core slowdown threshold

[***Deimos***]

Hi, I hope Jimmor or one of his followers is still around.

I read through virtually every thread and article on the internet on the topic, but there are no instructions, except for those provided by Jimmor:

However, as an example for a 6600 type bios with core threshold temp of 145 it is easier if using NiBiTor's hex editor you first look for the binary string '04100905'. Having found the string, the binary part you are after is the four bit string '1009' sandwitched between the '04' and '05'. Now to get the two bit binary value that equates to 145 decimal, you just take the first and fourth bits of '1' and '9' and transpose them to become hex value '91'.

So for instance to set 100C only requires the hex '91' to become '64'. Or to put it another way, change the original '04100905' string to '04400605' !!

I have several 6800 BIOS files and noticed that some have 125C and some (like the one using currently) 105C. Lets say hypothetically, that the temperature reported is not the actual temperature used for thermal trip (wouldn't be a surprise, since having seperate thermal diode is common practice on CPU). So, even though at most, I only hit 62C load, there may be some hotspot near the thermal trip kicking it in. So, I'm trying to modify core "slowdown threshold":

I have 105C, as seen in nvidia driver. So according to Jimmor instructions that would be 04900605h, where 96h is transposed 69h = 105C. Similarly, 125C should be, 04D00705, where D7h is transposed 7Dh = 125C.

However, I can't find either of these strings in my BIOS. Not surprising, his instructions were actually for a 6600, and I have 6800. Jimmor, if you're there.. can you shed some light on this.

[***Deimos***]

I think I may be on to something:

I devoted the whole morning and afternoon today to testing my video card. The premise was simple enough.. poor heat sink contact => throttling and artifact problems.

First, I made startling observation after removing existing AS5 off the NV5 and NV40.. there is a tiny chip near one corner! This might cause the problems I'm having.

About 18-20 times, I removed the NV5, cleaned and reapplied, or just redistributed the AS5. Each time I made small variations. I tried tightening the top-left screw more. The top-right...etc. I noted the idle temps at stock and overclocked settings as well as overclocked results. Each time I methodically followed the same procedure, of loading windows, opening rivatuner, starting monitoring temps, overclock load 3DMark2005, and try to do 3 runs. Each of the 18-20 attempts failed miserably. I got anywhere from nearly working (25fps at the begining.. ie throttling), some artifacts (purple walls, black guns, techno colors), to horrible artifacts (guys missing, huge tearing, garrbly gook and 1-3fps).

And then it happened.. I was just about to give up. I quickly replaced the NV5 heatsink, as I needed the PC to do something. For kicks, I thought I would just as well try it one last time. It worked! The thing actually ran all the way through not one but 2 whole 3DMark2005 runs. Temps were very similar so at best I can only guess the cause as forgeting the screw that holds the video card to the case. Can it be that simple?

I was going to try the video card in a NF2 system my brother is running, but he had some problems with Windows activation. I replace the 6800 in my system, screwing it in. 1-3fps... WTF!! I remove the screw, while the computer is still running.. run 3DMark2005 again...33fps (much better). I screw it back in.. still 33fps.

I am now almost entirely convinced that either this video card is the work of the devil here to curse me and toy with me, or that there is some problem of poor contact somewhere, perhaps due to bending of the PCB when screwing the NV5 too tight.

[***Deimos***]

The legend continues:

Just ran a few more experiments:
Athlon64 @ 800Mhz!! (vs 2.4-2.6Ghz I typically run)
The stupid flicker+stutter+throttling still happens, so now after switching so many power supplies, and all this, I'm getting pretty confident the PSU is not the problem.

So I get the trusty old multimeter and try to catch the video card "in the act" so to speak. Booting.. 1.1..V, windows loading 1.2..V, in windows 1.316-1.317V (I have the 1.3V BIOS flashed). Ok, lets give nVidia's detect clocks thingie a try - woah, dips down to 1.309V.. hmm.. maybe thats why it always detects such low values.

So I procede to hold the probes there for the following 2 hrs. Within that time, I discover, that when 3DMark2005 is loading game tests, the voltage fluctuates all over the place, and soon after the "shuttle" lands and doors open, the voltage takes a dip to 1.308V. Also, in GT2 when the green light shows up, and many other places. In GT3, near beggining, when cannons are shooting, and when monster opens mouth. - in all these, the voltage dips to 1.308V, wheras it is typically 1.310-1.311V, even sometimes jumping up to 1.315V. Now, I dont think its a coincidence, that when I get throttling or artifacts, it starts happening in these same areas.

