Windows 7 64 bit dual 5970 rig question

[Herodes]

OS: Win 7 64 bit, phoenix 1.4, SDK 2.4

I have some issues getting all the gpu's to work properly. And thus I have some questions:

I read in this thread: http://hashcat.net/forum/archive/index.php/thread-421.html

It says there that you need 4 monitors connected if you want to use all cores (4 in total) on 2x5970. Can anyone confirm this?

Also a possible solution was stated in that thread to have the cards connected with crossfire, and then just have one monitor
connected to one of the cards.

I wold think it was logical to have only one monitor connected to each card to make the mining work on all cores, do I really need to get
dummy connectors for all DVI-outputs?

I have two cards, both 5970, let us call them A (d0,d1) and B(d2,d3).

d3 never semmed to work properly with any miner that I tried. It worked for a short while then froze.

Please see this picture for the behaviour:
http://imageshack.us/photo/my-images/859/gpuactivity.png/

I tried to remove A and insert B only. then d2 and d3 both performed well with phoenix. Once I insert A, I have issues with d3 again. It just starts, runs for a few seconds, then freeze. There is one monitor attached to a cable from each card (I tought this was sufficient).

I'm stuped as to how this problem occurs, and I'd like to set up the source for phoenix to run some debugging with it to see what really happens here, but not sure which environment to use, and how exactly to do that, but if anyone gives me any pointers. I'd figure it out.

I am tempted to write off d3 as an hw issue, but won't give up just yet.

I'm interested in knowing which software versions and how cables and/or dummy plugs are connected to your cards if you run a dual 5970 setup under windows.

Also would I possibly have more success if I used Ubuntu? My next move will be to install Ubuntu to see if I can get all cores working under Ubuntu. I really don't care if I use windows or Ubuntu, all I want is to get this damn thing to work. Again, if anyone helps me to solve this issue, I'd gladly donate a sum that will make you feel it was worthwhile helping me. I've been looking into this on and off for weeks, but intensified the efforts the lasts days, so far with limited luck.

I was thinking of alternative solutions like restarting d3 mining every time it locked up (would not be optimal, but better than nothing), but when I restart the miner, nothing happens, a reboot is actually required.


I run one bat-file for each worker, the content of each file is as follows, with device ranging from 0 to 3.

Code:

cd c:\bitcoin\phoenix
phoenix -u http://mymail@hotmail.com_1:thepassphrase@server.com:8332 -k poclbm DEVICE=n -v VECTORS BFI_INT AGGRESSION=11

I I run poclbm from the command line with no parameteres, it too will freeze. Gpu-Z or GpuCaps will not run either. So I suspect there might be something wrong, and I am not sure if the hw or the sw is to blame. But for the record, d0,d1 and d2 runs just like a charm, it's only the 4th gpu, device 3 that's not working. That'll be the slave gpu on my 2nd 5970.


Thanks in advance for all the help you can provide.

[1713994098]

1713994098

[1713994098]

1713994098

[1713994098]

1713994098

[1713994098]

1713994098

[1713994098]

1713994098

[commlinx]

I only run a single card in my regular PC that doesn't have very good cooling for a GPU and I had a few problems initially with lock-ups that I solved by lowering the mem clock and 1000Mz and the GPU to 750MHz. It might be worth a try just in case you have a problem with marginal power or thermal issues that only crop up when all 4 GPUs are running full tilt. In my case the temperatures were within spec so it took me a while to think of trying it, so just thought I'd mention it as something quick to try that might be worth a go.

[gigabytecoin]

Yes, you need to plug in dummy connectors to all outlets when running a windows system. Don't ask me why.

All I know is that this is not the case when it comes to Ubuntu.

+1 for Linux!

[commlinx]

Yes, you need to plug in dummy connectors to all outlets when running a windows system. Don't ask me why.

I always thought from what I read that for users with that problem the GPU wouldn't be recognised full stop? This guy seems to have a problem where it just doesn't keep running long.

[Herodes]

First of all dishware, thanks for your answer.

You didn't mentioned at what clock speeds you running.

Tried both stock speed and up until 800Mhz, no difference in how the 'bad' gpu behaves.

You said win 7 64, sdk 2.4. have you installed all the things useless to mining coming with the driver?
11.4?

