ATI 4850/4870 voltmod thread

[dejanh]

Well I have been using the Diamond 4870 XOC BE and ASUS TOP BIOS. No real difference here..

Hmm, the 900MHz core I can't get stable. I can get Furmark stable but then Crysis will fail, or the opposit.

Looks like the best possible all stable is about 880mhz.

The mem I havent messed with yet. Stock/water is only accepts about 1025, but I am considering modding the mem as well.

Quite dissapointet that the vmod only gave me about 30 Mhz from the non mod 850mhz oc I had before Though I am quite sure it is possible to bench much higher, as I was able to run ati tool artifact free at 940Mhz.. Maybe not..

Is my problem normal or? Is it possible that a mod will remove the OVP (If thats what causing my problem?)

What do you cool your card with? And how musk can I expect to gain in mem mhz when max stock is about 1025? (Know its hard to predict, but if it is just 25Mhz, I wont even bother). To me 24/7 and stability is nr.1

My cooling is a HR-03GT with a 92mm fan on top with stock HR-03 sinks on memory and VRM. I am looking to install this guy http://www.thermalright.com/new_a_pa...50aHVzaWFzdA== (an HR-09 Type 4) onto the VRMs. I already tried the HR-09 Type 2 but its wider 15.5mm base is too wide and cannot fit properly onto the VRMs. The Type 4 guy has a 12mm wide base so it should fit. I found that the card is extremely sensitive to upping the voltage on the core and the memory. When you do both you get BS much faster. I think in this case it is some kind of OCP kicking in. You will see that yourself if you try to voltmod the memory in addition to the core. With such a high core at 880MHz or 900MHz you will not be able to OC the memory much from your 1000MHz, or at all. I would say that you will get artifacts as soon as you hit over 1050MHz so if 50MHz is worth it for you to mod things go ahead. Unfortunately the 4870 XOC BE does not OC that well. I am not sure about other cards but I think most 4870 have a problem getting over 850MHz core and about 1125MHz - 1150MHz memory.

I also found the same thing you did in terms of ATITool/FurMark and games. If I stabilize ATITool can be artifact free with high voltage at high core clocks with good cooling. However, FurMark crashes at those voltages so it forces you to go with lower voltages if you want higher clocks for FurMark. These higher clocks at lower voltages however cause instability in games as your lower volts now are not sufficient for stable operation in games so you need to lower your clocks as well. It is a really stupid cycle that I found. Hence the reason why I had to stabilize at the clocks I mentioned before. I have worked on this card for > 12 days straight with constantly the same results using three different cooling systems and several cards. All were with same results.

To summarize, what you will find is that if you up the memory voltage and memory clock you will have to lower your core clock and potentially core voltage, and vice versa. The highest stable overall combination will be something in between 850/1150 and 875/1125 (notice the trade-off between core and memory clocks ). Ironically, I found that in games the best performance was at 825/1125 clocks and that anything higher did not yield anything significant in terms of FPS (some higher clocks even yield less ) as well as that the performance at higher clocks did not justify the added heat output and power consumption of the board.

Edit: One other thing to note is that the Qimonda memory chips are 40X, so they are rated for max 4000MHz with critical maximum voltage being 2.0V (stock is 1.501V). Running them at 4400MHz or higher is already 10%+ OC on the memory so it is not that strange that it would be causing problem when trying to push it very high. Some people can run very high since they do not care about 100% stability and they do not even notice some artifacts in games so for them going 4600MHz - 4800MHz with high core clocks (> 850MHz) is okay.

Edit 2: I would not hold my breath on the OVP/OCP being disabled in the future using any BIOS or other hard-mod method really. I found in the past that there is usually just too much work involved in doing this with mixed results at the end. If anything, it severely increases your risk of blowing up the board.