I also measured the GDDR memory, but nothing as interesting there. pretty much stays 2.71-2.72V.. dont have high resolution for that voltage range.

Ofcourse, my crummy multimeter is not so precise and has poor bandwidth, so these variations might very well be quite a lot larger. In one case, right after the 3 GT ended, I swear I noticed Vgpu drop to 1.279V. I think this just might explain why on quite a number of ocasions after clean GT runs, I get horrbile artifacts on the blimp in the CPU tests.

I did capacitor mods before for both the Vgpu and Vmem, but perhaps I was using poor quality capacitors, and that's why it hardly helped. I'm trying to get some nice low ESR 4700-6000uF capacitors this time.

[jimmor]

Hi, I hope Jimmor or one of his followers is still around.

I read through virtually every thread and article on the internet on the topic, but there are no instructions, except for those provided by Jimmor:



I have several 6800 BIOS files and noticed that some have 125C and some (like the one using currently) 105C. Lets say hypothetically, that the temperature reported is not the actual temperature used for thermal trip (wouldn't be a surprise, since having seperate thermal diode is common practice on CPU). So, even though at most, I only hit 62C load, there may be some hotspot near the thermal trip kicking it in. So, I'm trying to modify core "slowdown threshold":

I have 105C, as seen in nvidia driver. So according to Jimmor instructions that would be 04900605h, where 96h is transposed 69h = 105C. Similarly, 125C should be, 04D00705, where D7h is transposed 7Dh = 125C.

However, I can't find either of these strings in my BIOS. Not surprising, his instructions were actually for a 6600, and I have 6800. Jimmor, if you're there.. can you shed some light on this.

Theres actually not much difference between 6600 and 6800 bioses. So for a typical 6800 bios, with Core and Ambient slowdown temps of 135C, you would be searching for two strings of "04700805".

And as a help to anyone interested in playing with these functions, both are located in the binary data just before that of the 2D/Thrtl/3D speed tables; with the first being for Core slowdown temp and the second for Ambient.

EDIT:

And ***Deimos***, if the above explanation doesn't work for your particular 6800 bioses, then the only way I can help further is for you to post a bios for me to check?

[***Deimos***]

it all sounds great.. now what is your email address.. or how do I post an attachment?.. maybe just post the ASCI (lol)?

[***Deimos***]

I also find the "force performance level" feature in rivatuner facinating. Throughout past months I always ran with it enabled, and forced to "performance 3d".

However, recently I did a series of controlled experiments with it disabled. I set the 2D, 3D throttle, and 3D performance all to different clocks. I used rivatuner to monitor the clocks, and 3DMark2001 Nature, and 3DMark2005 for test.
@425, which is highest I've seen work without artifacts, what happens instead is it either briefly goes to 425, and then quickly back to 2D clock. Subsequent times you run 3D app, it doesn't budge from 2D clocks at all. However, once the 3D application is finished, the clocks suddenly jump to 3D speeds... weirD!
I tested this also at 410, 400, 390 and it still happened. Perhaps its doesnt like the clocks being changed over what's in the BIOS (3D=370).
Interestingly, after running 3D application, it takes a considerably long time 15-20sec, to step down to 3D thottle clock, and then to 2D clock (each). BUT, if you close the application, it steps down to 3D throttle within 5sec, and then 2D about 1 sec later. I repeated this experiment many times to make sure its no fluke.

So you all are probably wondering, what happens when you set 2D clocks to 3D speeds. Well obviously the clock stays up. BUT!, the performance is still similar to 2D performance.. This even happens with 2D=throttle=3D all set to the same high level (even if in the BIOS).. but after repeated attempts sometimes it performs like it should. Not entirely consistently.. still haven't found a pattern.

This is similar issue to that experienced with "force performance level" on, except in that case, performance changes in the middle of the test, and you can spot the instant, by the stutter and a flash.

[***Deimos***]

Fixed?!

So, just when I though removing that screw etc, had got things back how they were (throttling sporadically, but not always), I wake up this cool morning to find out that 425/950 doesnt work at all.. I get that 3-4fps and whole screen is garbled.. it doesnt even give me the decency to do the 27fps thottling thing.

I try pushng the card in all sort of places.. I get it to work ok 2 times.. but out of like 50+ tries. Hardly a victory. I try torqueing the heatsink to each side... similarly dismal results.. although at first I was convinced I was getting different artifacts when tilting heatsink each way, without any pressure I get all different artifacts so I guess artifacts are just random.