I mostly made express installs, and if there's 'too much' stuff installed, please enlighten me as to how that would negatively affect my setup.  I run 11.1, not 11.4. I tried 11.3 which is the newest from here: http://support.amd.com/us/gpudownload/windows/previous/Pages/radeonaiw_vista64.aspx

The win 7 64 bit selection leads to that page, I think the name (vista) is just misleading.

11.3 did not work well, it made the system crash, believe it or not. So I went for 11.1 which seem to work well.

It has been told many times, by many members that it is best to go for less than 11, thats 10.10 or 10.XX with ati stream sdk 2.1 for HD 5000 series cards.

Yes, I've read that. But does it also apply to win 7? Anything less than 11.1 on the ati drivers page has no mention of windows 7, only vista, but perhaps the vista and win 7 drivers are interchangeable? Besides, the rig works with the current setup, also with the 'bad' gpu if the card with the 'bad' gpu is left alone in the box.

If over clocked beyond limit with out looking out temp, also has the possibility that one gpu in that card blown out.

I hope that's not the case. :\ Fan is running at 100% with temps 70-73 celcius on gpu 2 and gpu 3, and the 'bad' gpu and the other gpu on that card is running at aprox 50 celcius.

Better run only one card & see which card's gpu giving problem. Isolation of problem is first step in trouble shooting.

Already did that. If a single card runs alone. Then it works fine. The trouble is just the one 'gpu' when both cards are running.

You can see that by device 0 & 1, after finding the gpu, mark the card, then uninstall all the AMD drivers &
then run latest version of "driver sweeper" from www.phyxion.net & analyse & clean AMD-Display drivers.
Then install 10.xx & 2.1 & check again to confirm THE gpu is blown or not.

I  might just try that, altough I have this feeling it will not work. Not because it would not work in theory, but because I think there might be some hw issue at play.

Always try to run fan at 100% speed, if u can able to with stand sound.
Coz, cost of fan is very less than the cost of gpu's.

Exactly. Thanks for your most helpful reply!

[Herodes]

Yes, you need to plug in dummy connectors to all outlets when running a windows system. Don't ask me why.

I always thought from what I read that for users with that problem the GPU wouldn't be recognised full stop? This guy seems to have a problem where it just doesn't keep running long.

Exactly. As you can see from the imageshak graph. The 'bad' gpu runs well on Phoenix for a very short while. It is hashing. Work is gotten from server, and even submitted. Then suddenly, just bom, GPU percentage use goes to 0 and phoenix miner freezes.

So honestly I do not know if it would have made a  diff if I had dummy plugs in all outlets on the back of the card.

[Herodes]

I only run a single card in my regular PC that doesn't have very good cooling for a GPU and I had a few problems initially with lock-ups that I solved by lowering the mem clock and 1000Mz and the GPU to 750MHz. It might be worth a try just in case you have a problem with marginal power or thermal issues that only crop up when all 4 GPUs are running full tilt. In my case the temperatures were within spec so it took me a while to think of trying it, so just thought I'd mention it as something quick to try that might be worth a go.

I run it at 800Mhz core clock and 925MHz Memory clock. And temps are good. Do you suggest any changes to these temps? For the 3 working gpu's this setup have been running stable for days at a time.

[dishwara]

It seems, you getting problem only if u run all the gpus, here 2*5970, thats 4 gpu.
I suspect the PSU, power supply not giving enough power.
For to run the above spec, u at least need, 300 for each card with out OC.
So in total u must have at least 300+300+250= 850 Watts without OC.
I recommend go for at least 2 * 750 W (Recommended by AMD, not cheap ones), which give 1500W, or 2*650 W, if money is problem, which give 1300 W, which enough to give power to all cards & cpu also. & also OC.

I mostly made express installs, and if there's 'too much' stuff installed, please enlighten me as to how that would negatively affect my setup.  I run 11.1, not 11.4. I tried 11.3 which is the newest from here: http://support.amd.com/us/gpudownload/windows/previous/Pages/radeonaiw_vista64.aspx

Instead of express install, go for custom & only install these 3 or 4.
many times VC++ grayed out.
ATI display driver, ATI catalyst install manager, APP 2.3 or SDK 2.1 & VC++

Vista drivers will work for win 7.
For 5000 series card, you must at least try 2.1 & 10.12 or less.

[Herodes]

Thank for you suggestions dishwara. I will get back to this thread once and if I get things sorted out.

