Rounding errors in Prime95

[crush3r]

I built a new pc with an e2180 and p35-d3sl gigabyte motherboard.

I went into the bios and loaded the fail safe defaults to test for a baseline. I ran Prime95 V25.4 small-fft for a few hours...and everything passed. I then started a Blend test and I kept getting "FATAL ERROR: Rounding was 0.5, expected less than 0.4" in less than a minute of each test. Pc responded fine after it failed, no blue screens, reboots, or freezes.

I switched over to Orthos ran the same small-fft and blend tests and it ran fine on each test for a few hours.

Any suggestions on what might be causing this? Is this a bug in Prime95 or is my pc unstable even at fail-safe settings?

Thanks

 

[Evilonigiri]

That BSOD means it's most likely a ram issue. It means one of two things: either your ram is bad or it's unstable.

Run Memtest and if it passes your ram is good. Then run blend test for at least 8hours, 12hours recommended, and if it fails, tweak your ram around to get it stable.

 

[crush3r]

I'm afraid you misread what I wrote. When Prime95 fails nothing bad happens. I do NOT get a blue screen, reboot, or freezup.

That being said, I ran Memtest V 2.01 a bunch of times and I keep getting errors during test #5. I tried different bios settings such as increasing the voltage and decreasing the frequency.

I have Corsair XMS2 TWIN2X2048-6400 memory sticks. Does this mean that one of my sticks is bad? or am I just missing something?

Thanks for the help

 

[llnyc]

When Prime95 fails something is wrong, it doesn't have to go BSOD.
Your memtest tells you one of the sticks is bad. Try test one stick at time, try different slots.

 

[crush3r]

ok. I tested each memory stick separately and one stick passes all tests, while the other one has 18-28 errors at test 5 everytime.

I take it this means I have a bad stick of memory and probably has to be rma'd with corsair?

edit: ok i've been testing the "good" stick of ram. It still fails Prime95 and when i do memtest again I get errors on test 6 now.

 

[Mondoman]

Much RAM today is sold as a factory-overclock, requiring a voltage higher than the DDR2 standard 1.8V in order to operate to spec. You likely need to go in the BIOS and manually set the DIMM voltage and timings to those in the RAM's spec.
If that doesn't solve the problem, please list the model numbers of *every* component in your system, including things like the CPU cooler, in order to give us a better picture of what's in your system.

 

[llnyc]

Since you are running everything at spec settings, which you should have already double, triple checked, I'd say something is wrong with the following basic components, from high to low likelihood: ram, mobo, psu, vid card, cpu.

 

[crush3r]

All right, I've been testing all morning and I cannot get past 5 minutes of blend tests in Prime95. (Keep getting FATAL ERROR: Rounding was 0.5, expected less than 0.4")

I havent seen any problems yet with Windows. So far I've ran Firefox and Team Fortress 2 without any errors. But it still annoys me that Memtest and Prime95 will not pass.


My System is-
Windows XP 64bit SP2
Antec 430W Earthwatt
Intel Penitum 2180 - AC Freezer 7
Gigabyte P35-D3SL - Bios F7
Corsair XMS2 TWIN2X2048-6400
Nvidia GeForce 8800 GS
Seagate 250GB HD (SATA)
IDE DVD Burner (I forget which model/brand....its a few years old)
120mm Fan

CoreTemp shows my idling at about 25 C. With Prime Running, I top out at about 40-45 C before it fails

I've been adjusting the bios settings all day. I've increased the DDR2 voltage to +.1V (Memory is rated for 1.9V). I've lowered the memory multiplier to 2 (so its 1:1). I've lowered all memory timings a hair, and I still cannot get Memtest and/or Prime95 (blend test) to pass with out any errors.

Any help is appreciated

 

[Mondoman]

The EW 430w has a 30A limit on total +12V; perhaps that's an issue.
More likely, I note that you're not using the stock Intel CPU cooler or similar "downdraft-style" cooler. Thus, you're missing the secondary cooling from the CPU cooler, which normally blows air onto MB areas near the CPU, such as the northbridge and PWM areas. Either of these overheating will reduce stability; stressing via Prime95 would especially affect these two areas.
Try installing the stock Intel CPU cooler (which should be plenty even for OCing the e2xxx).

PS - If you're getting errors in memtest86+ and Prime95, your system is definitely not running correctly. While Windows may seem to work now, its files are slowly getting more and more corrupted. Once you get the issue fixed, don't forget to reinstall Windows (or at least do a "repair install").

