HD 5670 overclocking using CCC

cosmiconion

Distinguished
Jun 6, 2010
48
0
18,530
Hi Guys,

I am using CCC to overclock my Sapphire 512mb HD 5670 (I have the most up to date drivers / CCC version).

Before using the ATI overdrive function my clocks are:

- GPU clock 775mhz
- Memory clock 1000mhz

Now while my GPU is in power saving mode it is running at:
- GPU clock 157mhz
- Memory clock 300mhz

When fully "overdriven" utilising the Overdrive facility my clocks become:

- GPU clock 850mhz
- Memory clock 1050mhz

However this is where my query applies...

As soon as I OC the card my power saving mode seems to become less restrictive:

- GPU clock 400mhz
- Memory clock 1050mhz

The 400mhz / 1050mhz power saving clocks also apply even if I only increase the clocks by one increment.

The first image here:

http://www.tomshardware.com/reviews/radeon-hd-5670,2533-18.html

shows the same full overclock but with the clocks returning to the default power saving levels.


Is their a particular reason for my clocks staying considerably higher once OC'd?


Cheers
 

JunoNH

Distinguished
Aug 1, 2011
22
0
18,510
sorry to dig up an old question, but it comes quite high up when searching google for overclock 5670

the 5670 utilizes ULPS, and when the parameters are changed, like when you overclock, the card has trouble returning to state 0 (157/300), so most often it will give you this state instead. a couple of tips if you really want to get the most out of a 5670, all you need is 3 things, a 2b pencil, some ram heatsinks ($15 for a bag of 10), and the TRIXX utility.

trixx will let you go over the CCC clocks limit and can be set to force ULPS off, which is essential for a good oc. i'd advice leaving the memory clocks at 1000 to begin with, even if the stock is 950 you can set it at 1000 without trouble. make sure you put the fan upto 100% spin all the time, as the memory is cooled by the recycled gpu air being blown over it. those ramsinks go best on the 2 exposed modules under the heatsinks right side. theres about 10mm clearance there so go ahead and just stick them on, if you have steady hands and a pair of tweezers you don't even need to remove the HSF.

you should be able to get to 850/1000 perfectly stable without doing anything obtuse. alot of people fail to realise that because of the lack of cooling on the memory, OCing that could cost you alot of GPU clocks that could be had with little fuss. the next part is probably the most important.

http://img232.imageshack.us/img232/3679/s6303495.jpg

put your card infront of you. find the resistor that is marked here as vcore up.

take your 2b pencil and focus the point down, applying gentle pressure, but not letting the pencil loose contact with the resistor, go up and down 10 times. this took me a bunch of times to get right without slipping off constantly, and i totally recommend practicing drawing on a line of solder or a straightend paper clip first.

ok now here's the warning. 10 times should be enough, but thats all down to how much contact the pencil got on the resistor, and how much graphite was left. this isn't an absorbent surface like paper, some strokes will wipe off more graphite than they lay, or atleast not lay as much as others, so it's up to you how many strokes you feel will give you a .1 or .2v increase. The only real way to tell, is to have a multimeter, or failing that look at your temps on gpu-z or HWmonitor. if you were at 30c idle before and now it can't drop bellow 40 with full fan on, that would suggest you did too many lines. a hearty volt mod should raise your idle temp 1 degree, maybe 2 if you don't turn 100% fan on or have generally bad airflow. a safe volt mod shouldn't even affect your idle's.

also, if you smell burning plastic, hold down that power button until everything turns off.

last tip, the little black heatsinked area just next to the top of the fan by the box saying r47, feel free to give that a touch AFTER you've turned off your computer (or crashed it). if it's burning hot the instant you touch it, thats a bad sign. i have the heatsinked version, but not all 5670's do have the mosfet heatsinked, so i can only speak from experience on this, but i've found if by the time my computer turns off, that heatsink hasn't gone down to a temperature i can keep my finger on for a few seconds without burning it, then there's a good chance those components are running to hot, and i need to bring the volts down a bit more.

hope this helps. sorry about the essay, but in practice this whole routine might only take 10-15 minutes if done cautiously, and can give you much more than you paid for.
 

JunoNH

Distinguished
Aug 1, 2011
22
0
18,510
not to sound patronising, but volt modding doesn't raise the clocks itself, it just gives the card more power to reach higher clocks than before.

here's a list of errors you may encounter and what they mean

http://www.playtool.com/pages/artifacts/artifacts.html

to add to that, i'll just say, if your screen get grey or coloured vertical bars, that is most likely a vga ram issue (overheating or not enough voltage)

sudden crashes will either be far too many volts or a psu failure due to not enough wattage being supplied to the pcie lanes, thru the 12v rail. this is especially prone to heavily oc'd systems that are running <600w PSU's, or low rated 12v lines. a highly OC'd 5670 will be drawing ~100w's from the pcie lane.

freeze then crash, or driver failure and recovery without crashing are most likely GPU errors. again, overheating or not enough volts.

benching on anything but furmark won't really test your OC's stability. so remember, just because it ran in a bench fine doesn't mean you wont get one of the errors i listed after 10 minutes on crysis.

assuming you had infinite watts and perfect cooling on all area's, you'll know when your coming to the limit of your card when higher clocks are giving you worse benches than lower ones.
 

JunoNH

Distinguished
Aug 1, 2011
22
0
18,510
if you've laid plenty of graphite, and your still getting crashes at pretty tame clocks, the culprit may be an overcharged mosfet burning up your card.

bumping up your vcore will stabilize your mosfet. transfering more of the voltage regulation over to the processor. this will result in higher CPU temps, but in return you'll get lower gcard temps and more stability at load. if you lowered your vcore earlier to save energy or keep the processor cool, you may have to take it back up again when OCing this card, however the stock settings with UPLS enabled takes low vcore quite well.

the best clocks might not be the highest scoring ones, but the most vcore effecient by performance gain. a vcore of 1.5v is the maximum safe for most modern processors without specialist cooling. aslong as you have some watts available, 1.5v on vcore should hold pretty much any OC this card can give you, but my advice would be to find a stable clock at 1.5, then drop 50 mhz on both and take down the vcore a little. final test on any gcard OC should be to play something graphics intensive for a while, then turn off and touch the mosfet to see how it's holding up. warm or very warm is fine, hot is not good and burning hot means you have to either turn up vcore or take off some lines.