5870 memory overclock? What is safe or optimal?

[Jay85]

I currently have it sat at 1250 and gpu clock at 900 max. Did clock test and both passed of course. Im wondering if theres any point in clocking the memory any higher though?

 

[hudson86]

i'm wondering the same thing, just fitted a Zalman VF3000 to my Sapphire 5870 and about to start messing with the clocks.

Have you bios modded yours to amend voltages or its yours already good to go on that ( MSI etc )

 

[straxusii]

No dont bother, it's tricky to know with the 5870 as the memory just clocks back with error correction and you can get worse performance over 1250.

 

[Jay85]

Mines a referrence card but it doesn't support voltage tweak but neither do i want to flash the bios on the gpu. Guess i'll stick with my current clocks if its worse performance over 1250.

 

[bobert50]

My 5870 pcs memory doesn't like going above 1250 but the core seems fine at 930 at stock volts.

 

[Greebo]

The only way to tell with error correction is keep upping it and benchmark it with something like vantage. Your score will go down as error correction kicks in

 

[Mr Krugga]

The only way to tell with error correction is keep upping it and benchmark it with something like vantage. Your score will go down as error correction kicks in

Did this several times but the results were inconsistent.

Stock voltage core/clocks. Memory clocks/3D Mark Vantage GPU Performance Points:

• 1100/14,560
• 1150/14,595
• 1175/14,715
• 1200/14,687
• 1225/14,788
• 1250/14,776
• 1275/unstable

3D Mark Vantage Custom (Extreme@1680x1050):

• 1250/unstable
• 1225/9016-9075-9125
• 1200/9043-9046-9035
• 1175/9041-9164-9095
• 1225/9119-9090

I just stick to 1225 on memory (afaik Radeon 5850s use the same modules).

 

[Greebo]

Vantage can be inconsistant and you can have point score differences between each run even at the same settings.

However, if you can be bothered, doing several runs for each setting and taking an average will give you good enough results.

Also don't forget, if you are on the limit/over it, the memory correction will give different results each run as you will have a different number of errors on each run. You may have a "lucky" run where there were a low number of errors and the extra speed of the memory outweighs the correction taking place.

If your extreme results are 3 results for each memory speed (?) I am tempted to say that at 1225 memory correction is kicking in.

1175 seems to have the highest average score?

 

[Mr Krugga]

I'll do some more testing today, maybe with Unigine Heaven as well? I'm only sure that 1250 was definitely unstable

 

[Greebo]

Unstable is one thing which means even the error correction cannot keep up and I am not saying 1225 is unstable.

Just saying that it looks like lower average speeds than 1175 and hence there is more error correction going on.

I think I might need to spend some time checking mine as well.

Heaven may be a good one to check with, not sure how consistant that is between runs.

Maybe crysis benchmark since it does 3 runs and averages?