(4) 5970's on Win7 X64 Inconsistent

[Cobra]

After searching the forum the closest thing I came to was this issue with guiminer and very similar to what I am seeing, don't want to bring that old thread back to life. My hashrate is the same as what this person is getting, no confirmed fix. https://bitcointalk.org/index.php?topic=37911.0

Running Asrock Fatal1ty 990FX board, 8 core cpu, 16GB of memory, corsair gt ssd, 1300 watt Rosewill lightning power supply, and all powered pci-e extenders with 4 Diamond 5970's. Before anyone asks on the hardware this is my main desktop.....running in an open frame setup in the next room. Driver is 12.1, SDK 2.6. To solve this problem I have tried fresh Windows installs with 11.6, 11.8, SDK 2.5 and saw the same results. I really would like to keep the current driver as well since I game on this machine as well from time to time.

When running Cgminer 2.1.2 with -I 9 - k phatk worksize 256 vectors2 all 8 GPU's would only put out about 200 MH/sec each.  The strange thing is if I change to -I d on all 8 of them they would all put out 340 MH/sec. I have all of the cards overclocked to 830/300 which has been very stable, stock voltage.

If I simply remove 2 of the cards and run them with settings above at -I 9 all 4 GPU's will put out a stable 386 MH/sec and run that way for days with no crashing. To maximize the potential of these cards I moved 2 of them to another board and still powering them with the 1300watt power supply and they are doing great as well. It only seems to be a problem with more than 2 cards, had the same problem with 3 cards. Also tried 2 5970's and a 6950 and that didn't make a difference either.Temps are low,65C on all GPU with large fan pushing air to them.

I know most people that have 3 or 4 5970's are running Linux.....I am just not at that point yet with my comfort level. I would like to use this single board to its fullest potential since the hardware should support it. I have it powered with a 220v TrippLite PDU so I can actually look at power consumption via that, I have not had a chance to dig into that yet and configure but assume the 1300W is fine, single rail.

Any advice is greatly appreciated.

[1714779926]

1714779926

[ssateneth]

are you sure you are mining on the correct devices? also monitor gpu usage; are they 99% or hovering around 80-85?

[ummas]

Check affinity!
Set all those GPU to one CPU core.

Is it possible to run 4x5970 under windows ??!!

[Cobra]

Thanks for the feedback guys. I am sure it was the right devices. Windows 7 supports 8 GPU's Max. GPUZ reports 99% when running the 2 cards, or when all 4 were in there it was about half load of GPU at -I 9. I am baffled why -I d would give higher hashrate than -I 9 with all of them. 340 vs 200 when the 4 cards are in there, and 386 with 2 cards.

I didn't think about the CPU affinity but it is at 0-1% while mining so didn't think that would be an issue, I can give it a try.

I tried guiminer with 2 5970's and a 6950, the 5970's ran at full hash of 380's and then when turning on the 6950 driver crashed.  When I did that test I was on 11.8 and SDK 2.5. I'm going to try and see what my power consumption is at as well.

I have searched and could not find any post regarding 4 5970's on Win7.....sure on Linux.

[ummas]

1. use -I 8 insted 9
2. What are the temps of them ??
3. check affinity.
4. first GPU -I d , any other 8

[Starlightbreaker]

1300 watt Rosewill lightning power supply

ಠ_ಠ

[ummas]

http://www.jonnyguru.com/modules.php?name=NDReviews&op=Story3&reid=258

seems legit... o_0

[rjk]

1300 watt Rosewill lightning power supply

ಠ_ಠ

It's a good PSU, I have one - but I wouldn't want to run ALL FOUR 5970's on it. You are probably running out of power.

[Cobra]

the 4 cards running from 2 different boards seem to do fine with the single power supply. It still may be the issue though. I currently have only 2 cards in the machine and under load it is pulling 2amps on 220v.  I found a deal on a PDU TRIPP LITE PDUMNH30HV that I couldn't pass up.  Unfortunately I can not see how many watt an outlet is using, only amp of each of the 2 banks, I assumed it would show that but looks like a limitation of that device. I need to get a kil-a-watt.

I think I may just leave 2 cards in the main rig and call it good. From feedback here sounds like 4 cards is too much for a 1300W.  I'm going to play it safe and stick with this 2 card config which is a good balance.

It is solid with 2 cards at 386 MH/sec per core. Total average of 1523 MH/sec. I am using -I d,9,9,9 Vectors2 and worksize of 256 I tried 8 for intensity and it reduced hash by 2 so no cpu hit by leaving it at 9. Efficiency has been in mid 90's, .17 stale. With only 2 cards machine has been stable while under heavy load with no drop in hash rate while encoding video with all cores or running multiple VM's. The more I think of it, I seem to agree with the recommendation that 1300W is not enough for 4 especially with all the other stuff going on in this rig. If this was a dedicating mining rig it may be worth pursuing 4 cards on one board but I don't want to press my luck. I think I underestimated the power requirements thinking the 300mhz memory underclock would give enough headroom.

Can someone verify step for setting affinity with cgminer
In my kickoff cmd, I put /LOW /AFFINITY1 before cgminer.exe?  After that I have --config "path to config file" --quiet

I just assumed someone out there has ran 4 5970's under Windows 7 and was looking for an example config of such an environment.

Thank you all for your feedback

[rjk]

You could also try 3 cards instead of just 2.

[k9quaint]

It is probably too much draw on the PSU. Try 3 cards instead.

[Cobra]

It is probably too much draw on the PSU. Try 3 cards instead.

I have tried 3 cards.  Looks like no problems in Windows same results with 11.8 and 2.5 as 12.1 and 2.6. The 3 cards will work but with same config at -8 gets only 200 mh/sec for the 6 GPU's or 340 if I use -I d on all of them. All 4 cards did work with -I d at 340 but I know the core is capable of 386 because that is what worked with 2 cards.

When more than 4GPU's are in the same system even if I tell the miner to use only a single GPU or even a single core the hashrate will be lowered. I have tried Phoenix and Cgminer, both of them seem consistent.

When running 2 motherboards and all 4 gpus from same 1300W power supply the primary board that the PSU is powering with get the expect 386 hashrate on all 4 cores.  On the other motherboard with 2 cards and -I 9,9,9,9 (dedicated miner) those hash rates fluctuate from 330-340 with same overclock, GPUZ says all are running 99-100% on each GPU. temps 65c or under. Same driver/sdk.  The secondary board has a 550W Antec HE PSU that is only powering the board. If I was really running out of power wouldn't GPUZ of the second board show a reduced GPU load? No crashes on that, seems stable. The secondary board is an older Windows7 install and could have cruft slowing it down. I plan to format that one and start over.

[chungenhung]

how do you set up "vectors" and stuff like that on cgminer?
I usually use "VECTORS BFI_INT FASTLOOP=false AGGRESSION=13 WORKSIZE=256". But I have no idea how to put them when using cgminer.

[Cobra]

how do you set up "vectors" and stuff like that on cgminer?
I usually use "VECTORS BFI_INT FASTLOOP=false AGGRESSION=13 WORKSIZE=256". But I have no idea how to put them when using cgminer.

I went into settings first (one of the menu items) and told it to write conf file. I tweaked the conf file so it looks like this. The last line of the conf is where I specified vectors.

Code:

"intensity" : "d,9,9,9",
"gpu-engine" : "500-830,500-830,500-830,500-830",
"gpu-fan" : "0-85,0-85,0-85,0-85,0-85",
"gpu-memclock" : "300,300,300,300",
"temp-cutoff" : "95,95,95,95",
"temp-overheat" : "85,85,85,85",
"temp-target" : "70,70,70,70",

"algo" : "c",
"api-port" : "4028",
"expiry" : "120",
"gpu-threads" : "2",
"log" : "5",
"queue" : "3",
"retry-pause" : "5",
"scan-time" : "60",
"temp-hysteresis" : "3",
"worksize" : "256",
"kernel" : "phatk",
"vectors" : "2",

[TheHarbinger]

Well, you are for sure trying to pull more power than you have.  Even with down clocked memory, at 850 GPU you are trying to pull over 300W per card.  Add in a CPU, some RAM, a HDD or two a few things on the USB bus and you are well above the 1300W rating of your pwer supply.

Also, take a read through the cgminer thread.  Your configuration looks like you are only setting 4 GPU cores, not the full 8.  If I remember correctly, with dual GPU cards you have to set up the cores, not the card.  (4 5870s = 8 cores)

Try this;

"intensity" : "d,d,9,9,9,9,9,9",
"gpu-engine" : "500-830,500-830,500-830,500-830,500-830,500-830,500-830,500-830"


This may have changed in in different versions of cgminer though, check the thread.

[Cobra]

Sorry for the confusion. My current configuration is for 4 gpu across 2 motherboards and all 8 gpu'sare pulling 386 Mh/sec at 830/300.

All is working great stable for days with the 1300w psu powering 4 5970's. I just couldn't get it to work off same board but I am happy with my current configuration. I will report back when I get a kill a watt.

Thanks for the advice.