CCminer(SP-MOD) Modded GPU kernels.

[go6ooo1212]

So I share some result with my other 970s ( more OCed) MSI_Gaming and Palit Jetstream...

[bensam1231]

So I share some result with my other 970s ( more OCed) MSI_Gaming and Palit Jetstream...

Include a photo of the same setup running Epsylons?

[go6ooo1212]

So I share some result with my other 970s ( more OCed) MSI_Gaming and Palit Jetstream...

[_img]https://i.imgur.com/2pRBwwl.jpg[/img]

Include a photo of the same setup running Epsylons?

Epsylon's crashing the driver with such clock.

[antantti]

Epsylon's crashing the driver with such clock.

If you drag that power limit slider all the way to the right and memory all left?

-edit- Sorry, with sp_ miner

[bensam1231]

So I share some result with my other 970s ( more OCed) MSI_Gaming and Palit Jetstream...

[_img]https://i.imgur.com/2pRBwwl.jpg[/img]

Include a photo of the same setup running Epsylons?

Epsylon's crashing the driver with such clock.

Yeah you have to downclock quite a bit for Epsylons compared to other miners, not sure exactly why. It pays off though.

[Nikolaj]

So I share some result with my other 970s ( more OCed) MSI_Gaming and Palit Jetstream...

Reference 970 @ 1505 MHz, tpruvot 1.8.1, maximum frequency. 20MHz more it crashes


The tolerance of the frequency could be related to the chip itself, to evaluate the overclocking headroom we should at least compare with the same miner on the same settings and VGA, with at least 3 measurements and average readings. ASIC quality 64%, so technically speaking a lower average frequency (I've got a custom liquid cooler).

With the same frequency it should be almost the same result, because of there's a gap of 86.5MHz on the core.

+15% should be 185.51 MH/s. The interesting thing it's the TDP amount on GPUz, I should test with the wattmeter with the 2 kernels because it seems that it's less stressful than tpruvot 1.8.1

[bensam1231]

So I share some result with my other 970s ( more OCed) MSI_Gaming and Palit Jetstream...

Reference 970 @ 1505 MHz, tpruvot 1.8.1, maximum frequency. 20MHz more it crashes


The tolerance of the frequency could be related to the chip itself, to evaluate the overclocking headroom we should at least compare with the same miner on the same settings and VGA, with at least 3 measurements and average readings. ASIC quality 64%, so technically speaking a lower average frequency (I've got a custom liquid cooler).

With the same frequency it should be almost the same result, because of there's a gap of 86.5MHz on the core.

+15% should be 185.51 MH/s. The interesting thing it's the TDP amount on GPUz, I should test with the wattmeter with the 2 kernels because it seems that it's less stressful than tpruvot 1.8.1

You can't just throw a completely different configuration at a different benchmark and compare the two. The whole reason I asked him for a screenshot running Epsylons. I have 970s and have also ran Epsylons, it's not the same thing.

[antantti]

Afterburner 106% gpuz 123.2%?

[alrami]

So I share some result with my other 970s ( more OCed) MSI_Gaming and Palit Jetstream...

[_img]https://i.imgur.com/2pRBwwl.jpg[/img]

Include a photo of the same setup running Epsylons?

Epsylon's crashing the driver with such clock.

Thank you, go6ooo1212!
Your info is very useful.

[Zekku]

Lots of fire in this thread lately. Haven't been around. My 2x 970s worth mining on nowadays?

[antantti]

Lots of fire in this thread lately. Haven't been around. My 2x 970s worth mining on nowadays?

They do at least 300 units on lbry. Do the math.

[Zekku]

Lots of fire in this thread lately. Haven't been around. My 2x 970s worth mining on nowadays?

They do at least 300 units on lbry. Do the math.

Thanks for the tip. I had been testing the waters on lbry but need to sit down and oc again.
Recently formatted and reset everything to stock.
Zzz..

[go6ooo1212]

@Nikolaj: Which exact model of ZOTAC is that you have , mate: https://www.zotac.com/product/graphics_card/gtx-970/all
If I know right - ZOTAC is te brand with the most OverClockable cards ,which I wish to take a hand on...

