FSX & SLI

[Black ZR-1]

[quote]Alright. I think we confirmed your running ok. Usually a GPU idles around 35-45c depending on the card. The newer cards

[DizZa]

Since we're talking about temperatures, in a stress test my X850XTPE reached 100degrees Celcuis with no artifacts, completely stable.  :D

I don't really care if it dies. That gives me an excuse to go buy a 7900GS or 8600GTS.

[Nick N]

[quote]Since we're talking about temperatures, in a stress test my X850XTPE reached 100degrees Celcuis with no artifacts, completely stable.

[Black ZR-1]

So I'm assuming the over-clocking stage is complete?

[justpassingthrough]

[quote]So I'm assuming the over-clocking stage is complete?

[ATI_7500]

Never knew you could push an 8800GTS *that* high without it making trouble...

[Nick N]

[quote]So I'm assuming the over-clocking stage is complete?

[Nick N]

Never knew you could push an 8800GTS *that* high without it making trouble...

Based on the card being from the first PNY run, the components being used will do that clock, and most likely higher. If the card is from their second production run then it may be at its max or very close to it right now. In either case, you assume the responsibility for raising those sliders. There is a right way and a wrong way to do it. I just happen to know what the memory chips and other components on the card will do so we skipped about 10 steps in testing and went right for the meat and potatos.



The correct way to do it is use a base increase (for that card) of about 50Mhz per clock, then raise it 10-20Mhz steps using a scan tool to check for temp and artifacts between each increase. The correct process takes about a day to do it right and about another 1/2 day to load scan the final numbers to insure no overheat and artifacts may appear.

[Black ZR-1]

Did you raise traffic to 25/60 and fly real world weather?

I haven't changed any settings other than fiber. I wanted to get the clocks right before I moved on.

[Black ZR-1]

I think the video card is set as high as it should go and still remain in safe limits based on the manufacture components and how they are currently working right now.

Will it go higher? It may go another 50+ on the core and 100 on the memory but in order to do that it requires special attention to setting it up using software to load it down and scan for artifacts all the while it monitors the temp.

Unless you would feel comfortable running a scan tool that will check for artifacts while it heats the card up so we can find the max core and memory on the card, I would say the clock on it is where it needs to be.

Be aware, finding and locking higher values and running for hours will most likely push the card very hard and although I have done it in the past just to see how far things will go and then reduce the clocks based on the MAX values instead of hardware estimate, it also comes with much higher risks. I will leave that up to you because we are already 150MHz+ on the core and 350+ on the memory past factory specs.

I also want you to understand, thus far there has been no real load placed on the card in a looped stress test, nor has the card been exposed and run a DX10 application so the current clock may be fine for FSX and DX9 but be over the top for a full blown DX10 run. That part I can not estimate without observing a test using the criteria mentioned to evaluate the situation.

I think I'm happy where I'm at right now with the card. I don't want to overdo it. Like you said I'm 1500/350+ the factory specs and any more would just put me back in the bottleneck..

[ATI_7500]

I just happen to know what the memory chips and other components on the card will do so we skipped about 10 steps in testing and went right for the meat and potatos.

Ah, this explains a lot.

I was wondering, since you kicked the chip frequency to 600 in the first run - seemed a little odd to me.

The correct way to do it is use a base increase (for that card) of about 50Mhz per clock, then raise it 10-20Mhz steps using a scan tool to check for temp and artifacts between each increase. The correct process takes about a day to do it right and about another 1/2 day to load scan the final numbers to insure no overheat and artifacts may appear.

Do you think the ATI Tool can do this well enough?



P.S: Since the 8800GTS are all based on one and the same reference design...could those values also be used for an XFX 8800GTS?

[Nick N]

I think the video card is set as high as it should go and still remain in safe limits based on the manufacture components and how they are currently working right now.

Will it go higher? It may go another 50+ on the core and 100 on the memory but in order to do that it requires special attention to setting it up using software to load it down and scan for artifacts all the while it monitors the temp.

Unless you would feel comfortable running a scan tool that will check for artifacts while it heats the card up so we can find the max core and memory on the card, I would say the clock on it is where it needs to be.

