Avalon ASIC users thread

[GenTarkin]

Ive lost about 1GH on the latest firmware(20130814)... After 9h, my avg hashrate reports only 82000, previously it was about 83000.
Avy previously seemed to stay between 350-360, now it looks like it hardly gets above 350 =(

This is at nighttime too, when temps are 25, 36, 38C - Batch2

Do like the % changes to miner.php tho =)

Any reason for hashrate loss conman? Was there a change in autoclock algo?

No, I fixed a major latency issue, along with potential for duplicate shares (which would only show up as rejects). It's always hard to sell something that looks slower on the surface...

So, although slower, its working more effeciently? ... hence being better utilized .. hence a bit faster? =)

[1714213836]

1714213836

[1714213836]

1714213836

[1714213836]

1714213836

[-ck]

Ive lost about 1GH on the latest firmware(20130814)... After 9h, my avg hashrate reports only 82000, previously it was about 83000.
Avy previously seemed to stay between 350-360, now it looks like it hardly gets above 350 =(

This is at nighttime too, when temps are 25, 36, 38C - Batch2

Do like the % changes to miner.php tho =)

Any reason for hashrate loss conman? Was there a change in autoclock algo?

No, I fixed a major latency issue, along with potential for duplicate shares (which would only show up as rejects). It's always hard to sell something that looks slower on the surface...

So, although slower, its working more effeciently? ... hence being better utilized .. hence a bit faster? =)

LOL, overall about the same. However less shares will be lost across block changes, which affects p2pool a lot more than regular pooled mining.

I've spent a lot of time fighting with myself about the right approach to finding the balance between latency and apparent hashrate (aka marketing).

[aneutronic]

Very nice data.  Where is your tip jar?

Thanks and thanks. No tip jar for so little

I consider my Avalon batch#3 unit a nice done product. It came with a 850 W PS. It's an Enermax Gold able to put 840W on 12V. Out the box Avalon unit can run much higher then specifications.
But I'm not so sure that it will be able to pay itself, even overclocked like this
So I'm willing to try to push it more. I believe there are two problems which need to be solved:

• Cooling is less then optimal. The modules (heatsinks) are much hotter at the end. But the exhausted air it's only warm. I have some ideas of how cooling can be improved.
• Voltage on the chips must be raised a little. It's now around 1150mV (both from schematics and measurements). Specification of the chip is 1200mV. I believe it will be no problem to push it around 1240mV, which I hope it will allow it to run at 375MHz. But this will mean ~25% more power, so cooling has to be solved first. And this would be a hardware mod, 40 resistors to solder.

And during the time the unit is upgraded, it does not produce, or at least not at full capacity

+1  and ditto, you're reading my mind!  There's an overclock thread started here: https://bitcointalk.org/index.php?topic=271575.0

[GenTarkin]

Very nice data.  Where is your tip jar?

Thanks and thanks. No tip jar for so little

I consider my Avalon batch#3 unit a nice done product. It came with a 850 W PS. It's an Enermax Gold able to put 840W on 12V. Out the box Avalon unit can run much higher then specifications.
But I'm not so sure that it will be able to pay itself, even overclocked like this
So I'm willing to try to push it more. I believe there are two problems which need to be solved:

• Cooling is less then optimal. The modules (heatsinks) are much hotter at the end. But the exhausted air it's only warm. I have some ideas of how cooling can be improved.
• Voltage on the chips must be raised a little. It's now around 1150mV (both from schematics and measurements). Specification of the chip is 1200mV. I believe it will be no problem to push it around 1240mV, which I hope it will allow it to run at 375MHz. But this will mean ~25% more power, so cooling has to be solved first. And this would be a hardware mod, 40 resistors to solder.

And during the time the unit is upgraded, it does not produce, or at least not at full capacity

Can a voltage increase be safely done w/o drawing to much current through the power circuitry? Cuz they are only rated for something like 20 amps, and I thought at default voltage / clocks we are pushing pretty close to that....

[ProfMac]

I am running 0703 and kind of scared to update to 0813/0814.  One person got bricked, several others report decreased hash rates.  Anyone got positive results yet?

Me?