[sp_]

With LBRY sp-mod private #1 my gigabyte g1 gaming does around 190mhash @ 1600core. G1 gaming has a tdp of 250watt and 4pin+3pin.
I'll set up a screenshot later when I have time.

My kernel has a 0% memory controller load and can be used in dual mining with etherum. The opensource has 26% memory load and use more power and cannot be used in dual mining.

In release #2 I will include a new intensity switch for dual mining

[sp_]

So I share some result with my other 970s ( more OCed) MSI_Gaming and Palit Jetstream...

Your cards doesn't get enough power. If you reduce the memclock to the maximum you give more power to the core. Run with higher intensities than the default f.ex -i 26 or - 27 and overvolt( increase fhe power limit)

From the screeenshot you can see that the performance cap is power.

[Amph]

you are telling me that i'm talking about assumtions and speculation, when there is no guarantee either that it can be improved, nice argument you have there, you are contraddicting yourself

actually it's the opposite, there is no proof that it can be improvement with a great % of optimization

and no he didn't improve anything beyond 1-2%, he was talking about maxwell card(and i don't care about to be honest), not pascal, read again what he said

WTF are you talking about? He literally said there was a 10-15% improvement and people have been confirming it the last couple pages, including a guy on this page.

So Maxwell doesn't count now? You're trying to make a moving target here. He literally specified that line of cards.

No one knows what you 'care about' until you mention it. It's not my responsibility to read your mind or type out your responses for you.

That aside, I have no idea why you think 10-15% improvement isn't also possible on Pascal, it's very similar to Maxwell and pretty much all other algos except for weird ones (like Ethereum) show about the same improvements from Maxwell to Pascal (around 90%).

it's not possible for pascal only because it was maxwell that was underperforming in fact a 980ti should be very akin in speed to a 1070 and the previous hash was 190MH for it now it's 300 which make more sense since the 1070 is near that too

pascal on the other hand is on its limit on this algo, and again i don't care about maxwell yes, my whole argument was about pascal since the beginning, but you ignored it and because of this you were talking crap thinking that i was wrong

[Nikolaj]

@Nikolaj: Which exact model of ZOTAC is that you have , mate: https://www.zotac.com/product/graphics_card/gtx-970/all
If I know right - ZOTAC is te brand with the most OverClockable cards ,which I wish to take a hand on...

ZT-90105-10P

with this:

http://www.xtremehardware.com/cooling/liquid/recensione-dissipatori-vga-aio-a-liquido-arctic-accelero-hybrid-iii-120-e-hybrid-iii-140-2016053011458/

Later on I'll publish an article with this beauty





I've used this model of 970 in comparison also with 1070s and a 1080, on Neoscrypt/Blake2b/LBRY and Lyra2ReV2, here:
http://www.xtremehardware.com/index.php?option=com_content&view=article&id=11648

You can't just throw a completely different configuration at a different benchmark and compare the two. The whole reason I asked him for a screenshot running Epsylons. I have 970s and have also ran Epsylons, it's not the same thing.

I've just posted a screenshot with all the infos, and I was responding to him. Relax. If you have data to show, it's gladly accepted, but I prefer to let the data speak for me. I suggest you do the same.

Other than that, the platform it's insignificant because of:

- the OS it's still on windows, so no difference
- my VGA is connected to a 16X PCIE slot, I suppose that he've done the same
- the VGA chips are the same
- the driver identical

the only difference it's the kernel typology and frequency, but it's all shown in the images, and frankly speaking if you have got something different, why don't you show it? Give me some data, not speculations. It's pretty clear that the power consumption it's different, just look at the values reported by the softwares.

Afterburner 106% gpuz 123.2%?

That's why the only good reading it's at the wall, with a good watt-meter

With LBRY sp-mod private #1 my gigabyte g1 gaming does around 190mhash @ 1600core. G1 gaming has a tdp of 250watt and 4pin+3pin.
I'll set up a screenshot later when I have time.

My kernel has a 0% memory controller load and can be used in dual mining with etherum. The opensource has 26% memory load and use more power and cannot be used in dual mining.

