Topic: [ANN] Bitfury ASIC sales in EU and Europe - page 76. (Read 250482 times)

Well firstly I mean turning autotune off, that's a typo.

I let the harware run for about 24h, grabbed the best.log and copied it as follows:

And started manually changing the speeds in /opt/bitfury/best.cnf based on a large number of results with 5 or more minute intervals, thus using a new .stat.log every time. The log only holds a five minute sample, so I collected a large number of notes on each chip using paper and pen. When a chip constantly had 0 errors I pushed it's speed higher and when a chip constantly showed over 10% errors I lovered it. I also re-enabled a bad chip that autotune had shut down, it's not working properly, but is producing 2/3 valid shares without interfering with the other chips, using a speed of 52.

All other chips are now between 54 and 56 speed, 57 was a now go on every chip because of the hightened error rate.

The final modifications (so far) were made based on 20 samples and I believe that this is optimal for now, I will propably try altering the operating temperature next, with a modifiable make-shif enclosure and see where that leads me.

I am by far not sure the 10% rule is optimal, someone else will propably figure out the optimal number there.

The results of my current settings measured from poolside over 6 shifts (9h) put my hashrate with only the 2-board starter kit at 50,68 Gh/s.

You can see it in your router's web interface -> DHCP Clients List.

What is the IP address of the web interface of the Rpi

Totally right!

Thanks for the info, I missed it.

The basics are covered in http://punin.com/README_BFSB_v0.1.txt
Please note that everything does not apply for the kits your getting, you will have a preinstalled image on the sd card.
But what does apply is how to read the logs and configs and how to modify the latter.

To put it short, I disabled autotune after a 24h run for all chips and used the best run generated as a starting point. I tuned the chips based on about 50 5 minute runs, rising the speed of a chip if it had 0 errors and lowering it if it had 10% or more. Takes forever and is anal as hell.. most of you guys will love doing this .

Any update on this? My device should arrive tomorrow, I don't want to risk anything here..

Did I miss the option to tune the h-boards? If I understood correct you are tuning on your own, non-h-board, correct? Otherwise would you share this information?

I can confirm that the boards really do not seem to like cooling. Im even partially blocking any natural air-movement to them and have achieved the best results seen thus far:
speed": 1766, "noncerate": 50.967, "noncerateperchip":1.593, "hashrate":52.690, "good":3560, "errors":148, "spi-errors":2, "miso-errors":0, "jobs":375, "record":50.924

Which translates in to a constant average hashrate of just under 49,5 Gh/s.

[edit]
In should propably in all fairness also say that achieving this also required turning the autotune off and hand tuning the chips for about two working days... the problem with 5 minute samples is that you do need a lot of them to make the right adjustments.

Not a problem with the core voltage that is on your boards. Our chip has been proven to hash even at such extreme temperatures that the solder melted Then it started floating and shorted itself, ending hashing Some of our users are actually reporting decreased hashrates after adding fans. YMMV.

Congratulations on getting all the August orders shipped out guys! 

Now that they are all out of the way, could you please tell me what the lead-time is on the bare chips, and will they be available at a price that makes rolling your own boards a reasonable option compared to just buying the H-boards?

Cheers.

I received my starter kit today. I am running it now (about 41 GH/s) without any fan and the ambient temperature is 24C. Do I have to be concerned or running without any fans with a kit with just 2 H-boards is not a problem?

Hashing at a steady 41-42 GH/s.
Now helping to find the last blocks at 65mil diff. Will be much higher soon...

An odd thing that happened: When I connected to BTCGuild and went off to work, the hashrate kept dropping until it reached about 20GH/s... maybe I should have restarted the miner after I set the minimum difficulty to 32 for my worker?
Any way, it's steady after a restart now! I wonder if the order of the boards makes a big difference? When I get some other H-Boards in October, should these slower boards go in the slots in the back?

And Punin, I would like to thank you personally for this sexy device Ordered 25GH/s, got 41GH/s. Thats 164% of what I wanted 

Ok then, I stand corrected on that. Good to know.

My assumption was based on older fanless Thermaltake PSU, that was very pricey and died after a few months. The warranty was voided because in the manual there was a requirement that there must be an intake fan in the PC case.

No, this is different. There can be shorted pins on underside of M-board. The H-board capacitors are ok and joined pins on chip don't matter except if they are IO pins. I will post some photos very soon to help you identify problems!

I was saying this few days ago...
https://bitcointalksearch.org/topic/m.3020099

I have a 4 year old Fanless Seasonic that has been running in a case with zero air flow.  I used to write PSU reviews for XCPUs, and many of the high-end fanless PSUs do not need any substantial air flow, even at 75-80% of max load, for years on end.

Your assumption, simply put, is wrong. 

ATTENTION!!! PLEASE PERFORM OPTICAL INSPECTION OF BOARDS YOU'VE RECEIVED!! THERE HAVE BEEN REPORTS OF SOLDERING ERRORS! I WILL SEND INSTRUCTIONS TO DO THIS SHORTLY!!

500 GH/s.

Correct ... unless you use a 1500 Watt PSU for a 400 Watt job (and it is not sitting near window with direct sun light).

Most "fanless" PSU rely on the airflow forced by the case fan(s). They are overpriced and will die soon in true fanless environment.

could you tell us the ghz for ur setup?