CCminer(SP-MOD) Modded GPU kernels.

[chrysophylax]

@crysx

This seems to be probleem with quark on yaamp. I am the only one mining to this address but it shows 105 miners. Also total quark miners rose to 32K & hashrate dropped to 100 Mh/s, when miner started throwing bad shares with "Invalid job ID" and finally miner was not able to connect with the pool.

Yeah...I had about 30 miners. It was kicking me and reconnecting showing it as a new miner. I just quit it altogether. I think yaamp may be ready to crash again. Seems they can't keep it going over there.

I think yaamp needs to be rebuilt from the bottom up...and 1% pool charges...not 10% 

+1 ...

exactly ...

#crysx

[chrysophylax]

"stock clocks" varies from one brand/version of card to another.

exactly ...

the gigabyte oc versions are actually factory overclocked - but not by much ...

#crysx

I have one by gigabyte  crysx that is factory super overclocked and it's gpu clock reads 1350  750ti sweet spot. Of course it doesn't need to be oc any more.

was that the special edition? ...

i heard that they brought out a superclocked version - but couldnt find it anywhere to be purchased easily ... at least not here in australia that i know of ...

what are the hashrates for the algos that are mined? ... like x11 and quark and such ... if you have tested it on different algos of course ...

#crysx

[joblo]

Yeah, 750tis take a lot of space, even if they're slightly better, which is why I never bought any. I'm currently using 970s. 980s are just a waste of money as they're too expensive for the little bit extra performance they offers (much like the 290 vs 290x back in the day).

I did a hash/price(cad) comparison and the 970 is definitely in the sweet spot Table below). The hash rate
seems to scale linearly with the number of cuda cores, even between the sm 5.0 750ti and the sm 5.2 9xx
series cards. I would have expected the 9xx cards to provide more hash at lower power and I have no
clue why the 750ti is more power efficient. Maybe cuda 7 will show more improvement in sm 5.2.

Consider the time it would take to burn enough power to equal the cost of a card with the 750ti
and 50 watts as an example. The card goes for around CAD 180 and I'll assume CAD 0.10 /KWH
for electricity. $180 would pay for 1800 KWH (1800000 WH).

At 50 W the 750ti would have to run for 1800000/50 hours, or 98 years to spend the same on power
as the cost of the card. Even including the power for a rig with just 1 750ti, about 400 W, would have
to run for 12 years for the power burned to equal the card's cost.

I've never really understood why there is so much attention paid to power consumption when the
cost of the equipment is so much more significant. The only time the GPU cost isn't a factor is for
gamers who already have the HW. Serious miners don't have another use for their HW.

Just my thoughts on the subject.

Here's the full comparison:

980     = 25.7 Kh/s/$
970     = 39.7
960     = 36.0
750ti   = 33.3
TitanX = 22.5

I won't be buying a TitanX anytime soon.

[joblo]

Yeah, 750tis take a lot of space, even if they're slightly better, which is why I never bought any. I'm currently using 970s. 980s are just a waste of money as they're too expensive for the little bit extra performance they offers (much like the 290 vs 290x back in the day).

I did a hash/price(cad) comparison and the 970 is definitely in the sweet spot Table below). The hash rate
seems to scale linearly with the number of cuda cores, even between the sm 5.0 750ti and the sm 5.2 9xx
series cards. I would have expected the 9xx cards to provide more hash at lower power and I have no
clue why the 750ti is more power efficient. Maybe cuda 7 will show more improvement in sm 5.2.

Consider the time it would take to burn enough power to equal the cost of a card with the 750ti
and 50 watts as an example. The card goes for around CAD 180 and I'll assume CAD 0.10 /KWH
for electricity. $180 would pay for 1800 KWH (1800000 WH).

At 50 W the 750ti would have to run for 1800000/50 hours, or 98 years to spend the same on power
as the cost of the card. Even including the power for a rig with just 1 750ti, about 400 W, would have
to run for 12 years for the power burned to equal the card's cost.

