Avalon ASIC users thread - page 211.

fcmatt

legendary

Activity: 2072

Merit: 1001

Quote from: jgarzik on February 08, 2013, 08:40:55 PM

Quote from: Bogart on February 08, 2013, 12:58:19 PM

Quote from: jgarzik on February 08, 2013, 12:14:12 PM

Another reliability update:

Now that difficulty is sufficiently high, no longer seeing machine or cgminer restarts.

The most common symptom now is a cessation of mining; cgminer and machine are both responding to status queries, but no work is occurring.

This symptom occurs every 24-48 hours.

A simple machine restart fixes the problem immediately.

Sure sounds like a memory leak.

If you could check and record every hour or so, the free memory reported by 'free', and maybe capture which processes are using how much with 'ps axu', you may be able to find more definitive proof.

'free' is happy as a clam. The previous behavior can be attributed to a memory leak.

Now that difficulty is sufficiently high, the box reaches a condition where the controller (linux kernel, cgminer) are active and accessible remotely, but no work is progressing.

The box will restart if the memory leak condition is reached. The box does not restart upon this no-mining condition.

I am using a semi new version of cgminer and that behavior is what i am seeing. Program is running but no actual work gets to btcguild. Restarting cgminer every few weeks fixes it. No reboot needed on windows 7 with gpu.

Cant you drop in a new version of cgminer but somehow keep the critcal bits for those asics? Or are the drivers compiled into one big binary? Just thinking out loud. I am prob wrong.

Bogart

legendary

Activity: 966

Merit: 1000

Quote from: jgarzik on February 08, 2013, 08:40:55 PM

Quote from: Bogart on February 08, 2013, 12:58:19 PM

Quote from: jgarzik on February 08, 2013, 12:14:12 PM

Another reliability update:

Now that difficulty is sufficiently high, no longer seeing machine or cgminer restarts.

The most common symptom now is a cessation of mining; cgminer and machine are both responding to status queries, but no work is occurring.

This symptom occurs every 24-48 hours.

A simple machine restart fixes the problem immediately.

Sure sounds like a memory leak.

If you could check and record every hour or so, the free memory reported by 'free', and maybe capture which processes are using how much with 'ps axu', you may be able to find more definitive proof.

'free' is happy as a clam. The previous behavior can be attributed to a memory leak.

Now that difficulty is sufficiently high, the box reaches a condition where the controller (linux kernel, cgminer) are active and accessible remotely, but no work is progressing.

The box will restart if the memory leak condition is reached. The box does not restart upon this no-mining condition.

I see. Does 'dmesg' report anything new once this condition is reached?

jgarzik

legendary

Activity: 1596

Merit: 1100

Quote from: Bogart on February 08, 2013, 12:58:19 PM

Quote from: jgarzik on February 08, 2013, 12:14:12 PM

Another reliability update:

Now that difficulty is sufficiently high, no longer seeing machine or cgminer restarts.

The most common symptom now is a cessation of mining; cgminer and machine are both responding to status queries, but no work is occurring.

This symptom occurs every 24-48 hours.

A simple machine restart fixes the problem immediately.

Sure sounds like a memory leak.

If you could check and record every hour or so, the free memory reported by 'free', and maybe capture which processes are using how much with 'ps axu', you may be able to find more definitive proof.

'free' is happy as a clam. The previous behavior can be attributed to a memory leak.

Now that difficulty is sufficiently high, the box reaches a condition where the controller (linux kernel, cgminer) are active and accessible remotely, but no work is progressing.

The box will restart if the memory leak condition is reached. The box does not restart upon this no-mining condition.

-ck

legendary

Activity: 4088

Merit: 1631

Ruu \o/

It's a damn shame because the memory leak was very real and trivial to fix and fixed in the next release of cgminer. I'd suggest just using ozcoin (or a pool like it that supports setting high static diffs) to mine with, and set difficulty to 10,000. Then the number of shares submitted will be miniscule and it will take much longer to run out of memory. The only cost of doing that would be some more variance.

Bogart

legendary

Activity: 966

Merit: 1000

Quote from: jgarzik on February 08, 2013, 12:14:12 PM

Another reliability update:

Now that difficulty is sufficiently high, no longer seeing machine or cgminer restarts.

The most common symptom now is a cessation of mining; cgminer and machine are both responding to status queries, but no work is occurring.

This symptom occurs every 24-48 hours.

A simple machine restart fixes the problem immediately.

Sure sounds like a memory leak.

If you could check and record every hour or so, the free memory reported by 'free', and maybe capture which processes are using how much with 'ps axu', you may be able to find more definitive proof.

