DoubleMark (Prime Benchmark) - Open Source - Overclock.net

Licht · 01-18-08

Welcome to the most posted in topic in coding and programming!

After much work and allot of delays and errors i have finally finished build 3.

8, 30th : Next version is version 1. Stress testing is near finished and has been released to testers, proven to be highly effective and efficient. All glitches from last version are already fixed. Next version is from scratch so it'll be a lot cleaner, way different. Drawback being it takes a while to recode everything.

Duckydude · 01-18-08

Nice work

, worked perfectly for me

. This sounds like a really cool project, would be a nice new benchmarking utility that we could compare around here.

Here are my results for 25K:

Licht · 01-18-08

Quote:

Originally Posted by Duckydude

Nice work

, worked perfectly for me

. This sounds like a really cool project, would be a nice new benchmarking utility that we could compare around here.

Here are my results for 25K:

I'm working on build 3 for you Q6600 guys, will be done very soon. Should support the full 16 cores. (I will have to do a bit of math before anything so i can optimize it.)

dskina · 01-18-08

Running at stock everything, atm.

Licht · 01-18-08

Quote:

Originally Posted by dskina

Running at stock everything, atm.

I don't think B2 counts prime numbers incorrectly (cross thread integer changed must throw it off.)

The Bartender Paradox · 01-20-08

A question and a suggestion-

What method are you using to check for primes?

Also perhaps you could include some info on how many prime numbers there are for our given length chosen, or spit out an error when the number of primes found don't match the number of primes there actually is. So we can see if our computer screwed up calculating.

Oh, and a B2 screeney: (stock Q6600)

Edit: err.. compare this to Duckydude, I found 2 more primes

Licht · 01-20-08

Quote:

Originally Posted by The Bartender Paradox

A question and a suggestion-

What method are you using to check for primes?

Also perhaps you could include some info on how many prime numbers there are for our given length chosen, or spit out an error when the number of primes found don't match the number of primes there actually is. So we can see if our computer screwed up calculating.

Oh, and a B2 screeney: (stock Q6600)

Edit: err.. compare this to Duckydude, I found 2 more primes

Yeah i already confirmed that it was counting prime numbers wrong. It has something to do with cross thread value accessing. It is an easy fix and will be resolved in build 3. My method for detecting prime numbers is a simple and efficient method, however i am keeping it a secret until V1.0 when i go open source. (Unless plans change, although i promise to give that code up on V1.0 regardless, it is done in a single line of code!) Also, yeah i am also working on a sum check for Build 3 or 4.

Thanks for the help man, based on your noticing the 2 number difference in detected primes i can now confirm the error is within the crossthread memory access. I will find a way around it in build 3, shouldn't be too hard.

Ok, long story short. Memory access is bugged, working on the issue be done soon. All scores are indeed valid the error is in the memory not in the system itself, i personally guarantee all scores are 100% accurate up to a 10th (It is impossible on Windows to measure a length of time more precise then this) of a second (the score variances between tests are caused by other threads interfering with my applications processing, the more programs you turn off the less it will vary. If i forbid this it could cause system instability, this issue is present in every benchmark.)

ThePope · 01-20-08

Sweet Licht! I Cannot wait to see more of your work

Polska · 01-20-08

Cool stuff will keep an eye on this. Here is one result of mine. This is on stock e6850 (having system stability issues trying to determine if its to do with overclocking

).

Also, after I ran a test, the radio buttons for the number of threads all became accessible (screenshot 2).

I don't know if you detect how many cores the system has (because mine only had 1 and 2 available on program launch) but once they were available I ran it with 4 threads (on a dual core) and the program crashed. I assume you want the other options disabled, just wanted to let you know they are available after running the benchmark once.

Licht · 01-21-08

Quote:

Originally Posted by Polska

Cool stuff will keep an eye on this. Here is one result of mine. This is on stock e6850 (having system stability issues trying to determine if its to do with overclocking

).

Also, after I ran a test, the radio buttons for the number of threads all became accessible (screenshot 2).

I don't know if you detect how many cores the system has (because mine only had 1 and 2 available on program launch) but once they were available I ran it with 4 threads (on a dual core) and the program crashed. I assume you want the other options, just wanted to let you know they are available after running the benchmark once.

I had forgotten about that hickup, it came after i ran into some autoscaling issues with my adaptive threading. I meant to lock them off but apparently overlooked that before releasing (i was rushing quite a bit.) Now i am working on finishing up my adaptive threading code, when i am done it will be capable of any amount of threads no matter the number without losing any performance. This was intended for build 2 like i said but the rush left me locking the controls then forgetting to remove said renabling code.

No, i am not detecting how many cores the system has right now since that could leave the program locking off controls for rigs in the future when CPU access may be changed. Since you can run any amount of threads on any CPU it doesn't matter, so i classified those controls as thread controls instead of core usage controls. Although you can run 16 threads on a dual core for example, you will get better performance with 2.

PS : Just realized something, don't touch more threads then you have cores. The way i have thread affinity set up for adaptive threading will crash it after all...

		Overclock.net - Overclocking.net > Software, Programming and Coding > Coding and Programming
DoubleMark (Prime Benchmark) - Open Source

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)