Sunday, December 22, 2013

Radeon R9 290/290X Litecoin Mining: New Settings and Revised Expectations

Over the past several weeks I've had a flood of comments and emails about R9 290/290X hashing performance when mining Litecoins (or any other scrypt-based cryptocurrency). Initial experience with the cards suggests that the Radeon R9 290 can hit up to 900KHash and the Radeon R9 290X can hit 1000KHash, but the key part of that is "up to". In practice, the cards vary, sometimes wildly.

There are a lot of 290X cards that seem to top out in the 800-850KHash range, and short of adding better cooling I'm not sure how much farther you can push them. R9 290 tends to be more like 750-800KHash on a lot of cards. I think if we could get voltage adjustments working, we might be able to improve on those results for a lot of cards, but we probably won't have that for a few more weeks if not months.

One of the keys to understanding what's required for tuning performance on the new Hawaii R9 290/290X cards is that AMD has built in safety protocols to keep the cards running "safely". A lot of people have tried my original settings and come up with far lower performance than expected. This is why I provided batch files to help you tune and tweak performance.

Scrypt hashing by design is far more memory intensive than SHA256 hashing. The idea was to make it impossible or at least impractical to do scrypt hashing on GPUs, let along FPGAs and ASICs. So far, no one has yet released any FPGA or ASIC hardware for scrypt devices, and I suspect we're probably still at least a year away from such hardware becoming publicly available. (There may be early hardware undergoing "testing" for month in China or elsewhere first, of course.) So when you're looking to fine-tune your GPUs, start with the memory clocks and push those as far as you can while maintaining stability -- and maybe add a few extra cooling fans to help out! If you want to hit extremely high hash rates, you'll want the memory clock at 1600MHz or higher, and most cards won't do that without some sort of additional cooling or other modification.

Once you've found the maximum reliable memory clocks for your GPUs, only then should you start tuning the GPU clock speed. What's interesting is that even though most cards will happily accept GPU engine clocks of 1000MHz or higher, in reality they'll often run at far lower clocks -- this is AMD's safety protocols kicking in. So if you start off at 950MHz core and 1350MHz RAM and a thread concurrency of 32000 (give or take), and you find your hash rates are closer to 600 or 700KHash rather than 800-900KHash, don't despair!

Start with a lower GPU clock of around 800MHz (on R9 290/290X) and then start bumping it up 25MHz after about 10 minutes or more testing at each clock speed. You'll most likely find that 850MHz ends up being the sweet spot, and then you can play with clocks a bit more bumping it up/down in 5MHz increments to see if you can squeeze a bit better performance out of your system. When all of that is done, only then should you bother with tweaking your thread concurrency.

And with that said, here are my revised settings for R9 290/290X graphics cards. I'm going to include cgminer settings for four cards as an example this time, just to illustrate what you would use. I've also included three pools, the first two being multi-coin pools, since that's what I use these days. I'm also going to try toying around with BFGminer to see if I can coax anything better out of it. For the time being, these settings are more or less "stable".

cgminer.conf settings for ~810 KHash R9 290X / ~770 KHash R9 290
{
"pools" : [
{
"url" : "http://stratum01.hashco.ws:8888",
"user" : "trogdorjw73.tester",
"pass" : "tester"
},
{
"url" : "stratum+tcp://middlecoin.com:3333",
"user" : "13saHMFfUAv1crYp3QLbFr9aP1JYnb89oh",
"pass" : "x"
},
{
"url" : "http://coinotron.com:3334",
"user" : "trogdorjw73.tester",
"pass" : "tester"
}
]
,
"intensity" : "20,20,20,20",
"worksize" : "256",
"kernel" : "scrypt",
"lookup-gap" : "2",
"thread-concurrency" : "24272,24272,24272,24272",
"gpu-engine" : "850-850,850-850,850-850,850-850",
"gpu-fan" : "45-85,45-85,45-85,45-85",
"gpu-memclock" : "1475,1475,1475,1475",
"gpu-memdiff" : "0,0,0,0",
"gpu-powertune" : "50,50,50,50",
"gpu-vddc" : "1.100,1.100,1.100,1.100",
"temp-cutoff" : "100,100,100,100",
"temp-overheat" : "95,95,95,95",
"temp-target" : "85,85,85,85",
"api-mcast-port" : "4028",
"api-port" : "4028",
"auto-fan" : true,
"expiry" : "120",
"failover-only" : true,
"gpu-dyninterval" : "7",
"gpu-platform" : "1",
"gpu-threads" : "1",
"hotplug" : "5",
"log" : "5",
"no-pool-disable" : true,
"queue" : "1",
"scan-time" : "60",
"scrypt" : true,
"temp-hysteresis" : "3",
"shares" : "0",
"kernel-path" : "/usr/local/bin"
}
And just in case this isn't clear, with the above settings you would launch cgminer.exe with the following command:
cgminer.exe --auto-fan --config cgminer.conf --failover-only