For remote/automated restart capability, I like these: http://www.digital-loggers.com/lpc.html

Photon939

sr. member

Activity: 452

Merit: 250

That's rather poor performance TBH, I imagine it will quickly become frustrating to manually restart your miner every/every other day. My GPU miners easily went over a month with no restarts or failures (on Windows even)

Too bad nobody can do anything about it since they haven't made any of the code available. Disappointing after their "it's not for the money" statements.

jgarzik

legendary

Activity: 1596

Merit: 1100

Another reliability update:

Now that difficulty is sufficiently high, no longer seeing machine or cgminer restarts.

The most common symptom now is a cessation of mining; cgminer and machine are both responding to status queries, but no work is occurring.

This symptom occurs every 24-48 hours.

A simple machine restart fixes the problem immediately.

os2sam

legendary

Activity: 3583

Merit: 1094

Think for yourself

Quote from: lastbit on February 07, 2013, 12:57:13 PM

Quote from: Hawkix on February 07, 2013, 11:11:54 AM

Seymour Roger Cray from Cray Research would love that HW design ...

Nah, he would have made the chips from Ga-As, include a small radiator in each chip's package and submerge the whole shebang in a force-pumped fluorocarbon bath.
And he would have designed the case like a flower with one petal missing, just for the lulz.

+1 on the Gallium Arsenide

kano

legendary

Activity: 4592

Merit: 1851

Linux since 1997 RedHat 4

Quote from: jgarzik on February 07, 2013, 12:30:07 PM

Quote from: kaerf on February 07, 2013, 12:08:27 PM

jeff, does the cgminer interface have any configuration options? can you change the launch command?

i suppose you can ssh/telnet into the machine and run cgminer from the command line...would that manual instance show up in the GUI?

That is what the machine does during boot: run cgminer from the command line. That is necessarily how it must be run under Linux

One only needs to edit /etc/rc.d/S99cgminer. Here is a sample of S99cgminer:

Code:

DEVS=`find /dev/ -type c -name "ttyUSB*"  | sed 's/^/-S/' |  sed ':a;N;$!ba;s/\n/ /g'`
PARAMS=" $DEVS $POOL1 $POOL2 $POOL3 --avalon-options 115200:24:10:45:282 -q --api-allow "W:0/0" --api-listen "

ntpd -d -n -q -N -p 0.openwrt.pool.ntp.org \
	-p 1.openwrt.pool.ntp.org -p 2.openwrt.pool.ntp.org -p 3.openwrt.pool.ntp.org && \
start-stop-daemon -S -x $APP -p $PID_FILE -m -b -- $PARAMS

Just in case you didn't realise ... "W:0/0" means anyone with local or remote network access to the device can do whatever they like ... including adding their own pools/workers, changing pool priorities ... etc

Though that also means you yourself can also do any API commands to it from any computer with network access (including restart/quit/...)

Normally you'd set it to --api-allow "W:127.0.0.1" or --api-allow "W:127.0.0.1,W:192.168.0/24" (or whatever your subnet is)

... and in case you didn't know ... with your current setup try this from another Linux box

echo -n 'summary' | nc IPOfAvalon 4028 | tr "|," "\n"

salfter

hero member

Activity: 651

Merit: 501

My PGP Key: 92C7689C

Quote from: Photon939 on February 06, 2013, 06:24:34 PM

Normally, yes. Since Avalon is using QFN packages for their chips, there is a significant layer of plastic over top of the actual IC silicon. The bottom of the chip has a thermal pad and the circuit board usually has a bunch of copper plated vias underneath the chip. This means most of the heat transfer is done through the bottom of the chip. In most cases the PCB itself is used as a crappy heatsink but since these chips are dissipating a fairly large amount of power in a package not really designed for high power applications, they are likely using thermal pads and the large heatsink to pull heat from the back of the PCB.

BFL was going to do something similar, according to the photos that have been posted to date. They ended up switching to FCBGA packaging to get better heat removal, but that ended up putting their shipping schedule even further behind than it already was.

(Whether they would've been first-to-ship if they hadn't made this change is anybody's guess at this point.)

MrTeal

legendary

Activity: 1274

Merit: 1004

Quote from: ProfMac on February 07, 2013, 02:14:02 PM

Quote from: goxed on February 02, 2013, 01:25:08 AM

Quote from: ngzhang on February 01, 2013, 10:20:54 PM

Chip's power consumption is <6.6W/Ghs

ATX power loses(82%), DC/DC power loses(87%), Fan(5~20w), etc…

So…

650W * 0.82 * 0.87 - 20W = 443W, About right.

This is a business opportunity to make a high efficiency wall to dc converter.
Making a very focused product, well specified load and voltage, can allow the efficiency to be much higher.

http://www.newegg.com/Product/Product.aspx?Item=N82E16817377002
http://www.newegg.com/Product/Product.aspx?Item=N82E16817292058

There's a lot of server PSUs with a single 12V rail that are extremely efficient. You won't easily find a power supply that rectifies 120/240V AC down to 1.2V directly. It just becomes too difficult to have such a large difference between input and output voltage, and to deal with 500-1000A of current on the 1.2V rail.