 

[crush3r]

I already checked the PSU and I'm almost certain its not the issue.

I dont think its an over heating issue. Im running at stock speeds (10 x 200MHz FSB) which my motherboard should be able to handle easily.

I got an OEM processor so I dont have the Intel heatsink. Although, to test your theory, I did flip the case fan around (so its blowing on the mother board) and took off the side panel. I ran Prime95 and Memtest and I still get the same errors.

Also, I went through and tested each stick individually. One stick gets errors on Memtest # 6, while the other gets errors on test #5 and a couple on #6.

 

[crush3r]

Does anyone have any more suggestions on what I should do? It seems to me that it is either the memory or motherboard that is bad. Are there any more tests I can run to help me figure out which component is at fault?

 

[xringx]

While I dont know the exact specs of your mobo, my Gigabyte runs memory at 1.8v

If I did the correct look up on your memory, it should run at 1.9v

You may just need to raise the voltage on the memory to the correct number..

 

[Mondoman]

I already checked the PSU and I'm almost certain its not the issue.

How did you check the PS?

I dont think its an over heating issue. Im running at stock speeds (10 x 200MHz FSB) which my motherboard should be able to handle easily.

Yes, if it is properly cooled. However, you're using a non-standard CPU cooler and we can't see the case config. What kind of temp readings are you getting for CPU, northbridge, and PWM areas?

Although, to test your theory, I did flip the case fan around (so its blowing on the mother board) and took off the side panel.

Just to clarify, if you mean that the 120mm is now mounted above the CPU area of the MB, blowing air down onto the MB northbridge and PWM areas (rather than parallel to the MB surface), then I agree that should be a reasonable test.

Finally, did you ever just run the RAM on all "auto" settings?

 

[llnyc]

Only thing I can think of is to "Borrow" another stick of compatible ram, and another PSU from another PC or friends. I think you would have already tried it if you have spares laying around. You can also test your ram in another pc.

You should have already stripped down the PC to the bare minimum--cpu, 1 stick of ram (try diff slots), vid card, 1 IDE drive. Also disconnect the usb/1394/audio headers, clear cmos memory through jumper and power cycle.

 

[crush3r]

How did you check the PS?

I used a multimeter and followed the instructions here http://jongerow.com/BFGPower/troubleshooting.html

All the values were within 5% while idleing and under a load.

Just to clarify, if you mean that the 120mm is now mounted above the CPU area of the MB, blowing air down onto the MB northbridge and PWM areas (rather than parallel to the MB surface), then I agree that should be a reasonable test.

No, I had it so the fan blow across the motherboard. I'll have try it the otherway tonight.

Finally, did you ever just run the RAM on all "auto" settings?

Yes, the auto SPD settings for the ram were 5-5-5-18 @1.8v

Only thing I can think of is to "Borrow" another stick of compatible ram, and another PSU from another PC or friends. I think you would have already tried it if you have spares laying around. You can also test your ram in another pc.

Yeah...problem is all my friends who are local either have ancient (ie no DDR2) PCs or have laptops. If I were back at school, I wouldnt have this issue.

I was considering buying a stick of this ram off newegg http://www.newegg.com/Product/Product.aspx?Item=N82E16820134044

Its cheap, is on my motherboards compatible memory list, and would hopefully help me figure out which component is bad in my pc.

Right now I'm thinking its the memory because each DIMM has its own set of errors that is independent of the slot I place it in the motherboard. One stick always has error bits 00008000 on test 5 and 6, while the other always has error bits 00000080 on test 6. If it were another component, I would guess that the errors would be similar for each stick of memory or more random.

Again...thanks for the all help everyone.

 

[xringx]

You for sure have the ram at 1.8v??? Spec per manufacturer is 1.9v

That could very well be your problem. Yes, .1v makes a difference

 

[crush3r]

Ok so finally after a couple days I was able to get one stick of memory stable.

I set timings to 6-6-6-18 @1.9V and set the memory multiplier to 2 (so it was DDR2-400). I ran it through 2 passes of memtest with no errors and ran a Prime95 blend test for 2+ hours without any errors.

Afterwards, I tried bumping the multiplier to 3.33 (for DDR2-667) and I immediately got errors in memtest.

I think this shows that my memory is bad and it needs to be RMA'd.

I was considering buying some new ram while I send in for the RMA. Any suggestions on what to get?

Thanks for all the help.

 

[Evilonigiri]

Transend Bad AXE DDR2 800 ram OCs like crazy. Seems to be very good.