Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 589. (Read 2347641 times)

4690k on that rig.

Latest sp-mod#2 does 175-180MH with tdp limited 8-pin card, I might start playing with bios if it is boring night in cryptoland.

What cpu do you have?  thx

265 *1.05 = 278.25 (on average, not considering too high settings, that are simply useless in a multigpu environment given the absence of rock solid stability)

@bensam: you're not the only one with a "decent" amount invested in cryptos, but there's also a thing callededucation and mutual respect. I understand your frustration in such discussions, but it should be better if you both ignore each others; it's a nonsense to continue like this in every thread you both collide.

@sp: would you be so kind to post hashrates with a human, achievable, frequency? A common 1070, even custom, barely reaches 2000MHz in stability without thermal and power throttling (in a multigpu setup obviously, due thermal complications). Interesting findings about the memory controller btw. Have you measured the difference between the reference values (TDP 150) and  TDP to 170 + forced p0 state? Such tests would be much more interesting with watt measures. What's you model of 1070?

Thanks

My card is trottleing at high speeds I only have a 8pin connector and no powered riser.
My kernal #2 does 293 at a 2100 core and -1000 on the memory (1400)(with nvidia inspector). Tdp changed with the nvidia-smi tool to 170 (limited in the bios) and p0 state.

Anybody know a good tool to reduce the memclock to 500 on the 1070?

sp can you post screen about 1070 +5% boost, which would put a 1070 around 300 already

Run with f.ex -a lbry --dual 32 -i 20

And in the genoils ethminer .bat you add --cuda-schedule auto

Dual mining is under developement, and the --dual doesn't work properly in #2.   (cannot set intensities for card 0 in the rig., shedule flags are wrong. etc)
I will improve it in release #3

selling... not releasing

this current version that sp is releasing for fee + donation

i got 1070 gainward
i thought -b 0.0.0.0 q working in first 4 mins...after that i can see hashrate of each card in rig... I want to see only sum of hashrates not single card hash, thats why I was using -b 0.0.0.0 q
btw,
can you give us example for dual mining?
thanx

#2 gave me some headroom to play with. My 8-pinners were at 165-170 area with #1, this one lets them do 175-180 before power limiters kick in.

remove the -b 0.0.0.0 q it has no effect.

If you run ccminer --benchmark it will exit after 4 minutes.

There is a bug in the intensity and dual intensity. The Gpu #0 intensity will not be set. I will fix in release #3.
What cards do you have?

Compared to what version? SP already mentioned he's working on Pascal performance... we already know that, that's nothing new nor a 'down point'. A 10-15% improvement is under performing? And I have problems with reader comprehension?

You make it true every time you open your mouth, just like with the Ethereum fix, now believing there is a 1-2% improvement with cards, which was disproven before you even mentioned it... Now you're desperately trying to dance around the whole issue to make it seem like you're right.

"really a waste of time talking with you" Hey look, something I said about you weeks ago... Vomiting it back up. Only reason I took you off ignore is because you keep throwing inaccurate information around that completely muddles the conversation and confuses people that are talking in threads.

You want to be a big boy, grow the fuck up and fact check your information before you open your pie hole... And if it's something you can't prove or disprove, use some common sense and look at what information is available (like a vetted and well known developer like SP saying he's capable of doing something and has even shown that he has done it (such as like with Maxwell) and other well known members in the thread confirming what he's saying).

It doesn't matter what Pallas, Wolf, Epsylon, any other developers say in passing about a algo they think is maxed out. Just the same as if SP says 'well no one can improve Cryptonote' for example and Wolf says 'I am working on a kernel that improves things by 25%', I'll believe Wolf0 over SP. Developers aren't god, they all don't pull from the same collective knowledge. Each of them has their own unique take on coding and code things differently, improving things over each other or hitting brick walls and thinking 'that's all that can be done', then another comes along like Nanashi... for a algo we thought was maxed out and doubles the performance.

