Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1089. (Read 2347601 times)

Joblo have you changed the intensity setting for the 980?  The default might be too low for the 980.

you're doing something wrong... the 980 should do around 1.5MH/s (at least with my version... not following much what is happening here)

So your OCed 750 with 1 GB of RAM, my non-OCed 980 with 4 GB, and an i7-4790k can all
hash about the same, but with significantly different TDP profiles. It looks like the 750 has
the best TDP.

I am disappointed with the performance on the 980. I'll have to do some wider testing to see
if any other algos show this non-linearity among different cards.  It will complicate profit
switching if the card model has to be taken into account. I had wrongly assumed that all
algos would scale linearly and just used the 750ti as a reference.

>> I stand corrected. I couldn't get over 900K on a 980
My junior gtx750 can do 890khs lyra with sp_'s #50 heavily oc'ed to 1480/1575 cpu/mem

I stand corrected. I couldn't get over 900K on a 980.

@CapnBDL you run 2 instances to divide hashrate between two pools, in case one goes down other gets full 100% other wise both get 50-50%. It was a simple way around lack of failover since cudaminer days.
I get 5700 khash/s on qubit w/o any switches try that, but you have to have CPU idle for that or reduce 200 khash/s for each core used. I tried upto 2 cores.

Ok...have quibit intensity set to 19.3. Close enough for the 750Ti?

I don't think the release done for nVidia cards has such bug...it's working for me. Vid ram is within ranges. Just switched me to Quibit....plain jane settings get 5200+. Now I need a setting for 750Ti mining quibit. Help?

right now it is .....-r 3 -R 10 -a qubit

gimme a i
gimme a g
gimme a........

What's that spell....hell, who knows    

serious..750Ti switches for quibit plz? ..Now it switched me back to quark. Those setting I already have entered.

edit; Just read the rest of your comment. Why would you want to run more than one instance? I don't understand? Wouldn't that saturate any cards mem?

edit2; Tried it! Duh...of course the mem gets saturated. So, I ask again...why would you want to do that?

is the miner affected by the nvida "bug"(they said that they have done it intentionally, but let's call it a bug) when i open more than 1 instance, and as a result i saturate all the ram available?

Lyra2RE does random accesses: some are even based on the previous results.
So it is a mix of latency and bandwidth but I'd say more latency than anything else.

Is that the latency problem all over again?

And Lyra2RE depends heavily on ram speed: just think that r9 290 and 290x are virtually equivalent, regardless 10% more shaders. Kinda like scrypt.
Thus, I don't think the 980 can do 2.2 Mh/s, probably half that.

By extrapolation from the 750ti the 970 could match that @ 145 watts and a 980 2.2 Mh/s @ 155 watts,

Looks like AMD gpus are best on lyra2re, see my miner in the signature (r9 290 almost 1.8 Mh/s and about 200W).

Yeah, so there really is no reason to manually limit the TDP or turn it down with the power slider compared to adjusting clocks and voltage?


Multipool profit %s (like on nicehash) are not always representative of how much you can make by mining with a certain algo compared to another one. They don't always keep up with miner development or certain generations of cards. For a long time Nicehash was completely inaccurate when it came to that. Obviously it doesn't include power usage either, which can be extremely misleading.

Is that on the multi-algo port? That only works with the opencl miners, not sure if it's only some of them.
ccminer certainly can't handle algo switching.  I don't know if the custom factors can be used on the individual
algo ports, which would allow a simple round robin bat file to do profit switching. If Miner Control can manage
algo switching with ccminer (and it seems it can) it can probably handle custom price normalization as well.

im sure you are correct ...

but i have yet to figure out how on westhash ...

#crysx

was running on yaamp as a test a little bit ago ...

only 3 threads ( 3 x cores ) actively test mining on this machine ( the only windows machine ) ...

- windows vista ultimate x64 ( yup - i know i - know ) ...
- quad core amd phenom ii 965 running 3.4GHz
- tpruvot cpuminer-multi-mingw64 1.0.9

cpuminer-multi-mingw64 -a lyra2 -u 1CTiNJyoUmbdMRACtteRWXhGqtSETYd6Vd -p x -o stratum+tcp://yaamp.com:4433 -b 0.0.0.0:4048 -t 3

runs quite well for a cpu miner ...

-------

[2015-05-15 14:13:36] accepted: 365/367 (99.46%), 294.01 khash/s yay!!!
[2015-05-15 14:13:59] thread 0: 2240475 hashes, 96.67 khash/s
[2015-05-15 14:13:59] accepted: 366/368 (99.46%), 293.20 khash/s yay!!!
[2015-05-15 14:14:01] thread 2: 4887637 hashes, 97.77 khash/s
[2015-05-15 14:14:01] accepted: 367/369 (99.46%), 292.67 khash/s yay!!!
[2015-05-15 14:14:12] thread 1: 5893801 hashes, 97.14 khash/s
[2015-05-15 14:14:15] yaamp.com:4433 sent lyra2 block 60339
[2015-05-15 14:14:15] thread 0: 1525634 hashes, 96.15 khash/s
[2015-05-15 14:14:15] thread 1: 253782 hashes, 91.62 khash/s
[2015-05-15 14:14:15] thread 2: 1294897 hashes, 96.27 khash/s
[2015-05-15 14:14:52] thread 0: 3681011 hashes, 97.38 khash/s
[2015-05-15 14:14:53] accepted: 368/370 (99.46%), 285.28 khash/s yay!!!
[2015-05-15 14:15:01] thread 1: 4481330 hashes, 96.29 khash/s
[2015-05-15 14:15:01] accepted: 369/371 (99.46%), 289.95 khash/s yay!!!

-------

if cpu can do this - how much more can be optimized with gpu? ...

#crysx

I think yaamp normalises prices based on AMD HW, as do most profit switching multipools. Some algos are way
off compared to ccminer.

The whole reason I asked about Lyra is because I'm using Miner Control to help keep my connection to yaamp. It works great, took me some time to get it to run, but does, according to price, switch to that algo.