Avalon Overclocking Thread - Share Your Results and Tips for others

[greaterninja]

Avalon Overclocking deserves its very own thread.
The purpose of this thread is for the community to share their tips and results on overclocking their Avalon's so we can maximize hashing output.


Please Post in this format:


Batch#:  3
Modules:3
PowerSupply:850w Seasonic Gold Modular
Extra CoolingSetup:  $20 Box Fan added; 75F Ambient
[Firmware Version] => 20130723
Chip Frequency(Default: Advance):  350(insane); 354 is actual read out
More Options(Default: --quiet):--avalon-auto
HashRate:   ~83GH/s is average; Max is 87GH/s but it eventually drops.
Additional Info(whatever I may have missed):
Uptime:  5 days

Thanks Guys, I believe we can push our units further!  Probably above 90Gh/s stable.

**Update 8/15/2013-
Please Post in this format:


Batch#:  3
Modules:4
PowerSupply: 850w Seasonic Gold Modular
Extra CoolingSetup:  $20 Box Fan added; 75F Ambient A/C Room Temp
[Firmware Version] => 20130723
Chip Frequency(Default: Advance):  350(insane); 354 is actual read out.
More Options(Default: --quiet):--avalon-auto
HashRate:   ~110.5GH/s is average; Max is 112.5GH/s
Additional Info(whatever I may have missed):
Uptime:   3 days (I reapply CGMINER settings each day as the miner does sometimes drop its GH/s to less than half its potential after about 24 hours.  I do this to keep production of BTC constant and maximize potential.)


***NEW UPDATE 8/19/2013  My 850Watt Seasonic GOLD PS failed. I was running 4 modules at 358mhz.  111.5 GH/S for 2 weeks straight.   I unhooked a 3rd module and put a 750watt bronze on it.   The AC in my place went out today and I live in AZ...so that could be why or it could be because I should be running 1000+ watts.  Going forward I am going to buy 2 x 1300w Gold Rosewill PS from newegg. 

[1713493823]

1713493823

[1713493823]

1713493823

[greaterninja]

Reserved

[aneutronic]

+1   If the other asics deliver on time in the next couple of months it will be tough to break even.    

Batch#:  3
Modules:3
PowerSupply:850w Seasonic Gold Modular  (curses at seasonic's damn 6 + 2 PCIE connectors that tripped me up   )
Extra CoolingSetup: 2x $20 Box Fan added drawing outside air when cooler and always exhausting hot ; ~ 60 - 80F Ambient so far, no AC, I'm poor.
[Firmware Version] => 20130723
Chip Frequency(Default: Advance):  350(insane)
More Options(Default: --quiet):--avalon-auto --avalon-freq 340-355 --avalon-fan 100
HashRate:   ~83GH/s is average, I can run it harder but there were too many rejects so a 355 limit holds these to about ~0.5% so far
Power: maxes out at ~730w

Additional Info
I hit a wall at about 355mhz even with < 60 ambient temps, I feel overvolting will be necessary to go futher.

burnin's thread has some overvolt data here:  https://bitcointalk.org/index.php?topic=179769.msg2738301#msg2738301

His board is pretty nice having the voltage software selectable, we're going to have to break out the tools to overvolt our avys.    

He hit 409 mhz on 1.28v stable on air cooling. I calculate a standard value of 8.87k for R10 should produce 1273mV, the only question is the current capacity of the regulator circuit. I have no experience in smps design but the inductor looks good for ~30A heat rating, not sure about the output stage.
I'm hoping there are some smps designers on this forum that can help out.

In any case before attempting this I thought it would be a good idea to make sure the chip to heatsink interface is optimum. The water cooling thread has some good info on replacing the questionable thermal pad with heatsink paste.  https://bitcointalk.org/index.php?topic=211825.msg2687268#msg2687268

 

[MXRider]

I'm interested in power consumption so I'd appreciate if someone could report that.

[GodfatherBond]

Batch#:  3
Modules:4
PowerSupply:1250w Corsair Gold
Extra CoolingSetup:  No extra cooling; 75F Ambient
[Firmware Version] => 20130723
Chip Frequency(Default: Advance):  350(insane)
More Options(Default: --quiet):--avalon-auto
HashRate:   ~108GH/s is average; Max is nearby 120GH/s but it eventually drops.

