MSI PM8M3-V rev. 1 - unstable gaming only

[trodas]

Before reinstaled, I installed the NETframework. Version 2.0 works, just install of v2.0 SP1 does not. But in both cases then the testing tools failed miserably:





So testing tools that require way too much are out of the question.


MSI PM8M3-V bois v1.4 (+ unlocked v1.4) and v1.5 (+ unlocked v1.5):
MSI PM8M3-V bioses.zip
DepositFiles
MSI PM8M3-V bioses.zip | Ulo?.to


But now I have the C (Win partition) backup w/o any drivers at all (no VIA 4+1, no sound driver and no Forceware driver) and after that installed. And the reinstall really helped. Call of Duty (1), Quake 3 or the cursed SoF2 not crashing on exit anymore!
Hoooray!

...yet still they crash during longer gaming, witch I blame on the cap and overheating of almost everything inside. Mainly the GPU VRM (71°C with just small ATItool cube!) and the NB. The NB I plan to fix using the Thermalright SI-128 SE cooler (just get the missing female screws for P4) and the GPU VRM? Well, I don't know why it overheat that much. But maybe the Chemicon 330uF 16V SMD d10 polymers are not as good as they should be and a replacement of the TWO input caps on the card (the third and last cap on the board is 100uF suxxka) could help?
Or add another caps for the voltage filtering of the 12V input?
Or add a serious cooper custom made heatsink on the GPU VRM?

[trodas]

Prime95 overnight CPU torture test - 7h 8min


No errors. Claims, that the machine is unstable in terms of CPU are dismissed. It must be the gfx card and/or it's powering.

[trodas]

I just realized why the Furmark does not work for me, even the 6800GT is supposed to support OpenGL v2.1:
NVIDIA GeForce 6800 GT AGP | techPowerUp GPU Database
Well, it is quite simple. The Forceware drivers started supporting the OpenGL v2.0 required for Furmark since version 77.72:
http://www.nvidia.com/object/winxp_2k_77.72
I use version 61.76, witch means that the most I could get is v1.5 of OpenGL. Everest helped me to realize my mistake...

So up for next - a GFX card test :D

[trodas]

Well, the more recent drivers worked and now the card support OpenGL v2.0 and hence - it run FurMark

Got a very low score of 547:
OpenGL Graphics Benchmarks Scores | oZone3D.Net

...but the problem is, that the FurMark did not crashed for about TWO hours straight MaxBurn test...



I was somewhat confused, that the image on screen did not update, but that it probably because FurMark is checking for bugs in the image, right?
Never the less, I have complain. On screen it write this:
Renderer: GeForce 6800 GT/PCI/SSE2

Obviously I have a problem with the "PCI" part, because the card is AGP and given some of the slow-downs it really looks like as it run on PCI speed, not AGP 8x...

No idea what this cause, but probably some wrong initialization of the card. Regardless if this works as I supposed right, then the card seems to be stable.

There are minor annoyances too, tough they are not new at all. First at all - opening overlay took for the first time eternity. (1 - 2 minutes) Dunno why that happen, never seen that in my life, but it does happen with different Forceware frivers (61.76 WHQ, 77.72 WHQ) as well, as with different VIA Hyperion 4in1 drivers (v5.24a, v5.11a, v5.04a). So it have to do something with what the card seems to initialize.

ALSO I cannot get the AGP page back in the drivers. It was there:



...and now, even I add to the registers the coolbites:

REGEDIT4

[HKEY_LOCAL_MACHINE\SOFTWARE\NVIDIA Corporation\Global\NVTweak]
"CoolBits"=dword:0000000f
"NvCplEnableHardwarePage"=dword:00000001
"NvCplEnableAdditionalInfoPage"=dword:00000001
"NvCplEnableAGPSettingsPage"=dword:00000001

It is just not there.
Also a minor problem is, that the gamma value in nVcpl can be set only between 0.5 and 3.61 - too sensitive, no way to set better value or get back to 1.0 :(

Kinda bad that FurMark run slow, yet not fail. I did not dare to take a screenshot of the "PCI", but the slow-downs in SoF2 looks like something is wrong with the speed...

[trodas]

Aaaargh! My card is really running in PCI mode:



...that is why the speed suxx and also why the nVidia drivers therefore CANNOT show the AGP page, when the card is running in PCI mode...

