Just purchased X1950Pro AGP

phxz31 · January 16th, 2007, 06:47 AM

OK, well I did a few more benchmarks... with strange results.

First, a new personal high score in 3DMark '06, just slightly higher than the last. I got 4078 points. Not a bad score for my system I think.

So then I installed 3DMark '05. First time I ran it was only 7109 points, which is much less than the score Gumbo posted, of 7699 points. I tried tweeking a couple of things and raising the FSB/memory frequency a bit, and my subsequent 3DMark '05 scores got even lower. Second result was 6621!!! Extremely low, especially considering I had RAISED almost all of my systems clocks.
Confused, I set the clocks all back to the same settings I obtained my high 3DM'06 score, and still got a score lower than my first run.. 7024!

Confused even more, I ran 3DM'06 again, and got 4076 points, right in line with the last two scores I got on 3DM'06.

So, here are the system specs (again) that got me the highest scores in both scenarios:
mobo: NF7-S (nforce2 ultra) @ 227mhz fsb
ram: 1GB (2x 512mb), dual channel, @ 227mhz CAS 2.0-2-2-11
cpu: Athlon XP 2500+ @ 2388mhz, 1.85volts
AGP bus: 87mhz, 1.8volts
video card: GPU @ 634mhz, Video RAM @ 1525mhz

Anybody have any ideas why my 3DMark '06 score would be this high for my system specs, and very consistent.. yet my 3DMark '05 score would be very low and very inconsistent? This sort of pisses me off.

gumbo · January 16th, 2007, 11:06 AM

Have you optimized the ATI drivers for performance and not quality ?
I found that changing AGP memory from 64 to 128 i bios gave positive effect on scores, same with Enhance PCI performance.
Are you sure setting a higher AGP clock helps ? I ingreased mine from 66 to 70 without any noticable increase in score...

When overclocking my x1950 to 614/753 today i got 7804 3Dmark05
http://service.futuremark.com/orb/re...=0&UID=7621277

Yupp, i know 3Dmark01 is way to old for my hardware, but it has been usable so far to see performance increase when tweaking settings, i'll use 05 from here on.

I also checked the minimum requirements for 3Dmark06:
"Intel® or AMD® compatible processor 2.5GHz or higher "
I think maybe our scores in 3Dmark06 are to much influenced by slow prosessors to see any noticably gains on X1950 tweaks...
Minimum requirements of 3Dmark05 is :
"x86 compatible processor with SSE support, 2GHz "

phxz31 · January 16th, 2007, 04:29 PM

Quote:

Have you optimized the ATI drivers for performance and not quality ?

Yep. When I run a benchmark, I set all the settings to "performance", although it seems that I get very negligeable increase in score over leaving the settings at "high quality", but it does give some increase.

Quote:

I found that changing AGP memory from 64 to 128 i bios gave positive effect on scores, same with Enhance PCI performance.
Are you sure setting a higher AGP clock helps ? I ingreased mine from 66 to 70 without any noticable increase in score...

Well I have always set my AGP aperature size to half of my system RAM as a rule of thumb. I used to have it at 256mb when I had 512mb of RAM, but when I upgraded to 1GB I changed the AGP aperature 512mb. I have not experimented with smaller sizes at all, perhaps I'll go back to 256 or even 128 and see if it changes anything.

As far as my AGP bus speed is concerned.... no. I do not get any noticeable change between running it at 66mhz or 83mhz+. I have always just done it because 83mhz gives the bandwidth equivalent to AGP 10X, which I figure would help eliminate the AGP bus from being the bottleneck... though as I already mentioned, it seems to do nothing really.
Actually, I'm quite impressed with the performance of the card on AGP, especially considering the bridge chip inbetween the AGP interface and the actual GPU core. Even running at AGP 8X (66mhz) my graphics test scores on 3DM'06 are almost exactly in line with scores from a brand new system. This really just indicates to me that AGP was killed off prematurely for economic reasons rather than for any real performance benefit. The strange thing is this: PCIe 16x performs much better than PCIe 8x boards, as indicated by older SLI / Crossfire systems that split the 16x into 2 8x PCIe lanes. However, even with similar bandwidth to PCIe 8x, AGP 8x performs closer to PCIe 16x performance wise.. (did that paragraph make any sense?)

Quote:

I also checked the minimum requirements for 3Dmark06:
"Intel® or AMD® compatible processor 2.5GHz or higher "
I think maybe our scores in 3Dmark06 are to much influenced by slow prosessors to see any noticably gains on X1950 tweaks...
Minimum requirements of 3Dmark05 is :
"x86 compatible processor with SSE support, 2GHz "

Well, yes, our processors are very old. Mine is whole system is almost 5 years old. However, I believe that 3DM'06 is optimized to take advantage of dual core CPUs, and since my CPU score is about half of that of a modern dual core CPU with the same clock speed, I'm actually very satisfied with its performance.
Actually, I know that the CPU is probably the largest bottleneck on my system, but I can prove that its not affecting the GPU tooo much because changing my GPU core clock has a significant impact on my 3DM'06 score, as well as game performance. This indicates to me that my CPU is more than adequate at feeding information to this GPU because I can crank the GPU speed up and the CPU just says "Ah! You want some more do you?"

gumbo · January 16th, 2007, 09:20 PM

Yepp, you should try 128mb AGP memory, it gave med a bit higher score than 64 or 256mb. You probably don't want to use AGP memory anyway as it's to slow compared to the 256/512mb videomemory (as you know).

I got texture corruption when i tried 72mhz AGP speed, it looked like the the GPU was clocked to high. Maybe high AGP speeds has negative effects on overclocking the videocard ?

So far i've settled on 7984 in 3Dmark05, it seems my CPU can run stable (prime) on 2500mhz (1.9vcore) now that i got an aircooled videocard. My previos videocard was watercooled and helped warm up the CPU's water.

Any improvements on your results ? As your memory and FSB is faster than mine, you should at least hit 7650-7750 points...
I reached 7854 (max) @ 2.4ghz, so my scores increased only 130 points with 100mhz extra CPU speed, which is why i still belive the CPU isn't the main bottleneck, and going from cas 2.5 to 2.0 does almost nothing with the scores either...

Dyre Straits · January 16th, 2007, 11:42 PM

IN my own past experience in overclocking AGP-based systems, I alway found that locking the PCI/AGP to 33/66 gave me overall better results than allowing the AGP to overclock at all. Of course, my previous boards were mostly ABIT boards which have some excellent OC'ing features.

I also concur with gumbo that you should try lower AGP Aperture settings with the newer motherboards. The reason being, pretty much as stated, that the onboard VRAM is faster than the AGP bus. So, give it a whirl at all the settings before settling on what works best for you.

phxz31 · January 17th, 2007, 01:31 AM

Obviously the system RAM is much slower than the onboard video RAM... but shouldn't it only fall back to using the system RAM when theres too much data to fit on the onboard RAM anyways? So, hypothetically, shouldn't it not matter what your AGP Aperature is set to unless your onboard video ram is too small or too low of bandwidth to service the GPU properly?

Anyways, I'm giving it a shot. We'll see what happens with 128MB AGP Aperature. I won't have a chance to bench it until later tonight though so I guess check back tomorrow and I'll have tested it by then.