Be aware, finding and locking higher values and running for hours will most likely push the card very hard and although I have done it in the past just to see how far things will go and then reduce the clocks based on the MAX values instead of hardware estimate, it also comes with much higher risks. I will leave that up to you because we are already 150MHz+ on the core and 350+ on the memory past factory specs.

I also want you to understand, thus far there has been no real load placed on the card in a looped stress test, nor has the card been exposed and run a DX10 application so the current clock may be fine for FSX and DX9 but be over the top for a full blown DX10 run. That part I can not estimate without observing a test using the criteria mentioned to evaluate the situation.

I think I'm happy where I'm at right now with the card. I don't want to overdo it. Like you said I'm 1500/350+ the factory specs and any more would just put me back in the bottleneck..

The bottleneck is open but you are still in it. The only way I can get you out of the bottleneck is to put you into the doghouse or on the sofa for a few months.  

[Black ZR-1]

[quote]The bottleneck is open but you are still in it. The only way I can get you out of the bottleneck is to put you into the doghouse or on the sofa for a few months.

[Nick N]

I just happen to know what the memory chips and other components on the card will do so we skipped about 10 steps in testing and went right for the meat and potatos.

Ah, this explains a lot.

I was wondering, since you kicked the chip frequency to 600 in the first run - seemed a little odd to me.

The correct way to do it is use a base increase (for that card) of about 50Mhz per clock, then raise it 10-20Mhz steps using a scan tool to check for temp and artifacts between each increase. The correct process takes about a day to do it right and about another 1/2 day to load scan the final numbers to insure no overheat and artifacts may appear.

Do you think the ATI Tool can do this well enough?



P.S: Since the 8800GTS are all based on one and the same reference design...could those values also be used for an XFX 8800GTS?

Not necessarily. The reference design is a BASE for the actual production run. The distributor -can- order upgrades to that design to fill their order but the card run must meet the minimum reference design standards placed on it by Nvidia. Some manufactures, like Gainward, and a few other will order the runs with upgraded components.

What I usually find is they play a game. The first production run may be ordered with upgrades so the initial release of the product has a significant impact in the review world and the chatter on the message boards. The second order the skimp on and make more money because by then everyone is buzzing about what that product can do... it increases sales and the cost is dropped so it increases profits. Since overclocking means you void the warrantee, they don't care. All they want it the marketing profit.


Yes, ATI tool -v26- will do the job HOWEVER it may NOT properly raise the memory clock on the new Nvidia cards and it MAY try to change things in its settings the card will not like so it needs to be set up right. v26 is said to correct the lack of 7900-8000 memory clock control but I can not verify that. You can not have ATI Tool running and use the Nvidia control panel to raise clocks at the same time. The tool must exit and sometimes the system rebooted to enable the NV panel to have control back for card clocks... at least that is how it worked in the past..

I also understand that with the 8800 the "find max core" and 'find max mem" features may not work. I do not use those anyway.

In the settings I set up the tool to monitor temps every 10 seconds and turn off the change PCI Latency feature in the misc settings area. I then set the core and memory for 2D, low 3D and high 3D, load them, save that as a new profile and then adjust the high 3D value and run the "SHOW 3D VIEW" which heats up the card. After a good 3 minute run in 'show 3D view" I click "Scan for Artifacts" and let that run for about 10 minutes. If artifacts are found the timer will reset to zero and you will know either the memory or core is too high.

Using the tool take practice and an understanding of how core and memory work for a specific card with respect to their estimated high run values.

If everything checks out I will then stop the artifact scan, raise the core and memory 10 or 20 MHz, and allow the 'show 3D view' to heat it up for 3-5 minutes, then run the artifact check and repeat the process.

Once I start to see the artifact timer reset, then I back off the memory and core by 50 MHz and then test each one separately until I find the max on each. Once I have those numbers I will then set up 20 MHz shy of the max and run a 1/2 day load and scan. If more errors show I back them both down by 10 MHz until not errors appear.

All the wile the heat must be watched. I will usually not run a newer Nvidia core past 80c in those tests because real use and a load test are 2 different things. If the core reaches that temp I will not even try to go further even if I can. At that point I would probably replace the HSF and install memory heatsinks.

[Nick N]

[quote][quote]The bottleneck is open but you are still in it. The only way I can get you out of the bottleneck is to put you into the doghouse or on the sofa for a few months.