LTC mining, GPU to Memory ratio?

[spinx]

Hi, i'am trying to optimize my new rig, right now im getting about 530Kh/s per card (7950's). I have read alot and trying to find the best settings, and once and for all, does LTC mining benefit from memory oc?

I have tried to following clock settings to find the "sweet spot" aka ratio between gpu and memory clock (0.57-0.60???)
Theese are my results so far:

825 / 1650 = 478 Kh/s (ratio 0,50)
850 / 1650 = 482
875 / 1650 = 501
900 / 1650 = 505
925 / 1650 = 510
950 / 1650 = 517
975 / 1650 = 525
1000 / 1650 = 529 (ratio 0,60)

No sweetspot found...

This is my bat file:
cgminer.exe --scrypt -o stratum+tcp://eu.wemineltc.com:3333 -u xxx -p xxx --shaders 1792 --intensity 20 --worksize 128 -g 1 --auto-fan --gpu-engine 1000 --gpu-memclock 1500 --gpu-powertune 5 --temp-target 75 --thread-concurrency 26624

*26624 is the highest thread-concurrency working.´
*intensity lower than 20 drops the Kh/s 50-75% to 200-250Kh/s.

Any input?
Thanks!

[logicus]

For 7950 sweetspot  is ~0.7 -0.8 
Gigabyte 7950 GV-R795WF3-3GD
1140 core and 1420 memory    608kh/s  1.09v
cgminer --scrypt -o stratum+tcp://stratum.give-me-ltc.com:3333 -u xxx -p xxx --shaders 1792 --thread-concurrency 10240 -I 13 -g 2 -w 256

[andrewsg]

Agreed with the above, I undervolt and run 925 core / 1250 mem - around 555/560kH per card. (ratio 0.74) (not that I particularly recommend that ratio, I haven't spent that much time optimizing, but just as a reference because I am running slower clock speeds compared to your numbers and still getting better results.)

Please do post your final settings if you can improve.

[lewlzminer]

1050 1500 = 640khs

[spinx]

1050 1500 = 640khs

Thank you i will try, so far i have not found the sweetspot, more like higher core = higher Kh/s.

1015 / 1450 = 533 Kh/s (0,700 ratio)
1025 / 1450 = 537
1030 / 1450 = 540
1035 / 1450 = 545
1040 / 1450 = 547
1045 / 1450 = 550
1050 / 1450 = 552
1055 / 1450 = 553 Kh/s (0,721 ratio)

1050 / 1500 = 559

I know the cores can do 1100+ and memory's 1650+ but i need better cooling. And with 1045 / 1450 the whole setup consumes about 1280W at the wall... Guess my 1250W PSU's crying

[trenal]

I am doing 1080 | 1500 with stock voltage on Sapphire 7970s and getting around 750kH/s on each of my 4 cards.

[spinx]

I am doing 1080 | 1500 with stock voltage on Sapphire 7970s and getting around 750kH/s on each of my 4 cards.

That about 0,72 ratio, but the 7970's has more shaders than my 7950's.

[tom_o]

I am doing 1080 | 1500 with stock voltage on Sapphire 7970s and getting around 750kH/s on each of my 4 cards.

That about 0,72 ratio, but the 7970's has more shaders than my 7950's.

Yeah I get highest hashrate at 0.7233333*r

1085/1500 Gigabyte WF 7970 718kH/s

[lewlzminer]

maybe try my settings, cant imagine how you get only 5xx khs 

"intensity" : "20",
"vectors" : "1",
"worksize" : "256",
"kernel" : "scrypt",
"lookup-gap" : "0",
"shaders" : "1792",
"thread-concurrency" : "16384",
"gpu-engine" : "850-1050",
"gpu-fan" : "100",
"gpu-memclock" : "1500",
"gpu-memdiff" : "450",

[HuuHachu]

Seems similar to my problem ...

I have a sapphire 7950.
There is a"sweet spot" at 850 core / 1250 memory, which gives ~510KH/s

Almost any other frequency go much lower than that ...
850 / 1251, 851 / 1250 or variations give ~450KH/s

same goes for 925/whatever, only 450KH/s

I have to substantially increase the frequencies (like 1100/1400) to reach 550 KH/s.

I usually mine under linux (13.4 driver with APP 2.. I observed that my card switchs between only two voltages (and ignores the one I want to force).
Up to 850/1250 included, it runs on 1.125V. Beyond that (even 1MHz for core or memory), it jumps to 1.250V
(I used a wattmeter and I effectively observed a clear jump in consumption over 850/1250)


Under windows, using trixx, I'm able to tune the voltage as I want. It allows me to reduce significantly the consumption (at 850/1250, I can go bellow 1V) ... but the hash-rate behaves exactly like under linux.

I'm considering a bios update ... but don't know which one would be better ...

[Edvin512]

7950 850/1250 = 525
        950/1250 = 575

for me

[Amph]

try int 19 and currency 8192 or 24000, also worksize 256

[HuuHachu]

try int 19 and currency 8192 or 24000, also worksize 256

At --thread-concurrency 8192 -I 13, I get something like 350KH/s  (850 core / 1250 memory)
Higher intensities (up to 18, after that I does not increase hash-rate) allows me to go up to 570KH/s ... BUT at the same time HW errors start to appear (a few at -I 14, but at -I 18, maybe 80% !)

Ha,  also found a "sweet spot" for thread intensity at 42000 ... (at that level I got the same performances as at 22000). At that level it consumes almost all the video ram (42000 * 128KB / 2 ~= 2.6GB)
(divided by 2 because of --lookup-gap 2)

[aggster]

I get 550kh/s with cgminer with intensity 17, but my pc runs real slow when mining and get allot of artifacts, don't think this is good for me gpu, is this normal when hashing at 550?

[computerparts]

I get 550kh/s with cgminer with intensity 17, but my pc runs real slow when mining and get allot of artifacts, don't think this is good for me gpu, is this normal when hashing at 550?

If you're getting artifacts then you aren't stable.

[aggster]

any suggestions for stabilising and maintaining speeds?

[computerparts]

any suggestions for stabilising and maintaining speeds?

increase voltage until it's stable but make sure temperature doesn't go too high.

[add1ct3dd]

This thread is full of wrong, the 7950's dont have the same sweet spot 7970's do,

Try 1150 / 1250 @ 16384 TC, i19. That should get you 660-680.

You can thank me later.

[computerparts]

This thread is full of wrong, the 7950's dont have the same sweet spot 7970's do,

Try 1150 / 1250 @ 16384 TC, i19. That should get you 660-680.

You can thank me later.

No two cards are alike. They are ALL different and optimized at different settings.

[spinx]

Now im running

cgminer.exe --scrypt -o stratum+tcp://eu.wemineltc.com:3333 -u xxx -p xxx --shaders 1792 --intensity 20 --worksize 256 -g 1 --auto-fan --gpu-engine 1100 --gpu-memclock 1500 --gpu-powertune 10 --temp-target 75 --thread-concurrency 26624

this gives me 610 Kh/s per card.

cant really go higher on any parameter, any ideas?