Radeon X800 Non-Pro/X800 XL/X850 Voltmods

Super XP said:

Ya, the Sapphire Radeon X800GTO2 all have the R480 Core. And when you flash the bios to gain the extra 4 pipes for a total of 16, then you can OC to @ least X850XT PE speeds. And even take it further to over 600 MHz on the core with water cooling.

They beat the new X1800 XL's by far in terms of performance.

I ordered a GTO2, should get it in 2 days or so

Click to expand...

Oh gees give me a break. The x850 isn't even in the same league as an x1800XL. I am well over 10K in 3DM05 with x1800XL's on air cooling.

Viper

Super XP said:

Ya, the Sapphire Radeon X800GTO2 all have the R480 Core. And when you flash the bios to gain the extra 4 pipes for a total of 16, then you can OC to @ least X850XT PE speeds. And even take it further to over 600 MHz on the core with water cooling.

They beat the new X1800 XL's by far in terms of performance.

I ordered a GTO2, should get it in 2 days or so

Click to expand...

The first part is true as far as I've seen, but the part about the X1800XLs...thats just not true.

The X1800XL is faster than the X850XT PE stock and also overclock very well.

not to mention shinier, prettier, less loss with AA/AF, and they have SM3.0

if you read reviews around you'll see some where the X1800XT gets like 7-8K and others they break 9-10 in 3dm05. It's cause a lot of those reviewers have dual core CPU's, and a dual core 1.8Ghz system means a mere athlon 3000+ level for the GPU to abuse... its not fast enough. so read the reviews with single core GPU's if you want to know how fast it really is

(FWIW: Nvidia have dual core drivers already (dodgy as they may be) so they arent hindered like ATI is there.)

ViperJohn said:

Oh gees give me a break. The x850 isn't even in the same league as an x1800XL. I am well over 10K in 3DM05 with x1800XL's on air cooling.

Viper

Click to expand...

I've seen reviews with the X850XT PE beating the X1800 XL, but now with the Cat 5.11 drivers, the XL comes on top.

What would be the easiest way to undervolt the GPU (with a VR, not VID) and memory? Plan to use my X800GTO in a quiet PC with passive cooling and I want to reduce heat output. I don't plan to overclock, will keep it at stock speeds (although it looks like I got an R480, will probably try to unlock the extra four pipes). Thanks.

undervolt eh? you'd need to cut the trace already there, solder in a VR pre-set to the original resistance, and raise away. The other guys here are better on this than me, so they should be able to tell you the best types to use.

for memory undervolting, just follow the resistor mods but connect the other end to +5v instead of ground, so the voltage regulator sees a higher voltage and reduces its output

wizz: you should mention that in the first post, and how much lower the voltage would be. i'm sure a few people would be interested.

I'd assume the same could be done with the Vgpu mod, just take the VR from the feedback pin to Vcc on the regulator? And in both cases start out with maximum resistance and lower to undervolt?

W1zzard said:

for memory undervolting, just follow the resistor mods but connect the other end to +5v instead of ground, so the voltage regulator sees a higher voltage and reduces its output

Click to expand...

WHOA!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

You do not go to +5V!!! Your B+ connection would go to the output of the Vdd and/or Vddq PS output(s) the regulator is regulating.

Virtually all voltage feedback circuits are the same. You have two resistors in series forming a voltage divider. The regulators feedback pin is connected in between the two resistors. One end of the series resistors goes to ground and the end of the series resistors goes to the output of the power supply the regulator is regulating.

For increasing the voltage out we are simply paralleling a trim pot across the low side resisitor of the voltage divider. To lower the regulators output you put your trim pot in parallel with the high side resistor of the divider.

Do not assume you can use the same trim pot values for doing a voltage lowering mod as a voltage increasing mod. The resistor values used in the voltage divider are seldom the same value so the resistances you parallel them to have the effect you want can be radically different.

Viper

So for Vdd/Vddq mods you take the feedback pin to a Vdd/Vddq read point through a VR. That'll result in a parallel resistance with the high side of the voltage divider. Start with a high resistance and decrease to reduce voltage?

Also, what exactly is Vddq? Will it be worth the time to wire up a mod for it and reduce voltage, or does the circuit not put out a lot of heat to begin with. Does it need to be proportional to Vdd (like Vtt for a motherboard memory mod for example) or could I undervolt Vdd and leave Vddq alone w/out any problems?

Also, could the same method be used for Vgpu. Take the same feedback pin as in the overvolt mod through a VR (set at maximum) to Vgpu? Lower resistance on the VR to decrease parallel resistance, therefore reducing Vgpu.

