Same clocks good with software but not with the bios

[octavr]

I have been using RBE 1.12 with a Sapphire HD3650 256 MB to set default settings for fans and clocks. I used Furmark and AMD GPU Tool to find the max stable settings, at 864/1044 (half an hour on the "Xtreme" test! Now I think it was too much... Bad settings crash in 2-3 minutes...). However, when those settings are applied to the original bios (only the fields 03 and 09, unlocked), they are not stable, the highest stable setting through the bios being 823.50 for the GPU (the memory setting is fine). Any ideas why?

I also have a suggestion. The Duty cycle min setting is very important also for the look-up table mode. It should be at the bottom I think, separated and never grayed out. (Works great anyway, thanks! I set it to 14, and the spin-up to 28. The spin-up setting I took from the updated bios from the Sapphire forums, which also has a different look-up table, which I however don't like. The Duty min cycle I determined with Rivatuner.)

[VuurVOS]

AMD GPU Tool is the cause of this problem. Its locking the states so the card can't lower his voltage or speed. When switching between those states your cards gets unstable.

And actualy a half hour testing isn't enough. Let it run altleast a few hours before you can call it stable.

[Hayder_Master]

did you try ati tool

[octavr]

Its locking the states so the card can't lower his voltage or speed. When switching between those states your cards gets unstable.

Why do you think that the card lowers its voltage or speed? After it exits Furmark it certainly does, but while the stress test runs it wouldn't be obvious why it would, and clock readings don't change. If indeed, which states would that be?

And actualy a half hour testing isn't enough. Let it run altleast a few hours before you can call it stable.

It is stable in comparison to a really quick crash. I have not seen an in-between crash, and I don't want to shorten the card's life. I understand "stable" should be justified, but then you would need to define "at least a few hours"!

did you try ati tool

You mean the one with the original furry cube? Not really, this time, why?

[VuurVOS]

Why do you think that the card lowers its voltage or speed? After it exits Furmark it certainly does, but while the stress test runs it wouldn't be obvious why it would, and clock readings don't change. If indeed, which states would that be?

The card switch constandly. Even when you see the loading screen, the card is running in a lower 3D state or even in a 2D state. In the lower 3D state there isn't so much voltage needed but when switching up it doesn't have enough voltage so it will crash. You can incease the voltage of it or you must realy set the cooler to run faster. More heat means more voltage at a certain speed but when your gpu is cooler it doesn't need so much anymore that why some of us is using dry ice for (extreme) overclocking.


It is stable in comparison to a really quick crash. I have not seen an in-between crash, and I don't want to shorten the card's life. I understand "stable" should be justified, but then you would need to define "at least a few hours"!

How long do you game? Mostly atleast an hour I think, so it must be stable for 2 or more hours. Because your gpu gets hoter and hoter during stressing you can say at a certain point it doesn't become any warmer. At that piont it must be stable for some time.


You mean the one with the original furry cube? Not really, this time, why?
It can find artifacts before you see it which means you shouldn't clock that high (864MHz could be far to high for a long time (HEAT!))

[octavr]

The card switch constandly. Even when you see the loading screen, the card is running in a lower 3D state or even in a 2D state.

Exactly! When it enters or exits 3D mode (or even the "low-load" 3D mode, as can be seen in the readings, but only under unusual circumstances, from what I've seen so far), it switches. If it indeed switched under constant full load in this test, I would say this would be quite extraordinary. I could say more about what-ifs, but I don't want to obfuscate.

Mostly at least an hour I think, so it must be stable for 2 or more hours.

Well, that's not really the topic but I've just thought of something... During all the time you use a card, you may as well play for thousands of hours. This does not mean you should put the card under extreme stress for thousands of hours. :pimp:

Remember that the not very big increase in heat is not important as shown by stability tests (the length of which is not important as long as it is constant) after overclocking with the AMD GPU Tool.

It can find artifacts before you see it

True, but the point is the stark difference of "crashes very soon" and "works and works and..." between AMD GPU Tool and the bios setting.

[VuurVOS]

Exactly! When it enters or exits 3D mode (or even the "low-load" 3D mode, as can be seen in the readings, but only under unusual circumstances, from what I've seen so far), it switches. If it indeed switched under constant full load in this test, I would say this would be quite extraordinary. I could say more about what-ifs, but I don't want to obfuscate.

