Radeon 9600 PRO Overclocking

[Dave Baumann]

This is billy bonkers. Its now up to 537.75, and maxed out on the multiplier on Rage3D! I can't quite believe this.

        GT4    PS2.0
400/300 18.7   27.9
520/300 23.1   31 
537/300 23.6   31.4

Ich - whats the reg setting to increase the multiplier?

 

[Neutrality]

This is pretty darn cool. Imagine the R300 on 0.13 um.

-Neutrality-

 

[Ostsol]

This is pretty darn cool. Imagine the R300 on 0.13 um.

-Neutrality-

Well, I'm guessing that it'd max out quite a bit faster, given that the R300 has TONNES more transistors.

 

[Neutrality]

This is pretty darn cool. Imagine the R300 on 0.13 um.

-Neutrality-

Well, I'm guessing that it'd max out quite a bit faster, given that the R300 has TONNES more transistors.

Ah, yes, but still. It cant hurt to go with 0.13 um.

-Neutrality-

 

[Prometheus]

Dave,what about the memory clock?

 

[gokickrocks]

HKEY_LOCAL_MACHINE\SOFTWARE\Game Util\ATI Radeon Overclocker

just change the UpperCoreX and UpperMemX keys to a larger multiplier

 

[mczak]

I'm not sure however that the memory overclocking would be indicative of retail cards - they could well use different memory chips.

mczak
btw to get the most out of that chip, you'd probably need a volt mod. Since it looks like it very easily does the required 400Mhz and it only needs a tiny hsf, it's likely ATI hasn't used the maximum feasible voltage for that chip, but instead used less. If you could measure the voltage and compare it to a 5600Ultra, I'll suspect it's lower on the 9600pro despite using the same process technology.

edit: typos

 

[Ostsol]

Anyone know what the transistor count on the FX5600 is as compared to the RV350?

 

[gokickrocks]

i dont remember where i put my 5800 diagram for vmodding, but i remember the vcore was a square ISL chip and the vmem was a rectangular ISL chip like on some ti4800s

edit: nevermind this post, i thought you said 5800 and not 5600

 

[Nebuchadnezzar]

I sense 600-700 with volt mod + 'fat' cooling 8)

 

[Dio]

Amazing what you can do on .13 with standard cooling and no power connector, innit?

 

[demalion]

Why would they go further? I see plenty of reasons for them not to even if they could get a lot of cards that could handle these speeds:

I think there is an inherent limitation because of how far apart the best possible and the very common case good yield clock speeds (9600 non Pro) are. If there was too much separation, they'd be wasting a lot of performance (i.e., opportunities to charge the user more for each chip...overclocks aren't included in pricing) on the cards that fell in the middle and couldn't reach these best case clocks, even if it was common.

If there is too much performance offered by these special cards, it would cannibalize opportunities to clear out inventory (a mistake ATi probably doesn't want to repeat).

Of course, they miss out on opportunities for charging even more for super clocked versions of these cheap chips (though card makers might not, I guess we'll have to see...and I still wonder about how cost effective a MAXX RV350 card would be), but there would still be a question of board costs and drawing people away from high end cards that make quite a bit more than they could charge for it (it would have to be clocked at about 650/620DDR to sell for the same price).

And they could just save "super" 9600 inventory for matching up with whatever "R390" launch comes in the fall (though the naming would present some difficulties ) and driving some newly invigorated margin injection into their sales..

 

[Colourless]

Impressive

 

[Prometheus]

Anyone know what the transistor count on the FX5600 is as compared to the RV350?

FX5600=80mil trans.
RV350=75mil trans.

 

[LeStoffer]

This is billy bonkers. Its now up to 537.75, and maxed out on the multiplier on Rage3D! I can't quite believe this.

My God Dave, have you been smoking something hallucinogenic?

 

[Dave Baumann]

Anyone know what the transistor count on the FX5600 is as compared to the RV350?

FX5600=80mil trans.
RV350=75mil trans.

AFAIK, its 60M for RV350

 

[Prometheus]

AFAIK, its 60M for RV350

You're right Dave.
Got the 75M number from Tomshardware.

 

[Ichneumon]

This is billy bonkers. Its now up to 537.75, and maxed out on the multiplier on Rage3D! I can't quite believe this.

...

Ich - whats the reg setting to increase the multiplier?

lol... and we upped the limits with 3.8...

Change the ratios in the following reg key...
HKEY_LOCAL_MACHINE\SOFTWARE\Game Util\ATI Radeon Overclocker
UpperCoreX
LowerCoreX
UpperMemX
LowerMemX

 

[martrox]

This is billy bonkers. Its now up to 537.75, and maxed out on the multiplier on Rage3D! I can't quite believe this.

OH MY GOD! And with that tiny heatsink?

 

[mczak]

Anyone know what the transistor count on the FX5600 is as compared to the RV350?

FX5600=80mil trans.
RV350=75mil trans.

AFAIK, its 60M for RV350

Is this an official figure? All other reviews I've seen indicate ~75 million transistors (doesn't mean they are correct of course, they probably had all the same source...). Though 75 million indeed seems a bit high considering smaller caches, no hierarchical-z, half the vertex/pixel pipelines, but OTOH 60 million seems a bit low