56 comments:

  1. Thanks again Jared...and great timing as I sat down to research some new configurations... For the second morning in a row I've woken up to the black screen of death...which is worse because it's just sitting idle. Not sure what is going on. I too am running 4 290's and will be testing out your ideas today. Why did you go to the model of stating each value individually for each card? And, I see you've gone back to auto-fan...thoughts?

    ReplyDelete
    Replies
    1. Auto-fan causes BSOD, but it's generally more reliable in my experience than letting AMD's drivers determine the fan speed. The four GPUs were listed explicitly because you could tune them this way, so GPU 1 might do better at a higher core clock, CPU2 might take higher RAM speeds, etc.

      Delete
  2. Guys, please check for HW errors. I can get hash rates above 800 kh without a problem, but my card obtains no valid shares anymore. Switching debug output on reveals the stellar increase in HW errors.
    What seems to be the key is the intensity. Above 13 my cards screws its shares.

    ReplyDelete
    Replies
    1. Is that with an R9 290/290X or a different GPU? What OS and what settings? Usually HW errors mean you need to update the OS, perhaps reinstall the drivers, and also install the chipset drivers. Maybe some other stuff as well? Anyway, I'm getting no HW errors with the above on a 4x290X system.

      Delete
    2. Weird, I'm running on RHEL, latest release, latest drivers, running headless ... good to know it is no general issue, have to dig into it a bit further ... first I'll try BAMT ... thx for your answer

      Delete
    3. Now I feel a bit dumb ... my HW errors were strictly related to my thread concurrency being too low ... I was using the --shaders options. With an increased thread concurrency my HW errors are completely gone ... thx mate

      Delete
    4. You can try changing thread-concurrency to higher values. I'm also gettin a lot of HW errors with ~24000 but 33792 works fine for me.

      Delete
  3. I'm waiting for the Tri-X versions to come out and then I'll be jumping into this head first.

    ReplyDelete
    Replies
    1. I actually thought about doing the exact same thing! But for what it's worth you might be better off just doing water cooling if you're seriously into more cooling for mining rather than waiting... The long you wait the higher the difficulty the less the payout. So by jumping in early you may be able to more quickly pay off any additional cooling method you use... Even if you only get 850KH/s, with current difficulty/price on LTC you would get ~2 LTC / week or about $34 (per card)

      I'm running 5x Sapphire 290's with just the reference cooler, temp-target @ 87*C, and a 20" box fan (<$20 @ wal-mart)... Temps stay where I want them, and I'm averaging about 907KH/s each (4.533MH/s) with rejects only ~2% or so. 4 of the 5 run about 917KH/s, 1 of them about 870KH/s (may need to do more fine tuning on this one). I left mine going all last weekend solid >48 hours without touching it no problem! I did stop it this morning to adjust some clocks down to get my rejects down lower, over the weekend they were ~4%...

