better screenshot so people arent confused.

Yes, cool n' quiet is on, the chips was running at 1.6ghz. The 32MB score was a direct result of CnQ throttling the cores causing the score to lag. he ran a 2ghz (I think) run that wasted that score above. I updated the main link to show the 2ghz run...

[GenTarkin]

Yes physical cores are always better then HT'd cores. Because they are just HT cores, really just running 2 threads at same time on one core. Only increases performance like 10-15% max in certain areas. AMD does not use any sort of HT on their chips.

[charged3800z24]

Yes physical cores are always better then HT'd cores. Because they are just HT cores, really just running 2 threads at same time on one core. Only increases performance like 10-15% max in certain areas. AMD does not use any sort of HT on their chips.

Yes, this is obvious, I was making a snood remark...well, cause it was brought up some time ago before Nehelam launch:p: AMD will implament a form of HT, it will be different implemintation of course. I even read soemwhere that the Phenom cores could have had it implemented but never was...maybe TDP reasons? I dunno, I wish i new where I saw that, it was posted here. It was a link to the white papers or what ever you call it.;)

[Chumbucket843]

How does the intel chip perform with SMT disabled?

HT adds about 10-15% more performance in multithreaded apps. even more if it is memory intensive.

[wuttz]

its cmt, cluster-multi-threading, works reverse of intels hyperthreading. thats all i understand about it.

[keithlm]

Actually the entire virtual core issue and hyperthreading has always made me wonder something:

When an application/benchmark is truly optimized for the number of cores a system has, and it truly runs all the cores at 100% then theoretically hyperthreading would not give any benefit at all because there would be minimal need for more threads. (I.e., this situation would be the "best case" most efficient use of the system.)

So the question then becomes what type of coding benefits the most from hyperthreading? A type that basically just throws a ton of threads out and let's the hyperthreading do the work of optimizing the task from the pool of waiting threads instead of doing the optimization itself?

[demonkevy666]

HT adds about 10-15% more performance in multithreaded apps. even more if it is memory intensive.

I'm just wonder how memory intensives would ad hyper threading


its cmt, cluster-multi-threading, works reverse of intels hyperthreading. thats all i understand about it.

I think it's more like the Barcelona 128 bit fpu's except it works just split up work better since they're be able to switch the work from being 4 way single core to two 2 way cores in one 128k L1 core caches.,


Actually the entire virtual core issue and hyperthreading has always made me wonder something:

When an application/benchmark is truly optimized for the number of cores a system has, and it truly runs all the cores at 100% then theoretically hyperthreading would not give any benefit at all because there would be minimal need for more threads. (I.e., this situation would be the "best case" most efficient use of the system.)

So the question then becomes what type of coding benefits the most from hyperthreading? A type that basically just throws a ton of threads out and let's the hyperthreading do the work of optimizing the task from the pool of waiting threads instead of doing the optimization itself?

you mean something likw continuously threaded programs verse optimality number of core count.
I imagine one wold be good for real the other better for virtual cores

[Clairvoyant129]

Um... these two CPUs are not even in the same market segment... :rolleyes:

[charged3800z24]

Um... these two CPUs are not even in the same market segment... :rolleyes:

Umm, how do you know where AMD is putting it. You are saying there can't be a dual socket system for a kick ass work station. No one will no for sure till launch really. It is still a single CPU with 12 cores. I was just showing a 12 threads (6 cores) vs 12 threads (12 cores) comparison. Both are in dual socket systems and both are 12 thread capable in a single CPU. ;)

[keithlm]

you mean something likw continuously threaded programs verse optimality number of core count.
I imagine one wold be good for real the other better for virtual cores

Basically yes.

What I was basically getting at was this: If you are creating benchmarks then you would need to create one for each of the two situations. If you only create a benchmark that measures one situation and ignore the other then you are not presenting every aspect of the situations.

BUT MORE: In some of the older threaded libraries that I've seen, on the Sun platform for example, the methods of multi-threading were very reliant upon the developer. Good development practices would have you attempt to split the work equally. Bad practices would have it random and not balanced. I've not been recently involved with some of the newer threading libraries so I'm not sure if it still pertains. (But generally things don't change that much. They come out with newer things like Java and .net etc... but things still work about the same as they did in the "C" days.)

[i found nemo]

so an 8 core amd should be as fast as .. 6 core 12 virtual intel ... ?

[Zucker2k]

so an 8 core amd should be as fast as .. 6 core 12 virtual intel ... ?Not if it's clocked like this (http://kingpincooling.com/forum/showpost.php?p=1184&postcount=11)

[informal]

^^ 5.66Ghz bench stable with excessive Vcore is only a tiny bit better than Bloomfield quad core with SMT-on ,done on LN2.
I'm not saying it's a bad results since we have 50% more cores,just not that impressive considering this is 32nm design.

[demonkevy666]

that cpu score seem really low for 12 threads and 5.6ghz.