Rewriting SuperPi

Any comments are welcomed here.

Do you think it would be wise for me to invest my time in trying to create a new "SuperPi", removing all the access to the disk (to hopefully decrease the time taken to calculate all of the digits), and renewing the UI so it is basic Windows, and maybe making it a more serious benchmark than it already is?

If so, would anyone like to recommend an algorithm or method to calculate pi to 33 554 432 places (32mb)?

I would also hope to be able to make it to be able to calculate even more, and be multi threaded.

Post anything, even post "no, leave it alone, its perfect, don't touch it".

Thank you.

I would say you should removing access to the disk so that we would not have to have 2 raptors in raid in order to get better times. Then it would just be cpu dependent.

Quote:

---

Originally Posted by cajer

I would say you should removing access to the disk so that we would not have to have 2 raptors in raid in order to get better times. Then it would just be cpu dependent.

---

Good to hear from you cajer, and hope to see you next week.

And, definitely. The main goal of removing disk access is because it breaks your calculation into 7 files, and stores it there.

As far as multi thread, the number of digits would be divided by 2 or 4 (depending on the amount of cores, 1 thread could be run too) and then calculated on each core.

I could see this going places.

Personally I would like to see it happen since the hdd accessing so much runs against the idea of a "cpu" benchmark. Multi threaded support (which hyperpi also has i think - except it runs 4 instances of pi calc. I dont think it divides the work of 1 calc between the cores) and a new UI would be a welcome addition. Most important things are stability/predictability (wont give widely deviant scores on same hardware/os setup) and efficientcy so that it scales well as clock speeds ramp up.

The key would be to have hwbot support behind it. I would talk to massman or richbastard about this. I think widespread adoption would happen instantly if hwbot accepts it for points (or even just for submissions to start or something).

Ehm isn't this talk just what Wprime was built on/decided on?

Just wondering lol.

Quote:

---

Originally Posted by Marvin_The_Martian

Ehm isn't this talk just what Wprime was built on/decided on?

Just wondering lol.

---

Yeah, wPrime is an awesome benchmark.
But with the recent fluctuation between versions, I don't know.

If you want to code a new benchmark try to do a own benchmark, not Super Pi Mod MuffinFlavored-version. ;)
Super Pi is so popular because it is like it is. :yepp:
IMHO a new benchmark should have a good runtime, new CPU architectures can decrease it very easy (just compare A64 - C2D :eek: ).

Quote:

---

Originally Posted by Hyperhorn

If you want to code a new benchmark try to do a own benchmark, not Super Pi Mod MuffinFlavored-version. ;)
Super Pi is so popular because it is like it is. :yepp:
IMHO a new benchmark should have a good runtime, new CPU architectures can decrease it very easy (just compare A64 - C2D :eek: ).

---

That is what I was aiming for. I wasn't going to just build on top of SuperPi.
I just meant something similar to it. A CPU benchmark that calculates, something.

Sounds good to me, especially multithreaded would be nice but honestly would it even work to split the calculations to different cores? Isn't that the problem with multithreaded nowadays, you have to let cores do specific tasks rather than try and focus several cores one thing.

How could this be done? I'm curious about it too. Anyways, sounds like an awesome idea.

Like the sound of it! will be interesting to see how much faster it works...

wPrime and superpi have nothing to do with eachother

wprime sucks in testing RAM......RAM has very little impact on wPrime where it's supersensitive to superpi :D

Quote:

---

Originally Posted by RPGWiZaRD

Sounds good to me, especially multithreaded would be nice but honestly would it even work to split the calculations to different cores? Isn't that the problem with multithreaded nowadays, you have to let cores do specific tasks rather than try and focus several cores one thing.

---

If you do a run of wPrime, and you set, say 4 threads, you will see it say "running sqrts of x to x on thread x"

If you set it to 1 thread, it will say "running sqrts of 0 to 3000"
If you set it to 2 threads, it will say "1. running sqrts of 0 to 1500 2. running sqrts of 1501 to 3000".

I know how to do everything, EXCEPT the benchmark. I need something to take up cycles besides n = n + 1

:p

I think it would be nice if it could use 4... maybe 8+ cores efficiently. Couldnt you make it access the memory instead of the disk? That would surely help make it a cpu only test, no?

Quote:

---

Originally Posted by ZOMGVTEK

I think it would be nice if it could use 4... maybe 8+ cores efficiently. Couldnt you make it access the memory instead of the disk? That would surely help make it a cpu only test, no?

---

I can make it access the CPU only (kind of).

I mean, I made a simple benchmark. (i = i + 1; square root of i; see how big i gets in 1 minute, that is score)
It creates a 100% load, generates heat.

But I want it to be more complex, more efficient.

Quote:

---

Originally Posted by dinos22

wPrime and superpi have nothing to do with eachother

wprime sucks in testing RAM......RAM has very little impact on wPrime where it's supersensitive to superpi :D

---

I was not talking about benching perspective but about programming perspective. Wprime was made to offer a multithreaded superpi equivalent. That there is to much score variations now due to diffrent versions and that the benchers ( you ) don't like how tweakable it is, was not wat I was saying.

Gl Muffinflavoured btw, you seem to get more positive reactions then I did ;)

Quote:

---

Originally Posted by dinos22

wPrime and superpi have nothing to do with eachother

wprime sucks in testing RAM......RAM has very little impact on wPrime where it's supersensitive to superpi :D

---

I agree, thats why SPI is 10x more fun than wprime!

I just need a way to benchmark the CPUs.
After that, I can make it less memory dependent, more CPU dependent, anything.

I just need a base.

Why don't you make it exctally like super pi 1.5 but remove the hard drive dependent part? That way it would be more likely for other people to accpet it. It might ever become super pi 1.6. ;)

yes please!

I just found out SuperPi uses the Gauss-Legendre algorithm to calculate pi.

I will hopefully have a beta up within a few hours.

this should be interesting... may i beta test?

I have a couple different pairs of ram including TG DDR2-10,400 (1300MHz) 2GB

look here, too.

http://gmplib.org/pi-with-gmp.html

Our Super Pi will stay our version, muffinflavored pi or whatever you want to call it to avoid any confusion.

But go nuts on making your version, Im not one to hold anyone back from succeeding, just dont do it thinking you are replacing Super Pi.

Quote:

---

Originally Posted by Qkjhfhaiguihfma

look here, too.

http://gmplib.org/pi-with-gmp.html

---

clearly issues there if Core2 cannot beat AMD chips at faster clocks even :rolleyes: