Prime95 rounding errors, not overclocking!

[MADhix]

Hello everyone.

I was hoping the people at AnandTech forums could help me with my problem. First let me list my system specifications.

CPU: AMD Athlon XP 3200+ 400MHz FSB
MOBO: ASUS A7N8X-E Deluxe BIOS Rev 1009 and 1011 (problem occurred with both versions)
RAM: Kingston ValueRAM Model KVR400X64C3AK2/1G 2x512MB PC3200
VIDEO: ATI RADEON 9800 PRO 128MB (built by ATI)
PSU: Antec TrueControl 550
HDD: Western Digital WD2000JB 7200RPM 8MB EIDE
COOLING: Thermaltake Extreme Volcano 12 A1745

I am not overclocking. I got the HSF so that my CPU can run cooler. My motherboard shipped with BIOS Rev 1009. I ran prime95 and it failed after 1 hours, 30 minutes with a "FATAL ERROR: Rounding was 0.4999661446, expected less than 0.4". It then would fail around a 4 hours mark. I updated the BIOS to Rev 1011 because I heard from these forums that 1009 and 1010 are bad. I got the error after about 7 hours after updating. I just reflashed Rev 1008 and currently testing.

It may be significant to note that I had this same problem with the first 3200+ I received. It had factory HSF and the temp at max load was around 58-59C. People said that the errors was most likely caused by heat. I then bought the cooler and it brought the max load temp to around 48-49C. I still received the rounding errors and I have received a couple resulting sum errors with the old 3200+. I put an 1800+ in my board when I had to RMA the 3200+ and the FSB was set to 133 and it had no errors and I ran it 65 hours.

I have been reading these forums and noticed that maybe this problem is caused by a BIOS Rev, voltages, I don't know. I am running everything at stock and I get these rounding errors. Please let me know any information you all may need to help.

I would greatly appreciate any help.

 

[edmundoab]

test memtest86 first and see if there is any errors with the memory.

then only try to figure out whats wrong with Prim95,
as prime involves other components as well.

 

[MADhix]

Sorry, I forgot to say that the max load temp with the new 3200+ is 45-47C (ambient temp alternates a lot).

 

[MADhix]

Woops, I forgot to say that. I ran memtest with my old 3200+ I believe and it ran 24 hours with no errors.

 

[ectx]

I'd try lower fsb and higher multplier to see if it is the problem of the mb... some mb just cannot handle high fsb well enough.

 

[MADhix]

I don't think I can change the multiplier.

 

[Gillbot]

I'm having similar issues with my SN45G running a 1.8 Duron using PC3200 Ram.

Memtest86 passes 100% and Prime95 passes at 1.8GHz. Temps are about 40°C full load and between 35 and 37°C idle. Stock voltage also.

I tried to run at 2.0GHz by changing the multiplier only and that resulted in max of 42°C loaded temps with idle temps the same. Failed Prime95 after about 4 hours and then hard locked after it reached 42°C. I never tried to push it anyfarther with voltage or anything, I'm just trying to keep it cool and stable at 1.8Ghz for now.

 

[ectx]

Oops. I think you are right unless you got an older chip.. Sorry about that. I don't assume you have another mb to test it on?

I'd suggest you run small fft test to test the chips only to see if you still have error -- if so I'd say you have problem w/ memory or mb - since you can pass memtest, more likely it is the mb...

I'd also slow down the timing of the ram just to see if it is the ram.

Have you tried any 3d-applications to see if it works ok. Mmy asus a7n8x+ and 8rda+ (both v1 nforce2 chips) can not handle 3dmark at fsb higher than 195 - they would pass p95 at about 196-198. Move the same parts to an abit nf7-s and everything is ok at 207fsb (never tried to push higher yet).

Anyway - my apologies for the dumb suggestion. Good luck to you.

 

[MADhix]

Right now I am running the test with 200MHz FSB, RAM synced with FSB, vcore at 1.65 (stock), vdimm at 2.6 (stock), timings at 3-3-3-8 (I think these are the standard timing for my RAM, all I can find on www.kingston.com is 3-3-3), agp voltage at 1.5 (default), agp aperture 64MB with BIOS Rev 1008.

It has ran 2 hours with no errors so far.

I have already RMA'd the CPU once, this sucks... =(

 

[MADhix]

Well, it just failed.

"FATAL ERROR: Rounding was 0.49858284, expected less than 0.4"

Does anyone have any ideas?

