SuperPi on GPU, were going CUDA

[rozzyroz]

No, it will thread itself to the cores without being prompted for number of cores.

will this type of code be able to be used in different programs?

[dinos22]

Progress!

Playing with threaded code now.

.........and?

[Lestat]

thats true.. the big reason I'd do single core runs is to test if certain cores are running slower than others. I haven't had any problems it's just something I'm doing when I'm ocing.




ok... I'm just curious about what options will be enabled in the new versions.

one core running slower than the other would be a 1 in a billion chance and even if it was you would never notice it unless that one core was 75% slower.

[safan80]

one core running slower than the other would be a 1 in a billion chance and even if it was you would never notice it unless that one core was 75% slower.

yeah I know.. I was just testing out why in some cases some cores in a prime run appeared to run slower, but they don't it's just overhead.

[Lestat]

yeah I know.. I was just testing out why in some cases some cores in a prime run appeared to run slower, but they don't it's just overhead.

good question to have, the reason is some prime tools are not multi threaded, but your talking about PRIME, this thread is about PI 2 VERY different things.
and XP is not multi thread friendly where Vista is a must more multi thread friendly OS, a perfect example is video encoding. run the same program under those 2 os's and your guarenteed to have more mutli thread usage under vista than XP, if the program is coded correctly that is.

in terms of PI, PI is not a multi threaded application so unless you set the affinity it usually always hits 1 core with a higher thread priority than the other core. thats just the nature of the beast.

Fugger is designing the app to be a video card PI capp which again as you probably know is a very different type of coding and multi threading scenerio.

you've said prime several times are you meaning prime or are you meaning PI

lastly if you are referring to PI and just mistakenly said prime, and you are setting the affinity for each run of PI to a different core, and 1 core is slower, well, Fugger might be able to shed some light on things or someone else maybe but i have not seen the slow down per single core that you are referring to... :\

[Marvin_The_Martian]

I have, core 3 from my quad is noticable and repeatedly slower then the other 3 cores, though that could be the ihs not being perfectly flat since that core is usually hotter ( hotter=slower? ).

Maybe Safan was saying it would be fun to be able to target specific hw ( say you know the nv architecture and want to bench one set of stream processors directly ( nvidia groups sp's into groups of 8 iirc? )).. That way you can find out if any of the shader groups is faster or slower then the other. Though I don't think you can do this with CUDA.

[angra]

Charles - is it public who is doing the coding, or is that being kept quiet? My research focuses in a related area and I could possibly contribute coding effort to this. I actually looked into doing this a year or two ago, but never got around to pursuing it in earnest.

[saaya]

I have, core 3 from my quad is noticable and repeatedly slower then the other 3 cores

i think thats caused by how the cores are prioritized between one another for cache and system requests in the crossbar that connects them to the interfaces.

[MuffinFlavored]

I am still excited for this program.

[safan80]

good question to have, the reason is some prime tools are not multi threaded, but your talking about PRIME, this thread is about PI 2 VERY different things.
and XP is not multi thread friendly where Vista is a must more multi thread friendly OS, a perfect example is video encoding. run the same program under those 2 os's and your guarenteed to have more mutli thread usage under vista than XP, if the program is coded correctly that is.

I do video encoding xp x64 is not slower than vista x64 at doing it. I have no clue why you would think this. When I'm talking about prime I'm talking about running the 64bit version of prime from the makers of prime95. I haven't run a 32bit OS in a while. There are cases where some cores run slower than others, but I've noticed it depends on the voltage one uses on the chip if the problem comes up.


edit: anyways Fugger get the new version of Superpi get out

[Lestat]

I do video encoding xp x64 is not slower than vista x64 at doing it. I have no clue why you would think this. When I'm talking about prime I'm talking about running the 64bit version of prime from the makers of prime95. I haven't run a 32bit OS in a while. There are cases where some cores run slower than others, but I've noticed it depends on the voltage one uses on the chip if the problem comes up.


edit: anyways Fugger get the new version of Superpi get out

your talking about a 64bit OS vs 32 bit --- once again a VERY different animal... VERY different in terms of core and memory utilization.

[Reznik Akime]

Psst, how goes it? Numbers need be crunched.

[safan80]

your talking about a 64bit OS vs 32 bit --- once again a VERY different animal... VERY different in terms of core and memory utilization.

no I'm not this file is the one I get, it's 64bit.

[RaZz!]

so will it be possible to recode superpi to run on a radeon (in an CTM environment) too? and why didn't anyone tried to do it for ctm yet?

even if it's through some middleware layer so that ctm could use cuda's extensions etc... would be cool though.

[MuffinFlavored]

Is there any progress to report that hasn't been reported?

[Yellowbeard]

Yes, there will be a special report later at an as yet unannounced time

[Marvin_The_Martian]

Well I'm waiting to hehe so hurry up...

Maybe abit OT but a good gpu benching tool which isn't 3d ( I mean, with 3dmark ect you can push the clocks to a point where you get artifacts and stll get a score but somehow I don't think the people over at Stanford would like people folding on those clocks ( well it should produce eue's but still I wouldn't mind a tool with which I can test the output besides atitool since I got false negatives and false positives with that in the past or furmark as I don't trust my eyes enough since were not talking about graphics but ggpu results ) is really needed, for reasons I somehow managed to put into one very long and badly formed sentence.

Anyway, we're waiting

[ZOMGVTEK]

I WANT THIS NOW!!!!!!!!!!!!!!11!!!!!!!!!!!!ONE!!!!!!!!!!SHIFT! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!11

Over one month later...

<_<
>_>

[safan80]

Over one month later...

<_<
>_>

lol it might be before duke Nukem Forever.

[Fixxxer]

And so?

[zabomb4163]

bit more difficult than once thought....?

[-X-hellfire]

is it necessary to calculate pi with the graphicscard?



it took William_Shanks 15 years to calculate pi with 707 digits by hand in 1873 but it had an error in it, the first 527 decimals was correct, which wasnt found until 1943 by ENIAC, one of the first computer's
http://en.wikipedia.org/wiki/William_Shanks

http://en.wikipedia.org/wiki/Pi
Memorizing digits
The Guinness-recognized record for remembered digits of π is 67,890 digits, held by Lu Chao, a 24-year-old graduate student from China.[35] It took him 24 hours and 4 minutes to recite to the 67,890th decimal place of π without an error



Stu's pi page
OF ALL THE ONE'S TESTED
THE FASTEST PI PROGRAM
that will run on your PC is....
http://home.istar.ca/~lyster/pi.html


Software for calculating pi
http://en.wikipedia.org/wiki/Softwar...ulating_%CF%80

[Nosfer@tu]

is it necessary to calculate pi with the graphicscard?

Nothing is nessesary!

Like this quote
Why would you put a jet engine on a motorcycle ? BECAUSE WE CAN

[-X-hellfire]

...would it then be possible to archive supercomputer performance on a pc?


but mostly pi don't need more than 100 digits anyway in any practical purposes...

so what is it useful for?

[bowman]

It's not useful for anything, jeez. It's just neat to see how fast our GPUs can calculate Pi.

Is there any use in running SuperPI over and over, or 3dmark over and over? No, it's just fun to test.