[ANN] Avermore miner - optimized sgminer (AMD) fork for x16r/x16s/xevan(beta)

[Marvell2]

got my hands on a few furys nanos , what hash rate do those get ?

I mined with this on one of our gaming rigs (with R9 Fury X) and it got around 7 Mh/s on x16s at stock clocks and 50% power limit. I don't remember what it was on xevan, but I can check it for if you like.

nice , why the low power limit tho ?

[Elder III]

got my hands on a few furys nanos , what hash rate do those get ?

I mined with this on one of our gaming rigs (with R9 Fury X) and it got around 7 Mh/s on x16s at stock clocks and 50% power limit. I don't remember what it was on xevan, but I can check it for if you like.

nice , why the low power limit tho ?

In most of the algorithms that I have tested the R9 Fury X gets 90% of the hashrate at 50% power as it does at 100% power. I'd rather take 90% performance at 170 watts then 340 watts for just 10% more.  Interestingly enough, it handles almost any game I've run at max settings (1080p) at 50% power too, so I just leave it at that power limit 24/7.

[Marvell2]

got my hands on a few furys nanos , what hash rate do those get ?

I mined with this on one of our gaming rigs (with R9 Fury X) and it got around 7 Mh/s on x16s at stock clocks and 50% power limit. I don't remember what it was on xevan, but I can check it for if you like.

nice , why the low power limit tho ?

In most of the algorithms that I have tested the R9 Fury X gets 90% of the hashrate at 50% power as it does at 100% power. I'd rather take 90% performance at 170 watts then 340 watts for just 10% more.  Interestingly enough, it handles almost any game I've run at max settings (1080p) at 50% power too, so I just leave it at that power limit 24/7.

last time i played with eth power limit on my nanos it seems alot slower at 50 percent power lim,ill check zcash results which is what i mine atm

[bluspirit]

This happens with v1.4

https://imgur.com/JRzScD7

[Cezarr]

One of my card (random one) goes SICK and then DEAD, even if I don't OC any of them... Any idea?
using
sgminer -k x16s -o stratum+tcp://rushhourmining.com:3665 -u RVuwFCFN2t3k2CrqRMjn76cvacT3XZ28tU -p c=REEF -w 64 -X 192

Tried -X 128, -X 192, -X 256 and always one of cards goes SICK and then DEAD...

[Marvell2]

yeah miner is just not stable, i swiched all 20 of my nanos and a vega rig to it

even with low powertune and intensity , random sick dead cards

[nyo_x]

Yeah same problem with my vega cards

[Marvell2]

Yeah same problem with my vega cards

its been stable for a few days pretty much luck, on the vegas i set fan speed to 3000

the nanos just started to work. lol, powelimit get to -40 on nanos

one rig i had to use x128

[msallak1]

strange, but I actually get lower rates with 1.2 on RX480 8G with Uber 3.1 mem timings
(running with  -g 2 -w 64 -X 256 --benchmark)

X16r    Avermore-1.1    8.4 mh
X16r    Avermore-1.2    7.743 mh

X16s    Avermore-1.1    8.366 mh
X16s    Avermore-1.2    7.711 mh

Interesting. Can you compare the Hamsi kernel hashrates between 1.1 and 1.2? (run with --benchmark-sequence=BBBBBBBBBBBBBBBB). It's possible some of the changes I made caused regressions in cards other than the RX580.

exits immediately when I use --benchmark-sequence=anything (even 0123456789ABCDEF) in both 1.1 and 1.2

edit:
I also noticed that I'm hitting the thermal card power limit and it starts throttling
in 1.1 the GPU clock gets throttled to ~1200Mhz and temperature stays around 80c, so it can probably reach even higher speeds with better cooling
in 1.2 the GPU clock is ~1300Mhz and temperature is ~75c

580 is basically the same card, with slightly higher clocks, so should perform similarly

I'll take a look into why --benchmark-sequence isn't working on Windows.

I did some testing on my test rig, and the results are weird.

The Hamsi kernel, when benchmarked by itself (i.e. repeated 16 times), is indeed 35% faster on 1.2 compared to 1.1 (4.2 MHs -> 5.7 MHs). Also, when I benchmark the rest of the algos (0123456789ACDEF), the hashrates are the same between the two versions (as expected, since the only change was with the Hamsi kernel).

However, when I test running all of the algos (0123456789ABCDEF), v1.1 is indeed slightly faster than v1.2 (7.9 MHs vs 7.7 MHs). I'm a bit stumped right now on why this would be the case. So anyways; if you're getting better results on v1.1, then stick with that version for now. I'll see if I can get a patch to v1.2 to address this issue.

All tests were done with -g 2 -w 64 -X 64, by the way. Found this to be the optimal xIntensity on my RX580. When running with -X 256, v1.2 did beat out v1.1 (7.1 MHs vs 6.8 MHs), but both were slower than with -X 64.