However, the whole day was not lost. I did make a surprising discovery, when removing heatsink, cleaning AS5, and reapplying AS5. I typically use cotton swab to clean around.. I use WD40 or contact cleaner to clean off the AS5. There were some dusty spots around the big capacitors, and lo and behold, the base of C143 seems to be loose!! I can nudge it a little left and a little right.. it doesnt come off at all, and I cant pull the capacitor off or tilt it or something like that.. but when you have strange problem like me, anything like this is news.

[jimmor]

it all sounds great.. now what is your email address.. or how do I post an attachment?.. maybe just post the ASCI (lol)?

In this forum, you just upload it via "Attach Files", which is part of the normal post adding/editing "Additional Options" facility. Which simply means "zip" the bios file, or rename it using any of the indicated as valid file extensions, and then upload it as an attachment to your posting by clicking on the "Manage Attachments" option !

[jimmor]

I also find the "force performance level" feature in rivatuner facinating. Throughout past months I always ran with it enabled, and forced to "performance 3d".

However, recently I did a series of controlled experiments with it disabled. I set the 2D, 3D throttle, and 3D performance all to different clocks. I used rivatuner to monitor the clocks, and 3DMark2001 Nature, and 3DMark2005 for test.
@425, which is highest I've seen work without artifacts, what happens instead is it either briefly goes to 425, and then quickly back to 2D clock. Subsequent times you run 3D app, it doesn't budge from 2D clocks at all. However, once the 3D application is finished, the clocks suddenly jump to 3D speeds... weirD!
I tested this also at 410, 400, 390 and it still happened. Perhaps its doesnt like the clocks being changed over what's in the BIOS (3D=370).
Interestingly, after running 3D application, it takes a considerably long time 15-20sec, to step down to 3D thottle clock, and then to 2D clock (each). BUT, if you close the application, it steps down to 3D throttle within 5sec, and then 2D about 1 sec later. I repeated this experiment many times to make sure its no fluke.

So you all are probably wondering, what happens when you set 2D clocks to 3D speeds. Well obviously the clock stays up. BUT!, the performance is still similar to 2D performance.. This even happens with 2D=throttle=3D all set to the same high level (even if in the BIOS).. but after repeated attempts sometimes it performs like it should. Not entirely consistently.. still haven't found a pattern.

This is similar issue to that experienced with "force performance level" on, except in that case, performance changes in the middle of the test, and you can spot the instant, by the stutter and a flash.

Stable running a 6800 above both 415 and 920 should only be attempted with core volts at 1.4v and using much better than stock heatsinking. And its very important to make sure that core and memory chips are making very good mechanical/thermal contact with the heatsink. Also, if not sure about quality of chip/heatsink contact, then maybe 415/900 would be much more stable for you?

And if you can't properly control heatsink temp to around max of 60ish, then you shouldn't be running memory above 920 without first increasing memory volts; and maybe also changing the bios's default memory timings?

Also, too many people still insist on using AS5 in spite of the fact that its "electrical" characteristics can cause instability and lockups, and, in a worst case scenareo, can also help to prematurely kill the vga card. So, if not already doing so, I recommend you only use Arctic Ceramique on your 6800. Its "inert" characteristics means it doesn't matter how sloppy you are when applying it to core or memory chips !

And no point in people posting their preference for AS5 because of a belief that it has much better thermal properties than Arctic Ceramique --> because its just not true. Thermal differencies, where they exist, will usually be very small and thus well within measuring technique or equipment inaccuracies, and are therefore of no real importance.

Also, since setting "2D", "Thrtl" and "3D" to same levels effectively disables a vga cards safety throttling functions, there should be no throttling or feeling that everything still operates as if stuck in a low 2D mode. So, your card's inability to run stable in such a situation indicates that your problem is likely to be either a Vga card fault, or a bad Driver, or a PSU problem?

[***Deimos***]

1st of all, I have NV5 silencer. And as I have reported this odd problem occurs whether its cold outsider (~50 load), or if I block the heatsink fan. (~70 load).

However, it does seem to get worse the longer I stress the card... although that doesnt necessarily mean its heat related.

