[ANN] YACMiner - AMD GPU miner for Scrypt-Chacha, N-Scrypt, and Scrypt coins - page 11.

ivanlabrie

hero member

Activity: 812

Merit: 1000

Quote from: regnevacs on April 06, 2014, 03:28:55 AM

hi, cant seem to find the settings for UTC for the current n-factor.. what does buffer size means?

It mean show much vram you're allocating for the miner to use.

You can more often than not find the sweetspot using 80%-90% of the total ram. Do the math and experiment, then set lookup gap to 1,2,3,4 following the guidelines in the readme.

For UTC you can still use lg 3 or 4 with most higher end cards, and more or less get away with -R equal to 4 times the shader count and -g 1. lg 3 will demand you cut down on -R but you'll have to experiment yourself.
I find getting more than 0 HW unacceptable.

regnevacs

full member

Activity: 182

Merit: 100

hi, cant seem to find the settings for UTC for the current n-factor.. what does buffer size means?

Thirtybird

hero member

Activity: 693

Merit: 500

Quote from: Hilux74 on March 26, 2014, 01:49:13 PM

Update:

I have had success in running 6x Sapphire R7-240 4GB with 8GB system ram with windows7 x64.

Buffer = 1519, g = 2, rawintensity=768, lookupgap = 2 >>>> each card hashing YAC at 2.9-3.0 khash/s. Any other intensity values gave less hash and more HW errors (768 = shadersx2).

Still would be better with more system memory as I get about 5% HW errors still. With 4 cards I could run Buffer=1600 and have one card with like 2% HW errors. WIth 6 cards if I set Buffer >= 1520 I get the kernel4 error and 1 or more cards switch to OFF.

I tried it with 16GB system memory and could run Buffer = 2700 before kernel4 errors. At this setting I got 0 HW errors on any cards...but needed that ram back in other systems. (I can't remember if I had g=1 or g=2 for this test).

I think with 12GB ram you could run 6x R7-240 4GB cards at max hash with 0 HW errors.

I think you did some math incorrect. Those cards have 320 shaders on them. I just got confirmation that a user has gotten 6xR7 240 4GB cards running on a windows machine with G=1 with 8 GB of memory. I'm doing that with 4 cards, so It's certainly in the realm of possibility with 8 GB. System memory only dictates how large a buffer you can assign - HW errors come from setting too high a -R within a single thread. When I ran mine with -g 2, -R should be 640. Minimum buffer size for -g 2 -R 640 at N=14 is 1280.

Hilux74

legendary

Activity: 912

Merit: 1000

Update:

I have had success in running 6x Sapphire R7-240 4GB with 8GB system ram with windows7 x64.

Buffer = 1519, g = 2, rawintensity=768, lookupgap = 2 >>>> each card hashing YAC at 2.9-3.0 khash/s. Any other intensity values gave less hash and more HW errors (768 = shadersx2).

Still would be better with more system memory as I get about 5% HW errors still. With 4 cards I could run Buffer=1600 and have one card with like 2% HW errors. WIth 6 cards if I set Buffer >= 1520 I get the kernel4 error and 1 or more cards switch to OFF.

I tried it with 16GB system memory and could run Buffer = 2700 before kernel4 errors. At this setting I got 0 HW errors on any cards...but needed that ram back in other systems. (I can't remember if I had g=1 or g=2 for this test).

I think with 12GB ram you could run 6x R7-240 4GB cards at max hash with 0 HW errors.

Thirtybird

hero member

Activity: 693

Merit: 500

Quote from: ozzy1926 on March 24, 2014, 03:03:26 PM

Quote from: Thirtybird on March 24, 2014, 01:56:33 PM

If you're mining UTC, that's fine - like I said, you're shader limited ... for now

When you hit N=12, you'll need to make changes.