Thanks all.

frostedflakes said:

So for Vdd/Vddq mods you take the feedback pin to a Vdd/Vddq read point through a VR. That'll result in a parallel resistance with the high side of the voltage divider. Start with a high resistance and decrease to reduce voltage?

Also, what exactly is Vddq? Will it be worth the time to wire up a mod for it and reduce voltage, or does the circuit not put out a lot of heat to begin with. Does it need to be proportional to Vdd (like Vtt for a motherboard memory mod for example) or could I undervolt Vdd and leave Vddq alone w/out any problems?

Also, could the same method be used for Vgpu. Take the same feedback pin as in the overvolt mod through a VR (set at maximum) to Vgpu? Lower resistance on the VR to decrease parallel resistance, therefore reducing Vgpu.

Thanks all.

Click to expand...

Correct although I would go across the resistor itself as I do not use leaded trimmers. I ofter use a 3 wire trimmer install so I can both raise and lower the voltages.

Vddq is the voltage applied to the memory chips I/O buffers.

If yu lower the Vdd you should lower the Vddq as well. With rare exceptions (the x800 XT-PE AGP cards were one) Vddq should always be run equal to or less than Vdd.

With memory chips Vref will normally run about 50-70% of Vddq (just depends on the chip) and must track with the Vddq. Vref is normally derived from the Vddq PS through a fixed SMR voltage divider to set the percentage and maintain the track.

Viper

Thanks for the help.

Just curious have you messed around with X800 undervolting? What kind of results did you get? Would be nice to know what to expect w/an R480 (400MHz/12 pipes) and Samsung 2.0 (490MHz).

frostedflakes said:

Thanks for the help.

Just curious have you messed around with X800 undervolting? What kind of results did you get? Would be nice to know what to expect w/an R480 (400MHz/12 pipes) and Samsung 2.0 (490MHz).

Click to expand...

As I said I often use 3 wire trimmer in first mod workups installs so I can have total control over the regulation setpoint. In the past there were times where reducing a memory voltage resulted in a higher memory OC.

Viper

Heh, interesting. Guess lower heat > more voltage in some cases.

Although my goal is purely to keep heat down, would be pretty nice if I could undervolt and squeeze a few % extra out of the GPU/mem.

I definitely plan to do the Vgpu undervolt mod, and maybe Vdd/Vddq. Probably be within the next couple weeks. I'll be sure to post my results.

Just wanna add up my own experiments of the last week, since I'm playing with pencil and asus ax800xt, which is agp x800xtpe.. ATItool0.24 is my friend too.

So my overclocks were going about nowhere.. The card from stock 520/560 went up to like 563/567 completely stable.. Not good on the ram, very disappointed I'd say, since it is some gc16 samsung's 1.6ns. :shadedshu Then I heard about softvmod of the beta0.25.. but Mr.W1zz confirmed me no softvmod with x800 series... errr again..
So let's take the DMM and the good old pencil..

Stock
vcore=1.39v
vdd=1.99v
vddq=?.??v(did not measured it at first.. wanted to skip this one but had to do, and did it the blind way.)

First objective is to get that 600mhz from the 1.6ns ram.. Checked samsung's docu and went bezerk with pencil.
Now
vcore=1.44v
vdd=2.16v
vddq=2.19v

Reminding you I have this card on stock cooling. Asus' solution.. That blows hot air inside. The single slot one.. Yeah, without ramsinks, and still getting over it... Errr...
On the ram.. With a 120mm blowing on the side, I keep the case open, I got 599.40mhz stable for 3h+ then I decided to unplug the sidefan. It did errored after 20-30min. And now it's at 597.38mhz since 20min+... I'll let it go for the night, let it burn, let it burn..
Then go for the 600mhz stable!!

For the core, just did some testings at 1.42v, 1.43v, 1.44v, and seem to gain about 3mhz on max stable o/c for each .01v increase.. From 563mhz/1.39v stock, I went to 573mhz/1.42v, 576mhz/1.43v, 579mhz/1.44v... That's where I am now..
Temps get a little hot wiithout the sidefan, between 72c-74c, but even stock it goes up to 70c... W/Sidefan, temps dropped to 62c-64c.

I'll have a connect3D x800gto in next week, gonna try that 2.0ns r423 and see how it does beside the more expansive ax800xt 1.6ns r420...

Many thank you to you all for the good info, I went thru the whole thread, and others, always good to share the knowledge.


