Topic: Avalon ASIC users thread - page 144. (Read 438600 times)

https://raw.github.com/ckolivas/cgminer/master/ASIC-README

Conman,

Will you make a post reviewing all the extra commands and how they work. (When you have time of course.)

I'm not sure of what all is available and the usage of each. If there's a separate readme that I missed just point me to it.

Thanks.

I've reuploaded it. If you're flashing, this is the md5sum of the proper 20130701 firmware:
a57305b42abe77d3335d44790e938d9e *openwrt-ar71xx-generic-tl-wr703n-v1-squashfs-factory.bin

Actually you shouldn't need to do this except when you're changing from very old firmware that used the old serial interface as that might have had commands which are no longer valid.

when installing your firmware, should we always not check the "Keep Settings" box?

Con,

Seems i got it working stable

Batch ONE unit with unmodified WR703 powered from molex/USB HUB just for power - PSU. Jumper 5 Removed, and FPGA connected directly to WR703 USB. ftdi/usbserial drivers removed, Wifi drivers removed also


The results

Elapsed 1day 16h 56m 52s   
Difficulty Accepted: 2907218.00000000 1183.30/m 84704.14 - MHS

Clock is set to auto moving between 357-362

Thank you

PS: you read my mind - two great options

--avalon-fan Set fanspeed percentage for avalon, single value or range (default: 20-100)
--avalon-freq Set frequency range for avalon-auto, single value or range

New avalon firmware:

http://ck.kolivas.org/apps/cgminer/avalon/20130701/

Changes:

I've changed the default hardware error target with avalon-auto to a bit lower so it hovers closer to 1.2% instead of 1.8%. The hashrate increase of the higher error rate was marginal at best but I'd prefer a little more headroom so that auto can be left on. (Yes it will give slightly lower hashrates, if you don't like it then set a manual value!)

Auto now will not raise the frequency if you are over the target temperature.
Auto will try to drop the frequency to minimum if it has to idle the avalon due to overheat.

New command line options:

--avalon-fan Set fanspeed percentage for avalon, single value or range (default: 20-100)
--avalon-freq Set frequency range for avalon-auto, single value or range

These of course have to be added to the "More options" box in the web interface if you're not running cgminer directly.

Note that if you give invalid options it will not start mining!

Your temps are fine and the pools all use different algorithms to determine hashrate and fluctuate wildly especially with higher diffs so that looks about right (mine reports +/-20% at the pool usually).

i'm still not getting full "passthrough" of my hashrates from ckolivas (4th down) to ozcoin?

are my temps ok?:

fans with an adapter from 4 pin molex to fan connector, or from the motherboard, HDDs straight from the 4 pin molex'

I would get a voltmeter and check the psu voltage output.

Your extra connector probably had different wiring than the ones for the Enermax and fed 12V were 5V or ground was expected....

I mean the 6 pin modular connector, the HDD was a 2.5" with sata power connector. I used an extra connector from one of my other modular PSU because the 850W enermax did not come with any extra connectors for the 6 pin connections.

How did you power fans and HDD and stuff? did you use a connector from another PSU like I tried?

I have it running in an old system, no problems here... weird , you say 6 pin connectors for the HDD?  doesn't it need a 4 pin molex?

NOTE  Re: Enermax 850W PSU that comes with some avalon batch 2.

I upgraded to a larger PSU while trying to troubleshoot some problems, it did not seem to be the problem, but I left the larger PSU in the avalon and thought I could use the 850W Enermax to power a small 2X7970 system... connect the 24 pin to the MB, connect one of each 7970 to the PCIe plugs, good to go, right?

It powered up just fine, but got no OS loaded error... doh! forgot to plug in the HDD... no big.

so then I pluged in the HDD to one of the 6 pin connectors on the PSU, no boot, no power on, no nothing... I tried a couple other of the 6 pin connectors, first one, same thing, second one, the system would power up, but I still get no OS error... boot into bios, no HDD detected... weird.