The same config with YACMiner will work (because I kept all the legacy settings available), you'll just have to get used to calling it scrypt-chacha

Code:

yacminer --scrypt-chacha --worksize 256 -g 1 --lookup-gap 2 -X 128 --buffer-size 2625 --gpu-engine 1040 --gpu-memclock 1250 --gpu-powertune 20 -o stratum+tcp://poolinfo -u username.worker -p password --nfmin 4 --nfmax 30 --starttime 1388361600

When N changes in 2 weeks, you're going to want to set --buffer-size as high as you can get it (while still keeping it in dedicated GPU memory - see my scrypt-readme file on my github for how to do that) and then if you can get it high enough, you can go to lookup-gap 3 to keep using -X, otherwise at LG=2 you're memory limited and will want to start using --rawintensity (-R) to set intensity instead of -X. IF that's as much as you can allocate, you're stuck with LG=4 when N=12.

i tried this setting for vertcoin but i am getting all rejects :/
also 15kh/s when mining ultra

Those settings were not for vertcoin, so of course it wouldnt work!
15 KH/sec on UTC with those settings - on an R9 290? Only way that's possible is if your thread is going into dynamic memory instead of dedicated memory - lower the buffer-size if you find that to be the case (I detailed that in the scrypt-readme I pointed you to). My laptop gets 35 KH/sec with those settings, so you've got something to figure out my friend.

ozzy1926

sr. member

Activity: 378

Merit: 250

Quote from: Thirtybird on March 24, 2014, 01:56:33 PM

If you're mining UTC, that's fine - like I said, you're shader limited ... for now

When you hit N=12, you'll need to make changes.

The same config with YACMiner will work (because I kept all the legacy settings available), you'll just have to get used to calling it scrypt-chacha

Code:

yacminer --scrypt-chacha --worksize 256 -g 1 --lookup-gap 2 -X 128 --buffer-size 2625 --gpu-engine 1040 --gpu-memclock 1250 --gpu-powertune 20 -o stratum+tcp://poolinfo -u username.worker -p password --nfmin 4 --nfmax 30 --starttime 1388361600

When N changes in 2 weeks, you're going to want to set --buffer-size as high as you can get it (while still keeping it in dedicated GPU memory - see my scrypt-readme file on my github for how to do that) and then if you can get it high enough, you can go to lookup-gap 3 to keep using -X, otherwise at LG=2 you're memory limited and will want to start using --rawintensity (-R) to set intensity instead of -X. IF that's as much as you can allocate, you're stuck with LG=4 when N=12.

i tried this setting for vertcoin but i am getting all rejects :/
also 15kh/s when mining ultra

Queripel

newbie

Activity: 19

Merit: 0

Quote from: Thirtybird on March 24, 2014, 11:19:31 AM

Your settings look pretty close to me. YAC favor cards with lots of memory, and relative to how many threads you have (2560), 4GB is actually not much i the way of memory.

Your ideal settings (no HW errors) will be one of these (based on 3600 buffer-size)
LG   rI
6   5296
7   6227
8   7200
9   7680
10   8694
11   9804
12   10240

You may need to round down to the nearest multiplier of --worksize (Depending on your driver). Obviously, the more you can allocate with buffer size, the higher your rI can go

Thanks - I'm just so used to my R9 290s significantly outperforming my GTX 780s I assumed I must be missing something! I also got a bit excited when I got over 10k on my first attempt before noticing all the hardware errors... Sad

What's the relationship between buffer size and raw intensity at the various lookup gaps? Besides LG 8 being double the buffer size of course, that one I can manage myself!

Thirtybird

hero member

Activity: 693

Merit: 500

If you're mining UTC, that's fine - like I said, you're shader limited ... for now

When you hit N=12, you'll need to make changes.