      Settings I'm running now:

      "thread-concurrency" : "32765", (note you need >4GB system RAM for this, at least I did..I'm running 8GB)
      "worksize" : "512",
      "gpu-engine" : "1039,1035,1015,1035,1039",
      "gpu-memclock" : "1473,1470,1345,1470,1473",
      "lookup-gap" : "2",
      "gpu-powertune" : "20",
      "gpu-threads" : "1",
      "shaders" : "2560",
      "intensity" : "20",
      "auto-fan" : true,
      "gpu-fan" : "67-90",
      "temp-cutoff" : "96",
      "temp-overheat" : "91",
      "temp-target" : "87",
      "temp-hysteresis" : "3"

      Donations welcome! :D
      LTC: LWHgLSxi3iLHGQ4ESME6Tj4STcpX5p3yTD
      BTC: 1DeAqKqi2QGw1bQNJQQu1W8KqWxUi2EkNi

      Delete
    2. Can You say which brand are each of this cards? I can't squeeze more than 878khash on my 2x Asus R9 290 Elpidia memory. Also do You have elpidia or hynix memory on them? I will try Your settings when I get back home tomorrow, however Your previous one's (from couple of day before were giving me very bad results.

      I run on just stock coolers and they are usually ~62% with 85-87C.
      My settings are, 1000/1500 clocks set by Msi AB and -25mv undervolted.
      "auto-fan" : true,
      "gpu-threads" : "1",
      "gpu-fan" : "40-90",
      "gpu-powertune" : "20",
      "intensity" : "20",
      "temp-target" : "88",
      "temp-overheat" : "93",
      "temp-cutoff" : "95",
      "temp-hysteresis" : "3",
      "vectors" : "1",
      "worksize" : "256",
      "lookup-gap" : "2",
      "thread-concurrency" : "32765"

      Delete
    3. All my cards (6 now) are Sapphire... The ones with free BF4. I think most (based on performance, and memory clocked up to 1500MHz), if not all, of my cards have the hynix memory. I've only checked the initial one I got but it definitely came back and said hynix, haven't checked any of the others though... I do have 2 cards that won't do more than ~880Kh/s, but I think those may be the ones hooked up to the monitor (2 rigs, 3 cards each). I've also tweaked my settings some more, can get up to 925+Kh/s out of some of the cards now running 1050/1500! :)

      Delete
  4. Hi Jarred!

    I currently have 3x R9280X-DC2T-3GD5 mining around 740kh/s. Do you have any configuration on those?

    Thanks!

    ReplyDelete
    Replies
    1. R9 280X is basically a newer 7970, so 740KHash is actually really good and probably can't go much higher without issues. (I assume you're meaning 740K per card, right?)

      Delete
    2. I was wondering if it can be pushed up a bit more. And yes I'm getting 740k/card.

      Delete
    3. I haven't personally used an R9 280X, so I don't know how much farther it might go, but considering many R9 290 users aren't getting above 800KHash I'd say 740 from a 280X is doing quite well!

      Delete
  5. Hi Jarred, using your first post as a template I'm able to get stable 890/900 kh/s with my 2 Sapphire 100362BF4SR r9 290s (ignore the third card, thats my 270 going at 400 kh/s) I also have a vortex personal fan going over them to keep things cool but I hit about 88-90c max on the 290s.

    0 HW errors on them too.

    "intensity" : "20,20,17",
    "vectors" : "1,1,1",
    "worksize" : "256,256,256",
    "kernel" : "scrypt,scrypt,scrypt",
    "lookup-gap" : "2,2,2",
    "thread-concurrency" : "32765,32765,12000",
    "gpu-engine" : "1025,1025,1000",
    "gpu-fan" : "45-100,50-100,50-100",
    "gpu-memclock" : "1475,1475,1350",
    "gpu-memdiff" : "0,0,0",
    "gpu-powertune" : "20,20,20",
    "gpu-vddc" : "1.110,1.110,1.110",
    "temp-cutoff" : "99,99,99",
    "temp-overheat" : "95,95,90",
    "temp-target" : "90,90,80",
    "api-listen" : true,
    "api-mcast-port" : "4028",
    "no-submit-stale" : true,
    "api-port" : "4028",
    "auto-fan" : true,
    "expiry" : "120",
    "failover-only" : true,
    "gpu-dyninterval" : "7",
    "gpu-platform" : "0",
    "gpu-threads" : "1",
    "hotplug" : "5",
    "log" : "5",
    "no-pool-disable" : true,
    "queue" : "1",
    "scan-time" : "60",
    "scrypt" : true,
    "temp-hysteresis" : "3",
    "shares" : "0",
    "kernel-path" : "/usr/local/bin",
    "api-allow" : "W:127.0.0.1"
    }

    ReplyDelete
    Replies
    1. Are you running Win7, Win8, or Linux? I'm starting to wonder if Win8 might be doing better, but it's probably just cooling of the VRMs that helps with clocks and stability, yielding better performance. Need to investigate this tomorrow! :-)

      Delete
    2. Thought I replied last night :) but guess I forgot too! I'm running Win 7, was going to try win8.1 and linux but still was learning about tweaking the cards themselves.

