4890 OC limit fixed!

[Kabauterman]

I want to add something to this which perhaps may disappoint some of you and bring in an "uncomfortable situation" the ATI company/designers...

Those 4890s, DO NOT have a 5 phase VGPU design as the reviewers have told you and ATI DIDN'T EVEN mentioned that (by mistake? ), BUT instead, it has ONLY THREE phases as for it's main core.

SO DON'T expect to overclock it ONLY with a bios fix (which personally I don't believe it'll help much or not at all).....
PLUS some/many/all of you may SEE frequencies going up as is, but the card works to the frequencies SHE WANTS and NOT as YOU have picked....
You NEED hardware mods to do so....

.......to be continued......

http://www.techpowerup.com/reviews/S...front_full.jpg

I see 5 Phases GPU and 2 Phases Mem...

[Vivi]

http://www.techpowerup.com/reviews/S...front_full.jpg

I see 5 Phases GPU and 2 Phases Mem...

indeed its visible, but i think they don't work together to form 5 phase power to the gpu, something must be wrong in the design somewhre, i know nothing of phases

[hipro5]

There are 3 phases for the MAIN CORE
2 phases (NEXT to the 3 core phases - 1.2Volts) for another circuit inside the core (VTT?)
1 phase for the MAIN RAMS
1 phase for rams VTT
1.82Volts for the Core (VTT?)...

PLUS Rivatuner VT1165 can provide you UP TO 1.45Volts REAL....
EVEN if you pick 1.7Volts, it gives you 1.45Volts REAL and that's it...

TIP for guys under LN2
USE ONLY the 850, 900, 950, 1000, 1050, 1100, 1150, 1200, 1250, 1300 as for the core and ALSO every 50MHz as for the rams
The XTAL oscillator on these VGAs (and the 4870 series) it's NOT 27MHz so to go every 27MHz step...
It's 100MHz, so NOT to get into problems, follow the RULE every 50MHz step...
For example - DON'T put the core 1070MHz....Put it at 1050 OR 1100...

[SNiiPE_DoGG]

Thanks for the insight Hipro

[DeDaL]

good information hipro5, thanks

[loopy83]

Thanks for sharing Hipro... very interesting.

[SF3D]

Those two phases are for memory controller in GPU. You need to raise that voltage as well, if you like to clock your memory well.

This bios does fix the limit, but it does not make any card to clock better. I think this is what people have missed in here.
Bios is from ATI engineers, so I am not talking crap here. We need to find the issue, which holds us back after 1250MHz clocks and no, it is not voltage.

Mod the card George, it will help for sure. Let's hope, that cores will scale well in your tests

[hipro5]

Those two phases are for memory controller in GPU. You need to raise that voltage as well, if you like to clock your memory well.

This bios does fix the limit, but it does not make any card to clock better. I think this is what people have missed in here.
Bios is from ATI engineers, so I am not talking crap here. We need to find the issue, which holds us back after 1250MHz clocks and no, it is not voltage.

Mod the card George, it will help for sure. Let's hope, that cores will scale well in your tests

Excuse me?...How high do people want to bench with these cards and there is a problem?....
If 4870 did ~900MHz on air and ~1150MHz under LN2, then this one must have core limit ~1250 - 1300MHz... No?...

Thanks Petri... I'm "almost" benching 1100MHz on air with one of them (the first I touched) the 2k6 canyon flight...1050MHz does it very easy...

[massman]

It's not voltage AND not temperature ...

[boblemagnifique]

There are 3 phases for the MAIN CORE
2 phases (NEXT to the 3 core phases - 1.2Volts) for another circuit inside the core (VTT?)
1 phase for the MAIN RAMS
1 phase for rams VTT
1.82Volts for the Core (VTT?)...

PLUS Rivatuner VT1165 can provide you UP TO 1.45Volts REAL....
EVEN if you pick 1.7Volts, it gives you 1.45Volts REAL and that's it...

TIP for guys under LN2
USE ONLY the 850, 900, 950, 1000, 1050, 1100, 1150, 1200, 1250, 1300 as for the core and ALSO every 50MHz as for the rams
The XTAL oscillator on these VGAs (and the 4870 series) it's NOT 27MHz so to go every 27MHz step...
It's 100MHz, so NOT to get into problems, follow the RULE every 50MHz step...
For example - DON'T put the core 1070MHz....Put it at 1050 OR 1100...

Very interessing georges thank

i hesitate to buy a HD4890 1go but i have see sapphire to propose soon the model 2go , Anybody have a news for the performance of overclocking or a difference at stock compared at the model 1go ?

[FlawleZ]

There are 3 phases for the MAIN CORE
2 phases (NEXT to the 3 core phases - 1.2Volts) for another circuit inside the core (VTT?)
1 phase for the MAIN RAMS
1 phase for rams VTT
1.82Volts for the Core (VTT?)...

