LOW Hash Rate with VEGA (WINDOWS)

[psychocrypt]

Some users reported that they have a low hash rate with cryptonight_v7 (current monero POW) with there VEGA GPU. I created therefore two test binaries and need feedback if one of those binaries creates better hash rates than the last release 2.5.0 If you report any hashrate please report the rate from 2.5.0 and the test cases here. Do not forget to mentioned the test case number. Please also report the config used. (only use one GPU/ but feel free to use two threads per GPU)

I will not give any help in this issue if you are not able to compile the source or start the binaries.

btw: the binaries are compiled without CUDA support!!!!

Test 1: source: https://github.com/psychocrypt/xmr-stak/tree/topic-vegaTest1 binary: https://mega.nz/#!RPhywa7S!uIa5MQ_BBFBsd28Ye8MlQQcIBQ5FdcpPs0ILcvwhyyM sha256: 052f1b1fc20361f49f4985a80ec4775da6ae950eeff39f9eacee3a70566947fa xmr-stak-2.5.0-vega1.zip

Test2: source: https://github.com/psychocrypt/xmr-stak/tree/topic-vegaTest2 binary: https://mega.nz/#!VOhASSrT!zaPwSOUMeCMAiXeknxG3cJZJ2zSXsUgcFddAhaW56vE sha256: bf77802d8baba244de0bbe2b76af67e2810d969ceb8188d99a5d64e7205cf966 xmr-stak-2.5.0-vega2.zip

[maxfreemax]

I am noob but this is my test:

with the parameter (blockchain driver):

{ "index" : 0, "intensity" : 1536, "worksize" : 8, "affine_to_cpu" : false, "strided_index" : 1, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : true }, { "index" : 0, "intensity" : 1536, "worksize" : 8, "affine_to_cpu" : false, "strided_index" : 1, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : true },

V7==> test in nanopool V8 ==> test on (killallasics.moneroworld.com:7777)

VEGA 56

vega1: cryptonight_v7 ~1850h/s ==> OK idem as usual cryptonight_v8 ~880h/s

vega2: cryptonight_v7 ~1850h/s ==> OK idem as usual cryptonight_v8 ~1100h/s

[daletho]

I'm seeing the same right now, big drop in hashrate testing against v8 on killallasics. I noticed the difficulty never increases (Used port 7777) and always sticks to 10000, the miner has to submit shares too regularly so I don't think the GPU's are being used optimally. Not sure if it would be a problem with the pool or a different routine in the new miner.

[psychocrypt]

With v8 you can not use the old configs. You need to find new best values. Also the hash rate will be lower. around 95% of the old config. That is normal.

[Q-Sharp]

Installed 18.6.1 driver (better than blockchain! for sure :D)

Vega 56 on v7 (v 2.4.7): ~ 2020h/s

Vega 56 on v7 (v 2.5.0): ~1870h/s Vega 56 on v8 (v 2.5.0): ~1700h/s Vega 56 on v8 (Test 1): ~1700h/s Vega 56 on v8 (Test 2): ~1700h/s

1407 / 950

config: { "index" : 2, "intensity" : 1792, "worksize" : 16, "affine_to_cpu" : 1, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 4, "comp_mode" : false }, { "index" : 2, "intensity" : 1792, "worksize" : 8, "affine_to_cpu" : 1, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 4, "comp_mode" : false },

-EDIT- Tested around with (normal) 2.5.0 (no test build) Vega 56 on v8 (v 2.5.0): ~1840h/s

{ "index" : 2, "intensity" : 1792, "worksize" : 16, "affine_to_cpu" : 1, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : false }, { "index" : 2, "intensity" : 1792, "worksize" : 8, "affine_to_cpu" : 1, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : false },

Could be optimized a little bit more, but hey its okay. :)

[gotminer]

Somehow I have achieved stability and a slightly better hash rate than I had with 2.4.7, with the 2.5.0 binary that I downloaded two days ago. Approx 960 h/s on each thread.

Vega 56

// gpu: gfx901 memory:3920 // compute units: 64 { "index" : 0, "intensity" : 1775, "worksize" : 8, "affine_to_cpu" : false, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : true }, // gpu: gfx901 memory:3920 // compute units: 64 { "index" : 0, "intensity" : 1775, "worksize" : 8, "affine_to_cpu" : false, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : true

[gotminer]

With v8 you can not use the old configs. You need to find new best values. Also the hash rate will be lower. around 95% of the old config. That is normal.

Why is a 5% loss in hash rate normal?

[gotminer]

After about 30 minutes, I'm down 5% percent in hash rate.

[gotminer]

Seems like we start off good and then after 30 minutes or so, hash rate gradually decreases.

[psychocrypt]

the 5% is v7 compared to v8. The v8 algorithm differ from v7 and therefore the hashrate is different. What you see with the 5% slow down over a long time can be the smoothing of the statistic. The long run contains all fluctuations.