It happend to me when I was underclocking my hd3870 in 2d mode. Everytime it came out a 3d mode (3d mark) it crashed but not when the system was idling at the desktop of Windows



Well, that's not really the topic but I've just thought of something... During all the time you use a card, you may as well play for thousands of hours. This does not mean you should put the card under extreme stress for thousands of hours. :pimp:


I ment not all the hours you are spending on gaming but the average of when you are playing a game on the computer. Do you play a game for 1 hour without stopping or have you enough of gaming after 3 hours behind each other.


Remember that the not very big increase in heat is not important as shown by stability tests (the length of which is not important as long as it is constant) after overclocking with the AMD GPU Tool.
The length is also important. Your case temp is increasing during the stressing. At a certain piont the card doesn't get enough cooling which can cause artifacts.

I can overclock without any problem after a cold boot my HD4870 to 860MHz. But when the card gets hoter it will crash. This can only be fixed with beter cooling (or let your cooler run faster)

[octavr]

It happend to me when I was underclocking my hd3870 in 2d mode. Everytime it came out a 3d mode (3d mark) it crashed but not when the system was idling at the desktop of Windows

Interesting. It is different though. No crash on entering the 3D mode. Actually I noticed a problem when it freezed in the END credits of 3dmark06. So I run 3dmark06 again and it passed. So I knew it wasn't some obvious 3dmark-drivers problem. Then I discovered Furmark.

Is the 2D setting you were underclocking also just the group 01 in RBE on hd3870? Did you have any problems with Overdrive actually using the correct settings for different modes? It took me some 20 attempts to get this part right. :o

I meant not all the hours you are spending on gaming but the average of when you are playing a game on the computer. Do you play a game for 1 hour without stopping or have you enough of gaming after 3 hours behind each other.

I'm not saying your 2-hours testing is wrong. However, imagine you tested for 2 hours, and then game 2 hours the first day, no crash, 2 hours 2nd day, a crash! 2 hours 3rd day, a crash! 2 hours 4th day NO crash. Does it mean you should have tested for 8 hours?

The length is also important. Your case temp is increasing during the stressing. At a certain point the card doesn't get enough cooling which can cause artifacts.

It's certainly worthy to keep in mind that the things don't heat up instantly. Particularly the case temp however is about the tests' order and such and not their individual length.

When I'm choosing a certain length and I'm seeing something happening under one condition but not the other, it's not important what length I have chosen. I always have to assume some length is enough. In the end I was doing 2 x 4 minutes Xtreme and stable. It's certainly possible it was bad luck but the difference seemed quite extreme.

I can overclock without any problem after a cold boot my HD4870 to 860MHz. But when the card gets hoter it will crash. This can only be fixed with beter cooling (or let your cooler run faster)

Again, interesting. But again remember, I ran for half an hour once and then with the bios mod it was crashing in seconds to one minute, I think... I haven't been so painstakingly precise with it after all... Hmm... Anyway I'm still quite sure, but then I don't really have more days to keep restarting this system... for this reason at least... :o

[VuurVOS]

Interesting. It is different though. No crash on entering the 3D mode. Actually I noticed a problem when it freezed in the END credits of 3dmark06. So I run 3dmark06 again and it passed. So I knew it wasn't some obvious 3dmark-drivers problem. Then I discovered Furmark.

Is the 2D setting you were underclocking also just the group 01 in RBE on hd3870? Did you have any problems with Overdrive actually using the correct settings for different modes? It took me some 20 attempts to get this part right. :o

I only changed the 2d and the lower 3d (clock info 2&3 and 4&5, I'm not sure if this is the same for your bios)


I'm not saying your 2-hours testing is wrong. However, imagine you tested for 2 hours, and then game 2 hours the first day, no crash, 2 hours 2nd day, a crash! 2 hours 3rd day, a crash! 2 hours 4th day NO crash. Does it mean you should have tested for 8 hours?

No, only 2 hours because it is average use.


Again, interesting. But again remember, I ran for half an hour once and then with the bios mod it was crashing in seconds to one minute, I think... I haven't been so painstakingly precise with it after all... Hmm... Anyway I'm still quite sure, but then I don't really have more days to keep restarting this system... for this reason at least... :o
I downloaded a Sapphire HD3650 256MB and the first thing I saw was a 2 naught clock info's. Try to increase the voltage to 1,25 of clock info 1&2 (those are 1 volt). After that you problem is mostly solved.