5 Attachment(s)
Vmod on XFX 8800GT in SLI
Hi Guys
I just finished the Vgpu mod on my two SLI’d XFX 8000GT’s late last night, and I thought I’d share my results here. I took a bunch of pictures along the way, and I have a pretty extensive electronics background, so if there is enough interest I can post up a full step-by-step guide with pictures. It seems like there are a handful of people on here who really know their stuff, and another handful of people who shouldn’t be allowed near a working toaster, let alone a $300 video card. Hopefully I can write up a guide that could help the people in between
Now for the results:
Pre-mod I had what amounted to a set of dud cards. I’m not sure if the SLI configuration makes things less stable, but the pair of cards couldn’t reliably break 675MHz. Both cards were on DD Maze5 waterblocks with idle temps between 18-22C and load temps between 28-30C with the help of a dryer vent and a very cold Canadian winter. Basically, there was no excuse for their poor results. I actually returned one of the cards after finding out that it couldn’t even run reliably at the stock 900MHz mem clock. I got the exact same pixilated crashed discussed earlier in this thread. I had to scale the mem clock back to 870 just to get reliable play time in Bioshock.
The Vgpu mod has certainly been a saving grace for both these cards. I’ve run the voltage up to a conservative 1.276V measured and adjusted under load from ATItool. Both cards are currently happy at 800gpu/1000mem, which is exactly where I wanted them to begin with. I’m sure they could go higher, but to be honest, I just wanted them to perform reliably at a solid clock speed. I’m not willing to trade stability for another 60MHz.
My core temps have gone up 1C at idle and 2C under full load, which still keeps them below 35C at the highest. I have a bunch of ram-sinks all over the cards, covering the memory and mosfets, and a single 80mm fan blowing over the power supply end of the card. Everything is cool to the touch, except the RAM at the far end of the card, and the little SOIC chip on the bottom of the card, which is pretty toasty as well.
I was hoping that by keeping the cards below 1.3V under load, I could avoid the OCP mod, and so far it looks like I’m alright. I’ll perform some longer stability tests and report back on the outcome. I have a pile of SM resistors hanging around, so I guess if worse comes to worse I might give the OCP mod a shot.
Now, I have a quick question for anyone who is familiar with 3DM06. I’m finding that my scores for the video cards are really sub-par compared to what I’ve been seeing on here. I did a run with a single card fully stock, and the 2.0/3.0 scores were 4815/4915 respectively. At 800/1000, a single card pulls a more respectable 5650/5860, but I was expecting to be well into the mid 6000’s at that clock speed. What’s even worse is that the SLI’d pair turn in even more pathetic results. With two cards at 800/1000, and a Q6600 at stock clocks, I get an embarrassing 13696. I’ve had the processor up at 3.2GHz, and that helped quite a bit with the overall score, but the SM2.0/3.0 scores are still poor for the cards, regardless of where the processor is at. I’m running a 122-CK-NF63-TR motherboard with the latest 169.21 drivers. (there’s a 122-CK-NF68-A1 in the mail right now, so that should help with the Q6600) Any suggestions would be welcome!
Both my GT and GTS have Qimonda RAM chips. I wish I had some Samsung's! The way I understand it is that both the vRAM, and the memory controller are fed from the same voltage line. The vRAM by itself should respond better to more voltage, and I think the specs have them listed up to 2.5v. The problem arises b/c the memory controller doesn't respond well to voltage. So, if you start raising the voltage you are just making the controller crash, and that is seen as the memory avalanche.