I tried disabling pixel and vertex processors. I had the strangest results. It actually made things worse 90% of the time. I went through each vertex, trying to disable each one and looking at results. Sometimes horrible artifacts, but sometimes only throttling.. I actually had just as good as 16/6 performance with the hardware masked vertex unit 3 disabled, but when disabled in conjunction with another vertex unit, it was worse (artifacts). Then, when I tried it again, disabling vertex3, 3DMark05 was all garbled... However, what was interesting is that with some vertex or pixel shader disabled I was getting very very similar results to the "thottling" that I've seen. There is also a capacitor, C143 that I found that may be loose. As for AS5, I did get some of it under the BGA's as well as the GPU, but I tried my best to repeatedly clean it. However, I had this issue before that mess up.

Perhaps the problem is that because of insufficient power, one of the pixel shaders, or vertex shaders etc get turned off... or something along those lines.

[***Deimos***]

Oh and I think you also misunderstood the conditions under which my throttling occurs. With 1.316V, It happens predominantly at 425, but also at 410, 400, and even 390. It happens regardless if I have additional fans blowing at the card, and cool air, or not. Or whether memory is clocked at 950 or at 750 or even 700. But, if both memory and GPU clocks are low, it doesnt seem to happen (ie 370/750 is fine.. but hardly satisfactory when you know you can get 3 or even 4 clean 3DMark2005 runs in a row at 425/950).

[jimmor]

Oh and I think you also misunderstood the conditions under which my throttling occurs. With 1.316V, It happens predominantly at 425, but also at 410, 400, and even 390. It happens regardless if I have additional fans blowing at the card, and cool air, or not. Or whether memory is clocked at 950 or at 750 or even 700. But, if both memory and GPU clocks are low, it doesnt seem to happen (ie 370/750 is fine.. but hardly satisfactory when you know you can get 3 or even 4 clean 3DMark2005 runs in a row at 425/950).

The bios you linked has the core slowdown threshold stored in string "04902605" (69h=105C) and Ambient slowdown in "04700805" (87h=135C).

And noticed you added "2.5ns" to the bios message. However, if this means that your 6800 is actually fitted with 2.5ns memory chips, then let me state here and now that 2.5ns chips will not run stable at 900 plus levels without an appreciable increase in operating volts. Only 2.2ns types should be pushed beyond 900 with stock volts. However, stable 900 plus levels requires very good heatsinking; and preferebly temps contained to max of around 60C. When extremely overclocked the memory chips are very temp sensative.

And without exceptional heatsinking, which keeps core temp below 60C, it is also very unlikely that a basic 6800 will operate stable at 420 with gpu volts of only 1.3v. Which is why I usually only recommend 420 and above when running 1.4v. However at 1.4v it then becomes more important to keep a wary eye on temp of the core volts VRM components, since these components were not meant to support such a high loading !

Also, although you may just have a faulty card, if your throttling or overclock instability problems only occur at speeds higher than 370/750 then it's possibly a faulty psu issue?

Of course testing the card in another computer would likely be a much quicker way to confirm that your problems are totally card related?

And as for vga card testing (looking for artifacts), especially when overclocked, I suggest you try with ATITool and the Nvidia LUNA demo, since both will properly stress your cards memory chips !!

And I would be interested to know how you got on with the two modded versions of your bios I have linked below in the x684180.zip file?

[***Deimos***]

Perhaps I have Hynix 2.8ns chips.. maybe 2.2ns.. its irrelevant. Without volt mod, atitool runs for at least 15 min at 900+, with appropriate timings as high as 960. Volt modding 972 passes, and even 984, but strangely, more volts doesnt help beyond around 2.85V.

As I have already mentioned, I did interesting temperature test, by blocking the fan on the NV5, and increasing room temperature. Instead of ~60C under load, I had ~70C (when I started!). However, I was still able to do 3DMark2005 at full speed. If I recall, it locked up on the 3rd run. How can it do 2 whole runs at 70C+? And likewise, why is it that on a cold morning, when the GPU temperature before I start a test is only ~30C, why would it have artifacts or slowdown then?

I think after I research a bit more the Vgpu capacitor mod, as well as the Vmem capacitor mod, I will have better answers. I did both of these before, but I think I used inappropriate capacitors, as well as too long wires.

SilkaRippa suggested that it possible that I damaged some capacitors on my card when I was cleaning off some AS5 that got under the BGAs with "contact cleaner".

[***Deimos***]

I tried both, each many times. First, the 4180, as-is. I got good result. I had force perf level in rivatuner, so I turned it off for the rest of the tests to see any transitions. 410/800 worked 5 times in quick repetition with good performance. Next I tried it with 425/950, but there were horrible artifacts.. GPU was clocked down to 300 during that period.