[maxfreemax]

I am a noob in config, but if my experience can help some vega 56 onwer.

After make some change with configuration of my Vega 56: Installer 18.6.1 driver (blockchain driver before for V7) desactived HBCC (HBCC actived with V7) Wattman: -4% GPU fréquency ( -30% with V7) Memory 950 Mhz (950Mhz with V7) power limit: -20% (-20% with V7)

config on V8 xmr-stak-2.5.0-vega2:

// gpu: gfx901 memory:3920 // compute units: 64 { "index" : 0, "intensity" : 1792, "worksize" : 16, "affine_to_cpu" : false, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : true }, // gpu: gfx901 memory:3920 // compute units: 64 { "index" : 0, "intensity" : 1792, "worksize" : 16, "affine_to_cpu" : false, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : true

==>V8 ~1750h/s (first 10s)

hashrate seem OK but I dont know why this OK now!?

[Spudz76]

"v8" (true name CN2v2) has more math work to do math work that ASICs can't even try, ever We all lose 5-15% but ASIC will irrevocably lose by 400% to 1200% with no path to fix it (other than designing more of a complete CPU/GPU vs currently a little specialized chip with easy/quick fab and no prediction unit or parallel executions or the huge amount of registers and cache needed to keep the data off the main memory bus).

Thus, big win. Maybe never another fork until ASIC hackers get good at competing with Intel and AMD and nVidia (not happening).

Also when everyone drops 10% hashrate the same day, the payouts are all the same share (of global chain rate, which will also lose at least 10% - or more if there are hidden ASICs out there that could do "v7" aka CN1v2) The only variance from this is if you had "the best GPU for v7" it may not be the best anymore in v8 / the new code shook up gains and losses for various hardware so it will all be new rankings after the fork. Not every hardware has the same hashrate difference between v7 and v8, some are hit harder and some actually do the v8 faster than v7 (go figure).

[onweer]

some win10 version standalone graphics will link to integrated graphics。 Causing low computing power

[onweer]

@psychocrypt

[rb116]

I am a noob in config, but if my experience can help some vega 56 onwer.

After make some change with configuration of my Vega 56: Installer 18.6.1 driver (blockchain driver before for V7) desactived HBCC (HBCC actived with V7) Wattman: -4% GPU fréquency ( -30% with V7) Memory 950 Mhz (950Mhz with V7) power limit: -20% (-20% with V7)

config on V8 xmr-stak-2.5.0-vega2:

// gpu: gfx901 memory:3920 // compute units: 64 { "index" : 0, "intensity" : 1792, "worksize" : 16, "affine_to_cpu" : false, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : true }, // gpu: gfx901 memory:3920 // compute units: 64 { "index" : 0, "intensity" : 1792, "worksize" : 16, "affine_to_cpu" : false, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : true

==>V8 ~1750h/s (first 10s)

hashrate seem OK but I dont know why this OK now!?

Hi ,how can i deactivate HBCC if without AMD setting being installed ? I only installed driver bcos I am using ODT

[Wacholek]

Hi ,how can i deactivate HBCC if without AMD setting being installed ? I only installed driver bcos I am using ODT

You can try switch-radeon-gpu command line tool but HBCC function is only for Vega.

[rb116]

Thanks for the reply.  I will try this tool and may I know if enabling the large  page for my vega 56 will gain extra hash and stability ? I oredi enable compute mode. 

Sent from Yahoo Mail for iPhone

On Friday, October 19, 2018, 4:37 PM, Wacholek notifications@github.com wrote:

Hi ,how can i deactivate HBCC if without AMD setting being installed ? I only installed driver bcos I am using ODT

You can try switch-radeon-gpu command line tool but HBCC function is only for Vega.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[djp1701]

QoOdrink3r...we have a winner! I have been fighting with this issue now for 2 weeks with my new Vega 64. The driver 18.6.1 with HBCC disabled did the trick.

Mining cryptonight_v8 BlockChain driver:

GPU: 1408 MHz/905 mV Mem: 1100 MHz/905 mV

For both GPU threads //gpu: gfx901 memory 3920 // compute units 64 { "index" : 0, "intensity" : 1932, "worksize" : 8, "affine_to_cpu" : 1, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : true

Hash Rate was no better than 980 split between the two threads!

AMD Radeon Vega 64 driver v18.6.1 with HBCC disabled: All other settings listed above remain the same...

Hash Rate is ~1735 at 150w GPU chip, 64w GPU Core, 21w GPU Memory

Tried the above settings with driver v18.10.1 and received nothing but unrecognized results. THANK YOU!!!

[naymikm]

Thanks you so much! I had the same issue. Switching from the blockchain driver to 18.6.1 w/o hbcc worked!