Something odd.....

[Silicon Skum]

I've since overclocked both of my socket A XP chips to 2.5Ghz, and I've noticed something strange with the number of completed cands per XP chip compared to my X2 @ 2.7Ghz. The X2 system scores really high on benchmarks compaired to the XP systems, yet I have exactly the same output from both platforms.
Per core, the X2 has the same output as the XPs per CPU core.

Surely an XP CPU core at 2.5Ghz can't be able to produce the same output as a higher clocked 2.7Ghz single X2 core?? I would have thought there would be a higher output with the X2, even if it was just due to the 200Mhz faster clock speed, let-alone the much better A64 core / chipset design.

If this hold true, then I have noticed a few things about the D2OL node. It seems that RAM speed / bandwith makes very little difference to output, as does the size of the CPU cache. Acording to the guys running the D2OL project, raw Mhz seem to be the way to go, though the difference between the XP and the X2 core seems to indicate something else may be involved.

Anyone have any thoughts on this?

SS
_________________
my sig disappeared from the image host (?)
But at least I have a Josh Award!

[Google]

[fussnfeathers]

Well, are you running two instances of D2OL on the X2? If not, then it'll perform just about the same as your other procs. It's only utilizing one core. Try running two instances on all the machines for awhile, and I'll bet you'll see the X2 pull ahead by a bit, since each instance would be using a seperate core.
_________________
Big enough to scare you

[Silicon Skum]

Yeah, I'm running two nodes on the X2, one per core, what I mean is that I would expect a single core of the x2 to outperform an individual XP core, clock for clock.

After thinking about it for a while, I wonder if it is more to do with the way the D2OL program performs the maths involved, it might be that the program code will give a similar result for any processor at a given clock speed. In other words, a universal program not optomised for any perticular CPU platform, just relying on the number of iterations per clock to produce results.
AMD CPU cores are able to perform more iterations per clock cycle than an Intel CPU of a similar clock speed, hence the XX00+ name of the AMD cpu line that equates it against a higher clocked Intel chip. Depending on how / if the D2OL program code uses integer or floating point calculations, this could lead to some interesting results to increase output on dedicated crunching hardware. AMD CPUs have a weaker FPU than Intel CPUs, so if both intel and AMD perform similarly, then it is possible that the program code does not use floating point calculations alone / at all and may rely on Integer math.

Hey Weaze, if your reading this, you have both Intel and AMD crunchers, are you willing to help out in an experiment to see what gives the most output for a given CPU platform?

I think most systems are overclocked to provide the best CPU clock speed and still maintain a high RAM clock speed, a trade off between the highest CPU speed and RAM speed. I think that it may be possible to increase the node output by simply increasing CPU clock speed even if this means running RAM at a lower clock speed, at least from what my results show, RAM speed is of little importance to D2OL (within reason of course).
Using this way to overclock, it is possible to get the highest (stable) CPU speed at the expense of slow RAM, yet still be able to increase node output.

Obviously this is no use to anything but dedicated crunchers, if you only have one machine and use it for games, this is not going to be a good experience.

SS
_________________
my sig disappeared from the image host (?)
But at least I have a Josh Award!

[[KoG]^weaZel]

for what is your production like per node and type of cpu?

D2OL does better on Intel cpus than it does on AMDs. And it does like the highest cpu mhz possible.

I don't have any AMDs in my house anymore just Intel.
Just for comparison purposes my PEE 840 @ 3.6 (200X18) and memory at 800 mhz 4-3-2-9 will do 330-360 a day. And my DX's @ 3.0ghz DDR400 2-2-2-5 do 300+ a day. The P4 EE @ 3.4ghz does 180+/- daily and the standard P4 @ 3.0 does 150+/-.
_________________
I tweaked and it tweaked back! So I Tweaked some more!

"Barney is like the Michael Jackson of PBS." - James Tybeerious

[2old2care]

I have an XP 2000+ and a P4 2200 (400fsb) running side by side. Without any OCing they are almost identical in output. The P4's mobo's BIOS does not allow an OC and the 2000+ OC's wonderfully. Regardless, I beleive you are correct in it is just operations, period.
I have noticed though, that memory speed does play a minor role. Obviously with the XP chip if I run the memory as fast as can for any given cpu speed, I get more WU's out. Likewise, with a P4 tighter memory timings help a bit. I know this goes against conventional wisdom, but if I loosen the timings up to get let's say a 400MHz OC, I can get "X" WU's out. But if I can tighten the timings back up and get a 375MHz OC I can still get "X" WU's out.
I know this is just anecdotal evidence, but it is what I've seen.
_________________
.
Liquid-Cooled Q9450 and an EeePC
.

[[KoG]^weaZel]

you also have to take into account the candidates that each system is working on. These will play hell with your averages when comparing the different machines.
_________________
I tweaked and it tweaked back! So I Tweaked some more!

"Barney is like the Michael Jackson of PBS." - James Tybeerious

[ToggleHead]

With that said....it would be difficult to yield totally accurate results without knowing the method d2ol uses to calculate the data....and therefore simulating a test using the same benchmark
_________________

[[KoG]^weaZel]

to answer ^^^^^ that use Dbench.
_________________
I tweaked and it tweaked back! So I Tweaked some more!

"Barney is like the Michael Jackson of PBS." - James Tybeerious

[Silicon Skum]

per node I get about 35 -39 per update (140 - 156 per 24hrs) for all AMD machines, though the X2 nodes have once gone as high as ~42 per node / per update, but I think this was due to a "faster" batch of cands that day. The only other XP node I have is a 2200+ @ stock 1800Mhz (no longer overclockable due to electromigration) and that gets about 29 or so per update.
I do have two P4 3Ghz HT systems, but these are laptops so not a good banchmark to base a comparison between the two platforms. Not sure what they can do per day, Not had them running for a while.

I had a test running with the same node and cand on each machine, but there was a power brown-out tonight and it's screwed up my machines, I still have'nt got one to reboot yet, looks like the BIOS might be slightly screwed up, I'll try a reset and overclock it slowly to make sure all is well. I think something has screwed up on the X2 machine, my two seperate nodes in different directories have somehow become the same node and neither will run correctly. think the drive might have corrupted.
I can understand how the guys running the project feel when their power supply drops out and the machines go down. Unfortunatly I can't afford a UPS that could power a single machine, let alone all 5 curently active nodes. Also kinda like the D2OL project....

I will have a play with my settings and try to get some data on what affects the output. I'll use dbench as a benchmark. I wonder if I might can get some info to help out from the guys running the project, tho I feel they may be a bit busy at the moment. worth a try.

On a side note, has anyone been able to download new cands? I've not been able to get any for a few days and I'm almost out.

SS
_________________
my sig disappeared from the image host (?)
But at least I have a Josh Award!

[[KoG]^weaZel]

I will have to check mine tonight to see if they have any new cands. I always set my buffer to 2000 so that I am good for a while in case the server goes down.
_________________
I tweaked and it tweaked back! So I Tweaked some more!

"Barney is like the Michael Jackson of PBS." - James Tybeerious

[2old2care]