Then, I tried the 4190 BIOS, and likewise first tested as-is. Five consecutive runs and the GPU clock remained at 410, although this time there were dips where each test begins. However, the test results were good. I tried 425/950 but once again, artifacts, but this time it went to 400. When I repeateded this many times, it went either 300 or 400, and minor or horrible artifacts.. pretty much random. Next I activated nVstrap 16/6, and tried again BIOS clocks. It stayed at 410. I repeated this 5 times, and except for one slight low (4871), all other results were very close to each other.(4920-4950). Now this is not unprecendented.. afterall, this card has a mind of its own. I once had as many as 4 consecutive 425/950 runs, before things turned sour. I quickly tried 425/950, while it was still hot from last run.. and it WORKED! It was running 425 throughout all game tests, but then near the end, froze.. 15-30sec later, just as I reached for power buttonn, it continued.. it had clocked down to 300.

Now, perhaps there is something you changed, other than the timings, but the following time when I tried 425/950, it throttled the first game test down to 400, started at 425 for half the second game test, but then went down to 300. And amazingly started at 400 in game test 3, but then RAISED to 425. I have never seen GPU clocks raise during benchmarking. They seem to reset themselves when each new game test is run, and only fall/throttle down.

Finally, I tried to be a bit more conservative. I tried 414/900 several times and it seems to work fine (I probably need >10 runs to be sure). Likewise 410/936 seemed to be fine.. I left the machine idle for a few min, so the card already cooled some, ran 410/936 again - scores dropped, throttled to 400. Go figure.

Would it throttle down to 400 or 300 because of heat or defects? Is this something that increasing the slowdown threshold can help?

[***Deimos***]

also, why did you remove the 16/6 from the BIOS I sent you, and why doesnt the rivatuner patch script work to replace it?

[***Deimos***]

6600GT with dusty fan:


this is only ~2min with ATITool!!
Imagine what would have happened if I didn't chicken out.

[jimmor]

also, why did you remove the 16/6 from the BIOS I sent you, and why doesnt the rivatuner patch script work to replace it?

The bios you linked is at NOT unlocked to 16x1,6vp, which is why the modded bioses I posted back are also not unlocked to 16x1,6vp. However, if you want these bioses fully unlocked, then thats exactly what the Rivatuner "NV40BIOSHwUnitsMaskEliminator.rts" script file is for !

But just to make it a bit easier, you will find fully unlocked versions in the linked x684180u.zip file below,

[***Deimos***]

The bios you linked is at NOT unlocked to 16x1,6vp, which is why the modded bioses I posted back are also not unlocked to 16x1,6vp. However, if you want these bioses fully unlocked, then thats exactly what the Rivatuner "NV40BIOSHwUnitsMaskEliminator.rts" script file is for !

But just to make it a bit easier, you will find fully unlocked versions in the linked x684180u.zip file below,

he he, yeah, I know all about the patch scripts since like the last time I did the cap mod on the memory last summer..
I forgot that I had renamed the filed. The original not-modified one was xfxmod8.old.. I renamed it to .rom and vice versa for the unlocked.. complete confusion... Sorry for taking up valuable time to do something trivial like use patch script...

However, I have a very important question regarding my 6800 problem. Do you know how to increase the frequency of the switching power regulators on a 6800?

[jimmor]

I tried both, each many times. First, the 4180, as-is. I got good result. I had force perf level in rivatuner, so I turned it off for the rest of the tests to see any transitions. 410/800 worked 5 times in quick repetition with good performance. Next I tried it with 425/950, but there were horrible artifacts.. GPU was clocked down to 300 during that period.

Then, I tried the 4190 BIOS, and likewise first tested as-is. Five consecutive runs and the GPU clock remained at 410, although this time there were dips where each test begins. However, the test results were good. I tried 425/950 but once again, artifacts, but this time it went to 400. When I repeateded this many times, it went either 300 or 400, and minor or horrible artifacts.. pretty much random. Next I activated nVstrap 16/6, and tried again BIOS clocks. It stayed at 410. I repeated this 5 times, and except for one slight low (4871), all other results were very close to each other.(4920-4950). Now this is not unprecendented.. afterall, this card has a mind of its own. I once had as many as 4 consecutive 425/950 runs, before things turned sour. I quickly tried 425/950, while it was still hot from last run.. and it WORKED! It was running 425 throughout all game tests, but then near the end, froze.. 15-30sec later, just as I reached for power buttonn, it continued.. it had clocked down to 300.