I've never really understood why there is so much attention paid to power consumption when the
cost of the equipment is so much more significant. The only time the GPU cost isn't a factor is for
gamers who already have the HW. Serious miners don't have another use for their HW.

Just my thoughts on the subject.

Here's the full comparison:

980     = 25.7 Kh/s/$
970     = 39.7
960     = 36.0
750ti   = 33.3
TitanX = 22.5

I won't be buying a TitanX anytime soon.

My power numbers are wrong. I worked it backward and came up with about 4 years. I'm tired,
will figure it out tomorrow.

[joblo]

Yeah, 750tis take a lot of space, even if they're slightly better, which is why I never bought any. I'm currently using 970s. 980s are just a waste of money as they're too expensive for the little bit extra performance they offers (much like the 290 vs 290x back in the day).

I did a hash/price(cad) comparison and the 970 is definitely in the sweet spot Table below). The hash rate
seems to scale linearly with the number of cuda cores, even between the sm 5.0 750ti and the sm 5.2 9xx
series cards. I would have expected the 9xx cards to provide more hash at lower power and I have no
clue why the 750ti is more power efficient. Maybe cuda 7 will show more improvement in sm 5.2.

Consider the time it would take to burn enough power to equal the cost of a card with the 750ti
and 50 watts as an example. The card goes for around CAD 180 and I'll assume CAD 0.10 /KWH
for electricity. $180 would pay for 1800 KWH (1800000 WH).

At 50 W the 750ti would have to run for 1800000/50 hours, or 98 years to spend the same on power
as the cost of the card. Even including the power for a rig with just 1 750ti, about 400 W, would have
to run for 12 years for the power burned to equal the card's cost.

I've never really understood why there is so much attention paid to power consumption when the
cost of the equipment is so much more significant. The only time the GPU cost isn't a factor is for
gamers who already have the HW. Serious miners don't have another use for their HW.

Just my thoughts on the subject.

Here's the full comparison:

980     = 25.7 Kh/s/$
970     = 39.7
960     = 36.0
750ti   = 33.3
TitanX = 22.5

I won't be buying a TitanX anytime soon.

I have no idea how I got the power figures above but I'll try again.

At 50 W a 750ti will burn 50*24*365 watts hours in a year, or 438 KWH. The cost of the card will get
you 4.1 years of power for it.

A 400 W rig with 1 750ti would burn 350 KWH so $180 would only last about half a year.

Even a 6 x 750ti rig has about 50% power overhead so tweaking the power efficiency of the
miner would have little impact on overall power consumption. More hash is always better.

[hashbrown9000]

with my gigabyte 750ti cards, (has an extra 6-pin), using release 50 on quark, it takes an extra 80 MHz of overclock to reach 6.0 MH/s . +200 MHz gets me to 6.5 MH/s. anything past that and the miner will crash.

wish i could find those gainwards. google and newegg come up dry. but i do have my eye on some evga 750ti cards, as they seem to get a bit more hash, and are now cheaper to boot.

i can also confirm SP_'s power increase. From stock clocks (5.5 MH/s) to full OC (6.5 MH/s) there is about a 15 watt increase per card.

[bathrobehero]

SP-mod release 50 Mining quark

The non powered

http://www.gainward.com/main/vgapro.php?id=926&

Draws 40-45 watt. and does 5750Khash on standard clocks.

the powered (6 pin)

http://www.gigabyte.com/products/product-page.aspx?pid=4948#ov

draws 60 watt and does 5550Khash on standard clocks.

