Avalon ASIC users thread

[GandalfG]

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.

For 1.3V  simple add  47k resistor parallel to R10 (10.5k) - pin2 U1 to GND
Unfortunately you have to do for each sub-module 24 or 32
Work on 395MHz 1.2% HW

[1714197233]

1714197233

[1714197233]

1714197233

[1714197233]

1714197233

[GenTarkin]

1.3v ... wont that kill the VR? ... like was stated we are at 17-18amps already and it handles 20amps...

[lastbit]

1.3v ... wont that kill the VR? ... like was stated we are at 17-18amps already and it handles 20amps...

From where was the info that it can handle only 20?

[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.

@S M I L Y
I found module backplane schematics. Indeed each group is connected with 4 signals (2 in, 2 out).

[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.

For 1.3V  simple add  47k resistor parallel to R10 (10.5k) - pin2 U1 to GND
Unfortunately you have to do for each sub-module 24 or 32
Work on 395MHz 1.2% HW

Indeed 47k in parallel would give ~1280mV, a 11% increase over default.
Assuming chips are able to take also 11% more frequency, you'll need a ~40% more powerful PS, i.e. 1200W for 4 modules.
Not very power efficient    

[GenTarkin]

1.3v ... wont that kill the VR? ... like was stated we are at 17-18amps already and it handles 20amps...

From where was the info that it can handle only 20?

I dont know where the links to the schematics are anymore... but, it was when I Was reading them. I looked up the main VRM controller chip(fuck if I know what its called) and as per the manufacturers specification the max amps out was like 20amps...

[crazyearner]

lastbit in terms of what your doing you might want to try this as modding not uploaded all details yet but this s on a batch 2 unit and working out the layout for batch 3 will have details on my website when I get around to it.

Avalon before pictures







Avalon after air flow mods.






Temps before and after more or less the same but not using as much fan power and after mining 24 hours sold this is running nice and cool with oc to 84GH also fans never go above 1800 now so lot better results still doing other mods in time to it to get better air flow. This is with no extra fans added maybe get even better results with some delta fans in but their too noise


Have also taken unit upto 90GH however too many hw% errors however temps solid 50c so not doing too bad. Also added new 860w corsair digital psu working perfect never takes over 730w

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

Change your diff to 64 on 50btc under settings for miner is probs why you getting too many rejected shares.

[S M I L Y]

Thanks lastbit and GandalfG for the info.   I have access to a nice 12ghz oscilloscope and will take noise and ripple measurements of the output of the VR and see if it gets unstable past 20a.  

I am running a 1200w supply so I should be fine on the supply side. Going to be switching to 220v as well to gain a bit more efficiency.

I will be running coolant to a radiator outside for watercooling...  Unfortunately it has been 105f or 40c for the last month outside.  Which is even more reason I want to get the 800+watt this thing produces out of my house!

The main issue I see with cooling is getting the heat from the chips into the heatsink.  lastbit's photo earlier show the exact bowing away Im seeing.
As the board heats up you can watch it bow away.   I have a clamping solution im working on as well to hold the chips flat against the heatsink.
 

Just took some measurements.  At the top of the board closest to the VR im seeing 1.2v at the bottom im seeing 1.15v!  Im surprised to see so much of a drop!  They must have cheeped out on the amount of copper on the board.

[GenTarkin]

  I have access to a nice 12ghz oscilloscope and will take noise and ripple measurements of the output of the VR and see if it gets unstable past 20a.  

That would be awesome =)

[Ytterbium]

If all you have to do is lower some resistance to get it to 1.2v, or 1.3v - couldn't you actually just solder another resistor in parallel to lower the resistance the correct amount?  You wouldn't need to remove anything that way.

One problem, of course is that the boards would run hotter even at the same clock speeds...

[GandalfG]

Thanks lastbit and GandalfG for the info.   I have access to a nice 12ghz oscilloscope and will take noise and ripple measurements of the output of the VR and see if it gets unstable past 20a.  



Just took some measurements.  At the top of the board closest to the VR im seeing 1.2v at the bottom im seeing 1.15v!  Im surprised to see so much of a drop!  They must have cheeped out on the amount of copper on the board.

I measure my board with my 100MHz oscilloscope with 1.3V mod and did not see noise on power lines. I check it now after over 10 days mining.
About drop:
Standby: 1.298V
395MHz:
1.278 on coil VR and nearest capacitor  
1.268  first two chip
1.228 last chip.

I solder  silver-plated wire 0,8mm across  one board and check how its help.  

[GandalfG]

If all you have to do is lower some resistance to get it to 1.2v, or 1.3v - couldn't you actually just solder another resistor in parallel to lower the resistance the correct amount?  You wouldn't need to remove anything that way.

One problem, of course is that the boards would run hotter even at the same clock speeds...

Its no problem set 1.3V or higher. VR can supply more amps.
The real problem now are cheap PCB Too thin copper layer generate huge  voltage drop on end of board.

