Topic: Bounty[PAID OUT] : a bitstream for better utilizing the Cairnsmore1 157-294.5btc - page 5. (Read 22006 times)

@ cranky4u

I've got 2 boards and 1 of my boards does that. The other mines full throttle.
Reflashing sometimes helps. I am 50/50 on if it's a hardware or bitstream issue.

I do figure however since one works perfectly it atleast has a chance of being fixed by an improved bitstream.

Multiply U with 71.5827883 to get the MH/s.

so after some time...I stabilise around the U: 5/m ICA line in cgminer...is this right for the 190MHs bitstream?  basically what sort of rate should I be getting? My pool is reporting about 1200~1400MHs for 2 * CM1 190bit flashed systems

It probably did flash properly, it's just that one of the FPGAs on ICA 2 has frozen. Probably FPGA1, that always seems to be the one that falls over first on ebereon's boards.

Hi,

I've sent my second 10 BTCs bounty part.

https://blockchain.info/tx-index/14400138/b6f539f3d23fe868d1e204f88c8e91f417435134b62bd64f79ffbc3844c5ab88

spiccioli

running as pr above but on + poclbm gives (approx)
ICA 0: 360/360    U: 4.30/m
ICA 1: 355/360    U: 4.15/m
ICA 2: 355/355    U: 2.05/m
ICA 3: 360/360    U: 6.75/m

overall U: 17.3/m

hint? is ICA2 not flashed properly despite VM saying it was a success?

What are your per worker U values? U=17.5 on 2 boards looks like 1 fpga isn't hashing which is a common issue with that bitstream.

ok...after 18 hours of testing, well short of the 48 hours I hoped for, here is my 190 bit stream findings;

System - Win 7 64bit, cgminer 2.6.1 + phatk, cm1 #62-0432 & #63-0433
Average hash rate per cm1 = ~725MHs at pool & via cgminer Note: neither board is reaching the bounty threshold of 750MHs
Average U = 17.5~17.8/m
HW & R are very low. HW = 0 & R <5
Maximum run time between failures (resulting in need for system reset to recover e.g BSD or non-responsive) - 14 hours
Minimum run time between failures (resulting in need for system reset to recover e.g BSD or non-responsive) - 0.5 hours
Average restarts of cgminer before both boards detect successfully - 3
Identified conflicts - use of 2nd 6770 GPU for mining causes system to BSD on start up of cgiminer or intedpendent start of of phoenix miner when cgminer is FPGA only. Note: Can mine on 1st 6770 GPU though.

I hope this helps developers...
Tweakers, any idea on how to tweak to get up to 750MHs and improve stability?

For now the focus is getting past the main stability issues and getting the bitstream running on all 4 chips in some sort of reliable way.

Before we get up/down working we'll need to finalize a new communications protocol. The communications protocol needs to stay fairly robust and flexible. And I've gotten several suggestions/ideas on that from TheSeven, so I'm sure that will change somewhat with the collaboration we have here now.

That's the kind of thing you want to hammer down and get 100% right from the beginning.

Once the protocol is decided, we'll roll it into the controller and get it working purely with the controller and the 4 worker fpgas...

Once that's tested, and validated, adding the same on the up/down port is a fairly minimal addition.

So basically, probably not for a bit still. Even if I used my previously defined protocol as-is as a temporary measure, it still needs to be finished and tested on the workers (and we still don't have stable clocks yet).

Right now the clocks seem to be the last real barrier, and so that's what's being worked on to solve many of the remaining issues.

Great to hear!

While competition for sure is the key factor to maximize performance in the long run, in this particular environment collaboration will guarantee to get the best possible result in the shortest possible time. Just great you joined your forces now.

To all CM1 operators out there: now there is no reason to withhold your contribution to the bounty. If you agreed to support it initially but did not send your coins by now, please consider doing it. If you did not commit initially, still please consider contributing 1BTC for each board you have (this is less than one week's mining income from the additional hashing power you gained with the bitstreams already provided). Be fair, people are working hard and deserve some compensation. The 122 coins collected so far are only almost half of what was committed to - please consider supporting.