      Delete
    3. Mate are you OCing the engine too much?
      I'm afraid of OC the engine and memory too much so no i'm hashing at 800-850 at i-20, TC 24450 and still trying to find the best sweetspot for average 820+.
      I used to OC the memory to 1000 and screen got flashed. Let me know what you think? Also are you running the rig 24/7 at that settings?
      I'm running a 6x290 rig so couldn't risk too much.

      Delete
    4. Heya, I've been thinking about this and yes 24/7 for few days. I've been considering backing down the OC if only because I'm worried about the fans running at high RPM 24/7 and potentially burning them out. Right now I feel like I need to redo the OS because ATI Catalyst seems to ever 4-5 hours. I've even brought it down to 820 and still ATI Catalyst dies. Might be a good time to try linux or Win 8.1

      Delete
    5. So reinstalled windows 8.1 and its seems so much harder to get these things stable, even at a moderate settings of 700-800 kh/s it just does not like it. Probably go back to Win 7 in a next day or so if I can't figure it out. Or try Linux.

      Delete
  6. Hey Jarred . I have purchased the following things but Im not sure what risers I will need? If you could suggest that would be great... the 16x16risers are $50 which Im assured is a high price? Cant seem to get many places...(Obviously this is for 2 rigs) Should I just get 7 x16x16risers?
    3x Sapphire AMD Radeon R9 290X 4GB Video Card
    4x Sapphire Radeon R9 290 4GB
    2x Seasonic x-1250 80PlusGold 1250w
    2x ASRock FM2A88X Extreme4+ Motherboard
    2x Kingston SSDNow V300 60GB 2.5" SATA III SSD SV300S37A/60G
    2x AMD A6-6400K 2-Core Socket FM2 3.9GHz APU Processor
    Regards

    ReplyDelete
    Replies
    1. I've seen those price gougings myself!! I'm running 5x290's on 1 mobo (MSI Z77 GD65) just using powered 1X -> 1X risers... I got them on ebay for ~$4 each (non-powered, no presence detect), and wired up the molex adapters myself to make them powered. Modified the card side connector (x1 to x1 riser hack in link below) by cutting out the slot so it would plug into the cards x16 slot. Also added the presence detect wire, but I don't think that was even necessary for my mobo, but easy to add just in case

      There's some tutorials out on the web to find out what pins you would need to power (same for x1 as x16): https://bitcointalk.org/index.php?topic=76121.0

      May also need to add the presence detect pin: http://framer99.blogspot.com/2013/11/multi-gpu-machines-and-pcie-in-band-vs.html

      Donations welcome! :D
      LTC: LWHgLSxi3iLHGQ4ESME6Tj4STcpX5p3yTD
      BTC: 1DeAqKqi2QGw1bQNJQQu1W8KqWxUi2EkNi

      Delete
  7. Have you tried lowering the intensity? It seems my R9 290 does benefit from a lowered itensity yielding stable 800 kh/s now ... what seems to be beneficial as well is a work size set to 512
    clocks 1000/1350 with
    aticonfig --odsc=1000,1350,1,1

    cgminer --scrypt -o stratum+tcp://stratum01.hashco.ws:8888 -u kwip.2 -p y -I 19 -g 1 -w 512 --lookup-gap 2 --thread-concurrency 32765

    ReplyDelete
    Replies
    1. Work size of 512 has no effect -- just look at the scrypt file names it generates. For instance, here's what -w 512 --thread-concurrency 8192 gives:

      scrypt130511Hawaiiglg2tc8192w256l4.bin

      It's a Hawaii GPU, lookup gap 2, TC of 8192, and a worksize of 256.

