pre-GCN AMD Cards reported hashrate wrong

[dormancygrace]

im using old Intel xeon e5430, AMD Radeon HD 5770 1Gb. I achieved best performance with this config: "gpu_threads_conf" : [ // gpu: Juniper memory:384 // compute units: 10 { "index" : 0, "intensity" : 256, "worksize" : 16, "affine_to_cpu" : false, "strided_index" : 2, "mem_chunk" : 2, "comp_mode" : true }, With this one GPU gives ~135 H/s. Ive tried to change mem_chunk to 16 and got incredible ~400 H/s. But miner only reports high hashrate, on the pool i can see real hashrate and it`s ~60 (only CPU). XMR-Stak versions: 2.4.2, 2.4.3, 2.4.4

[Spudz76]

Yes, drivers newer than Omega 14.12 do not work nice on Terascale and even some GCN1 - also note it is only detecting 384MB of the 1024MB it should have. AMD broke the Terascale support in the driver after 14.x and same applies to Windows.

I use fglrx 15.x driver in Ubuntu 14 for mine (Redwood not Juniper, but still Terascale) and it shows 896MB which is still missing 128MB somehow, and if given 'bad' mem_chunk it will report strange hashrates outside of reality. Another thing it does with some settings is never report a rate in the 10s column but then the later two will show rates (odd telemetry bug). I should revert to 14.x driver, but currently don't have that card in the rig currently (testing old GT520 in that slot now) and running some Tahiti LE just fine in it on the 15.x driver also

I do like these older cards as they do not need any added power whip, all from the slot under 60 watts, and stay in a single slot width unlike anything newer. I am personally interested in improving their rates and digging around why the newer drivers appear broken and if there are any ways to workaround/fix it.

[Spudz76]

OK, reverted my Ubuntu 14 rig to 3.19.0-80-generic kernel and installed fglrx-14.501.1003 with patch from here

Same rates from the Tahitis, good. Redwood still says 896MB so I guess that's normal. Tried several settings but this gives 100H/s for it which seems pretty good (never got more than maybe 75 with any other combos)

  // gpu: Redwood memory:896
  // compute units: 5
  { "index" : 0,
    "intensity" : 225, "worksize" : 15,
    "affine_to_cpu" : false, "strided_index" : 1, "mem_chunk" : 2,
    "comp_mode" : true
  },

Note the possible harmonic for tuning I used, three times compute units. So on yours reporting 10 units try 10, 20, 30 worksizes with intensity 420 or around there. It will underutilize memory still, but I don't think there is enough horsepower in these to run a full memory load of blocks.

[dormancygrace]

I`ve already installed 14.4. Issue still exists

[Spudz76]

Yeah, don't use strided_index:2, I also had iffy luck with 0. Terascale do not enjoy for some reason.

Does it work with the settings I suggested?
Does it show better memory amount?
Did you redetect (delete/move amd.txt from any previous runs)?
clinfo output?

[dormancygrace]

It works best with my settings/ Yes, it shows more memory. but it`s not the case =)

[Spudz76]

Learned new, check #472

[dormancygrace]

It is different.

[dormancygrace]

[dormancygrace]

AMD Radeon HD 5770, 1Gb GDDR5