It honestly looks like you're confusing yourself. You don't even remember what you were talking about in your last response. Go back a page and reread what you already wrote and what I responded to.

SP,    -b 0.0.0.0 q    doesn't work after 4 mins in this miner... also i am getting better hashrate, but with more power...so improvement can be achieved with OC only. others?

it's underperforming compared to this version, jesus you are not even able to read right, really a waste of time talking with you

and again with this "regurgitating" no sense, keep repeating it like a retard will not make it true

So you don't have either GPU you're talking about, you're just randomly guessing based on random shit you've seen (regurgitating), which you aren't very good at... I see...

Yes, Maxwell is apparently underperforming because you say so... Check... You're not even offering reasons why. The whole reason I mentioned intensity, is due to you somehow believing maxwell is underperforming? Do you even know what underperforming means? It means it's not performing where it should be... And that's simply not true. 150Mh/s for the normal miner for a 970 and 250Mh/s for 1070 is exactly where it should be. Comparatively, Pascal actually looks to be underperforming with Epsylons miner.


That was a general statement.

That aside, too bad if you don't like me 'heating things up', I have to deal with shit like Amph who just randomly piss information of no value or merit around the thread and mislead people. It doesn't help keep things organized... Hence his 'Ethereum Pascal fix' that never happened, matter of a fact WDDU didn't even fix things for W7/8, it was a Nvidia driver update at the beginning of August and was told to 'upgrade to W10 because W8 and W7 are dead'. Guess what? Still using W8 and have no intention of changing over to W10 as there are still issues with mining on W10... The WDDU upgrade didn't fix all the problems.

I have a decent amount of money invested in cryptos, as I'm sure quite a few other people around here also do, sorry if I take things seriously. I don't like listening to kids spout random shit they have no idea about, completely fucking with others that are trying to run their businesses because they want to be the guy that people listen to.

I bet he's still recommending people mine Lbry... Basically Claymores Dual miner has been a replacement since it's release. Lbry on occasion earns more, but the markets for Dagger/SIA/DCR are much more reliable.


Yup, I've noticed this and mentioned it a couple weeks ago. Apparently while getting lost in the Lbry craze, people didn't take time to look at other miners and see what sort of TDP they use (Lbry is actually relatively conservative power wise). Almost no one is mentioning that they're running into the TDP cap. ~110 is too low, which is what most 1070 GPUs have for a maximum... Unfortunately. Lbry sits neatly under 110 for most GPUs, a couple I have even bump against the TDP cap and throttle a tiny bit with Epsylons, it's really hard to notice unless you look for it.

Good job on removing the MCU usage. That frees up avenues for a dual miner.


You increased the dev fee to 3.3%? What happened to 2%?


Yeah, only you are getting that.

My 1070 can do 290mhs with opensource, when will you beat this result, sp? ))

LBRY sp-mod #2 has been sendt to the donators.

-Faster hashing on low power cards (1x8pin 2x6pin) gtx 1070 +5%
-Faster hashing on high power cards
-Added experimental --dual parameter for kernel tuning. (3-1024)
-Benchmark can now be tuned with -i and --dual so you don't need to wait for the donate to finish.
-Donation has been changed to 4minutes for me and 116 minutes for you. (less connections to the pools)
-bugfixes

Aaah. Okay. I've just been using the small amount I do make to order upgrades to my pc lol. Motherboard next then I'll just pay cash for cpu since either way with only 2 cards it takes a very long time.

This was explained a few posts ago by antantti. The card doesn't have enough power. because 1 8pin or 2 6pin is not enough. Some 970 and most 980/980ti cards have 4pin + 3pin power cables, these cards will give you 10-15% boost on standard clocks and up to 30% with OC.
The same problem is on the 1070 cards. Running with -i 23 is faster than the default (-i 24).

I am improving it in version #2

on my 1070 card it help to downclock the memory..

My Palit 980(without Ti)  Super Jetstream is giving me 212-215MH @1653MHz, but it is working stange with the new miner.