 

[edmundoab]

The rounding error could be cost by many factors, frankly, I experienced it while overclocking, but since you are running it at stock. My other thought was the memory.
Have you tested it on memtest86 to see if your memory are getting enough voltage or stable ?

 

[Jeff7181]

Can SuperPi be made to loop indefinately? If so, I'd try that with a small enough test, like 256k or 512k so that it doesn't use RAM. Then you'll eliminate the RAM as a possible problem and still stress the CPU.

 

[MADhix]

I ran memtest for ~24 hours and It passed. I don't know exactly which settings I picked; although, I think it was the standard test or something. I don't really remember.

 

[MADhix]

I am running prime95 again except this time I changed the RAM to be 83% of FSB which yields 166MHz.

Couple of things I noticed. I set my FSB to 200MHz manually and yet CPU-Z says the FSB is 199.2. I also have my vcore set to 1.650 in BIOS and CPU-Z in a two minute time frame will read vcore of 1.632, 1.648, 1.664, 1.680, 1.696, 1.712, 1.728 and 1.744 in random order.

Is there anything that may come of this information?

 

[SickBeast]

What about the cooling? IMO Thermaltake makes the biggests POS heatsinks out there, yet people buy them in droves. Granted, I know nothing of the one you're using, but it may not be up to the task.

Chances are that's not the issue though. I would try this:

-set your RAM timings to 3-4-4-11
-raise the CPU voltage by 0.1v, then check for stability
-if it remains unstable, put the CPU voltage back to stock, and raise the RAM voltage by 0.2v

That should tell you if it's your CPU or your RAM. If it's neither of those then you pretty much know it's your motherboard, so you might have to RMA it.

It could be your cooling though; 47C is pretty hot, especially considering you're not overclocking. To give you an example, my CPU doesn't hit 50C, even while running at 2500mhz at 1.825v under full load.

 

[someone16]

Cooling can't be problem when its only in the 40-50 range (Unless the temps are wrong). Get a program called Speedfan and monitor the fluctuations of the power supply when running Prime95.

 

[AWhackWhiteBoy]

ASUS A7N8X-E Deluxe, i had this motherboard, and returned it. i had the same problem with not being able to hit 200fsb. set the FSB to 195 and run prime,if that still fails,try 190. i couldn't get past 192 with my Asus without it crapping out. i hated that board....

"IMO Thermaltake makes the biggests POS heatsinks out there"
i use his same heatsink with my barton@2.6ghz, never gets above 45C

but,yes, thermaltake tends to just make cheap rip-offs now. their products are getting much worse from what they were a few years ago.

 

[3LEMENT0]

I had rounding errors in prime95 too but passed all test suite: memtest, sandra burn-in, windows memtest, dr. hardware benchmark. I upped the voltage a couple of times and prime95 ran ok afterwards. It could be voltage related, NF7-S seems to be giving lower voltage than what's in the bios.

system specs:
NF7-S
AXP-M 2600+@2532MHz at 1.9v (actual is around 1.8x)
1Gig OCZ perf at 2.7v 11-3-3-3

Good luck

 

[Gamingphreek]

Im running the Asus right now. Its rock solid. Im running my 2500+ at 2.1GHZ at .025 over stock voltages. Pretty good M/B. Try maxing out your VDIMM (1.8volts i think).

Also 50C is fine. Mine is using stock cooling (soon to change) and im hitting load temps of up to 67C and im not having any trouble. Rock Solid. Albeit my case is sorta extreme with the temps and all but nonetheless it easily passes Prime 95.

Also Asus boards overvolt themselves in order to gain stability.

-Kevin

 

[myocardia]

...

 

[TStep]

Been a while since I fiddle much with the socket a / nforce2 combination. When I was screwing with it, I do recall that if I set the CAS to 3, most of the time I could not boot. IIRC it was an Epox 8RGA+ with Twinmos BH-5. Maybe the problem was specific to BH-5, but CAS 3 was definitely a problem.

It was always my understanding that CAS 3 and nforce2 chipset was somewhat incompatible.

 

[Jeff7181]

Originally posted by: myocardia
Kingston Value ram isn't meant to be run in dual-channel mode. Didn't you read the package it came in? You need to swap the stick that's in slot 3, and put it into slot 2, so it will be running in single-channel mode.

LMAO...

 

[Avalon]

I have the same problem with CAS 3. I can't set my memory to CAS 3 or I will lose the ability to boot.

 

[MADhix]

I have never heard of that.

 

[MADhix]

Oops, I did not see a second page. I was referring to myocardia.