Topic: WTF is going on with my GPUs (R9 290X and 6950) ? (Read 2478 times)

I second this. I have had a 6950 for a long time (used occasionally for mining). With some heavy tweaking I was able to squeeze about 420 Kh/s out of it. I ordered an R9 280x which I installed a couple weeks ago and they did not place nice together. With enough work I probably could have gotten them to work, however I just pulled the 6950, stuck it in my roommates computer and mine with it when hes at work.

I offer no solutions, but i always warn against mixing cards that like different --gpu-thread sizes.  The 280x / 7970s are rare in that they like -g 2, and for that reason mixing them with any other cards (just about all of them) that prefer -g 1 is asking for it.  In other words, OP, you are likely doing nothing wrong.

Got another machine you could throw that 6950 into?

thread concurrency should be set to 8000 on the 6950.
EDIT: And intensity to 18.

 well i've installed my r-9 280x, was abt 350 khps out of the box but a little tweaking and it's running good, abt ~670 khps  

They are PCIe 2.0 ...

I've just lowered the worksize from 256 to 64 on the 6950, and it gave me a 40 kH/s boost ... that's only 285 kH/s, but it's a start since nothing else works.

  this is just a thought, as idk what diff it will make, and i will soon find out as i should be seeing a new r9 290x in the mail in a few days ...
the pciex16 slots on your mobo - are they v.3.0?

 mine are v.2,  .....

My ASUS R9 280x's are kicking out about 760KH/s on cgminer 7.3.2 with latest beta drivers and sdk.
Here is my R9 280x config, hope it helps:

{
"pools" : [
        {
                "url" : "http://192.168.0.201:1524",
                "user" : "x",
                "pass" : "x"
        },
        {
                "url" :  "stratum+tcp://stratum2.wemineltc.com:3333",
                "user" : "x",
                "pass" : "x"
        },
        {
                "url" :  "stratum+tcp://eubackup.wemineltc.com:3333",
                "user" : "x",
                "pass" : "x"
        }
],
"failover-only" : true,
"intensity" : "13",
"vectors" : "1",
"worksize" : "256",
"kernel" : "scrypt",
"lookup-gap" : "2",
"thread-concurrency" : "11200",
"shaders" : "0",
"api-listen" : true,
"api-network" : true,
"api-port" : "4028",
"expiry" : "120",
"gpu-engine" : "1090",
"gpu-memclock" : "1500",
"gpu-vddc" : "1.08",
"gpu-dyninterval" : "7",
"gpu-platform" : "0",
"gpu-threads" : "2",
"log" : "5",
"no-pool-disable" : true,
"per-device-stats" : true,
"queue" : "1",
"scan-time" : "30",
"scrypt" : true,
"worktime" : true,
"shares" : "0",
"kernel-path" : "/usr/local/bin"
}

I tried older versions of cgminer, it's worse and it crashes instead of starting mining ..

I already tried to play with the worksize on the R9 (256 and 128), but I didn't see any improvement or loss in hashrate

I'm lost ... could it be driver issue ? (although I'm stuck with 13.11 or 13.12 on the R9 because it's not supported by earlier drivers)

Are your machines plugged directly into an outlet or are they plugged into a power strip or something like that?

I had this exact same problem with my 6950. I had my computer plugged into a power strip with a bunch of other things; like monitors, game consoles, tv, speakers, etc. Plugged into the strip, I was getting the exact same results as you. But when I plugged the computer into an outlet all on its own I managed to get it to 420 kh/s with an intensity of 18 and thread concurrency of 6010.

I think it'll all come down to a lot of tweaking for you.
I've spent the majority of my day just tweaking the crap out of the config file in order to push my 280x cards to some higher hash rates.
You'd be amazed how much something like the right intensity makes a difference.

Now, tweaked it the way I think is good enough, getting a stable 760 khash, I've got the intensity on 13.
But when changing the intensity to 14, so higher, the hashrate drops about 200 khash... Also lowering the intensity will drop it.

So tweaking it right is the key for you I guess, don't try to push your card too much.

Do you have CCC overdrive turned on or off and how much memory do you have in each system?

I forgot to mention it, but these environment variables are already set on both machines.

Either set the system environment variables for
   GPU_MAX_ALLOC_PERCENT    100
   GPU_USE_SYNC_OBJECTS      1

reboot then run cgminer or create a batch file to set them and then run cgminer. Some motherboard BIOS requires the variable to be there other do not.

Hi everyone.

I've finally decided to post here, because I've been fighting for days with my GPUs trying to tune them to get decent hashrates (or at least something that looks like what I could find from various sources like https://litecoin.info/Mining_hardware_comparison or https://bitcointalksearch.org/topic/consolidated-litecoin-mining-guide-for-5xxx-6xxx-and-7xxx-gpus-117221 ) and miserably failed. So I need some advices from more experimented miners ...

I have two machines, both are running quite similar software setup : Win7 64 bits, Catalyst 13.12 drivers and cgminer 3.7.2.

First system has a HD6950 with reactivated shaders (so 1536 SP like in a 6970) clocked at 840/1250. This is the system I have experimented the most with settings and my current cgminer command looks like this : --intensity 12 -g 1 --thread-concurrency 4032
This setup shows about 240 kH/s according to cgminer.

The problem is that I can't get any close to the recommended settings of 4*SP count for thread concurrency (6144) or to what I can find on the page linked above (between 6144 and 8192). If I set anything higher than 4032, cgminer will refuse to start with a -4 error.

I have the same problem with intensity setting : at intensity 12, I get a few HW errors (I get none at intenisty 10), but I won't be able to set anything like those found in examples (instensities up to 18 or 19).

What am I doing wrong ?

Are people reporting higher hashrates (in the 450's) with aggressive setting only lying about their parameters, and are only doing so to report big numbers, but with nothing productive (tons of HW errors and no valid work accepted) ?

The same questions and situation apply to my other machine with a R9 290X (allowing +50% of power draw and fan up to 100%). I'm only able to use these settings : --intensity 14 -g 1 --thread-concurrency 16384 which produce 635 kH/s ...
This is quite far from the numbers I found (intensity 18-20, thread concurrency between 24000 and 32768 which should give hashreate in the 800's).

Any thoughts or advice are welcome.