Am I right....? 3DMark 2001 freezes the computer

[alexruiz]

Well, I finally managed to put the system together to replace my loyal 1700+ and K7S5A. However, I am having some problem. I guess I already know the solution, but I need some confirmation and opinions of what I think. Also, I need some help to figure out a very low score. These are the specs:

- Athlon XP 2500+ at 3200+
- Biostar M7NCD Pro -NForce 2 400 ultra-
- Sapphire Radeon 9100 64 MB (soon to be replaced by a 9600)
- Buffalo PC3200 CAS 2.5 (winbond BH5 chips)
- WinXP home SP1, fresh install.
- Catalyst 3.7, directX 9.0b
- NForce driver ver 3.13 (Nov 03 ones)
- GC69 (galaxy GC119) heatsink with 3110KL NMB fan 0.13 A (28 CFM, 25 dBA) and OCZ quicksilver
- Powertek 500 W PSU titanium modded with 2 NMB fans at low speed.
- AGP aperture size set at 128M

Once I put the system together I left everything at default , system started normally and worked perfectly. I partitioned the disk. Temp remained in the HIGH 30s after a few minutes of operation!
After this I upped the FSB to 200 at default voltage and the system worked with no problem. Started to install windows and everythig went fine. Temp never went above 47 C with the CPU at 2.2 GHz. System seemed stable also.

I decided yesterday (after installing al the drivers) to test it with 3DMark 2001. It freezes the machine at test 1, or in some cases it crashes to desktop. I suspected either the memory or the overclock were not stable.

I tested then at 333/333 (bus/memory) and ran 3DMark for over one hour. Test completes with no problems. I then changed the setting to 333/400 to see if the memory was the problem. Again, 3Dmark ran for over 1 hour and no problem. I repeated the tests either by changing the settings in the BIOS or using the Nvidia utility. In both cases, the 333 bus has no problems.

I already tried to bump the CPU voltage to 1.675 with 400 bus. This time the system runs almost all the tests, but freezes at test 15. I am thinking that going with 1.70 V should make it stable. I will try it tonight. The temps after running 3dmark for over one hour with the 333 Mhz bus peaked at 49 C (taken from bios directly with an immediate reboot after the test was completed). I don't think heat is my problem, but I could be wrong.

Now in addition to the freeze, I found that the scores returned by the system at 2500+ are abnormally LOW. Running 333/333 the scores average was 6072!, running 333/400 the scores average was 6049! My old system (K7S5A, 1700+, radeon 9000 VIVO non pro) scores 6450! That same system with a Ti200 scored over 7000 once.

I was expecting something closer to 8800-9200 as the radeon 9100 is an 8500. What do you think is holding the score back? I am of the opinion that this is the case where 3dmark really helps, to see if the system is performing at what it should.

Thanks

Alex

 

[DAPUNISHER]

I just checked the reviews for that card@Newegg and 4 different posters all mentioned the following about 2 chips per side on the 64mb model. 1 even mentioned it was slower than his 9000np just like you :Q

If you are getting this 64MB version, beware: about a couple of months ago, I got one of these, and it came with only 2 memory chips per side. At first, I did not worry about it. But I was puzzled because this card was slower than my previous Radeon 9000 NP! After lots of testing, drivers, and even OS reinstallations, I found out that the memory bus was cut to 64-bit because of the missing chips. In the end, I had something that performed between a GF440mx and a 9000 non-pro. Not what anyone would expect or like!

 

[alexruiz]

Originally posted by: DAPUNISHER
I just checked the reviews for that card@Newegg and 4 different posters all mentioned the following about 2 chips per side on the 64mb model. 1 even mentioned it was slower than his 9000np just like you :Q

If you are getting this 64MB version, beware: about a couple of months ago, I got one of these, and it came with only 2 memory chips per side. At first, I did not worry about it. But I was puzzled because this card was slower than my previous Radeon 9000 NP! After lots of testing, drivers, and even OS reinstallations, I found out that the memory bus was cut to 64-bit because of the missing chips. In the end, I had something that performed between a GF440mx and a 9000 non-pro. Not what anyone would expect or like!

I also read the reviews, but I wasn't really thinking it could happen to me I'll check it tonight, and in case this is the case I will RMA it....

Regarding the overclock, what do you think is the problem. The CPU is an AQXEA 0330

 

[DAPUNISHER]

It certainly sounds like it's time to up the voltage and see if it stabilizes as the temps don't seem to be at issue. I'd suggest disabling 8x AGP and fastwrites in the bios as it solved problems between ATI and nF2 for me but since everything is stable@defaults I don't think that's a issue for you.

 

[alexruiz]

Originally posted by: DAPUNISHER
It certainly sounds like it's time to up the voltage and see if it stabilizes as the temps don't seem to be at issue. I'd suggest disabling 8x AGP and fastwrites in the bios as it solved problems between ATI and nF2 for me but since everything is stable@defaults I don't think that's a issue for you.

This is my first experience at overclocking, and it was going too easy....

What is the maximum SAFE voltage that I could go? The cooler is a chip one, and I thought temps were going to be a problem at first (installing the heatsink was a delicate operation that I had to repeat 3 times... I am out of practice) However, seeing that even with a lower CFM fan the system works below 44 C at mild load, I don't think my heatsink installation was that bad

Where can I download memtest86 or prime95? Any other test to stress stability (of course, after I can make it stable in 3Dmark...)

 

[DAPUNISHER]

Hey Alex, keep it 1.8v or less and no worries, besides some of us have found that our 2500+'s will do within 100mhz-150mhz on stock vcore of what they'll do on 1.85v so it's a personal choice what to do. Of course some run upwards of 2v but I'm not the Daredevil/Xtreme OC'er type myself. For anything from p95 to memtest86 and beyond, checkout this thread link

 

[alexruiz]

Originally posted by: DAPUNISHER
Hey Alex, keep it 1.8v or less and no worries, besides some of us have found that our 2500+'s will do within 100mhz-150mhz on stock vcore of what they'll do on 1.85v so it's a personal choice what to do. Of course some run upwards of 2v but I'm not the Daredevil/Xtreme OC'er type myself. For anything from p95 to memtest86 and beyond, checkout this thread link

Well, I bumped the voltage to 1.70 and the system now runs 3dmark2001 for over 2 hours with no crash.... so it works!

Regarding the video card, you were so right! Mine has only 2 memory chips per side.... I already requested the RMA. This really surprises me, as it seems the sapphire engineers overlooked this fact. I am pretty sure they are comunizing the memory chips from the 128MB cards, and they thought that using the same chips (16MB instead of 8MB) was better....

I decided that I will leave it stock settings for my wife (2500+, 333 memory) and when I use it for something more serious, I will use a software utility to reach the 3200+ level.... (400/400).

My final coment is the CPU temp. When I got the heatsink, I was advised against cheap coolers. I went against the advice and got a cheap GC69. Even with a fan that is LESS powerful that the stock fan I was unable to break the 50 C barrier until yesterday after the 2 hours for 3dmark at 3200+ (53 C reported by motherboard monitor)... temps at light usage remain at about 41-43 at 2500+.... impressed with the cheap cooler. I guess the 3 lungs clip design really helps. So if my wife ever complains about the noise, I can use a panaflo instead.

 

[Duvie]

Two things alex....

1) Never install OS on an oc'd machine and especially one tha have not even been able to stress test as being stable. May corrupt the windows install or any of the driver installs.

2) When ocing you wnat to isolate the cpu first so therefore run the memory at really conservative or unoc'd speeds until you understand the limit of the cpu....In the end raw mhz speed of that cpu will win out. Tweak cas timings amnd what not at the end.

 

[alexruiz]

Originally posted by: Duvie
Two things alex....

1) Never install OS on an oc'd machine and especially one tha have not even been able to stress test as being stable. May corrupt the windows install or any of the driver installs.

2) When ocing you wnat to isolate the cpu first so therefore run the memory at really conservative or unoc'd speeds until you understand the limit of the cpu....In the end raw mhz speed of that cpu will win out. Tweak cas timings amnd what not at the end.

Wow, good pieces of advice here. I did install the OS when it was overclocked..... The memory is DDR400, so it was runnning since the beginning at the rated 400 MHz. I have not messed with the timings, and maybe I won't. It is rated 2.5 CAS (2.5 - 3 - 3 - 8 at PC3200). I wonder if I could go faster (Buffalo DDR400, winbond BH5 chips) Give me a guess, Do you think the memory could run at CAS 2?

 

[alexruiz]

One more question. DDR400 CAS 2.5 should have no problems running at DDR333 CAS 2 at fastest timings, correct? (2 - 2 - 2 - 6)

 

[Duvie]

Originally posted by: alexruiz
One more question. DDR400 CAS 2.5 should have no problems running at DDR333 CAS 2 at fastest timings, correct? (2 - 2 - 2 - 6)

You would think!!! definitely do not set it at those timings during an install of OS...I would get a program called memtest86 and make a bootable cd-rom or floopy (if it can fit) out of it and then restart computer with cd-rom aor FDD as first boot drive. When it restarts it will go into running that program. Hit configuration go to cache and turn always on...then go to test and click all of tem...let it run 5-10 passes. no errors means you should be fine...