PLUS Rivatuner VT1165 can provide you UP TO 1.45Volts REAL....
EVEN if you pick 1.7Volts, it gives you 1.45Volts REAL and that's it...

TIP for guys under LN2
USE ONLY the 850, 900, 950, 1000, 1050, 1100, 1150, 1200, 1250, 1300 as for the core and ALSO every 50MHz as for the rams
The XTAL oscillator on these VGAs (and the 4870 series) it's NOT 27MHz so to go every 27MHz step...
It's 100MHz, so NOT to get into problems, follow the RULE every 50MHz step...
For example - DON'T put the core 1070MHz....Put it at 1050 OR 1100...

Ah, and hipro5 has spoken.

Great info indeed.

[TONY GAL]

thanks hipro5 (george)!!

[chew*]

Excuse me?...How high do people want to bench with these cards and there is a problem?....
If 4870 did ~900MHz on air and ~1150MHz under LN2, then this one must have core limit ~1250 - 1300MHz... No?...

Thanks Petri... I'm "almost" benching 1100MHz on air with one of them (the first I touched) the 2k6 canyon flight...1050MHz does it very easy...

Hipro, How was the performance of 4870 versus 4890 at say 1050 core versus 1050 core?

[[XC] gomeler]

So is it time to crystal mod hardware again?

[Gautam]

FWIW I've ran 4890 at 1272MHz through 01 nature, and yes it did score higher than 1250.

But, I'll let all you experts proceed.

[chew*]

FWIW I've ran 4890 at 1272MHz through 01 nature, and yes it did score higher than 1250.

But, I'll let all you experts proceed.

Interesting

[Martin.v.r]

ned vmod/high clock to CCC softmod

[dinos22]

There are 3 phases for the MAIN CORE
2 phases (NEXT to the 3 core phases - 1.2Volts) for another circuit inside the core (VTT?)
1 phase for the MAIN RAMS
1 phase for rams VTT
1.82Volts for the Core (VTT?)...

PLUS Rivatuner VT1165 can provide you UP TO 1.45Volts REAL....
EVEN if you pick 1.7Volts, it gives you 1.45Volts REAL and that's it...

TIP for guys under LN2
USE ONLY the 850, 900, 950, 1000, 1050, 1100, 1150, 1200, 1250, 1300 as for the core and ALSO every 50MHz as for the rams
The XTAL oscillator on these VGAs (and the 4870 series) it's NOT 27MHz so to go every 27MHz step...
It's 100MHz, so NOT to get into problems, follow the RULE every 50MHz step...
For example - DON'T put the core 1070MHz....Put it at 1050 OR 1100...

i am not sure i sure about that george

at least the testing i've done with a couple of cards didnt seem to follow your rule

have a look at card clocks mentioned and how it went up with volts while i was testing 01

http://www.xtremesystems.org/forums/...d.php?t=222421

i could and have seen improvements in smaller jumps

[STEvil]

FWIW I've ran 4890 at 1272MHz through 01 nature, and yes it did score higher than 1250.

But, I'll let all you experts proceed.

I think the memory controller and core speeds are unlinked.. core jumps every 50mhz, memory controller stays to what you set or jumps in smaller steps, not that you can set them separately though.

[hipro5]

i am not sure i sure about that george

at least the testing i've done with a couple of cards didnt seem to follow your rule

have a look at card clocks mentioned and how it went up with volts while i was testing 01

http://www.xtremesystems.org/forums/...d.php?t=222421

i could and have seen improvements in smaller jumps

dinos I don't say that other multies don't work - for example the x10.8 - but the x10.7 doesn't always works with rams on certain frequencies -

[W1zzard]

dinos I don't say that other multies don't work - for example the x10.8 - but the x10.7 doesn't always works with rams on certain frequencies -

can you construct an example at "normal" frequencies where this pll problem exists? i seriously doubt that you are saying the freuqency changes in 50 mhz steps, because that is certainly not the case

[hipro5]

can you construct an example at "normal" frequencies where this pll problem exists? i seriously doubt that you are saying the freuqency changes in 50 mhz steps, because that is certainly not the case

For example:
When I pick 1070MHz core and 1153MHz rams through Rivatuner, I get a GREY screen with tiny paralel lines on my monitor....When I pick 1080MHz core and 1153MHz rams, I don't get this....

[W1zzard]

any example with lower clocks that i can test myself?

are you jumping directly to that clock from stock or are you moving in steps?

[hipro5]

any example with lower clocks that i can test myself?

are you jumping directly to that clock from stock or are you moving in steps?

I boot windows having rivatuner startup at 1000MHz and then I pick.

[W1zzard]

try if moving in smaller steps to that frequency helps