But if you reflash the bios to the black edition, the hashrate is 6100 and with overclocking I can reach 6500. (The gainward is not able to reach these speeds because it needs more power)


so 1 MHASH more with 15-20 watt more consumption.. But if you don't overclock you loose the 1MHASH

Those figures cannot possibly be right. The Gainward card you've mentioned also have a TDP of 60 watts regardless of 6-pin or not (source). The PCI-E slot can provide 66W alone (5.5A on the 12V rail and 3A on the 3.3V rail). Additionally 6-pin can provide 75W and 8-pin can do 150W. The 38.5W TDP figure from the BIOS doesn't mean what we thought it would. Non-powered cards can pull 60 watts as well with stock BIOS. I both have ASUS and MSI non-powered cards and they do pull 60W. Besides, wherever you look even gaming benchmarks show close to 60 watt usage for all 750 Ti's.

Increasing the TDP should have zero effect for stock clocks, all it does is give headroom for overclocking. And in my case at least, overclocking is always more efficient as in it always scales better than power consumption. For example +150 Mhz on the core is 12% gain in hashrate and 11% increase in power consumption at the wall for x11, and it's +10%/+7% for quark and so on. Conversely, with downclocking the efficiency gets worse.

[pallas]

in general, overclocking should bring better power efficiency, unless its effect is counterbalanced by automatic or manual overvolting, or by additional power draw from the cooling system (fans).

[sp_]

Those figures cannot possibly be right. The Gainward card you've mentioned also have a TDP of 60 watts regardless of 6-pin or not (source). The PCI-E slot can provide 66W alone (5.5A on the 12V rail and 3A on the 3.3V rail). Additionally 6-pin can provide 75W and 8-pin can do 150W. The 38.5W TDP figure from the BIOS doesn't mean what we thought it would. Non-powered cards can pull 60 watts as well with stock BIOS. I both have ASUS and MSI non-powered cards and they do pull 60W. Besides, wherever you look even gaming benchmarks show close to 60 watt usage for all 750 Ti's.

Increasing the TDP should have zero effect for stock clocks, all it does is give headroom for overclocking. And in my case at least, overclocking is always more efficient as in it always scales better than power consumption. For example +150 Mhz on the core is 12% gain in hashrate and 11% increase in power consumption at the wall for x11, and it's +10%/+7% for quark and so on. Conversely, with downclocking the efficiency gets worse.

I forgot to add the idle watt. The card use 7-10watt idle..

I have implemented --gpu-memclock and --gpu-engine parameters in ccminer now.
Will messure the watt and get back to you.

[chrysophylax]

Those figures cannot possibly be right. The Gainward card you've mentioned also have a TDP of 60 watts regardless of 6-pin or not (source). The PCI-E slot can provide 66W alone (5.5A on the 12V rail and 3A on the 3.3V rail). Additionally 6-pin can provide 75W and 8-pin can do 150W. The 38.5W TDP figure from the BIOS doesn't mean what we thought it would. Non-powered cards can pull 60 watts as well with stock BIOS. I both have ASUS and MSI non-powered cards and they do pull 60W. Besides, wherever you look even gaming benchmarks show close to 60 watt usage for all 750 Ti's.

Increasing the TDP should have zero effect for stock clocks, all it does is give headroom for overclocking. And in my case at least, overclocking is always more efficient as in it always scales better than power consumption. For example +150 Mhz on the core is 12% gain in hashrate and 11% increase in power consumption at the wall for x11, and it's +10%/+7% for quark and so on. Conversely, with downclocking the efficiency gets worse.

I forgot to add the idle watt. The card use 7-10watt idle..

I have implemented --gpu-memclock and --gpu-engine parameters in ccminer now.
Will messure the watt and get back to you.

now THIS is watt ( pardon the pun ) we have been waiting for ...

#crysx

[hashbrown9000]

lol. watt's up doc ?!

[ltc_bilic]

found new bug: -g parameter breaks "Extranonce subscription" - can be verified on nicehash,....was wondering why my one of my miners had it checked the other one was missing, but same config/hardware/software 

sp thank you once again for you're hard work,...will send some beers into you're btc wallet 

[chrysophylax]

uploaded to git yet sp? ...

getting late here and was just wondering ...

wanted to try it before i slept ... can wait for tomorrow ...