EDIT: I've tried vdd=2.20v/vddq=2.22v, and got very bad results, seems like chips don't like the 2.20v thru them. So I backed down to vdd= 2.16v but increased the vddq=2.24v. I now seem to have a 600mhz memory stable. Does someone know if these are "safe" to run 24/7 without much cooling? I still get 69c-70c/48c-49c temps in ATItool0.24 on "find max mem". Ram chips can be touched like 5sec with the finger before feeling burning.

I finally got to doing the undervolt mods, did Vdd and Vddq as ViperJohn suggested from the inverting input (FB) to Vdd and Vddq read points posted in this guide. Seems to work fine, Vdd and Vddq undervolt as they should.

But for some reason I can't get the Vgpu undervolt working. Tried taking VCORE+ to the Vgpu read point posted through a 10k VR. Lowering resistance does not seem to effect voltage in any direction, it stays at 1.409V.

ViperJohn (or anybody else), you have any ideas? I'm not really too knowledgable with this stuff, but it doesn't appear this circuit uses a voltage divider like the Intersil chips for Vdd and Vddq, so do I need to mod it differently? What would happen if I removed the resistor between VCORE+ and Vcore (labelled R6 on the typical applications diagram on the spec sheet) and replaced it with a 1k VR. Set it to the default resistance of R6 and decrease resistance to reduce voltage?

Or if there's an easier way that wouldn't involve removing a resistor I'm all ears. Please help I'd like to get this all back together ASAP.

EDIT: Alright, think I got it figured out. Experimented a little and took VCORE+ to the VSS pin on the regulator and now it seems to be undervolting like it should. Currently have the GPU at 1.35V, gonna test for stability and then take it down a bit more. Needless to say I'm very very happy, glad to have everything up and running and getting the results I wanted. Thanks to all who helped, I really appreciate it!

EDIT2: I seem to be having the wierdest problems now. It seems like I cannot boot below ~1.375V. I've taken it all the way down to 1.3V while the PC was running, but anything <1.375V and the PC won't start up. The PC (sounds like it may be coming from the PSU?) makes a squealing noise and Vgpu reads very low (fractions of a volt). As I raise resistance on the VR, the squealing noise gets higher in pitch and Vgpu increases slightly until at about 0.3V Vgpu jumps up to ~1.375V, the squealing stops, and the PC proceeds to boot normally.

I'm stumped as hell. Was taking VCORE+ to VSS a bad idea?

Bump, anybody have any ideas? Is this maybe some sort of under voltage protection kicking in? If so, how could I get around it?

Thanks all!

Not many have done undervolting and its still the weekend in some places of the world give it a few more days.

Sorry, just anxious to get this all working and do some extreme undervolting.

P.S. If it makes a difference looks like I had soldered to a different pad than I thought, taking VSS through a 10k VR to ground actually undervolts (but has problems booting <1.375V). With VCORE+ to VSS the GPU does not undervolt.

Also I've been thinking, would it be possible to build some kind of time-delay relay circuit to fix the problem? Wire up a capacitor and the control side of a solid state relay to +5V for example and then hook the Vgpu mod to the switching part of the relay. Capacitor should resist voltage change right? So one could control the length of time between 5V being initialized and the relay enabling the volt mod based on the capacitance and resistance of the circuit. This way the mod would be enabled after the IC started up at normal voltage, as the only issue seems to be starting up at lower voltage, not running at lower voltage.

Seems kind of crude, but I don't see why it wouldn't work. Of course any other suggestions would be appreciated, such as a way to avoid the undervoltage protection altogether.

This is going beyond me, my first thought was "slap a capacitor in somehow" and a blank beyond that. I'll just wait and watch while someone with a clue helps..

Well honestly most of this stuff is kind of beyond me as well, that's why I have to beg for help.

I'm actually at college for electrical engineering, and in my intro to EE class we spent the whole semester mainly talking about resistor, capacitor, and inductor circuits, but not so much about semiconductors. So I think I have a pretty good handle on the basics of these parts but am pretty clueless when it comes to voltage regulation, op-amps, MOSFETs, etc.

EDIT: LOL, I'm such an idiot. I don't know why it took so long for me to realize this but I could just test how low the GPU voltage goes using the resistor mod I have hooked up and then use the permanent VID mod setting for this value to get around the booting problem I have. No need for any complicated relay circuits.

For those who might be interested, GPU is now at 1.135V. Dropped load temps about 20*C with passive cooling. Ended up removing the Vdd and Vddq mods as the difference in memory temperature was neglegable (I was able to take it down to ~1.75V).