Now, perhaps there is something you changed, other than the timings, but the following time when I tried 425/950, it throttled the first game test down to 400, started at 425 for half the second game test, but then went down to 300. And amazingly started at 400 in game test 3, but then RAISED to 425. I have never seen GPU clocks raise during benchmarking. They seem to reset themselves when each new game test is run, and only fall/throttle down.

Finally, I tried to be a bit more conservative. I tried 414/900 several times and it seems to work fine (I probably need >10 runs to be sure). Likewise 410/936 seemed to be fine.. I left the machine idle for a few min, so the card already cooled some, ran 410/936 again - scores dropped, throttled to 400. Go figure.

Would it throttle down to 400 or 300 because of heat or defects? Is this something that increasing the slowdown threshold can help?

First of all, if you wan't to "disable" the throttling functions of your card then just set the speeds/volts/fans levels of "2D", "Thrtl" and "3D" to exact same values !

Mind you, even with the throttling functions disabled, your card can still artifact, stutter or lockup if you overclock/overload it beyond what the gpu and/or memory can cope with?

The max overclock for any card is the highest level which is continually stable for all situations (games, testing, etc), and therefore NOT just an ability to pass a 3dmark test once or maybe only a few times ! Which when I overclock, usually meant a max of about 15-20 MHz below the level which started to display artifacts ?

And there is no point in setting a core or memory speed that is not compatible with the bios set volts or fan speed !! As posted before, a basic 6800 is generally good for 410 at 1.3v and 425 at 1.4v. With 2.2ns memory chips good to around 920 at default volts, and 950 with around 2.85-2.9. But all of this only with good heatsinking/cooling.

Also, with the modded bioses I posted, throttling right down to 300MHz probably means you are either, overstressing the core/and/or memory way to high, or, your PSU is providing UNSTABLE volts to the card ? So suggest you keep Core/Memory speeds below that which causes core throttling to 300MHz levels?

And with the bioses I posted, it is normal to notice core speeds occasionally switching between 410 and 400 levels. This is just a knock-on effect of minor overloading of the gpu operating conditions. However is unlikely to have any real affect on gaming, etc. Although, ideally, it would be best to only overclock 3D core/memory speeds that your card is totally stable at, ie no throttling whatsoever !

And finally, changing the core slowdown temp will not improve how your card works ----> And that's because your card should NEVER get to, or exceed, the stock 105C level unless your fan fails !

[jimmor]

..... Do you know how to increase the frequency of the switching power regulators on a 6800?

Sorry, never had any reason to consider changing core/memory power supply's set switching speed. But why would anyone wan't to ?

And if something just happened to be different/wrong in that area of your particular card, then it would obviously be a fault which wouldn't likely be resolved by any increasing of the switching frequency?

[***Deimos***]

are these the small dips to 3D throttle you are talking about (far left)? They dont always happen. Sometimes there, sometimes not.


I know about setting 2d/thrtl/3d to the same, and tried various variations of that.. the result is same as with rivatuner force performance level, or clocks different.. it will throttle regardless. When I modified BIOS with 04900605, which should be 150, it shows 127C. I know how slowdown threshold is "supposed" to work, but hey maybe they made mistake and it throttles at 127F not 127C.. hey I'm just trying to eliminate all the possibilites.

Ok, for the sake of argument, forget everything I said before, and just focus on these two examples:
Couple nights ago, I'm tinkering around trying to get 425/950 stable. I'm trying to capture measurments at the moment the flash, stutter, and throttling starts. (measuring various voltages on the PCB like Vref, Vgpu, Vddr, etc, +5V, +12V) So, I first heat up the card by running ATITool for ~15min at 410/900.. good, its nearly up to 60 C. I fire away 425/950 3DMark2005, expecting throttle any moment. 3 full runs, full-speed scores.. NOTHING. In the meanwhile, despite the fact I quikcly click to test again, rivatuner is showing the card is cooling. Finally, it happens during 4th run. Highest temperatures were recorded before the 3DMark runs, and each one slightly cooler... would you say this is an indicator that heat is not an issue? I leave computer idle for about an hour, come back and run same settings again, but it doesn't just throttle.. discoloration artifacts. Next morning (very cool, card cold) I try to resume my testing, but artifacts are even worse. So, would you agree heat is not an issue?