In release #2 I will include a new intensity switch for dual mining

Interesting, thanks. Looking forward to it, especially for pascal. (you mean 8+6 I suppose, a bit more than mine, reference, with 2x6pin PCIE). With this profitability by the way what we need to seek is efficiency, so pushing to the limit a maxwell card would probably lower the efficiency ratio, expressed in W/MH. The pascal architecture here it's the clear winner, especially the 1070s. The 1080 unfortunately cannot achieve the same results if you downclock them, they are tied to the '70s.

The fact that your kernel seems to be more energy efficient than the opensource variant, it's a clear advantage.

[bensam1231]

With LBRY sp-mod private #1 my gigabyte g1 gaming does around 190mhash @ 1600core. G1 gaming has a tdp of 250watt and 4pin+3pin.
I'll set up a screenshot later when I have time.

My kernel has a 0% memory controller load and can be used in dual mining with etherum. The opensource has 26% memory load and use more power and cannot be used in dual mining.

In release #2 I will include a new intensity switch for dual mining

Sounds good bro, keep hacking away. Consider dropping either the flat fee or the miner fee though, people were right about that. Don't let the haters get you down.

you are telling me that i'm talking about assumtions and speculation, when there is no guarantee either that it can be improved, nice argument you have there, you are contraddicting yourself

actually it's the opposite, there is no proof that it can be improvement with a great % of optimization

and no he didn't improve anything beyond 1-2%, he was talking about maxwell card(and i don't care about to be honest), not pascal, read again what he said

WTF are you talking about? He literally said there was a 10-15% improvement and people have been confirming it the last couple pages, including a guy on this page.

So Maxwell doesn't count now? You're trying to make a moving target here. He literally specified that line of cards.

No one knows what you 'care about' until you mention it. It's not my responsibility to read your mind or type out your responses for you.

That aside, I have no idea why you think 10-15% improvement isn't also possible on Pascal, it's very similar to Maxwell and pretty much all other algos except for weird ones (like Ethereum) show about the same improvements from Maxwell to Pascal (around 90%).

it's not possible for pascal only because it was maxwell that was underperforming in fact a 980ti should be very akin in speed to a 1070 and the previous hash was 190MH for it now it's 300 which make more sense since the 1070 is near that too

pascal on the other hand is on its limit on this algo, and again i don't care about maxwell yes, my whole argument was about pascal since the beginning, but you ignored it and because of this you were talking crap thinking that i was wrong

This is incorrect, a 980ti DEFINITELY does not perform on par with a 1070. It's about 10-15% slower.

Maxwell was never under performing, you never took time to tweak the intensity settings.

Yup, long term veteran coder 'Amph' on the job! He says it so, so it can't be done. You can say whatever you want after the fact dude, now you're just side stepping to not try and look like a moron... Which has already been proven time and again. This is the point where you quote a prior post that showed you were specifically talking about Pascal when the entire thread was about SPs changes which are right now Maxwell and prove me wrong... Oh wait, there aren't any of those? Which means you are wrong.

[Amph]

This is incorrect, a 980ti DEFINITELY does not perform on par with a 1070. It's about 10-15% slower.

Maxwell was never under performing, you never took time to tweak the intensity settings.

i don't have a 980ti or any maxwell, if it was only about intensity then this whole sp-mod is useless, but since it is not, it was underperforming

others also said so after trying this version(there is a screen of someone doing more), as usual bensam shooting crap everywhere no surprise

[Nikolaj]

With LBRY sp-mod private #1 my gigabyte g1 gaming does around 190mhash @ 1600core. G1 gaming has a tdp of 250watt and 4pin+3pin.
I'll set up a screenshot later when I have time.

My kernel has a 0% memory controller load and can be used in dual mining with etherum. The opensource has 26% memory load and use more power and cannot be used in dual mining.

In release #2 I will include a new intensity switch for dual mining

Sounds good bro, keep hacking away. Consider dropping either the flat fee or the miner fee though, people were right about that. Don't let the haters get you down.

Are you considering me an hater? I am just, kindly, discussing with other users. You tend to overdo, heating up, discussions, and frankly speaking it's not necessary .

I bought from sp some releases, and will keep support his work in the future, as every other good dev in the scene