I let 20130810-1 run overnight.  The frequency dropped until it was about 304 MHz, and was steadily dropping when I stopped it.  The stable frequency under 20130703 is 345-347.  I am mining under eligius, and the recent behavior is online.  I returned to 20130703 an hour or so ago.

My temperature ran at 49°C and I had the idea that there was a "<" vs "<=" problem somewhere, or it's evil twin ! > vs if "<" then .  However, I tried to raise the temperature set-point to 51, and this did not help.

Wrong one. 10-1 is known to be slow. He is asking about 20130814.

Updated: [Firmware Version] => 20130814
about 9:00 AM CDT today.
I am watching the pool statistics reported by eligius.

No opinion yet, I'll let it run for a while.
My temp1 runs about 20-22°C, my ambient temp in the basement runs 65-68°F.

[thejestre]

Updated: [Firmware Version] => 20130814
about 9:00 AM CDT today.
I am watching the pool statistics reported by eligius.

No opinion yet, I'll let it run for a while.
My temp1 runs about 20-22°C, my ambient temp in the basement runs 65-68°F.

I think I done fcked myself a bit with firmware , but not too bad.

Had some _really_ strange experiences flashing to 20130813-1 then trying to go back to 20130703.  Let's just say things were kind of strange - made sure to NOT check the "keep settings" box, but settings were held over. 

Anyway Prof, looks like your hashes are nicer than mine right now.  Mind posting the changes you made after you flashed to 20130814 [options set, flags, etc]?  I might try that firmware when I get home tonight, my Avalon is hashing around 70 Gh/s right now, and that feels a bit slow, considering it was closer to 80 before I mucked things up by trying to "fix" it.  Although I just read ck's comment about hash rate being more marketing than power.

Interested to see others' results.

Thanks,

_theJestre

[GenTarkin]

How is MHSav calculated? Is it based off valid diff1 work?
Meaning HW errors is already accounted for in MHSav?

By my calculations based on clockspeed, looks like it does account for HW% error.

Also does --avalon-temp clock down regardless if --avalon-auto is used or not? Also, how does that relate to --avalon-freq if setting it to one value... will --avalon-temp still clock down?

[ProfMac]

Updated: [Firmware Version] => 20130814
about 9:00 AM CDT today.
I am watching the pool statistics reported by eligius.

No opinion yet, I'll let it run for a while.
My temp1 runs about 20-22°C, my ambient temp in the basement runs 65-68°F.

I think I done fcked myself a bit with firmware , but not too bad.

Had some _really_ strange experiences flashing to 20130813-1 then trying to go back to 20130703.  Let's just say things were kind of strange - made sure to NOT check the "keep settings" box, but settings were held over. 

Anyway Prof, looks like your hashes are nicer than mine right now.  Mind posting the changes you made after you flashed to 20130814 [options set, flags, etc]?  I might try that firmware when I get home tonight, my Avalon is hashing around 70 Gh/s right now, and that feels a bit slow, considering it was closer to 80 before I mucked things up by trying to "fix" it.  Although I just read ck's comment about hash rate being more marketing than power.

Interested to see others' results.

Thanks,

_theJestre

I kept the old settings.  My file /etc/config/cgminer is very similar to the following.  I am set to fail-over, not balance, but I don't see that interpretation reading this file.  I usually have the chip frequency option set very low, perhaps 256, because the fans stop for a while when I reset things, and I want to stop building up heat.

Code:

config cgminer 'default'
	option pool_balance '  '
	option pool1pw 'none'
	option api_allow 'W:127.0.0.1,192.168.0.0/24'
	option pool1url 'stratum+tcp://mining.eligius.st:3334'
	option pool1user '18bLcVkviErQi75zB8X39jZXxHNpSZggdC'
	option miner_count '24'
	option more_options '--avalon-auto --avalon-freq 256-350 --avalon-temp 50'
	option chip_frequency '347'

[lastbit]

Can a voltage increase be safely done w/o drawing to much current through the power circuitry? Cuz they are only rated for something like 20 amps, and I thought at default voltage / clocks we are pushing pretty close to that....

Around 17-18 Amps @350Mh/s and 1150mV I reckon. Transistors are not a problem. Inductor seems to be more closer to the limit. In schematics it's 1uH. But on board I have 1.5uH, which might be specified at only 27A.
Anyway, I'm willing to take the risk of 25% more power.
 

[lastbit]

So, here is my first step in overclock batch#3 quest.

I have took out back panel. To much of my surprise, temperature droped 2C !!!
With original fans=100%, @358MHz:
Avalon unit as it came: 24/66/46
Back panel removed: 24/64/45
I have put, as a proof of concept, some pieces of cardboard to force air through fins and at the surface of the boards, especially at the output region of modules. I have also obstructed air path above the modules (where modules cables connect to modules).
No back panel and some cardboard: 24/59/44.
I'd say a success!


Unfortunately this did not help at all regarding frequencies. Still 8%HW errors @375Mhz (like before).
 
I also think about replacing thermal pad with thermal paste. I have noticed PCB "bulges" between the screws, so I believe thermal contact is less than optimal with the pad:


I have also noticed modules are not parallel. they are slightly tilted, as you can see in the first image. If I would separate the twin connector PCB (for two modules), problem would be solved. I'll do it anyway when I'll work on modules to have only one module stopped at a time. But I believe it's not so important.

[S M I L Y]

@lastbit.

Have you reworked a VR with the 1.5uH?   How much current have you been able to push?

[cypherdoc]

hmmm, anyone ever see this?:

[lastbit]

@lastbit.

Have you reworked a VR with the 1.5uH?   How much current have you been able to push?

I haven't. I have to solve cooling first. I hope next week.

[lastbit]

Does anybody know if it's possible of not to clock each module independently?

[S M I L Y]

@lastbit.

Have you reworked a VR with the 1.5uH?   How much current have you been able to push?

I haven't. I have to solve cooling first. I hope next week.

Im in the process of watercooling mine.   Should be done next week as well.  I plan on switching out the sense resistors(R10 i believe) on the VR to get to 1.2v.  

Where can I find the schematics for this thing?

[lastbit]

@lastbit.

Have you reworked a VR with the 1.5uH?   How much current have you been able to push?

I haven't. I have to solve cooling first. I hope next week.

Im in the process of watercooling mine.   Should be done next week as well.  I plan on switching out the sense resistors(R10 i believe) on the VR to get to 1.2v.  

Where can I find the schematics for this thing?

Question more difficult than I believed
I have downloaded pdfs from somewhere, I believe it was from github. Unfortunately now I'm only finding this: https://github.com/BitSyncom/avalon-ref. Unfortunately files are in "ALTIUM DESIGNER" format.

R10 is 10.5k, connecting pin2 (FB) of TPS40193 to GND.

[S M I L Y]

Thanks lastbit!

Ill just pull the spec sheet for the controller.

Ill probably just drop to a 6k resistor and a 10k pot in series till I can figure out what voltage it takes to get to 400mhz.

Avalon chip spec shows 1.2v typical 1.32 max!

I wonder if I can shut off all of the individual modules except for the one Im working on.

[-ck]

How is MHSav calculated? Is it based off valid diff1 work?
Meaning HW errors is already accounted for in MHSav?

By my calculations based on clockspeed, looks like it does account for HW% error.

Also does --avalon-temp clock down regardless if --avalon-auto is used or not? Also, how does that relate to --avalon-freq if setting it to one value... will --avalon-temp still clock down?

MHSav is calculated from valid diff1work and does not count hw errors.
Avalon temp works regardless.
Avalon freq is the frequency range that avalon auto will use and ignores temperature unless it is over the optimal range - then it will clock down if the fans can't keep it down.

[WinTame2012]

Moved from bitparking to 50btc and now I'm constantly getting a lot of Rejected shares at Diff8  
Does anybody have fixed issue like this?

Edit: FW both 20130703 and 20130723

[lastbit]

Thanks lastbit!

Ill just pull the spec sheet for the controller.

Ill probably just drop to a 6k resistor and a 10k pot in series till I can figure out what voltage it takes to get to 400mhz.

Avalon chip spec shows 1.2v typical 1.32 max!

I wonder if I can shut off all of the individual modules except for the one Im working on.

By module I assume you mean a group of 10 asics with common VR.
Each of the big module is connected with 40pin. I believe you can visually and ohmmeter check what pins go to each group and make a custom ribbon cable with only that signals. And of course disconnect the other big modules.