Epoch

legendary

Activity: 922

Merit: 1003

Quote from: ProfMac on February 07, 2013, 02:14:02 PM

This is a business opportunity to make a high efficiency wall to dc converter.
Making a very focused product, well specified load and voltage, can allow the efficiency to be much higher.

Pragmatically, even if you manage to reduce power consumption by 45W you'd only be saving 1kWh per day. That's $0.11/day (depending on your local power costs, of course; I'm using $0.11/kWh).

Your yearly saving using such a device would be $40. A 'wall-to-dc converter' designed for a ~600W load would cost many times more than that. Sales volume would be low, since it would have little or no use beyond this niche application; so amortizing the non-recoverable engineering costs to design it would bump the per-unit price even further. Not to mention the expense of UL/CE certification that would be required.

Such a product would be unlikely to ever pay for itself.

ProfMac

legendary

Activity: 1246

Merit: 1002

Quote from: goxed on February 02, 2013, 01:25:08 AM

Quote from: ngzhang on February 01, 2013, 10:20:54 PM

Chip's power consumption is <6.6W/Ghs

ATX power loses(82%), DC/DC power loses(87%), Fan(5~20w), etc…

So…

650W * 0.82 * 0.87 - 20W = 443W, About right.

This is a business opportunity to make a high efficiency wall to dc converter.
Making a very focused product, well specified load and voltage, can allow the efficiency to be much higher.

lastbit

full member

Activity: 206

Merit: 100

Quote from: Hawkix on February 07, 2013, 11:11:54 AM

Seymour Roger Cray from Cray Research would love that HW design ...

Nah, he would have made the chips from Ga-As, include a small radiator in each chip's package and submerge the whole shebang in a force-pumped fluorocarbon bath.
And he would have designed the case like a flower with one petal missing, just for the lulz.

DeathAndTaxes

donator

Activity: 1218

Merit: 1079

Gerald Davis

How are the modules actually attached to the controller board? The Avalon site indicated expansion to 6 modules was possible before scaling it back to 4. Ia m just wondering if there is any limit ... with sufficient power and large enough chasis would 8 modules be possible, 10?

kaerf

hero member

Activity: 631

Merit: 500

cool. looks like it should be trivial to add a usb hub + USB miner and start cgminer with additional devices.

is the usb cable hot-glued into the tplink board?

jgarzik

legendary

Activity: 1596

Merit: 1100

Quote from: kaerf on February 07, 2013, 12:08:27 PM

jeff, does the cgminer interface have any configuration options? can you change the launch command?

i suppose you can ssh/telnet into the machine and run cgminer from the command line...would that manual instance show up in the GUI?

That is what the machine does during boot: run cgminer from the command line. That is necessarily how it must be run under Linux

One only needs to edit /etc/rc.d/S99cgminer. Here is a sample of S99cgminer:

Code:

DEVS=`find /dev/ -type c -name "ttyUSB*"  | sed 's/^/-S/' |  sed ':a;N;$!ba;s/\n/ /g'`
PARAMS=" $DEVS $POOL1 $POOL2 $POOL3 --avalon-options 115200:24:10:45:282 -q --api-allow "W:0/0" --api-listen "

ntpd -d -n -q -N -p 0.openwrt.pool.ntp.org \
	-p 1.openwrt.pool.ntp.org -p 2.openwrt.pool.ntp.org -p 3.openwrt.pool.ntp.org && \
start-stop-daemon -S -x $APP -p $PID_FILE -m -b -- $PARAMS

kaerf

hero member

Activity: 631

Merit: 500

jeff, does the cgminer interface have any configuration options? can you change the launch command?

i suppose you can ssh/telnet into the machine and run cgminer from the command line...would that manual instance show up in the GUI?

jgarzik

legendary

Activity: 1596

Merit: 1100

Avalon ASIC miner is currently surviving >24 hour stretches without restarting, now that we're on a stable pool with a reasonably high difficulty value (BTC Guild, difficulty 32.0).

However, have now seen the machine get "stuck" in a strange state, where it is not mining or restarting. The fans ramp up, then ramp down, in a cycle.

Power cycling fixes the problem immediately, and the machine goes back to mining in >24 hour stretches again.

As a workaround, some pools have an "idle worker" notification system, to let you know when your miner disappears.

jjiimm_64

legendary

Activity: 1876

Merit: 1000

Quote from: Frequency on February 07, 2013, 11:31:13 AM

Just looking THE Avalon threads up..

So this is still à USER thread instead of USERS thread .. Huh

Well i look up next week maybe someone else got their Asic by then ..!!!

GL Roll Eyes

when i name my db tables. i name do do NOT name them plural.

user table. not users table, because each record is one user.

user.id

Topic: Avalon ASIC users thread - page 211. (Read 438602 times)