      Delete
  8. I'm somewhat disappointed by the 290x i just got, doesn't seem to be able to handle the kh/s that I achieved with my 290s. I need to continue testing but if I had to keep purchasing GPUs it would be 290s at this point.

    ReplyDelete
  9. I just did a bunch of testing tonight on my R9 290x cards and posted the results from my research (Post 1043) here http://www.overclock.net/t/1437876/290-and-290x-litecoin-mining-performance/1040#post_21459996
    Alot of you may find that info useful and may get pointed in the right direction for your settings. Long story short try to find the highest frequency you can set your memory too, then divide that number by 1.44 and that should give you the optimal engine frequency setting. Also if you are only using 4GB of system memory try setting thread concurrency to 27400, I was able to set mine higher than that but then the kHs dropped.

    Nice site you have going here Jarred, I enjoy reading many of the articles you have posted here. I think I am going to give Frozen coin a try tonight yet or tomorrow.

    Happy mining!

    ReplyDelete
  10. Hi Jarred I tweaked my 290s a bit more, again I do have a personal vortex fan blowing air over them.
    Running at 920is each, even got my 270 to run approx 420.
    http://i.imgur.com/VrdvpCi.jpg
    Settings:

    "kernel" : "scrypt,scrypt",
    "api-listen" : true,
    "api-mcast-port" : "4028",
    "api-port" : "4028",
    "expiry" : "120",
    "failover-only" : true,
    "hotplug" : "5",
    "kernel-path" : "/usr/local/bin",
    "log" : "5",
    "no-submit-stale" : true,
    "queue" : "1",
    "scan-time" : "60",
    "scrypt" : true,
    "shares" : "0",
    "auto-fan" : true,
    "gpu-threads" : "1",
    "gpu-dyninterval" : "7",
    "gpu-engine" : "1025,1000",
    "gpu-fan" : "45-100,50-100",
    "gpu-platform" : "0",
    "gpu-memclock" : "1475,1475",
    "gpu-memdiff" : "0,0",
    "gpu-powertune" : "20,20",
    "gpu-vddc" : "1.110,1.125",
    "intensity" : "18,17",
    "temp-target" : "90,85",
    "temp-overheat" : "95,90",
    "temp-cutoff" : "99,99",
    "temp-hysteresis" : "3",
    "vectors" : "1",
    "worksize" : "256,256",
    "lookup-gap" : "2",
    "thread-concurrency" : "27400,16384",
    "no-pool-disable" : true
    }

    ReplyDelete
    Replies
    1. Yes, the fan idea is very useful -- it helps cool the VRMs, which I suspect are what's really holding a lot of people back.

      Delete
  11. Hi Jarrad I have a question.
    I'm running a 6x Sapphire 290 rig atm and hashing at 4.96M in total. I use an unified setting for all the cards
    TC 24450, engine 940 memclock 1250, i-20
    the hash rate varies among all these 6 cards
    GPU0 810
    GPU1 796
    GPU2 802
    GPU3 830
    GPU4 820
    GPU5 850

    All at temperature 80C and the wired thing is GPU4's fan speed is always 20-30% higher than other cards, say when others are at 60-70% fan with 75-79C it is at 80-83% fan with 79-80C. Any ideas?

    Another thing i noticed was the engine clock. I set it to 940 across the range but in CGwatcher all these cards were running at different speeds which was ranging from 898-940. Any ideas then?

    Also I used to OC the memory to 1000 and screen got flashed. Now i'm trying to find the best sweet spot to increase the total hash rate to 5.4M which means 850K average~ Do you have any useful TC to recommend?

    Thanks a lot!

    ReplyDelete
    Replies
    1. You're basically seeing the variance among GPUs -- I don't know how much of it is the GPU core, and how much is RAM and VRM and other component differences, but I've seen a pretty wide range of results for R9 290/290X GPUs now. I think tuning each card individually is about the only thing you can do to really improve the situation, and it's a painstaking process. If you see some cards dropping engine clocks down to mid 800s instead of staying at the 940MHz you set, it means AMD's card is seeing something it doesn't like and is reducing clock speeds to cope with the load.

