No, it doesn't. I now get "VPU Recovery" errors saying that the GPU stopped responding.
The problem may be software related instead of hardware, though it did seem like a possible power issue at first (and I should have posted this in the support subforum instead -- my apologies). I thought that Catalyst 11.12 came with APP Runtime 2.5, but it's bundled with 2.6 instead, which isn't supported on WinXP. I'm going to try moving to 11.12/2.5 and see if that helps.
Topic: Can't maintain full hardware acceleration mining on a Radeon 6950 (Read 1037 times)
April 11, 2013, 05:22:13 PM
Does it run fine at the standard clock? That would be my first test, reducing the overclock.
April 11, 2013, 03:13:16 PM
I haven't actively mined since 2011, but last week I had the good fortune to find a MSI 6950 on Craigslist for $100 and figured it was too good a deal to pass up. If I can earn back the cost of the card before the difficulty skyrockets, awesome; if mining becomes unprofitable for any reason, I'll still have a cheap upgrade to the 5770 currently in my desktop. I have a WHS system that is already running 24/7, so I put the 6950 in there. I installed Catalyst 11.12, which (to my knowledge) is the latest set of drivers with OpenCL that works on WinXP.
The card will only allow me to overclock it as high as 840MHz, and the lowest the memory will downclock to is 715MHz. My server PC is typically headless, but I connected it to a monitor to install the video drivers and make sure that everything was working. I'm using cgminer, but I haven't yet gotten it to run stably for more than a couple hours. The problem I'm having is that it'll crash, and the GPU will reset all of its settings, which includes placing the hardware acceleration setting near the low end of the slider. When this happens, it disables OpenCL, the TightVNC server stops working for some reason (stopping/restarting it has no effect), and RDP is useless as always in this regard. This means that I have to physically reconnect a terminal to the server in order to crank up the hardware acceleration back up to full and get everything working again.
I'm not yet sure why this is happening, so I want to double check some things. My only suspicion is the PCI-e x1 riser I had sitting around which I used to connect the video card to the PCI-e x16 slot in the motherboard, as there wasn't enough space to plug the card directly into the x16 slot. The 6950 has a TDP of 200W, and can get 150W through the two 6-pin PCI-e connectors. Can the remaining power be pulled through a PCI-e x1 connection? I know that the power pins are the same regardless of the bandwidth, but I also recall reading that for some reason PCI-e x1 connections are restricted to pulling 25W through the motherboard. Is that the case here? If so, I guess I should get a x16 riser instead. Thoughts?
Second, because I know this question will be asked, the power supply is a 500W OCZ. Other components in the system are a 95W CPU, the micro-ATX mobo, an optical drive which is almost always idle, 5x HDDss, 2x 80mm fans, and 1x 120mm fan. That's it. There are two 12V rails each capable of providing 18A, so I don't see any reason why the PSU shouldn't be capable of supplying ample power. The more I write about it, the more I think that it has to be a power restriction because of the riser...
The card will only allow me to overclock it as high as 840MHz, and the lowest the memory will downclock to is 715MHz. My server PC is typically headless, but I connected it to a monitor to install the video drivers and make sure that everything was working. I'm using cgminer, but I haven't yet gotten it to run stably for more than a couple hours. The problem I'm having is that it'll crash, and the GPU will reset all of its settings, which includes placing the hardware acceleration setting near the low end of the slider. When this happens, it disables OpenCL, the TightVNC server stops working for some reason (stopping/restarting it has no effect), and RDP is useless as always in this regard. This means that I have to physically reconnect a terminal to the server in order to crank up the hardware acceleration back up to full and get everything working again.
I'm not yet sure why this is happening, so I want to double check some things. My only suspicion is the PCI-e x1 riser I had sitting around which I used to connect the video card to the PCI-e x16 slot in the motherboard, as there wasn't enough space to plug the card directly into the x16 slot. The 6950 has a TDP of 200W, and can get 150W through the two 6-pin PCI-e connectors. Can the remaining power be pulled through a PCI-e x1 connection? I know that the power pins are the same regardless of the bandwidth, but I also recall reading that for some reason PCI-e x1 connections are restricted to pulling 25W through the motherboard. Is that the case here? If so, I guess I should get a x16 riser instead. Thoughts?
Second, because I know this question will be asked, the power supply is a 500W OCZ. Other components in the system are a 95W CPU, the micro-ATX mobo, an optical drive which is almost always idle, 5x HDDss, 2x 80mm fans, and 1x 120mm fan. That's it. There are two 12V rails each capable of providing 18A, so I don't see any reason why the PSU shouldn't be capable of supplying ample power. The more I write about it, the more I think that it has to be a power restriction because of the riser...
Jump to: