Bitcoin Forum

I get 900~ khs and 900~ wu for a solid 10 minutes (average) then i get a sharp decline (drop to 750 wu and khs always stays around 865khs), then it get slow decline by about 1 average WU per minute and then hash rate and then drivers crash, cgwatcher eventually restart the miner after that. For a while, i get alot of artefacts and black lines. Restarting or leave the gpu alone for a while clear that up. I dont get artefacts when i game at 1250 memclock nor 1300.

According to the logs, the process happens faster and faster until drivers insta crash like so :

[2014-03-12 12:02:32]        Total hashrate is below Scrypt cutoff 700 Kh/s (0 Kh/s); but am inside the miner's startup grace period (Elapsed:36 sec;GracePeriod:3 min) so did not attempt to restart yet, but I will if the condition exists after the grace period expires.
[2014-03-12 12:04:02]        SGMiner process closed outside of CGWatcher!
[2014-03-12 12:04:02]        The miner is closed outside of CGWatcher and you've requested for me to ensure it stays running. Attempting to start...
[2014-03-12 12:04:16]        SGMiner started successfully.
[2014-03-12 12:04:36]        Total hashrate is below Scrypt cutoff 700 Kh/s (0 Kh/s); but am inside the miner's startup grace period (Elapsed:26 sec;GracePeriod:3 min) so did not attempt to restart yet, but I will if the condition exists after the grace period expires.
[2014-03-12 12:06:16]        SGMiner process closed outside of CGWatcher!
[2014-03-12 12:06:16]        The miner is closed outside of CGWatcher and you've requested for me to ensure it stays running. Attempting to start...

I find it odd that it says the process was closed outside CGWatcher. But whatever, there it is.

So since i have CGwatcher now restart my miner after hashing drop too much. (As you can see at 700, but on the previous logs, by the time the grace period is over, its already down to 500s khs.

So in a nutshell. First critical drop occured after 40 minutes, then after 20, then 10 (its honestly that clean), then 10, 10, 10, 10, 20 (it did go back up), then it goes bad, 3 (was at 690 khs and 3 is minimum grace period), then 3 again (khs was at 300), then insta crash and restart every 3 minutes for about half an hour, then i wake up and stop this madness.

Here's my dxdiag

http://pastebin.com/XM5HW22H

This is my config:

"api-allow" : "W:127.0.0.1",
"api-listen" : true,
"failover-only" : true,
"gpu-threads" : "1",
"gpu-memclock" : "1250",
"gpu-powertune" : "20",
"temp-target" : "85",
"temp-overheat" : "90",
"temp-cutoff" : "95",
"vectors" : "4", (that and -w together is the only way to get khs and wu to a reasonable amount)
"lookup-gap" : "2",
"thread-concurrency" : "25601"

And these are the only 2 .bat arguments used --xintensity 512 -w 128 (its the -w 128 that yield high khs + high WU/kh ratio).

I use SGminer 4.1.0

I already spent days trying different things and settings and next step is changing a setting and not touch the computer for an hour to see if its more stable but i'd like to be stopped right here if there is an obvious issue, as there are prolly weeks worth of combinations to try still.

First of all your WU at the first few minutes (or even hour) isn't indicative of anything. It usually does start high, especially after a system reboot but will average out after a certain period of time (say a day) giving you its real value.

Second, from cgminer readme file, the vectors variable does nothing when it comes to scrypt mining.

Third, since you can't mine for more than a day (from my understanding), you can't possibly know what increases or decreases your WU (because it's an average value).

Now, you haven't specified your core clock speed nor if you have undervolted (if you have than restore volts to stock settings), so I'll type the settings I'm using for you to try since you got Sapphire Tri-X and those, if memory serves, come with HynixAFR memory type (thus our clocking speeds should yield same results).

Try: gpu-core 1015, mem-clock 1250, tc 20481, xint 400.

P.S. How did you come up with that TC value of yours? Usually what works best for most people is TC value equal to shaders*n+1 or in the case of R9 290: 2560*n+1.

Anyway, artifacting and driver crashes are usually caused by misconfiguration of your mem/core clock in relation to the voltage and high temperatures (I doubt the latter since you got tri-x but it can't hurt to check the VRAM temperature).

Thank you very much for the reply.

Yes you have good memory, it is Hynix. Reference is 1250 memclock, stock on tri x is 1300. I have issues when i so much sneeze as any memclock or power settings (like memclock and gputune).

I have a 130 khs gain/loss with 1250memclock and 1300 memclock. Artifacts only happen when i change any of those and last until i run a 3d application ( i have no idea why ) then if i so much sneeze at it or run mining with modified clocks/settings they start again. However, the flicker is not my concern. It is merely a symptom. The cause is the sharp decline in khash that lead to a driver crash, which happen after WU drops by from 900 to 600. As soon as it get there, the khash start plummeting and then crash.

For TC, i did shader x n + 1 i tried 1-14 for n, both extreme = doesnt run, lesser extreme = hardware errors, shader x 10 +1 yielded best results.

I use Sgminer, not Cgminer so i dont know if its different. When i use vector 4, i get the very high WU for 10 minutes and a 30mhs bump, vector 2 work bad, i get a big (100-150) drop in khash and vector 1 seem average. Also it is a scrypt-only setting in cgwatcher.

I didnt specify my settings further because volt is locked, stock bios doesnt let me ajust it. I heard somewhere its 1.225v or something like that. And i have (edit:) NOT touched core, which is at stock 1ghz, changing it to reference didnt fix the issue but i will try again (i have barely touched clock settings, except i been using gpu tune 20).

Lastly, i might have said something unclear. I can't mine for more than 40-60minute before issues start to arise. It takes about 3 hours before it become critical and i need to intervene, however performance after 1h30 is abysmal.

Now i just set -w to 512 and vector to 1(i believe thats what is autodetected), it yield much lower khash (low 700s) with 500-600 wu and it just crashed after 20 minutes at stock (not reference) settings.

Now i will run your settings.

 

There's no difference in that regard, vectors does nothing (I am also running sgminer 4.1.0 and don't even have it set up).

You should be able to change the voltage using MSI Afterburner or Trixx (Sapphires software). If you're going to use Afterburner than you need to check 'allow voltage control'.
But since you haven't touched your voltage than it can't be the issue.

-w 512 doesn't matter to R9 290 because the maximum size you can set for it is 256. You can confirm it by seeing the binary file that is being created and see that part of its name will contain w256 and not w512.

Anyway, did you check the temperature of the VRAM?

I keep forgetting to bring up temp ; its under 70c at all time with fan at 50-60%. Running borderlands 2 maxed net me 41c and a solid 60 fps with minimal gpu usage. No tearing. These looks pristine to me. I dont think the temp sensor is busted because it behave normally when i lower fan. The temp then sky rocket and i crash there and then.

I can re-say that volt is locked, thats what it says on the bios info, it use a different setting control. To unlock it, i'd need to install Stilt's bios but i get BSOD after i flash it and i havent gotten any reply to my post since yesterday. To compare, i can change volt just fine on my 2 other cards. It bothers me because i would like to undervolt to see if it solve my stabilities issues.

Vector is odd because i clearly get +60 mhs and +300 Initial WU and big boost for 10 minutes consistently everytime ( and i tried at least a dozen time) vs vector 1 or not vector set. My pool confirm that by paying me 200 extra mh for that 10min period. Regardless.

Then i guess setting 512 actually made me set it up from 128 to 256.

So anyways, i put your settings, effectively changed -w to 256(512), removed vector.

Two things happened:
-No artifacts from this OC/UC
-Crashed after 2 minutes, i will run it again after i check something

Khash was at upper 700s so fairly low. I'm getting a bit worried that the chain crash episode of this morning caused perm damage. I never, never crashed so fast so i will rerun it a few times.

I hope we find a fix or something change because i can't RMA it for 3 weeks and i fiddled with BIOSes so i dont know if they can access a log to see that i attempted to update the vBIOS.