...

#crysx

[rednoW]

found new bug: -g parameter breaks "Extranonce subscription" - can be verified on nicehash,....was wondering why my one of my miners had it checked the other one was missing, but same config/hardware/software 

It's not a bug. It was implemented in such way by sp_. He tries to fix sum bugs in -g mode with this trick ... but for now -g mode has no pros.
sp_ still giving advice to run with -g in solo-mining mode ... but ...

[CapnBDL]

in general, overclocking should bring better power efficiency, unless its effect is counterbalanced by automatic or manual overvolting, or by additional power draw from the cooling system (fans).

So with all this going on with these 750Ti cards, I take it that ovr clocking is the best thing to do? BTW, I just wanted to check so I pulled the cover, my card ASUS GTX750Ti 2g is the one /w the 6pin. I'm gonna guess that's makes up for a power drop from the dual fans? (I have seen what look like short card 750Ti's on fleaBay /w one fan). I don't have an accurate meter to do any testing on it. This card also seems to be hardcoded for automatic, BIOS voltage control. It fluctuates /w 1143mV being top range. I've tried AfterBurner and ThunderMaster but you still can't do anything about voltage control. I just ignore that setting (it don't change a thing) /w GPUTweak. I still get close to 6.5 /w quark and ~3.44 /w x11. I haven't tried any other, not sure if it will mine another coin.

Oppsss...got off topic..backup. OCing is best? There's the question.

Thank you

ps. This card was pretty cheap too...$89 on...I think it was newegg, but that was a while back.

[bensam1231]

SP-mod release 50 Mining quark

The non powered

http://www.gainward.com/main/vgapro.php?id=926&

Draws 40-45 watt. and does 5750Khash on standard clocks.

the powered (6 pin)

http://www.gigabyte.com/products/product-page.aspx?pid=4948#ov

draws 60 watt and does 5550Khash on standard clocks.

But if you reflash the bios to the black edition, the hashrate is 6100 and with overclocking I can reach 6500. (The gainward is not able to reach these speeds because it needs more power)


so 1 MHASH more with 15-20 watt more consumption.. But if you don't overclock you loose the 1MHASH

Those figures cannot possibly be right. The Gainward card you've mentioned also have a TDP of 60 watts regardless of 6-pin or not (source). The PCI-E slot can provide 66W alone (5.5A on the 12V rail and 3A on the 3.3V rail). Additionally 6-pin can provide 75W and 8-pin can do 150W. The 38.5W TDP figure from the BIOS doesn't mean what we thought it would. Non-powered cards can pull 60 watts as well with stock BIOS. I both have ASUS and MSI non-powered cards and they do pull 60W. Besides, wherever you look even gaming benchmarks show close to 60 watt usage for all 750 Ti's.

Increasing the TDP should have zero effect for stock clocks, all it does is give headroom for overclocking. And in my case at least, overclocking is always more efficient as in it always scales better than power consumption. For example +150 Mhz on the core is 12% gain in hashrate and 11% increase in power consumption at the wall for x11, and it's +10%/+7% for quark and so on. Conversely, with downclocking the efficiency gets worse.

So I'm actually curious why everyone talks about the TDP slider all the time. Why would it matter what the TDP slider is set at? It should just be set at the highest value just like with AMD cards. It wont use more power unless you turn up the mhz and even then it'll eventually crash unless you add more voltage. It just seems like a unnecessary variable as the TDP slider doesn't really increase efficiency if you decrease it as it will decrease your clockrates too, unless I'm mistaken.

What's the point of using the TDP slider over clocks/overvolts? Is it just my card that has a unlocked voltage modifier?


A extra six pin power connector doesn't necessarily mean it uses more power either. It can have a power connector and have lower core voltage, but uses the connector for cleaner power delivery instead of trying to pull everything through the bus. If it has a higher TDP, that also means it has higher core voltage, although it's not a absolute unit of measurement as each unit can be different (between bioses and versions).