The same config with YACMiner will work (because I kept all the legacy settings available), you'll just have to get used to calling it scrypt-chacha

Code:

yacminer --scrypt-chacha --worksize 256 -g 1 --lookup-gap 2 -X 128 --buffer-size 2625 --gpu-engine 1040 --gpu-memclock 1250 --gpu-powertune 20 -o stratum+tcp://poolinfo -u username.worker -p password --nfmin 4 --nfmax 30 --starttime 1388361600

When N changes in 2 weeks, you're going to want to set --buffer-size as high as you can get it (while still keeping it in dedicated GPU memory - see my scrypt-readme file on my github for how to do that) and then if you can get it high enough, you can go to lookup-gap 3 to keep using -X, otherwise at LG=2 you're memory limited and will want to start using --rawintensity (-R) to set intensity instead of -X. IF that's as much as you can allocate, you're stuck with LG=4 when N=12.

ozzy1926

sr. member

Activity: 378

Merit: 250

Quote from: Thirtybird on March 24, 2014, 01:00:10 PM

Quote from: ozzy1926 on March 24, 2014, 12:26:40 PM

@Thirtybird
do you have config for sapphire 290 trix oc?
also can we use this miner for ultracoin?

no, I do not - you can work it out though. You need to post the maximum memory you can allocate via --buffer-size and what coin you're mining. If you're allocating around 3552, it will likely be LG=8 and rI of 6912 or so for YACoin. With the same buffer-size of 3552, UTC would be shader limited, and I would recommend xI of 32 or xI of 64 or so.

we can. In fact, I do mine UTC with it. (and YACoin [of course], and vertcoin, and middlecoin, and wafflepool, and even Thorcoin and BitLeu)

If it's scrypt, Adaptive-N-Scrypt, or scrypt-chacha(even if you call it scrypt-jane), it'll mine it.

my rig has 24gb ram and using this config with cgminer:

Code:

"api-port" : "4028",
"expiry" : "10",
"kernel" : "scrypt-jane",
"kernel-path" : "/usr/local/bin",
"log" : "5",
"queue" : "0",
"scrypt-jane" : true,
"sj-nfmin" : "4",
"sj-nfmax" : "30",
"sj-time" : "1388361600",
"gpu-threads" : "1",
"gpu-fan" : "50",
"gpu-engine" : "1040",
"gpu-memclock" : "1250",
"gpu-powertune" : "20",
"scan-time" : "30",
"intensity" : "18",
"temp-target" : "88",
"temp-overheat" : "90",
"temp-cutoff" : "95",
"temp-hysteresis" : "2",
"vectors" : "1",
"worksize" : "64",
"lookup-gap" : "2",
"thread-concurrency" : "42000"

and getting 177 kh/s

ivanlabrie

hero member

Activity: 812

Merit: 1000

this should be the default bamt miner.

Thirtybird

hero member

Activity: 693

Merit: 500

Quote from: ozzy1926 on March 24, 2014, 12:26:40 PM

@Thirtybird
do you have config for sapphire 290 trix oc?
also can we use this miner for ultracoin?

no, I do not - you can work it out though. You need to post the maximum memory you can allocate via --buffer-size and what coin you're mining. If you're allocating around 3552, it will likely be LG=8 and rI of 6912 or so for YACoin. With the same buffer-size of 3552, UTC would be shader limited, and I would recommend xI of 32 or xI of 64 or so.

we can. In fact, I do mine UTC with it. (and YACoin [of course], and vertcoin, and middlecoin, and wafflepool, and even Thorcoin and BitLeu)

If it's scrypt, Adaptive-N-Scrypt, or scrypt-chacha(even if you call it scrypt-jane), it'll mine it.

ozzy1926

sr. member

Activity: 378

Merit: 250

@Thirtybird
do you have config for sapphire 290 trix oc?
also can we use this miner for ultracoin?

Thirtybird

hero member

Activity: 693

Merit: 500

Quote from: ilostcoins on March 24, 2014, 11:05:07 AM

Thank you for keep improving the miner, Thirtybird.

I've just updated though haven't really tried to do more tuning.

I'm getting like 1.6-1.7kh/s from a R7 240 2GB (buffer size 1752, lookup gap 3, g 1, w 256, raw intensity 1280, gpu engine 780, gpu memclock 780, gpu powertune -17), which is some way off what others are getting. I'm underclocking the RAM because performance is somehow very similar when I use stock settings.

