Topic: [1050 TH] BitMinter.com [1% PPLNS,Pays TxFees +MergedMining,Stratum,GBT,vardiff] - page 366. (Read 837101 times)

Now running latest version with bug fixed. It was a race condition that would happen rarely, but when it did it would cause a whole group of rejected proofs of work.

Parallel programming is fun once you find and remove the race conditions.

The pool server is now lean and mean. A good foundation to build on. For the future I'll add a new reward system, and probably support for roll-ntime.

Figures. It produced less stales on the test network, but does the opposite on the regular bitcoin network.

I have reverted back to the old version until I can debug this. Sorry for the inconvenience.

BitMinterD 1.0 (the mining pool software) up and running.

Changes:

• Reduced use of server resources considerably - for future scalability
• Notify long pollers faster at block changes

So, only internal changes that you should not notice beyond slightly lower stales. It's been tested on the bitcoin test network and seems stable enough, but let me know if you experience any problems.

wow amazing luck for such a small pool. another block haha

Yeah, we're on a lucky streak now 

and 2 short blocks very nice

Everything back up. Sorry for the downtime!

There's a problem with the pool backend... I'm working on it.

Finally! Good job, people. Always good when those slow blocks are finally solved. 

The GTX 580 has 16 CUs (compute units). Apparently there is a big difference between 460 and 580 according to the mining hardware comparison charts.

The reason you suddenly got a lower hashrate is probably because you forgot to turn off vectors this time. I forget that myself too. I'll have to make vectors off the default for nvidia soon - the way it is gives people a really bad impression of the miner on nvidia hardware.

On the topic of the 580, how many 'CU' does it have? The 460 has 7.


EDIT: In fact, it's now even LOWER! I barely pull 20 MH/s now on your client. Maybe I'm doing something wrong?
Further: I do notice there are new drivers out... I'll try them.

Maybe the GTX 580 is that much faster than the 460? I dunno. I run it now at stock speed with 127 Mhps.

Yeah, the CPU mining isn't really intended for use the way it is now. Next step for that is to implement OpenCL CPU-mining. This should be a significant boost, but I don't expect it to beat the fast CPU miners. After that it may useful to implement CPU-mining based on assembly language. The question is whether it will really be worth the effort. But I do want to have decent speed CPU mining.

Keep in mind that this is WITHOUT my OC, just a quick picture I took and cropped for uploading.

http://img8.imageshack.us/img8/4382/mhsh.png

Although, what I'm more interested in is how you pulled off this when my clock was faster than this. Using the exact same settings I got half the speed.

Finally, your CPU mining could use a biiiiiiit of optimization. I get .46 Mh/s and on rpcminer-4way (Quad-core CPU) I pull 12.46 Mh/s.
This does however show promise, given how quick you were to respond.

Wow, is CUDA that much faster on nvidia hardware than OpenCL? I will have to look into adding CUDA-support. I have to finish Radeon 5xxx optimizations first, though. Thanks for the info.

Amazing, it actually worked. I'm assuming it was the 'BSOD' that caused the Muffins to mess up, but... I dunno.

Unfortunately, it's slower than a CUDA miner for me. :[ The Gfx card goes at about 57-60 MH/s at 851 Mhz Core clock with 64 WGS, 50 ms interval, no vectors, BFI-INT enabled. Perhaps I should try increasing the WGS? (Work group size)

On GUIminer's CUDA miner, I use -gpugrid=512 -gputhreads=700 -aggression=8 and get 80-90 MH/s.

Yeah, they are both "1.0 beta". I will change this soon.


Hmm, I've never heard of this before. Since it crashes the whole system, I guess there's no crash dump left by Java on your desktop. Can you run other miners ok? Are you using the latest nvidia drivers? I would guess this is a crash in the drivers.


Sorry, there's a bug in there somewhere that I haven't been able to find yet. On rare occassions some users would get a completely blank login window. I put in additional error checking, which is causing the message you see about being unable to load settings. I guess this could be related to the "proceed" button not reacting. I will look into that. Hopefully I can finally squash this bug.

You could try deleting the BitMinter muffins, then restarting the client and trying again. Java WebStart muffins, on Windows, are located somewhere like:

C:\Users\USERNAME\AppData\LocalLow\Sun\Java\Deployment\cache\6.0\muffin

Muffins are kind of like browser cookies. It's where the BitMinter client stores its settings, like username and password.

As a general note about nvidia and 100+ Mhps: you have to turn off vectors, otherwise there will be a performance hit. And vectors are on by default (I'm going to change that for nvidia GPUs).


Perhaps the wording of this message should be changed. What it means is that your nvidia drivers only support OpenCL 1.0, not the OpenCL 1.1 interface. This means BitMinter is unable to detect optimal work group sizes for your GPU and warn you if you use a sub-optimal one. Not really a big deal, so it can just be ignored.

Nvidia had an OpenCL 1.1 driver in beta for over a year, without 1.1 support making it into the regular drivers. But I believe I read somewhere that the latest nvidia drivers have OpenCL 1.1 support, or that the next release will. Can't find where I read that now, though.

According to the logfile, I'm using the 'beta' 1.0 version (Even though it's the one from the non-beta link on the page). I can't make this thing work for the life of me. One of two things happens.

1. I log in successfully, see briefly 'GPU Device 0 has no space left, wating for confirmation' or something like that (it's not there long enough to read), then the screen goes black, and the computer restarts as if it had a BSOD (And windows considers it one, it makes a minidump).
2. The 'Proceed' button on the log in doesn't work. I can click it forever to no avail.

I'm using an AMD Phenom II X4 965 for my CPU and an NVidia GTX 460 as my GPU. The prospect of 100+ MH/s was enticing (I can overclock even higher than the value listed above and still see stability)... but it doesn't look like the client will work.


EDIT: Just noticed this is now popping up whenever I click 'start'
EDIT2: It now is claiming my card doesn't support OpenCL... but it certainly does. Both the official page and GPU-Z say it does, at least.
What on earth is going on here...?

I'm working on this right now. You can try the beta version (see the miner thread) - I believe it is about 1% faster on Radeon 5xxx. I'm still working on improving this further, though.

That is a good point.

So either it takes a long time to get paid, or you have many shares you don't get paid for. Neither sounds perfect. But it seems miners at mineco.in don't care too much about unpaid shares.

If anyone has a preference on this, let's hear it.

High N (2*difficulty): small chance shares don't get paid, but payments still coming for old shares after 12 days
Small N (difficulty/2): high chance shares don't get paid, but payments for a share finish within 3 days or so

If anyone has any questions about PPLNS or another reward system, feel free to ask.

note that it might take 10 days to get paid at 2x difficulty PPLNS because you can still get a payment for a block that is 3.6M shares later which in this pool is a while
mineco.in uses 1/2 difficulty for this reason
eligius can use 8*difficulty because they can blow through 10M shares fairly quickly