So then I thought I'd try another HDD, maybe that one died in the 1 week that it's been disconnected?

Bad move.

Sparks and then FLAMES started shooting out of the HDD. I kinda freaked/zoned out so it took me a few seconds to pull the plug.

Has anyone else tried to use the Enermax 850W with the chinese writing for a normal computer? Is it possible these PSUs are wired differently? I thought it was a standard ATX? maybe I just had 2 bad HDD in a row? (not willing to risk a third) How long 'till my apartment no longer smells like blue smoke?

well, duhhh 

You are on balance mode and not failover. Half your hashes are going to your 2nd pool.

here's ckolivas firmware with 325 Mhz setting.  but on ozcoin i'm only registering a mere 39416 MH/s.  what gives?

ADVANCED OVERCLOCKING TECHNIQUES

After typing out the above - relatively safe - method, I thought we might explore some best practices for finding the true optimum speed of your individual Avalon.

As I understand it, the -auto option will simply increase clockspeed slowly until an arbitrary number of HW errors are produced. This is a quick and easy way of doing things and very useful, but it's not perfect...

Overclocking hardware for long periods of time is all about balance. Anyone familiar with regular CPU overclocking will know that you will normally hit a wall where increasing voltage will begin to experience diminishing return with respect to clock speed. This wall is usually parabolic and once you approach this level you will require excessive amounts of voltage and cooling to go any further. This may be acceptable when trying to reach a world record on a "suicide run" but it will not cut it for a system that you want to be running 24/7 for as long as it remains profitable.


OK, that's all fine and dandy with a regular CPU, but in the case of a regular CPU, just one error can be enough to freeze the system and force a reboot.
- these ASIC's seem different. They are much more forgiving of errors and will happily continue to try their calculations even if they get half of them wrong, lol./

-- the thing I'd like to determine, is whether or not the HW error rate is parabolic? (I would expect that it is, from prior experience)

BUT  ...
There is a chance that when dealing with so many chips, it may be possible that only a few chips are giving HW errors and that you may be able to reach much higher clock speed and effective hash rate by exceeding the arbitrary limit set in the -auto function of the firmware. Conversely it may be possible that reducing your clock speed by only a small amount could result in a drastic drop in HW errors and effectively give you a higher accepted hashrate.

The only way to know for sure is through thorough testing. Every device, every chip will respond slightly differently. This is why I love overclocking. except ram timings, don't even get me started on ram timings  

This is how I did it. (was confirmed by CK to be a sensible way to determine your individual device's optimum speed)

- make sure that you have some free time and that you can actively monitor your device for the next hour or two (not necessarily mandatory, but prudent)
- in the cgminer configuration tab, in the "more options" window, enter --avalon-auto and save and apply (personally, I usually do a hard boot after this, just to be sure)
- watch your chip frequency climb. (I'd recommend refreshing to check your speed and temps every 10 minutes or so. The software is designed to find a sweet spot where your speed is high and your HW errors are at an acceptable rate. It is not mandatory to actively check, but anything is possible and neither myself nor ckolivas are responsible if this sets your avalon on fire  )

- after a while your frequency should settle into a certain range (for me this was around 349-351)
- at this point you can remove the -auto command and just put in whatever clockspeed you deem reasonable (I recommend you set your clock a little lower than where the -auto setting started to level off. Personally, I chose 345 to give a little breathing space.)
- reboot your device to ensure that the new settings are applied and you should be good to go.

A couple notes:
- this will likely reduce the lifespan of your Avalon.
- there is always a risk involved in running a chip beyond its designed specifications.
- make sure you continue to monitor the temperatures and your accepted/HW error ratio - it is possible that a lower clockspeed may result in less HW errors and effectively get you a better score at your pool. I recommend watching the reported speed on your pool closely to see if it starts to decrease for an extended period (this number will likely fluctuate greatly, but if it continues to drop, better check your speed)

Again, I take no responsibility if this fries your Avalon, and it will most definitely void your (non-existent) warranty.

Have fun!