Thats a good config. That R7 240 can clock up to 1100 MHz though, and that's where the difference is I'm sure. ~2.1KH/sec is about right for that card with a good overclock. I've found very little difference in speed with RAM speed changes - for me it's all been about the core GPU speed.

Thirtybird

hero member

Activity: 693

Merit: 500

Quote from: Queripel on March 24, 2014, 10:16:30 AM

I'm really struggling to get a decent YAC rate out of my three MSI R9 290 (4Gb) cards; the best I've managed so far without a load of hardware errors is a little over 5kH per card which seems too low to me!

I've tried to go through the tuning guide in the readme, have tried 2, 4, and 8 for lookup gap, but still can't do any better. Can anyone help please?

My "best" settings so far:

Code:

"device" : "0,1,2",
"rawintensity" : "7600",
"buffer-size" : "3600",
"lookup-gap" : "8",
"gpu-engine" : "1000",
"gpu-memclock" : "1250",
"gpu-threads" : "1",
"worksize" : "256",

"api-port" : "4028",
"expiry" : "20",
"failover-only" : true,
"no-submit-stale" : true,
"scrypt-chacha" : true,

"queue" : "0",
"scan-time" : "10"

Edit: I'm running Windows 7 64-bit with 8Gb of system RAM, using the above settings Task Manager shows 1419/8153 memory in use (so well over 6Gb of physical memory available)

Your settings look pretty close to me. YAC favor cards with lots of memory, and relative to how many threads you have (2560), 4GB is actually not much i the way of memory.

Your ideal settings (no HW errors) will be one of these (based on 3600 buffer-size)
LG   rI
6   5296
7   6227
8   7200
9   7680
10   8694
11   9804
12   10240

You may need to round down to the nearest multiplier of --worksize (Depending on your driver). Obviously, the more you can allocate with buffer size, the higher your rI can go

ilostcoins

sr. member

Activity: 274

Merit: 250

Thank you for keep improving the miner, Thirtybird.

I've just updated though haven't really tried to do more tuning.

I'm getting like 1.6-1.7kh/s from a R7 240 2GB (buffer size 1752, lookup gap 3, g 1, w 256, raw intensity 1280, gpu engine 780, gpu memclock 780, gpu powertune -17), which is some way off what others are getting. I'm underclocking the RAM because performance is somehow very similar when I use stock settings.

Queripel

newbie

Activity: 19

Merit: 0

I'm really struggling to get a decent YAC rate out of my three MSI R9 290 (4Gb) cards; the best I've managed so far without a load of hardware errors is a little over 5kH per card which seems too low to me!

I've tried to go through the tuning guide in the readme, have tried 2, 4, and 8 for lookup gap, but still can't do any better. Can anyone help please?

My "best" settings so far:

Code:

"device" : "0,1,2",
"rawintensity" : "7600",
"buffer-size" : "3600",
"lookup-gap" : "8",
"gpu-engine" : "1000",
"gpu-memclock" : "1250",
"gpu-threads" : "1",
"worksize" : "256",

"api-port" : "4028",
"expiry" : "20",
"failover-only" : true,
"no-submit-stale" : true,
"scrypt-chacha" : true,

"queue" : "0",
"scan-time" : "10"

Edit: I'm running Windows 7 64-bit with 8Gb of system RAM, using the above settings Task Manager shows 1419/8153 memory in use (so well over 6Gb of physical memory available)

Hilux74

legendary

Activity: 912

Merit: 1000

That's interesting. Part of the fun of Scrypt-Jane is that it is much more system dependent. I have no issue with 6x7950(2GB) doing normal Scrypt with only 4GB of system ram. The error code I get is the same as you mentioned bitcodo.