Looks to me as simple yet devastating HW issue with the cursed not replaced cap. Now let's source a replacement cap... :(

[trodas]

Replacement CPU arrived (a HUGE thanks must go to OC AU user Datsun 1600 for creating the missing screws for the Thermalright SI-128 SE and giving me great P4 CPU for neat CPU upgrade! Overclockers Australia Forums ), so I went gaming and... again the PCI/AGP troubles, slow performance, etc. And it crashed during my gaming once again :( Same way as before. Changed box cooler to Thermalright SI-128 SE (greatly temps get reduced) and also now the 120mm fan is blowing to the NB heatsink - yet still no help at all. Crashing. New CPU at least feel much faster now:

Old MSI PM8M3-V (VIA P4M800) board, new Pentium 4 540 (3.2GHz, 1MB L2) CPU :) Faster a "bit" - 45sec compared to a 72sec with Celeron D suxxka :)


CPU-Z Validator 3.1

old: View image: Super PI 1 M 1 12 641 Celeron D 336

So it is definitively NOT the CPU. The CPU might be dead-slow, yet worked just perfectly. No help in sight, except maybe replacing the last cap with something good, like 10uF elyte change to 47uF tantalum cap?

[trodas]

Now every software claims that the card run in PCI - the damn card must changed the slots by itself, and just forget to tell me about it :D


After all it more and more looks like a HW issue related to the capacitor AND mosfets around the AGP. Usually when instability happens, increasing voltage fix the problem. In this case, increasing voltages = increasing prblems. The mobo is now fairly stable - that does not mean it is not crashing anymore, it does crashing, but not that much often - (even in light gaming like SoF2, CoD (1) and Quake 2 (KMQuake port with 300MB of hi-res textures) when I use the lowest AGP voltage possible.
That lead me to believe that the capacitor is the culprit, that overheat the mosfets and they might even need resoldering to the mobo... So next step - obtaining a replacement cap. I think that tantalum 47uF 10V cap ( Invalid Request ) would be a good and suitable replacement. Plus maybe adding some little heatsink on the little mosfet?

Should be also noted, that I increased the FSB a lot from 133 to 200Mhz!
CPU-Z Validator 3.1
CPU-Z Validator 3.1
And it does not affect the instability of any way. If the problem is the chipset, being unstable for FSB 133, then it would go VERY bad when raising the FSB to 200MHz by using full P4 CPU in the place of Celeron D crap. But it did not happen. Therefore the chipset it cleared from any wrongdoing.

To show my appreciation I give it a great laped heatsink AND fan blowing at it now:



So there is no more excuses for the crashes that could be attributed to the chipset.

[trodas]

As for the 10uF d4 SMD cap, I will replace it (shortly, fingers crossed) with a 220uF 4V Tantal-polymer KEMET cap, so that should fix the problem. If it still fail, then the NIKOS poor mosfet is up to the blame and ... what replacement to choose for it??? Anyone?



Replacement suggestions?

The AGP regulator seems to be: NIKOS P3055LDG
http://products.niko-sem.com/images/...5182145127.pdf

This is, in short:
TO-252 (DPAK) 25V, 50mOhms, 12A, Gate charge 15nC, Gate treshold 1.2V

...so, what better with same gate charge/treshold and possibly lower resistance when opened (50mOhms are a bit HIGH IMHO) could anyone suggest?

[KaptCrunch]

P3055LDG - N-Channel Logic Level Enhancement Mode Field Effect Transistor - List of Unclassifed Manufacturers Manufacturer & Supplier

[trodas]

But... KaptCrunch, this is the same mosfet? :O


Looking at the IPS1011RPbF,
it have 6V input (max. step down 28) witch is fine (input is 3.3V) but a 10mOhms, should do 18A and the rise time and shutoff time are simply uncomparable with 50uS and 330uS, compared to the 6nS and 20nS :)
NIKOS owned!

Sadly, there is gate treshold 1.7V (a bit too high to replace a 1.2V) and also there is absolutely NO WORD about gate charge :(
http://www.irf.com/product-info/data...ta/ips1011.pdf

That suxx. Something else is probably necessary :(

Semiconductor NCV8401
42V, 23mOhms, 33A, rise time 41uS, shutoff time 164uS
but gate treshold 1.8V :(
http://www.onsemi.com/pub_link/Collateral/NCV8401-D.PDF

Infineon BTS 3142D
42V, 28mOhms, 30A, rise time 60uS, shutoff time 60uS
but gate treshold 1.7V :(
http://www.infineon.com/dgdl/BTS3142_DS_13.pdf

ST Micro
40V, 35mOhms, 12A, rise time 350nS, shutoff time 450nS
gate treshold (forward on voltage) should be 0.8V, witch is pretty good
gate charge is not listed there, but todal is 36.8nC
http://www.st.com/web/en/resource/te...CD00002219.pdf

...that one could be it...?