I can confirm that I got +8 mh/s with my MSI RX 580 MK2 Armor using stock settings, no bios mod or overclocking. I used these settings : -g 2 -w 64 -X 64  , miner version 1.4
https://i.imgur.com/V9GjWNF.png

[Marvell2]

strange, but I actually get lower rates with 1.2 on RX480 8G with Uber 3.1 mem timings
(running with  -g 2 -w 64 -X 256 --benchmark)

X16r    Avermore-1.1    8.4 mh
X16r    Avermore-1.2    7.743 mh

X16s    Avermore-1.1    8.366 mh
X16s    Avermore-1.2    7.711 mh

Interesting. Can you compare the Hamsi kernel hashrates between 1.1 and 1.2? (run with --benchmark-sequence=BBBBBBBBBBBBBBBB). It's possible some of the changes I made caused regressions in cards other than the RX580.

exits immediately when I use --benchmark-sequence=anything (even 0123456789ABCDEF) in both 1.1 and 1.2

edit:
I also noticed that I'm hitting the thermal card power limit and it starts throttling
in 1.1 the GPU clock gets throttled to ~1200Mhz and temperature stays around 80c, so it can probably reach even higher speeds with better cooling
in 1.2 the GPU clock is ~1300Mhz and temperature is ~75c

580 is basically the same card, with slightly higher clocks, so should perform similarly

I'll take a look into why --benchmark-sequence isn't working on Windows.

I did some testing on my test rig, and the results are weird.

The Hamsi kernel, when benchmarked by itself (i.e. repeated 16 times), is indeed 35% faster on 1.2 compared to 1.1 (4.2 MHs -> 5.7 MHs). Also, when I benchmark the rest of the algos (0123456789ACDEF), the hashrates are the same between the two versions (as expected, since the only change was with the Hamsi kernel).

However, when I test running all of the algos (0123456789ABCDEF), v1.1 is indeed slightly faster than v1.2 (7.9 MHs vs 7.7 MHs). I'm a bit stumped right now on why this would be the case. So anyways; if you're getting better results on v1.1, then stick with that version for now. I'll see if I can get a patch to v1.2 to address this issue.

All tests were done with -g 2 -w 64 -X 64, by the way. Found this to be the optimal xIntensity on my RX580. When running with -X 256, v1.2 did beat out v1.1 (7.1 MHs vs 6.8 MHs), but both were slower than with -X 64.

I can confirm that I got +8 mh/s with my MSI RX 580 MK2 Armor using stock settings, no bios mod or overclocking. I used these settings : -g 2 -w 64 -X 64  , miner version 1.4

-G 2 does not work on systems with more than six cards , whats another option for that ?

[4ward]

-G 2 does not work on systems with more than six cards , whats another option for that ?

2 instances of sgminer

[Marvell2]

-G 2 does not work on systems with more than six cards , whats another option for that ?

2 instances of sgminer

tried that it just crashes

I also had the line -d 0,1,2,3

to try and start the first instance

[4ward]

-G 2 does not work on systems with more than six cards , whats another option for that ?

2 instances of sgminer

tried that it just crashes

I also had the line -d 0,1,2,3

to try and start the first instance

and the second with -d 4,5.. ?

[brianmct]

Avermore miner v1.4.1:

• Optimizations to Echo (+7%) for an overall 1% speedup in x16.
• Added code to automatically select AMD OpenCL platform by default. This means that setting --gpu-platform is no longer required for rigs with integrated graphics or Nvidia cards.
• More sane default parameters. Changed default xIntensity to 256, from a default intensity of 8.

https://github.com/brian112358/avermore-miner/releases/tag/v1.4.1

[Marvell2]

-G 2 does not work on systems with more than six cards , whats another option for that ?

2 instances of sgminer

tried that it just crashes

I also had the line -d 0,1,2,3

to try and start the first instance

and the second with -d 4,5.. ?

switched to the other miner fork for now

[Elder III]

I'll download the new version and try to test it on one of our rigs in the next couple of days. I'll post results for some Rx 580 8GB GPUs after testing it out.

[coin4addict]

I used to mine RVN with R9 280x and RX 570 with avermore 1.2 on windows 7. But on windows 10, my RX 570 is not starting at all and declared sick. I tried 1.4 still the same problem but my R9 280x is doing fine.

[Marvell2]

I used to mine RVN with R9 280x and RX 570 with avermore 1.2 on windows 7. But on windows 10, my RX 570 is not starting at all and declared sick. I tried 1.4 still the same problem but my R9 280x is doing fine.

yeah same here having lots of issues with stability and this is with a powerlim of -25

[Marvell2]

there is something wrong sometimes with the miner after a dev fee it starts not propperly connecting to stratump and starts using curl commands

please look into this

[Kaezar]

Hi. brianmct, can you compile linux version with Vega's support?
Now compiled linux release works with vegas with x10 lower hash rate.