I'll add another 4GB of RAM and adjust the paging file and see what happens...

bitcodo

hero member

Activity: 509

Merit: 500

Can't upload avatar

You need a lot of system RAM. I have 8GB and sometimes I get low disk space warning (small disk) so it might be that is trying to make bigger paging file. I also sometimes get -4 warning from miners (something about kernel and queue), but everything works OK. I am mining YAC with msi r7 240 4gb low profile + microcoin with 7950.

edit: r7 can do --buffer-size 3750

Thirtybird

hero member

Activity: 693

Merit: 500

Quote from: Hilux74 on March 21, 2014, 02:40:24 PM

Quote from: bitcodo on March 14, 2014, 04:15:19 PM

YAC: R7 240 4GB

Code:

--scrypt-chacha --buffer-size 3376 --no-submit-stale  -w 256 --rawintensity 1280 --lookup-gap 2 --gpu-threads 1

2.9 Kh/s

So I just put together another mining system, this time with R7-240s for YACoin and having some trouble on my initial test runs.
MB: Gigabyte 990A UD3P (has 4x PCIE16x, 2x PCIE1x)
Memory: 4GB
GPU: 6x Sapphire R7-240 4GB (direct on MB for the 4 PCIE16x, 2 powered risers for the 2x PCIE1x)

If I use values like supplied by Bitcodo, on load yacminer flashes some error about the kernel and all 6 video cards are 'OFF' as soon as YACminer loads.

IF I set --buffer-size between 1200-2000 4-5 of the cards will run while 1-2cards are OFF.

In order to get the cards running I have to lower --buffersize all the way down around 1000 which seems wastefully low. Then they will hash at ~1.8khash but with pretty much equal amount of accepted and HW errors. If I drop --rawintesity down to 1024 it seems to run without HW errors but hashrates of 1.2 make it a waste of time.

It sounds like pretty much everyone with the 4GB cards are all Sapphire brand so I assume it must be something with system not the videocards that is screwing up the performance. My guess is that maybe 4GB system memory isn't enough for 6 4GB cards?

4GB is not enough for even 2 4GB cards. Wtih 4 GB system RAM, you can run 2 4GB cards with 2 threads (-g 2) and 1 2GB cards with -g 1. This is on Window 7 and from personal experience. Now that I have 8 GB system memory, I can run 4 R7 240 4GB cards each with 1 thread, and life is wonderful. Whether I could do 6 or not, I do not know. Someone said to try increasing your virtual memory (i.e. paging file), which may help you get it kicked off. Seriously though, 24GB in graphics cards on 4GB system RAM is probably too disparate a ratio.

Hilux74

legendary

Activity: 912

Merit: 1000

Quote from: bitcodo on March 14, 2014, 04:15:19 PM

YAC: R7 240 4GB

Code:

--scrypt-chacha --buffer-size 3376 --no-submit-stale  -w 256 --rawintensity 1280 --lookup-gap 2 --gpu-threads 1

2.9 Kh/s

So I just put together another mining system, this time with R7-240s for YACoin and having some trouble on my initial test runs.
MB: Gigabyte 990A UD3P (has 4x PCIE16x, 2x PCIE1x)
Memory: 4GB
GPU: 6x Sapphire R7-240 4GB (direct on MB for the 4 PCIE16x, 2 powered risers for the 2x PCIE1x)

If I use values like supplied by Bitcodo, on load yacminer flashes some error about the kernel and all 6 video cards are 'OFF' as soon as YACminer loads.

IF I set --buffer-size between 1200-2000 4-5 of the cards will run while 1-2cards are OFF.

In order to get the cards running I have to lower --buffersize all the way down around 1000 which seems wastefully low. Then they will hash at ~1.8khash but with pretty much equal amount of accepted and HW errors. If I drop --rawintesity down to 1024 it seems to run without HW errors but hashrates of 1.2 make it a waste of time.

It sounds like pretty much everyone with the 4GB cards are all Sapphire brand so I assume it must be something with system not the videocards that is screwing up the performance. My guess is that maybe 4GB system memory isn't enough for 6 4GB cards?

Topic: [ANN] YACMiner - AMD GPU miner for Scrypt-Chacha, N-Scrypt, and Scrypt coins - page 11. (Read 47386 times)