LTC mining with Reaper on a 7970

[philips]

Thanks... but anything I try that yields a buffer size over 512MB doesn't work.  My card is supposed to have 3GB too :-/

Have you tried the GPU_MAX_ALLOC_PERCENT variable?

[Tomatocage]

change thread_per_gpu to 1

I've played around with this setting from 1 to 5, and it doesn't affect my problem at all

[Tomatocage]

Thanks... but anything I try that yields a buffer size over 512MB doesn't work.  My card is supposed to have 3GB too :-/

Have you tried the GPU_MAX_ALLOC_PERCENT variable?

This sounded promising, but alas it doesn't appear to allow me to go beyond 8192 for the gpu_thread_concurrency

[Tomatocage]

Anybody with a 7970 getting 650-700 kh/s out of their cards?  Could you post your exact litecoin.conf file (minus your miner info, of course)?  Also, what is recommended for core/mem clock to hit those numbers?  Thanks.

[superfastkyle]

only tacotime and he is probably just trolling you. its usually a good sign of lying when people use interjection words like personally, honestly, seriously etc Its using language to "prove" a lie to yourself so you will have an easier time making other people believe it.

Anybody with a 7970 getting 650-700 kh/s out of their cards?  Could you post your exact litecoin.conf file (minus your miner info, of course)?  Also, what is recommended for core/mem clock to hit those numbers?  Thanks.

[meebs]

only tacotime and he is probably just trolling you. its usually a good sign of lying when people use interjection words like personally, honestly, seriously etc Its using language to "prove" a lie to yourself so you will have an easier time making other people believe it.

Anybody with a 7970 getting 650-700 kh/s out of their cards?  Could you post your exact litecoin.conf file (minus your miner info, of course)?  Also, what is recommended for core/mem clock to hit those numbers?  Thanks.

Tacotime is responsible for me being able to get 550-590 out of my 7950 using this: PLEASE note that running this setup on a card connected to your monitor WILL make the computer nearly unusable. The screen will literally refresh only once every 2-3 seconds. If it is on a computer you are "remoting" into, it'll be fine. Also, running the monitor on another card or IGP you'll be fine to. Also, reaper will consume 1.5gb of system ram and gpu ram PER card.

worksize 256
vectors 1
aggression 20
threads_per_gpu 1
sharethreads 32
lookup_gap 2
gpu_thread_concurrency 24576

[tacotime]

The memory bus is the same size for the 7970, so use threads = 1 and thread concurrency = 384 * 64 = 24576, aggression 20.

Words like "personally" mean that's what I got, and that your mileage may vary.  You can get 700kh/s on a 7950 though, so I'm not sure why anyone would think getting that much from a 7970 would be difficult.

[Tomatocage]

Tacotime is responsible for me being able to get 550-590 out of my 7950 using this: PLEASE note that running this setup on a card connected to your monitor WILL make the computer nearly unusable. The screen will literally refresh only once every 2-3 seconds. If it is on a computer you are "remoting" into, it'll be fine. Also, running the monitor on another card or IGP you'll be fine to. Also, reaper will consume 1.5gb of system ram and gpu ram PER card.

worksize 256
vectors 1
aggression 20
threads_per_gpu 1
sharethreads 32
lookup_gap 2
gpu_thread_concurrency 24576

Hmm, thanks. I still can't get past my inability to use anything beyond 8192 for gpu_thread_concurrency in litecoin.conf.  Is there any other file I should be editing?  My reaper.conf file looks like this:

kernel reaper.cl
save_binaries yes
enable_graceful_shutdown no
long_polling yes

device 0

worksize 256
aggression 13
threads_per_gpu 1
sharethreads 32
lookup_gap 2
gpu_thread_concurrency 8192
long_polling yes

mine litecoin

[tacotime]

Hmm, thanks. I still can't get past my inability to use anything beyond 8192 for gpu_thread_concurrency in litecoin.conf.  Is there any other file I should be editing?  My reaper.conf file looks like this:

Your reaper.conf should look like:

Code:

kernel reaper.cl
save_binaries yes
enable_graceful_shutdown no
long_polling yes

mine litecoin

your litecoin.conf should look like:

Code:

worksize 256
vectors 1
aggression 20
threads_per_gpu 1
sharethreads 32
lookup_gap 2
gpu_thread_concurrency 24576

[meebs]

also keep in mind that memory clock matters.. core clock does NOT have to be as high as BTC mining formax hash rate.

on my 7950 my LTC hash rate is the same if my core is set to 1000 or 1175. Keep that in mind.. If i drop the core down to 930, then I lose about 2-3% speed. Note that this results in a pretty good power savings over BTC mining as it takes quite a lot less voltage and wattage to run the card at 1000 vs 1175

also with the settings above on two cards I am getting a combined speed of 1150 khash.

