GPU Silicon Lottery - 7950 crossfire card OC differences

[Oblivion3333]

Hi,

I bought two MSI Twin Frozr IIIs 7950s for crossfire and while testing them both I noticed on stock volts of 0.993v card one can do 975MHz before errors appear in OCCT and card two does 925MHz. (These MSI cards ship 880MHz stock).

I have tested them both by applying more voltage and can do 1100MHz so no problems once overclocking with extra voltage, however is it normal to have such a difference between two identical cards? Do you think the worse card on stock volts can effect stability and overclocking potential when in crossfire with the better one?

I'm also thinking maybe card one isn't bad as such, it could just be my second card is a rare performing chip.

The main thing I want to know is are these differences normal between two cards?

Thanks

 

[suddenstop]

In my experience with 20 or 30 over clocks this is very normal. Silicon just out of the gate has more variability, with all obviously performing flawlessly at stock. As the process matures and possibly another stepping, over clocks tend to become much more consistent. Also strange as it may sound, the card that doesn't do as well on stock voltage may actually out clock the other card when over-volted.

 

[Chainzsaw]

Hello,

Yes the differences are normal for somewhat newish 7xxx series. AMD has stated the updated 7XXX series will use lower voltages compared to the cards that were just released 2-3 months ago.

What I mean by updated is that during manufacturing process, they find better ways to create the chip, hence newer cards requiring less and less power requirements.

Some cards will perform better than others - even within the same models. It's similar even with CPU's - some CPU's are "gold" meaning they OC very well, and are highly stable.

 

[Maximus_Delta]

These differences are perfectly normal.

Also, do your cards have the same stock voltage? 7970s, for example, come in a few flavors of stock voltage (1.175V, 1.112V, 1.050V, 1.025V) and that has an effect on what exact settings you need for your overclock(s).

For example - my Lightnings. One is a 1.175V card and one is a 1.112V card. The latter will do 1200 MHz on 1.175V (it also has a higher amount of OC headroom - it will go to 1250 MHz), but the former needs 1.22V to get 1200 MHz stable (and 1200 MHz is its top 100% stable clock).

You don't have a pic of your Lightnings in CrossFire do you ? Would really like to see how that looks.

Back on topic, I agree these differences are pretty normal. One of my 7970s overclocks higher, albeit only about 20mhz, on stock volts and also one draws more wattage at full load on stock volts. Just factor these sorts of things into which card you put on top e.g. the one which runs coolest.

 

[Maximus_Delta]

Magnificent thanks

 

[Oblivion3333]

Wow looks pretty awesome! Epic set up!

 

[Oblivion3333]

Yeah not sure why they choose those colours for the fans, black would have been better.

What are the temps like at your oc of 1205? I noticed my top card runs ten degrees hotter, which limits my max OC in crossfire. But I can run 1100 without it getting too hot. Haven't gone higher because of temps and also not done the enable unofficial overclocking thing yet so 1100 is the max the slider goes to in AB.

 

[Oblivion3333]

I see. Mine are at about 84c on top card and 74c on bottom card at 1100... but i have just 1X200mm fan blowing at them from the front and then my top and rear fans in my 650D case are all exhausting... Your set up seems to be better though as your fans are forcing the heat downwards, where as mine the top card gets cooked a bit by the bottom but the airflow does get rid of all the heat quickly out the top and rear of the case. I have a H100 by the way so prefer the fans to exhaust out the top of the case rather than blow hot CPU air onto the cards.

I totally agree that these cards dont like OCCT, them seem to get saturated by the 100% GPU load and cant get the heat away quick enough. As you said in game they perform very well, but they just dont like these extreme power virus tests. By the way what do you think the max safe 24/7 voltage for these cards is? I mean reviewers are putting up to 1.25 through them, but i do want mine to last 3-4 years ideally. I have so far got 1100 running on my 7950s at 1.17v but havent tried to reduce that voltage yet to see how low they can go.

What is interesting though is that there can be such a varience between two cards that are meant to be identical. Especially when the lightnings are meant to be cherry picked and selected as being good overclockers. Until you told me i didnt even know that cards could come with different stock voltages.

I have to say though once again, amazing set up, i do like the lightnings and wouldnt of minded those if they were released when i bought mine, although i dont think they would fit in my case as a ihave a fairly chunky sound card in the top PCI X1 slot and the "domes" on the lightnings would have hit the soundcard.

 

[Oblivion3333]

EDIT after some testing:

Here are my findings then... what do you think?

Some points before reading the results below:

- Bottom card is primary and running the monitor and testing done in Crossfire

- Top card runs at most 10c hotter than bottom (due to air flow in the case etc i think)

- Overclocking and voltages were set in MSI Afterburner and readings taken with Hardware Info (i hope they are correct as there are many readings shown in this software)

Testing results:

**IDLE - 300 Mhz / 0.993v**

Top Card - GPU 0 - 0.945v / 8w
Bottom Card - GPU 1 - 0.984v / 5w

**Load - 880 Mhz / 0.993v**

Top Card - GPU 0 - 0.972v / 54w
Bottom Card - GPU 1 - 0.984v / 55w

**Load - 1025 Mhz / 1.093v**

Top Card - GPU 0 - 1.062v / 86w
Bottom Card - GPU 1 - 1.075v / 78w

**Load 1100 Mhz / 1.174v**

Top Card - GPU 0 - 1.083v / 191w
Bottom Card - GPU 1 - 1.156v / 187w

Now this bottom card is the one that can only do 925 Mhz on stock volts and the top card can do 975 Mhz on stock volts.

 

[17seconds]

Have a little fun and test the ASIC quality of your GPU:
- Download and run GPU-Z
http://www.techpowerup.com/downloads/2137/mirrors.php

- Right click on the top title bar and select "Read ASIC quality..."

That will give you an idea of the quality of your GPU and it's overclocking abilities.

 

[Oblivion3333]

Ok I tried that, 78.5% and 78.8%.

However I heard that test is not accurate and not proven to relate to OC ability...

 

[Maximus_Delta]

If you are not using OCCT to test stability at max load / heat what are you using? My cards are actually fully stable at 1025/1500 on OCCT shader complexity 8, but the top card is going as high as 96/97 degrees before the fan goes 100% on auto to bring it back down below 95 degrees. In games the top card only gets to 85 degrees at 1025mhz. I would like to push higher but I have been relying on OCCT to verify its stable...

Also I too only have a 4 degree temp spread between my two cards currently. Both also indicate to be 1.12v cards checking via Afterburner monitoring (one is 74.8% ASIC the other 79.2%). However I did notice one was drawing more watts on full load (OCCT) and I put that one on the bottom, before doing this the temp spread was 6-7 degrees.

 

[Maximus_Delta]

Thanks, I think I will switch to your method of testing, makes sense. However I did some testing on OCCT with some interesting / concerning results.

In OCCT @ 1025/1500 – 1112mv both cards draw around 180w

In OCCT @ 1080/1500 – 1149mw both cards draw 215-220w

However, if I set to 1200mw going for 1100mhz+ after a few moments it crashes with the top card shutdown...

Even if I set to 1174mw (the highest stock 7970 voltage) I still get the above crash after maybe 5-10mins with the top card shutdown.

Any ideas? I was wondering if my PSU is too weak but I am hardly using extreme voltages above.

 

[Oblivion3333]

BigMack, how did you get MSI AB to go past 1100? Did you have to switch the Bios switch on your cards to unlock it or did you use a .dll edit thing?