@Glasswalker: now since you work together, how's the chance to get up/down functionality supported soon? I figured that half of the instability I have is due to USB cable and / or hub problems. If the related controller FW can save me from buying new cables and hubs (again), I will increase my contribution by another 25BTC.


Thank you all for the enormous efforts and the great outcome.

Yeah sorry didn't mean to imply you wouldn't be working on any other projects but this one only that we'd be putting our heads together on this one (in addition to any other side projects any of us have on the go) lol.

Truly great news,

I'll double my BTC later today, early tomorrow, so +10 to the bounty.

spiccioli.

Great news, ill be adding 2 to the bounty in that case. Ill be sending it later today after work.

My two sent~!

Thanks

Doff

Great news for everyone involved - cooperation is better than competition. As soon as I get my board (T minus 6 weeks) I'll send some coin your way. 

Nice to see you guys working together. Look forward to what comes from this.

Just a quick note to (mostly) confirm this. I may still be releasing my own bitstreams in parallel to working on the new project - they're focused on very rapid testing of small incremental improvements, whereas developing a new hashing core is really hard - but regardless of which "wins" in the end I'll still be splitting the bounty down the middle with Glasswalker. The duplication of effort was seriously getting out of hand and I didn't realise quite how much so until recently.

I have a new development to post...

As you all know makomk jumped in with some of his icarus improvements, and that resulted in a mostly stable bitstream which had some great jumps in performance for everyone.

In addition, I was working on trying to work around some of the board level issues on the cairnsmore and also making several other improvements to the bitstream as a whole. (some new fixes just came down the pipe this morning from Enterpoint which might help improve stability further).

Lastly, I've been working on another mining core on the side which I had hoped to make some money off of.

Well in light of all this, I had a chat on IRC with makomk, and we have decided that I'm going to release all my work so far, (even though it's still unstable, we're rapidly working on fixing that). And makomk will share all his "tricks" and work so far. Together we're working with some additional help from TheSeven as well. To collaborate on a single opensource bitstream project which hopefully will have many improvements for everyone.

Initially this will be targeting the Cairnsmore, but we're going to try and keep it "hardware agnostic" if possible. And of course 100% opensource, including project files required to build it, and it can be built purely in Xilinx ISE (no third party tools).

In addition, I'm now officially open sourcing my new bitstream (the new hashing core) which is still incomplete, but has huge potential I think. So with all this combined, this bitstream has the potential of being a great new boost to the bitcoin FPGA mining community as a whole.

Anyway, with this agreement, we've agreed, that if the community decides that we "qualify" for the bounty, we will be splitting the bounty down the middle 50/50.

This way we're no longer competing, but instead cooperating to deliver a new improved bitstream to the community. Which is good for everyone.

Hopefully makomk will pop in here shortly and confirm this from his end with a post.

And of course, here is the shiny new github where all our work will now live directly in the public eye:
https://github.com/pmumby/hashvoodoo-fpga-bitcoin-miner

So if any of you find problems/issues, please use the issues tracker on github to submit them, so they can be worked on as they are found. If you submit issues, please provide as much detail as possible.

Also if anyone has anything to contribute, ideas, suggestions, or wants to work on porting to another board, let me know, I'll let you get involved as a collaborator.

I'll post more news once I have it!

I've programmed 20 odd boards with the Makomk 190 Mhz bitstream.  They are mostly all hashing at over U: 5 / m.  There are a few ~10% hashing painfully slow 1 or less.  I won't have time to debug those, so I'm just letting everything run.

Startup was painful in a few cases, but generally, replacing cables and power cycling the boards seemed to resolve those problems.

Makomk, regardless of whether you with the bounty, there will be significant bitcoins coming your way from me!

Do you think you can push it farther?

Best of luck, All it takes is a little refresher.