[btcwalker]

Something strange are going.

worksize 256
vectors 1
aggression 20
threads_per_gpu 1
sharethreads 32
lookup_gap 2
gpu_thread_concurrency 24576

I got ~600khash\s with that config on 7950 xfx black 1000mhz core \ 1375mhz memory

but

I got ~360khash\s with that config on 7970 msi lightning 1160mhz core \ 1400mhz memory

and the maximum I got from 7970 is ~550khash\s when the following config was used:

worksize 256
vectors 1
aggression 13
threads_per_gpu 4
sharethreads 32
lookup_gap 2
gpu_thread_concurrency 8192

May be I am doing something wrong? Why 7970 with greater clocks and hardware power are worse on the same config?

[tacotime]

try these thread concurrencies with aggression 20 and worksize 64, threads_per_gpu 1

16384
17408
18432
19456
20480
21504
22528
23552
24576
25600
26624
27648
28672
29696
30720
31744
32768
33792
34816

at some point on the low end you'll get lots of HW errors and at some point on the high end you'll get a really slow speed.  at some point in between you'll find your optimum thread_concurrency.  unfortunately no one can tell you what that is, you need to experiment.

[btcwalker]

The following results have been achieved:

http://imageshack.us/a/img266/5644/7970reapertests.png

The best config for 7970 is (565khash\s):

worksize 256
vectors 1
aggression 13
threads_per_gpu 1
sharethreads 32
lookup_gap 2
gpu_thread_concurrency 34816

Maybe I will continue experiments to find the better config. But it seems that 7970 is better for BTC mining because it is not hard to get ~700mhash\s

[Badonkadonk]

was having some issues with my 7950`s when i switched to ltc mining, low speeds and loads of rejected / stale shares.

i solved my problem with going solo minig and changing to cgminer with "cgminer --scrypt -o 127.0.0.1:9332 -u username -p password --shaders 1792 -I 18 -w 256 -v 1 -g 1"

clocking my cards at 1000/1775 im getting ~600kh on each but tuned it down to 900/1675 to lower my heat issues and pumping out ~550kh each...

you should be able to pump out ~700kh with those settings

[tacotime]

The best config for 7970 is (565khash\s):

worksize 256
vectors 1
aggression 13
threads_per_gpu 1
sharethreads 32
lookup_gap 2
gpu_thread_concurrency 34816

Maybe I will continue experiments to find the better config. But it seems that 7970 is better for BTC mining because it is not hard to get ~700mhash\s

Maybe a stupid question, but have you tried aggression 16? This is about 30% faster than aggression 13 on my 7770.

[btcwalker]

Maybe a stupid question, but have you tried aggression 16? This is about 30% faster than aggression 13 on my 7770.

yes, I did, and the result is 100% hardware errors reported by reaper. 13 and 20 aggressions are working well only.

[tacotime]

if you can mine at aggression 20 you should be able also to mine at aggression 16.  but i guess if that's not the case that's not the case.

you could try,

Code:

worksize 256
vectors 1
aggression 16
threads_per_gpu 1
sharethreads 32
lookup_gap 2
gpu_thread_concurrency 34816

If the thread concurrency is within the correct range and threads_per_gpu set to 1 you should not get stales with high aggressions.  But these cards act really weird and it's hard to tell why they do the things they do sometimes.

[btcwalker]

if you can mine at aggression 20 you should be able also to mine at aggression 16.  but i guess if that's not the case that's not the case.

you could try,

Code:

worksize 256
vectors 1
aggression 16
threads_per_gpu 1
sharethreads 32
lookup_gap 2
gpu_thread_concurrency 34816

If the thread concurrency is within the correct range and threads_per_gpu set to 1 you should not get stales with high aggressions.  But these cards act really weird and it's hard to tell why they do the things they do sometimes.

I just retried this agression with config you provided. It seems that previous hardware errors on that agression, I said about before, were because of something different. But this config doesn't provide any benefit for 7970, the result is 370khash\s.

[tacotime]

bump with settings for 760kh/s

http://i.piccy.info/i7/eef9270eec21626fce47de4a5b0bfafe/4-47-398/10015963/760.jpg

using catalyst 12.6 and app sdk 2.6

[DiCE1904]

bump with settings for 760kh/s

http://i.piccy.info/i7/eef9270eec21626fce47de4a5b0bfafe/4-47-398/10015963/760.jpg

using catalyst 12.6 and app sdk 2.6

link isnt working for me