7970 Loses 85MHz of OC With New Z77 Mobo?

[crashtech]

I've been playing with my new Sapphire 7970 and have casually overclocked it to about 1135MHz with few issues, and it seemed to have more in it if I could get figure out a safe voltage and temp, etc.

This was on my old HP xw8600 with twin Xeon E5472s on the 5400 chipset. Pretty robust setup for its day which still does very well in multi-threaded apps, but its single threaded performance was just not there, so;

I found a good deal on a Gigabyte Z77X-UD3H and borrowed an E3-1230 to check it out. Contrary to what I've been told, the E3 can't be even slightly overclocked using Turbo multipliers, at least not with this board, but that's a subject for another thread.

What really concerns me is that my 7970 will only make it to about 1050MHz on this new board using the same techniques (minimal overvolting, stock cooler). I'm sure I could get it to do better with some more attention, but what I am curious about is why a mid-range Z77 board from a decent mfr. would be so much worse than ~6 year old tech?

So far I have tried the old Delta power supply from the HP on the new setup with identical results, so I don't think it's PSU related. The PSU I've been running with the new setup is a Seasonic M12II 520W, which I believe should be adequate for my single card setup, though perhaps not with too much headroom.

Also all BIOS settings on the new board are stock, which was the next thing I suppose needs looking at. Is there anything specific that could aid or detract from the ability of the 7970 to OC? What do you suppose is going on here?

 

[3DVagabond]

That Seasonic is fine. Plenty of headroom.

You might check and see if there's any other BIOS available for the mother board?

 

[crashtech]

Just checked, the only newer BIOS is still in beta. Hmm.

 

[crashtech]

GPU-Z reports a very different stock voltage than Trixx does, when Trixx is set to 1137mv GPU-Z reports 1061. This needs some looking into. I'm getting more clocks out of the card by crankiing the volts up, but it would be good to know the actual value. May need to fire up the old rig and pay better attention to this detail.

 

[Plimogz]

depending on what you're doing to test stability, it could be that the better single threaded performance of the new CPU is keeping the GPU fed closer to 100% of its capacity.

So, if you were CPU bottlenecked before, it would be possible that removing that bottleneck with the new system is allowing the 7970 to really stretch it's legs and revealing an instability in your previous OC voltage/clock settings.

Is the GPU stable at stock settings in your new board? if so, I say just start OC'ing from scratch and see if you can get that 1050 (or 1135, rather) stable again with slightly higher volts.

 

[crashtech]

I was getting 1135MHz stable in my old board under a variety of circumstances. On the new board the driver would crash at 1050MHz just mousing around on the desktop.

Right now I am getting 1110MHz stable (and increasing in small increments) but with the voltage slider in Trixx all the way up, indicating 1300mv, but GPU-Z indicates 1182mv. So either the new board requires me to give the 7970 more voltage, or it is causing the voltage to be lower while the readings are bunged up compared to the old board.

I'm tempted to put the old xw8600 back in service temporarily to see if the old setup also displays the same sort of voltage readout discrepancies. I didn't think the mobo had anything to do with GPU voltages...? Probably need to find yet another piece of software to access the sensors, too, or find the pinouts on the card for a DVM.

 

[ShintaiDK]

It might also be the move to PCIe 3.0 that does it.

 

[alcoholbob]

Z77 I found was definitely el cheapo compared to the old x58 platform. Had to get ac powered USB adapters just to keep all my USB devices powered (just 2 hdd and a controller) otherwise random devices get disconnected whereas my evga board had no problems even with all the USB ports used.

 

[ViRGE]

GPU-Z reports a very different stock voltage than Trixx does, when Trixx is set to 1137mv GPU-Z reports 1061. This needs some looking into. I'm getting more clocks out of the card by crankiing the volts up, but it would be good to know the actual value. May need to fire up the old rig and pay better attention to this detail.

GPU-Z reports the measured voltage instead of the VID. In many situations this would be the more useful metric, but in the case of video cards the VID is probably more meaningful for overclocking.

 

[crashtech]

Thank you all for the interesting replies! Overclocking this new card has proven more compelling (addicting?) than I had imagined.

I've had it running at 1110Mhz pretty well now, but the voltage selected is 1250mv, which on its face would seem to me more than desired, but then there is GPU-Z giving an actual of 1143 under load.

@ViRGE, I don't understand why the actual would be less useful than the setting?

At any rate, I'm happy to have most of my gains back, though it seems like I have to push the card harder to do that, though without putting back in the old machine I won't really know that.

Do you guys think 1230mv at idle is excessive? That's what it pops up to when not under load.

 

[(sic)Klown12]

Do you guys think 1230mv at idle is excessive? That's what it pops up to when not under load.

Yes. It should downclock to 300/150 at 0.806/0.850v depending on which BIOS you have.

 

[crashtech]

Yeah, I think that is actually my fault. I'm away from the machine right now, but I believe there is a check box checked that should not be. I'll surely be asking more noob OCing questions this evening...

 

[ViRGE]

@ViRGE, I don't understand why the actual would be less useful than the setting?

In a word: vDroop. Users see their voltage dropping (and never matching the VID) and think something is wrong with the card, when in fact that's intentional. What users really want to know is whether their card is being overvolted as they've set it to, which VID can tell them without causing them to freak out.

 

[crashtech]

I have some things to learn about overclocking, obviously, thanks for the heads up, ViRGE.

But so far there is still no explanation for why the card runs faster with a lower VID setting on the old board. I'm still tempted to put the old xw back in service briefly and see if the actual loaded voltages are different than they are with the new setup. I don't see why they would be, though. Same card, same driver, hell it's the same install that I just ran sysprep /oobe /generalize on and stuck it on the new setup.

 

[crashtech]

Must have been a driver issue, I can get 1160MHz now with the 13.3 Beta 3 driver, but now I have it at 1130MHz and 1180mV, this seems to be a good compromise between heat and speed.