Currently running with the following PSU: CorsairHX-850 watt

According to this thread, 850 watt should be enough:
http://www.tomshardware.co.uk/forum/286977-10-which-dual-5970

But I am no expert in this subject.

I will try all the things suggested. Thanks.

[dishwara]

Thank for you suggestions dishwara. I will get back to this thread once and if I get things sorted out.
Currently running with the following PSU: CorsairHX-850 watt
According to this thread, 850 watt should be enough:
http://www.tomshardware.co.uk/forum/286977-10-which-dual-5970

But I am no expert in this subject.
I will try all the things suggested. Thanks.

Corsair HX-850, is not in the list of certified PSU for not only 5970, but totally to 5000 series.
http://support.amd.com/us/certified/power-supplies/Pages/listing.aspx?p=Corsair
http://support.amd.com/us/certified/power-supplies/Pages/details.aspx?item=498

You can email to Corsair itself & ask them.

[Herodes]

Ah, ok, thanks! I will see what I can make out of this!

[Herodes]

I currently run my dual 5970 rig with the following PSU:

PSU: CorsairHX-850 watt


As written earlier in this thread, one GPU will not work at all, it starts working for a few seconds, and then it just dies off. Apparently the card works fine (both gpu's) when it is run alone in the rig (other 5970 taken out). I wonder if my PSU is insufficient for my dual 5970 setup?

The motherboard is a Asus rampage III extreme socket-1366.

What are your opinions about putting in a stronger PSU with more watts, would it possibly solve the problem with the GPU that does not perform?

[Herodes]

I attempted to fiddle around with Catalyst Control Center, as I've read that I might have some kind of crossfire issue since I can't use that 2nd gpu on the 2nd card of mine. I disabled "Enable Surface Format Optimization" from the Catalys A.I. box under Gaming -> 3D application Settings, and then both gpu's on my second card stopped working for mining.

But i reversed this change and the first gpu on card 2 worked for mining again.

So could it simply be a software issues? Some settings I have to make?

[bobR]

are you connecting a monitor to each card ?
If not you may need a dummy vga plug in the ones without monitor

[Herodes]

are you connecting a monitor to each card ?
If not you may need a dummy vga plug in the ones without monitor

Yes, I have a monitor connected to each card. Also tried to connect two monitors to the card with the GPU that does not work well. Same issue stil. Thanks for your reply!

[Nicksasa]

The PSU is enough, i have the same one. If you guys would do the trouble on googling an review you would know that you can pull 1KW from this psu and still keep efficiency around 81%.
Why the f would you use an amd recomended psu list ? Corsais uses the best oems availble and are pretty much the best psu's to buy, that's why they cost a bit more.


To the op, open catalyst and search for a "powertune" slider, not sure if 5 series cards have that option but anyways slide that to the max. If it doesn't help then i suspect the vrm's are overheating.

[Herodes]

I will see if I can find the powertune slider. Now I have the 4th gpu working somewhat.

Installed latest drivers 11.5 and latest sdk 2.4, and the 4th gpu is now running at 11Mhash/s, still about 300Mhash/s less than it is supposed to do, but from a "dead" cpu to this it is progress. Also in the progress of getting ubuntu up and running and see if that will take care of things.

Thanks for all your input. And I've googled quite a lot. Just is not an expert on overclocking and hw in general.

[allinvain]

I don't think you need a monitor or a dummy cable for each card. If you use quad crossfire it should work just fine. The only time where you need a dummy plug is when you are NOT using crossfire/SLI. The reason being is that when the OS detects that no video load is on the card's connectors it shuts off the card. A dummy plug tricks the card into thinking it's got a monitor attached.

[Herodes]

I have a monitor connected to each card at the moment. GPU1, 2 and 3 working fine. But GPU 4 just working at 11Mhash/s. I can't try the crossfire cable right now, as this is a remote machine. Then I will have my partner attempt to put on crossfire cable later on. thanks for the input.

[allinvain]

I have a monitor connected to each card at the moment. GPU1, 2 and 3 working fine. But GPU 4 just working at 11Mhash/s. I can't try the crossfire cable right now, as this is a remote machine. Then I will have my partner attempt to put on crossfire cable later on. thanks for the input.

Ok now that is odd. GPU 4 it seems is not really a GPU but instead the miner is using your CPU!!

Do you see 4 Cayman cores in guiminer (try guiminer as a test)?