DF CPU comparison?

[Ragnarog]

Here are my findings so far:

Athlon 1330: 38 structures / minute

Athlon-4 1000 (Pally core): 35 structs/minute.

Ok, thios could be a bit wrong...I donno...

What I'm after are Pentium 3 and P4 scores/minute please....

Rag

[nettnut33]

You may want to think in terms of how many structures per hour instead of minutes. Each structure is different, and some take longer than others (sometimes MUCH longer). As an example, I was trying to benchmark a PIII @ 733 for you. After 30 minutes it had produced 860 stuctures. At that rate it should produce 41280 structures per day, but it doesn't. However, it has been running for exactly 24 hours and has produced 37125, divided by 24 equals 1547 per hour. Soo, here are my benchmarks:

Athlon XP @ 1725mhz, winME, 512mb DDR: 3800/hour 92k/day

Athlon TB @1400, winME, 128mb DDR: 2920/hour 70k/day

PIII @ 733mhz, win98se, 128mb sdram: 1547/hour 37k/day

PIII @ 650mhz, win98se, 128mb sdram: 1350/hour 32k/day

These are averaged, with the -qt switch enabled (no display). It seems to add about 5k per day on my Athlon 1400, and 7.5k per day on the XP. DF is optimized for Intel chips.

Hope this helps!

NI!

[Ragnarog]

Yeah, I noticed that and did a few averages before posting....but gotta admit, nothing as exact as your figures :)

Thanks a lot...now we only need a P4 score cause DF seems to be performing extremely on those...

From your scores a P3 and Athlon TB @ 1 Ghz score 50000/day while the Athlon XP at 1Ghz is doing 53000.

Rag

[Fallguy]

Well I did a little post over at OCN on CPU performance, lets see if I can find the link.....

http://www.piratefleet.com/forums/cgi-bin/ultimatebb.cgi?ubb=get_topic&f=6&t=006311

There we go.

I always try to take performance on a daily basis, and then average back to find an hourly basis. In fact even the daily results tend to fluctuate quite a bit, so I usually average a weeks output.

My P4 1.6Ghz with 256MB SDRAM (work machine) averages 50K per day.
My Duron 1Ghz with 256MB SDRAM (home machine) also averages 50K per day.
My Celeron 1.2Ghz with 256MB SDRAM averages 40K per day.

From what I have seen, the code is not really that optimised in favour of the P4, it simply is bandwidth restricted. This makes the P4 with DC RDRAM the fastest around, NOT because of any optimisation, but simply because the data can be fed to the CPU much faster.

Fallguy

[Ragnarog]

Judging by your numbers that makes sense yeah...

Tho im slightly concerned about your slow Celeron 1.2 Ghz performance....40k/day

2nd post here in the thread a 733 P3 averages 37k/day....

Maybe the smaller L2 cache on the Celeron makes it perform so badly?

Rag

[Fallguy]

The 1.2 Gig Celetin (Celeron-Tualatin) has 256K L2 cache, and in almost every way is a match for the old P3. The unfortunate exception is the bus speed, where the Celeron only uses a 100Mhz bus. This limits the bandwidth badly, and hence lowers the performance. In most instances the 1G P3 on a 133 bus will outperform the 1.2G Celetin at DF, even though the Celetin has hardware prefetch and slightly improved cache structuring and buffering.

Its a bandwidth thing. Of course if this Celetin would just clock to 133Mhz bus (1600Mhz) it would be the king of all P3 systems.... but at the moment it won't :(

Fallguy

[Ragnarog]

Doh, of course....i feel stupid now...should have remembered the 1Ghz+ Celerons are Tualatins...

Still a BIG difference there in scores..

even if you look at the 650 Mhz P3 (which must be a 100*6.5 model)...

Something's fishy :)

Rag

[nettnut33]

The PIII @ 650 is a Dell laptop. I'm pretty sure it's running pc100.

NI!

[Fallguy]

Yeah,
my 1.2Ghz Celeron is running at PC100 as well (815 doesn't allow +33 :( ).

I think the answer to the small difference in performance between the P3 650 and the Celeron 1.2 is the bandwidth bottleneck again. The 1.2Ghz CPU is obviously very data starved, so its is probably losing alot of its cycles waiting for data to reach it.... :(

This shows how bad the situation is I think, and if I am right the Athlons are even more data starved than the P3's (i.e. if the coding was shrunk to fit in the 384K of cache for an Athlon, it would crunch alot faster). This reminds me alot of the old SETI program, when the code required 512K L2 cache to run, otherwise it took a huge performance hit (which made Xeons and P3 Katmai's the kings). Then the code was shrunk enough to fit in 256K cache, and suddenly the Coppermine came into its own.

Fallguy