Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 132. (Read 2347599 times)

If the miner crash, there is something wrong with your setup. My rigs are running stable for weeks, If a rig starts to crash, I troubleshoot at once, because if not, the board or risers can be fried. Power issues are normally the cause of the failures. PSU too weak / power risers draw too much power, cards are to close to each other and create to much heat. etc

For 2% fee you should mine 4+ hours for those 5minutes at start. If the miner crashes earlier - than it is 3-4-5-10...% devfee.

Read the post for the user Wacko a few posts back. You need to change the voltage curve of your cards. In afterburner select a card and press ctrl+f. With proper tuning, your boost clock won't fluctate so much anymore.

In bitcore the order of the 8 last hashing algos are mixed in a "randomish" order. This cause the boost clock to be confused because the load on the gpu is different. Note that in the sp-mod #4 the maxhashrate will vary depending on the order of the algos. If you run my mod for a longer period of time, you will see that the MHASH sometimes increase by 0.5-1MHASH (gtx 1070) and stay there for a while until the algo order is being switched again..

6 x 1070 with 100% tdp - never crash. I even used to put them to 110% tdp but decided to lower it. Very stable.

Almost all my rigs are W8, I haven't updated them nor had a reason to. W8 used to get better hashrate then W10 in some specific algos, not so much anymore, but I'm not going to take the time to update them if there is no reason.

I have EVGA, Gigabyte, Asus, Zotac, PNY, and MSI. The MSI cards are overvolted from the factory so they almost never crash (and use more voltage). The EVGA cards seem to have the most problems dealing with bitcore and the higher intensities. Although it seemed as though randomly machines would crash even if stable for a few days.

Like I mentioned it has to do with the clocks fluctuating up and down. Bitcore is the only algo that does this. This means the card doesn't have time to settle on a lower clock and voltage that is completely stable as Nvidia BOOST applies more voltage at higher clocks and steps back when it's not stable. However, since the GPU load flucatuates so much it never has time to reach a completely stable clockrate (even at stock). This will vary between manufacturers and even between different models in the same product line (like 1070). They use different voltage/clock ramps.

A way around this is if you could find a way to maintain a static workload on the GPUs which would have little variance when cycling between the algos (10 if I remember right) in bitcore.


Something is holding your cards back then, i 20 on #2 will give around 19.3 and i 22 will give around 20.6 give or take on #4. My cards are OC'd although the clocks are all over the place on this algo so it's tough to compare that. Roughly 1900mhz and they all run at the highest TDPs available to the cards.

It's a 2% fee, I said earlier it wasn't as I thought it was based off of a hour, it's not.

exactly all my rigs are 6x 1060 and i cant put stock setting or i get crash always. with tdp 70% work as well 24/24 never stop 11/12mh each card

Sounds like a power issue to me.. The single 4pin is not giving enough power to the card. If it is a power issue. 1 or 2 cards will work fine with 100% tdp at full speed, and 6 cards will fail.
The Bitcore kernel is using alot of power compared to other algos..

you have to use low tdp (70%)or you will get crash

What is the error message in the crash. Misalligned adress? Or is it just a reboot?
If you use -i 25.2, you get more speed..

i use official pool and always is precise

bitcorepool.cc

But did you check the biggest pools. Look at this yiimp graph (last 24 hours). For some reason some of the biggest pools are not paying you as much as you should have got...

Once again. 1080 - SP4 crashes on higher intensities with the same OC. I need to lower OC => No advantage
1080ti - SP4 is 3-4% faster on -i25. But all this advantage is eaten by ridiculous 5minutes devfee at the beginning.

So, don't lie.

i find that the fee kill the advantage of the spmod 4 over the spmod2 on lower tdp like 70%, maybe you can work more on that with the 5...

Open gpuz and check the memory usage of the two exe files.
SP2 is using more memory than SP4 on the same intensity setting.
This meens that if you use SP4  you need to increase the intensity to have the same memory usage as SP2.

Comparing the intensities between two kernels with different memory multipliers is a waste of time... As you can see in my example, the default intensity setting of the sp-mod #4 on the gtx 1070 give you 13.4% more hash than running with -i 20.

And here we come to the problem of many rigs. You need to remember each profile, each algo it is capable running with etc. You have 5 profiles at least 4 types of rigs and ~20 algos.
That's why it is impossible if you switch coins frequently and have many rigs.
I just set maximum OC that is capable of mining all algos without need to change OC.
Yes if sometimes I mine some coin for rather long period I finetune the OC.

I don't mine one algo. There are 5 "slots" for different profiles in MSI AB. I have a different set of curves saved for each slot. All it takes is just to select a different profile before switching to another algo.

Still this is only manageble if you mine one algo. I for example switch coins couple times a week and dont remember when I was mining the same coin for 14 days. Each algo stresses your card differently so the curves are not transferable unless you set it really low which kind of nulifies the effort.

If it takes you a lot more than 30 seconds then you're doing it wrong. With curves in MSI AB all you need to do is set one point precisely, smth like [email protected]. All the other points you just drag superfast down to random clocks, it doesn't matter which, they simply all need to be lower than the only clock you actually set. Then you hit "apply" and your curve is automatically made straight by the program. I've just done it with one of my cards in the test rig, and it took me ~ 30 seconds. Again, if it takes you a lot more than you're either doing it wrong or might want to learn how to handle a mouse. I mean, there's plenty of people out there who actually drag all the points to make a straight line or something. That takes a lot of time and is completely unnecessary.

As to applying the same curve to different cards — I don't know whether it's possible. Doesn't make much sense to me since all the cards are different and curves are by their nature designed for precise tuning of each card. Those that don't care about this kind of precision might just go with +/- clocks/limit adjustment, which can be synced between all the cards in the rig.

As I said before sp4 on 1080ti with i25 on same settings is only 3-4% faster then sp2.
On same intensities sp4 is worse than sp2. But yes it is capable of running on higher intensities... but as with my 1080 rig I need to lower OC in order to make it work. Which leads to even less advantage.
So taking insane 5minutes devfee into account at the start sp4 will make less for the miner.

You can talk fairy tales about 30 seconds to someone else. It is 3 clicks and vs several dosen . But anyway it is possible to set curve and really save it and spread on other cards by editing file directly.