      What I've found is that tuning of TC is based on the RAM and Engine clocks, and the ratio they have, so if one card runs really well at 940/1250 with TC 24450 but another card is better at 900/1350, you'll need different TC values as well. You can try my batch files, and basically only run on one device at a time (to find appropriate clocks and TC), but that will take some time!

      Delete
  12. Hello, sorry for my English, I'm using google translator.

    First thanks for all the information of your blog is very useful.

    I am new to the world of mining, I am writing to see if you can work out a question.

    I have a r9 290 new sapphire, and managed to make it stable 500kh / s with the following settings:

    {
    "pools": [
    {
    "url": "+ stratum tcp :/ / gigahash.wemineltc.com: 3334"
    "user": "Rencore.1"
    "pass", "x"
    }
    ]
    "intensity": "13",
    "vectors": "1",
    "worksize": "256",
    "lookup-gap": "2",
    "thread-concurrency": "24272",
    "gpu-threads": "1",
    "scrypt": true,
    }

    The problem is when I try to climb the 800 kh / s, when I go up the intensity or thread-concurrency the computer shuts down instantly.

    I think the problem is the power supply is 650W (f650 coolbox Poweline)

    You Might confirm that the problem is the power supply?

    Thanks!

    ReplyDelete
    Replies
    1. If you're only using a single GPU, the PSU shouldn't be the problem. What clocks are you running on the GPU? Anyway, not all PSUs are created equal, so if you can test with a different PSU you should be able to see if that's the problem. I have been running a single HD 5870 on a 450W 80 Plus Bronze PSU for 2.5 years now and the PSU is still working fine, and 5870 uses about as much power as a modern R9 290, but again it was a quality PSU when I bought it.

      Delete
    2. I'm using the default one. 1250/947, if I upload the computer shuts down instantly.

      Next week I receive AeroCool Templarius Imperator 1150W 80 Plus Gold and one 290 more.

      Thanks for reply.

      Delete
  13. Hello,

    I have 2 290x's. I'm running Windows 7. For some reason, only one gpu shows up on the command line after I run cgminer. It cranks to to 850kh/s or so for awhile (and goes down, but that doesn't matter in this discussion). Data for the other gpu doesn't display while this is happening. I don't know if it is not working, disabled, or is not activated by the thread scrypt. I had them crossfiring, but in the end only one gpu worked.

    My steps to try and resolve this:

    1) I plan on uninstalling and reinstalling 3.7.2
    2) rung cgminer -n to find out what the gpus numbers are
    3) Use the same pool, but tuning values separated by commas
    4) Use the same pool for both and username and password separated by one comma and tuning values not separated by a comma
    5) Use the same pool for both and pool username and password separated by one comma and tuning values separated by commas
    6) Use different pools, but username and password separated by one comma and the tuning values not separated by a comma
    7) Use different pools, but username and password separated by one comma and the tuning values separated by a comma
    8) Repeat 4-7 with -d 0 and -d 1

    and if that doesn't work:

    combine 2 incidences into 1 .bat file (I have to find a post about it)

    Hardware:

    2 XFX 290x's
    Sempron 145
    ASROCKS Motherboard
    4GB Hard Drive
    2 G skills
    Seasonic Power Supply

    Software:
    Windows 7

    If anyone has an ideas on ways I can resolve this, I would GREATLY appreciate it. There might be something simple that I forgot in the command line, or something to do with the hardware or software. I don't know at this point until I try this. But I wanted to see if anyone has had the same issue, and if so, how to troubleshoot it.

    Thank you!

    ReplyDelete
  14. Thanks for these settings! I had to remove some of them though, as cgminer would refuse to start up otherwise (Linux).

    Anyway, my question here is, are 85° target temp not too high? Do you consider it safe to run your gpus on that temperature for many month nonstop? Evidence?

    ReplyDelete
    Replies
    1. I've had 7970 and 7950 running at 85C for months on end without trouble, and Hawaii tends to run a bit hotter, so it's just a target. You can target 80C if you want, or put a big fan on the system to keep it cooler, and that certainly won't hurt.