If I could get your 410/900 perfectly stable I would be 100% happy. However, I play BF2 for ~0.5hr and it throttles.. Ok, so I lower to 400.. ~5min later throttle. Hmm.. so I change mem to 800 (410/800). Nope, still throttle. Lower both? Yes! 400/800 is working.. but after ~0.5hr it does it again. So this pretty much explains why I typically ran the card at 370/740 (below the lowest driver test failed on for each.. ie although driver test typically fails around 800 for memory, it once went as low as 754). Even though sometimes 3DMark05 finishes as high as 425/950, and many times at 410/900, it will still throttle if either or both GPU and MEM are clocked high. Especially sensitive to Splinter Cell: Chaos Theory (HDR, SM3 more power demand?).

[***Deimos***]

I dont think its a coincidence that throttling happens just as non-local memory fills up, or sharply decreases.

THE THROTTLING IS ON THE FAR LEFT

the others are me alt-tab and checking the graph while playing BF2

[jimmor]

are these the small dips to 3D throttle you are talking about (far left)? They dont always happen. Sometimes there, sometimes not.


I know about setting 2d/thrtl/3d to the same, and tried various variations of that.. the result is same as with rivatuner force performance level, or clocks different.. it will throttle regardless. When I modified BIOS with 04900605, which should be 150, it shows 127C. I know how slowdown threshold is "supposed" to work, but hey maybe they made mistake and it throttles at 127F not 127C.. hey I'm just trying to eliminate all the possibilites.

Ok, for the sake of argument, forget everything I said before, and just focus on these two examples:
Couple nights ago, I'm tinkering around trying to get 425/950 stable. I'm trying to capture measurments at the moment the flash, stutter, and throttling starts. (measuring various voltages on the PCB like Vref, Vgpu, Vddr, etc, +5V, +12V) So, I first heat up the card by running ATITool for ~15min at 410/900.. good, its nearly up to 60 C. I fire away 425/950 3DMark2005, expecting throttle any moment. 3 full runs, full-speed scores.. NOTHING. In the meanwhile, despite the fact I quikcly click to test again, rivatuner is showing the card is cooling. Finally, it happens during 4th run. Highest temperatures were recorded before the 3DMark runs, and each one slightly cooler... would you say this is an indicator that heat is not an issue? I leave computer idle for about an hour, come back and run same settings again, but it doesn't just throttle.. discoloration artifacts. Next morning (very cool, card cold) I try to resume my testing, but artifacts are even worse. So, would you agree heat is not an issue?

If I could get your 410/900 perfectly stable I would be 100% happy. However, I play BF2 for ~0.5hr and it throttles.. Ok, so I lower to 400.. ~5min later throttle. Hmm.. so I change mem to 800 (410/800). Nope, still throttle. Lower both? Yes! 400/800 is working.. but after ~0.5hr it does it again. So this pretty much explains why I typically ran the card at 370/740 (below the lowest driver test failed on for each.. ie although driver test typically fails around 800 for memory, it once went as low as 754). Even though sometimes 3DMark05 finishes as high as 425/950, and many times at 410/900, it will still throttle if either or both GPU and MEM are clocked high. Especially sensitive to Splinter Cell: Chaos Theory (HDR, SM3 more power demand?).

Random throttling between 3D and low power 3D levels is quite normal and thus very common; and not that important as long as they are kept within 20-30MHz of each other. However, as said previously, it is always better if you keep the 3D overclock at a level which doesn't cause any throttling at any time !!! ---> which in your case would probably mean limiting your max 3D overclock to no more than 400MHz ?

Also, since your card is designed to run 2D, Thrtl and 3D memory speeds at same levels all the time, i'm a bit confused as to why your 3D memory speed appears to be switching levels ??

And max value that can be set for slowdown threshold level is capped at 127C ---> and no design mistakes were made ---> the slowdown threshold function works just fine !

And unfortunately for everybody, heat is always an issue. However, its never the only or necessarily the most important parameter involved? So basically there is rarely a single parameter that outright determines how your card will operate on a minute by minute or day by day basis; especially when pushed outside its spec. Therefore, even with safety throttling functions effectively disabled (by setting all three modes to same levels), other driver/bios monitoring/controlling functions, will for example reduce the max output capability of the VRM supplies in an attempt to protect the card from excessive abuse. Where in your card's case, the abuse is YOU overclocking your card well beyond its spec.

And just for interest, if the card starts to throttle after 0.5hr when set at 410/900, then that is typically either an indication of GPU/memory overtemping, or CPU/mobo going unstable, or PSU going unstable ?

A strong fan blowing directly at the card would obviously help to prove if there is a card overtemp issue. But this will only be meaningfull as long as it is known that the GPU and ALL memory chips are making good contact with the heatsink ?

Of course, testing the vga card in another computer would make it much easier to isolate if it is only a card problem?