[lastbit]

I measure my board with my 100MHz oscilloscope with 1.3V mod and did not see noise on power lines. I check it now after over 10 days mining.
About drop:
Standby: 1.298V
395MHz:
1.278 on coil VR and nearest capacitor  
1.268  first two chip
1.228 last chip.

I solder  silver-plated wire 0,8mm across  one board and check how its help.  

Thanks for the info. It looks like it's more work to do. Loosing 3% hashrate for voltage drops would be sad.

[Ytterbium]

Well, you could always buy blank Klondike boards, then remove the chips and re-solder them .  A 3-module Avalon would take 15 K16s.  If you could run them at 400, you'd get a ~14% improvement over 350.

[GandalfG]

If  migrate, only to  Burnin board.

1 -20 chip instead 16.
2 - Very stable work at 400MHz
3 - programmable VR

[lastbit]

lastbit in terms of what your doing you might want to try this as modding not uploaded all details yet but this s on a batch 2 unit and working out the layout for batch 3 will have details on my website when I get around to it.

Thanks crazyearner. Indeed something along the lines of what you did. But batch#3 id different, each 2 modules are arranged in mirror with heatsinks facing one another. And I'd like to keep high airflow on the pcb and chips surfaces too because I believe it may play an important part in cooling.

[ProfMac]

lastbit in terms of what your doing you might want to try this as modding not uploaded all details yet but this s on a batch 2 unit and working out the layout for batch 3 will have details on my website when I get around to it.

Thanks crazyearner. Indeed something along the lines of what you did. But batch#3 id different, each 2 modules are arranged in mirror with heatsinks facing one another. And I'd like to keep high airflow on the pcb and chips surfaces too because I believe it may play an important part in cooling.

The pattern of holes attaching the heatsink to the bottom plate is different between batch #2 and batch #2.  Batch #2 uses 7 holes, Batch #3 uses 5 holes.

Has anyone made a hole matching plate, perhaps 3 mm aluminum with countersunk holes?

[GenTarkin]

Thanks lastbit and GandalfG for the info.   I have access to a nice 12ghz oscilloscope and will take noise and ripple measurements of the output of the VR and see if it gets unstable past 20a.  



Just took some measurements.  At the top of the board closest to the VR im seeing 1.2v at the bottom im seeing 1.15v!  Im surprised to see so much of a drop!  They must have cheeped out on the amount of copper on the board.

I measure my board with my 100MHz oscilloscope with 1.3V mod and did not see noise on power lines. I check it now after over 10 days mining.
About drop:
Standby: 1.298V
395MHz:
1.278 on coil VR and nearest capacitor  
1.268  first two chip
1.228 last chip.

I solder  silver-plated wire 0,8mm across  one board and check how its help.  

Have any amps / watts per board power measurement?
Also, was this a b2 or b3? Did they change anything on the b3 miner boards?

[S M I L Y]

I measure my board with my 100MHz oscilloscope with 1.3V mod and did not see noise on power lines. I check it now after over 10 days mining.
About drop:
Standby: 1.298V
395MHz:
1.278 on coil VR and nearest capacitor  
1.268  first two chip
1.228 last chip.

I solder  silver-plated wire 0,8mm across  one board and check how its help.  

Thanks for the info. It looks like it's more work to do. Loosing 3% hashrate for voltage drops would be sad.

50mv drop is horrible!

adding a couple thick wires from the output of the VR to the last chips should help a lot.     although the problem could be a return path issue.  I dont have my unit in front of me at the moment.  I assume the bottom layer is Ground and the top is VCC?    There may be an issue with the ground vias/pins not being able to handle the current.

[GandalfG]

Thanks lastbit and GandalfG for the info.   I have access to a nice 12ghz oscilloscope and will take noise and ripple measurements of the output of the VR and see if it gets unstable past 20a.  

Just took some measurements.  At the top of the board closest to the VR im seeing 1.2v at the bottom im seeing 1.15v!  Im surprised to see so much of a drop!  They must have cheeped out on the amount of copper on the board.

I measure my board with my 100MHz oscilloscope with 1.3V mod and did not see noise on power lines. I check it now after over 10 days mining.
About drop:
Standby: 1.298V
395MHz:
1.278 on coil VR and nearest capacitor  
1.268  first two chip
1.228 last chip.

I solder  silver-plated wire 0,8mm across  one board and check how its help.  

Have any amps / watts per board power measurement?
Also, was this a b2 or b3? Did they change anything on the b3 miner boards?

I did not measure amps. It is Avalon clone based on B2 design that I build self. Buy PCB and parts from Strombom.
Actually have only 50 chip and patiently wait for supply from Bitsyncom.

50mv drop is horrible!

adding a couple thick wires from the output of the VR to the last chips should help a lot.     although the problem could be a return path issue.  I dont have my unit in front of me at the moment.  I assume the bottom layer is Ground and the top is VCC?    There may be an issue with the ground vias/pins not being able to handle the current.

Yes that huge drops are very undesirable. As I stated I solder wire and make measure again. I'm a little surprised that the original B1/B2 have a similar problem.
How about B3 machines that have changed design ? I see on one photos hash unit are probably grouped in 4 on one PCB and added thermal sensor on it.