      Delete
    2. Okay, well, i have one R270X at 85° and one R290 at about 80°. I can get both slightly cooler, but the fans are maxing then which try to avoid mainly because of the noise... ;-)

      Delete
  15. Hi Jarred,
    Thank you so much for creating this blog. I find the cryptocurrency world fascinating, so I have decided to have a go myself following your guides! I've got 3 R290s running under Windows 8.1, but I can't mine due to an ADL openGL mismatch in CGminer. I think it is my using of the integrated Intel graphics to run a monitor that is causing it, which CG miner identifies as platform 0, with the three cards as platform 1. I have modified the config file to just consider platform 1 but to no avail. Do you have any ideas that may be able to help me? Thanks again, Andrew

    ReplyDelete
  16. Hey Jarred - So, I'm almost there. I built my first computer rig ever. I have four Sapphire r9 290x cards fired up and ready to go. Problem is - I installed windows and the screen is all scrambled. Even the bios screen was a little scrambled at first, but looked better w/the fourth start. Windows looks 8 bit, but worse and when you move the mouse, it just erases the screen.

    I'm sure it might be due to the drivers needing to be installed for the cards or even the drivers for the mobo...but I can't load the drivers since I can't see the screen.

    I'm doing it via the hdmi off of card/pcie #1. I tried using the dvi connection on the card w/a diff monitor, but there was no signal at all. I am forced to use an hdmi and a tv/monitor. I have enough power [1500w], a great cpu (amd3), 12gb ram (three 4gm cards], and a good mobo (gigabyte 990fx ud5). Everything is new and I wore a bracelet the entire time to ensure no static shocks.

    Any thoughts?

    ReplyDelete
  17. I got windows going by disconnecting all but one. Now I'm having a problem adding them. I'm guessing my prob might have been a x16 slot that was only x8 in power... So, now my prob is figuring out how to add the other 3 cards. Even one more is tough. Can't get it to work.

    ReplyDelete
  18. Hi,
    I´m getting this error "error -4: enqueueing kernel onto command queue."
    "gpu 0 failure, disabling !" when I put setx GPU_MAX_ALLOC_PERCENT 100. But when I put 90 it works...

    Other thing, at the moment I don´t have any thread concurrency. I tried between 20000-25000 but appears a lot of HW. Can you help me? I have one 290x

    Thanks

    ReplyDelete
    Replies
    1. What clocks are you running, and what are the specs for your system (RAM, CPU, motherboard, PSU)? Also, what drivers and OS are you running? It could just be some other program running in the background that's causing the problem, or a lack of RAM, or several other possibilities. What hash rate are you getting now with no TC specified?

      Delete
    2. I have P8H67M pro, i5 2400, SilverStone Strider Essential 500W, 4gb ram 1333. The drivers are 13.11, I think... I have a good hash rate 915, but I think it could be better. Clocks 1000, 1500, thats my sweet spot, I tried many others.

      Delete
    3. 915KHash on R9 290X is better than what most are managing, so possibly you're clocks are a bit too aggressive. Also, there's a relationship between intensity, thread concurrency, and HW errors -- set I=20 and if your TC is too low you'll get HW errors; set it too high and the same can happen, though that's not as common in my experience. If you don't give a specific TC, the value selected will usually be too low and you end up with HW errors, so try something like 24000, or maybe try my TC tuning batch file.

      Delete
    4. I have already I=20 and I don´t have HW. I already tried 24000 but appear error and more errors... Strange. I think it could be memory. I´m about to buy more 4gb, because I have some programs running in the background.

      Delete
    5. If you're only using 4GB system RAM, that might be the problem -- the GPU has 4GB RAM, so my feeling is you need twice that amount of system RAM (though that's just a guess). If you look in your cgminer folder, you can also see what the default TC is being set to by looking at the file names. You should see something like:

      "scrypt130511Tahitiglg2tc21584w256l4.bin"

      That's for a Tahiti (7950) GPU running TC of 21584 with work size 256. For your 290X the file name should be:

      "scrypt130511Hawaiiglg2tc[YOUR TC]w256l4.bin"

      As to the periodic error with not being able to enqueue, again RAM is a potential cause, but sometimes drivers or just slight instability. I have seen that error pop up before, and if a reboot of the PC doesn't fix it, reinstalling drivers usually does the trick.