[CapnBDL]

I give up...all I know is that it works. Here's a great question; since all these new mods have been introduced, is anybody gonna take the time to update a readme file so we know wtf is going on /w the new switches?


edit; if the readme had some suggested parameters would really rock.

Tell ya what...if I get help from this forum, I'll do it. If everyone would test 1 or prefer 2 algo's and just send 'em to me /w what card or cards they tested. And if the devs will send me what the new switches are, what they do, etc., I'll take the current readme and compile a new readme /w a git posting. I could even add it to the .zip, I think. I know zip will allow adding a file so...what do ya say...anybody in for this? Like a concise version.

[CapnBDL]

Wheeew! Just tried Lyra and my vid driver got blown outa da water!! Anyone got some rough seting for Lyra2RE?

[joblo]

I give up...all I know is that it works. Here's a great question; since all these new mods have been introduced, is anybody gonna take the time to update a readme file so we know wtf is going on /w the new switches?


edit; if the readme had some suggested parameters would really rock.

Tell ya what...if I get help from this forum, I'll do it. If everyone would test 1 or prefer 2 algo's and just send 'em to me /w what card or cards they tested. And if the devs will send me what the new switches are, what they do, etc., I'll take the current readme and compile a new readme /w a git posting. I could even add it to the .zip, I think. I know zip will allow adding a file so...what do ya say...anybody in for this? Like a concise version.

Good idea but I'm not sure the ccminer readme file is the best place for card specific data. The data will have to
be maintained as new cards are introduced. If ccminer development slows the data wil become obsolete pretty fast.
There is already some very obsolete information in the readme and help referring to coins that are long dead.

What you're suggesting sounds like a small database considering the number of algos times the number of different
cards. Even within one model and board manufacturer there are several variations of memory and clock speeds.
Keeping it out of ccminer allows it to be updated independently of ccminer development.

I suggest starting out by compiling the current data and posting here. If you or someone else feels ambitious
they could put it in a database and put it up on a web site.

[CapnBDL]

I give up...all I know is that it works. Here's a great question; since all these new mods have been introduced, is anybody gonna take the time to update a readme file so we know wtf is going on /w the new switches?


edit; if the readme had some suggested parameters would really rock.

Tell ya what...if I get help from this forum, I'll do it. If everyone would test 1 or prefer 2 algo's and just send 'em to me /w what card or cards they tested. And if the devs will send me what the new switches are, what they do, etc., I'll take the current readme and compile a new readme /w a git posting. I could even add it to the .zip, I think. I know zip will allow adding a file so...what do ya say...anybody in for this? Like a concise version.

Good idea but I'm not sure the ccminer readme file is the best place for card specific data. The data will have to
be maintained as new cards are introduced. If ccminer development slows the data wil become obsolete pretty fast.
There is already some very obsolete information in the readme and help referring to coins that are long dead.

What you're suggesting sounds like a small database considering the number of algos times the number of different
cards. Even within one model and board manufacturer there are several variations of memory and clock speeds.
Keeping it out of ccminer allows it to be updated independently of ccminer development.

I suggest starting out by compiling the current data and posting here. If you or someone else feels ambitious
they could put it in a database and put it up on a web site.

Well, as long as the forum helps out it really wouldn't be all that hard to put it in a spreadsheet type of thing. But that's going very deep. I was more or less looking at updating the readme that's with currently. It doesn't have to be exact. And be put in a concise form, similar to what is there. Get rid of the old crap, add the new mods, switches, etc. and a short form of, not recommended or highest, but starting places for people. EX. What just happenned to me...give me a rough idea of where to set a card for Lyra2RE......edit; we could easlier lowball  the numbers. Even newbies know this is just a starting place, not all emcompassing. (fat fingers..lol)