And as prevously posted, the stability of your card is heavily dependant on the stability of your PSU. So if the supply lines of your psu spike or dip for any reason, a knock on effect will likely be a triggering of a built-in card protective function ---> which to you will just appear as a stutter, instability or artifacts problem. But of course this sort of issue only becomes important where a card is extremely overclocked, ie on the edge between being stable and unstable. However, proving whether a PSU is faulty is easy, just replace it ?

Mind you, maybe you should just accept that your 6800 is unlikely to work totally stable at extremes of 425/950, if only because most of your testing so far appears to indicate that such levels are very close to your particular card's max capabilities. And because of the protections built into the cards design and its drivers, it won't do any good just assuming that increasing volts levels will automatically allow you to overclock that bit more. Because unfortunately, when your card's design limit is breached, further increases of core/memory volts just usually causes more instability, lockups or, worst case, the premature death of the card ?

Of course all of your testing may only be proving that your particular card is actually faulty/unstable in such a way that just isn't going to allow any meaningfull overclocking?

And not that I would recommend it for anything over 400 with a basic 6800, but have you done any testing with GPU at 1.4v ?

[jimmor]

I dont think its a coincidence that throttling happens just as non-local memory fills up, or sharply decreases.

Non local memory completely filling up, with likely knock-on of sometimes causing stuttering and/or lockups, will always be an issue for cards with less than 256Mb; especially when running most modern games, or testing with programs like 3dmark06. Because when you fill up both local and non-local memory you are just then forcing winxp and the cpu to continually keep moving data around, including for some, the very slow back and forth route between main memory and HDD. And since there is obviously a lot more work involved it is not surprising that all this data shifting causes the gpu to be easily overstressed to the point of shutting down due to overloading, or maybe just triggering a built-in throttling safety features.

But obviously as said before, you can disable your cards basic throttling functions by setting 2D Thrtl and 3D at exact same levels (eg, all at 400/900 with all fans at 100%). And although doing this can't be taken as cure all for every overstress issue during modern games, etc, it should definately make things a bit more stable.

And also to help minimise such issues it is important to have minimum of 1Gb main memory, AGP aperture at min of 256Mb; and winxp virtual ram of reasonable size (1-2Gb) on fastest HDD ! Or off course, just get yourself a vga card with 512Mb of memory ?

[***Deimos***]

On the top graph, I first ran without modifying clocks (BIOS values), then clock to 425/950 to simulate throttling.

Once again, I know about force-perf-level, or setting all clocks same in rivatuner, or setting all clocks same in BIOS. The same thing happens regardless. Either the card runs at full performance, or it runs at nearly identical performance to GPU=300Mhz (where would it get this clock speed from?), even though rivatuner continuously shows the same high overclock (ie 410), or terrible artifacts. When clocks are different, and there is no force-perf level, artifacts ALWAYS correspond to it switching to 300Mhz, and sometimes to 400 (using your bios).

I would be happy to let 425/950 go. I have nearly come full circle, and have re-investiaged the famous 6800/NF3 connection (yeah, I have NF3). So, I re-enabled FAST WRITES in mobo BIOS. Now, it might be just a coincidence, but anything north of 370Mhz, causes stuttering, pausing, blank screen, and yes even artifacts. However, the strange thing is, when 3DMark05 run for first time (starting from 2D clock), it will blank screen/pause resume, then pause again etc... when you quit test, it clocks up to 410 (even though you're just on the desktop). Quick subsequent run (before it throttles back down), it doesnt pause/blank at the very beginning.. still happens later on though. But rivatuner shows that higher clock was used for the first little while until it pause/blank and there is corresponding throttle to either low-power 3D or 2D.

I went over all the 6800/NF3 "solutions" again to make sure I didn't miss something: AGPx4, AGP=55, AGP=75, sideband=off, user mipmap=off, blitting, aperature=32, aperature=512 (normally 256), lower RAM speed 10%, CPU at stock, CPU @ 800Mhz, loosen all RAM timings, ...

Only thing that "works" is fast-writes = off. However, I think its only lowering the occurance of the problem. Quite similar symptoms show up, but much more rarely with fast-writes off. So the only way to get consistenly 3D clocks is 370/740 with fast writes off (370+ and/or 800+ will sporadically cause throttling)

EDIT: I'm already on 4th power supply!! And, I'm still trying to get this card working on my brother's computer.. last time just installing the card messed up windows and his ati drivers really bad.. he had to reinstall windows