      Delete
  19. Hi Jarred,

    thank you SO MUCH for all this help. Your modified 3.7.2 works AMAZINGLY. It does not crash any more with the gpu-XXXX params. A few questions I have for you:

    1. How do i read the scrypt1305HawaiXXXXXX bin file? text editor shows a bunch of junk

    I ran cgminer only with
    "intensity" : "20",
    "vectors" : "1",
    "worksize" : "256",
    "lookup-gap" : "2",

    and it had 30592 in the bin file when i search for thread, but that causes the "error -4: enqueueing kernel onto command queue.". NOw i only have 4gb of ram and tomorrow I will go get 4GB more. Would that solve it?

    2. With your thread-count utility, when I run it , it says TC is XXXX , just take that #? or let it run?


    here is my current configs . Should Add anything else?

    "intensity" : "20",
    "vectors" : "1",
    "worksize" : "256",
    "lookup-gap" : "2",
    "thread-concurrency" : "29000",
    "gpu-engine" : "1025",
    "gpu-fan" : "40-70",
    "gpu-memclock" : "1350",
    "gpu-powertune": "20",
    "temp-cutoff" : "99",
    "temp-overheat" : "95",
    "temp-target": "90",
    "apt-port" : "4028",
    "expiry" : "120",
    "failover-only" : true,
    "gpu-threads" : "1",
    "log" : "5",
    "queue" : "1",
    "scan-time" : "60",
    "temp-hysteresis" : "3",

    ReplyDelete
    Replies
    1. 1) The bin files are compiled OpenCL executables I think, so there's nothing in the file to look at -- just the tc[xxxxx] part of the file name tells you what thread concurrency you were using. Most likely you're having problems because of the limited system RAM, so when you have 8GB that should help -- but you'll probably still need to do some fine tuning.

      2) You might find that lowering your GPU engine to 900 or even 850 may improve performance, so that's something to try. Otherwise it looks like your settings should be okay.

      Delete
  20. hi,

    id appreciate any help at his point. I have NO idea whats going on. today is 2/8 I have the latest AMD drivers installed, here are my settings

    "intensity" : "12",
    "vectors" : "1",
    "worksize" : "256",
    "lookup-gap" : "2",
    "thread-concurrency" : "32765",
    "gpu-vddc" : "1.150",
    "gpu-engine" : "1000",
    "gpu-fan" : "40-55",
    "gpu-memclock" : "1500",
    "gpu-powertune": "20",
    "gpu-vddc" : "1.100",
    "temp-cutoff" : "99",
    "temp-overheat" : "95",
    "temp-target": "85",
    "apt-port" : "4028",
    "expiry" : "120",
    "failover-only" : true,
    "gpu-threads" : "1",
    "log" : "5",
    "queue" : "1",
    "scan-time" : "60",
    "temp-hysteresis" : "3",
    "scrypt" : true,
    "kernel" : "scrypt"


    my cgminer constantly crashes with amdocl.dll .... can you please help doesn' matter what I do with thread count.. I have tried 32792, 32000, 35000...i have 8 gb of ram ..

    I have tried intensity at 14 12 20 21 ... idk what to do .. please help

    ReplyDelete
    Replies
    1. Do you have a 290X or a 290 or some other card? Also, 1000/1500 clocks might not work on your card -- try 900/1400 as a safer starting point. You have gpu-vddc in the config you posted twice as well -- I'm not sure if that matters. So put intensity back on 20 and try with lower GPU clocks and see if that works. Otherwise send me an email and I can try to help you.

      Delete
    2. thank you SO much much answering ... im new to this.. how can I send you can email? what is your email address.. ?? where can I get it from?

      Delete
    3. Also I have 290x with 1000 1500 I get about 956kh/s with intensity at 20 , but I lowered the Intensity so I could lower the fan speed at night, but with intensity at 20 or intensity at 12 it keep scrashing. It would be amazing if I can email you and discuss this.

      Delete
    4. My email is my name with a period between the first and last names, at gmail. (I'm trying not to get lots of spam so I don't generally post it.) When you say you keep crashing, is it just cgminer crashing, or does the whole computer crash? Because we can set something up to restart cgminer in the event of a crash.

      Delete