[aneutronic]

I'm interested in power consumption so I'd appreciate if someone could report that.

Good idea... added power reading.

[aneutronic]

There's a good post about overclocking here by lastbit who also has concerns about reaching break even:  https://bitcointalk.org/index.php?topic=140539.msg2934835#msg2934835

I'd be interested to know if you measured more than one regulator and if so how close were the voltage readings between regulators?

[S M I L Y]

I'm interested in power consumption so I'd appreciate if someone could report that.

Some power measurements for Avalon.

Unit is batch#3 w/4 modules, fw 20130723.
Temperature target 65C. Keep in mind this is equiv with 45 for batch#1&2.
Temp1 (input temperature) is 21C .
Temp3 (output temperature) is roughly 15C lower then Temp2.
Wall is wall power in watts.
Errors is percentage of HW errors versus diff1 shares.

freq	wall	fan1	fan3	temp2	errors
256	685	2280	2040	60
270	722	2280	2160	60
282	755	2400	2160	61
300	803	2400	2160	63	0.94%
325	873	2520	2280	66	1.08%
350	943	3480	3240	63	1.35%

Produces arround 109-110Gh/s @350MHz, confirmed by btcguild (more or less).
At 375MHz unit cannot maintain 65C temperature target. Hash is less, around 105Gh/s, due to large percentage of HW errors. 

[GenTarkin]

AFAIK from reading and looking at design schematics, the main regulator or w/e it is is only capable of around 20A output. We may be gettin pretty close to that on just stock overclocks...
So, overvolting safely may be out of the question.

[jermwerty]

Batch#:  3
Modules:4
PowerSupply: 850w Seasonic Gold Modular
[Firmware Version] => 20130723
Chip Frequency(Default: Advance):  350(insane); 354 is actual read out.

Umm....

850W with 4 modules at 350+mhz?

I hope you have another PS laying around for when this one goes up in smoke.

For comparison, I have a good 910W PS on a B3 @ 325mhz and it pulls ~915W from the wall, a 65W increase from 300mhz which was 850.  My estimates put 350Mhz B3 4 module at 980W give or take from the wall, or basically you need a 1KW+ supply guaranteed! 

[GenTarkin]

Batch#:  3
Modules:4
PowerSupply: 850w Seasonic Gold Modular
[Firmware Version] => 20130723
Chip Frequency(Default: Advance):  350(insane); 354 is actual read out.

Umm....

850W with 4 modules at 350+mhz?

I hope you have another PS laying around for when this one goes up in smoke.

For comparison, I have a good 910W PS on a B3 @ 325mhz and it pulls ~915W from the wall, a 65W increase from 300mhz which was 850.  My estimates put 350Mhz B3 4 module at 980W give or take from the wall, or basically you need a 1KW+ supply guaranteed! 

He may be running close but under the PSU's max. Seasonics are tank PSU's.
Assuming your PSU is 90% effecient at that sort of powerdraw(which its likely less) that roughly a 90watt overhead from the wall. 915 - 90, leaves actual load near 810watts. His seasonic should have no prob with that.

[jermwerty]

He may be running close but under the PSU's max. Seasonics are tank PSU's.
Assuming your PSU is 90% effecient at that sort of powerdraw(which its likely less) that roughly a 90watt overhead from the wall. 915 - 90, leaves actual load near 810watts. His seasonic should have no prob with that.

Oh I agree SeaSonics are good.  I have an older one that mined a lot on GPUs.

What I was pointing out was 915W is @ 325mhz.  From the post above mine that reported draws at different speeds, I think its safe to say 950W-975W appears to be the "wall" draw @ 350mhz.  That puts him right at 860-885W PSU rating, which while it may work now WILL reduce lifespan of the PSU.  Just think about it, there is a reason Seasonic rated it 850 NOT 900 

I think the one benefit of being a former GPU miner is I personally feel I have a mini-bachelor in PSU performance....  Multi Rail, Single Rail, efficiency, ripple, brand, ratings... all plays into the mix.  At home today I have SilverStone, Seasonic, Thermaltake, PC Power & Cooling, Enermax,  CoolerMaster running today all 700-1KW.

But I digress, as long as it works, and you have a spare, go ahead and run that bad boy hard!

[greaterninja]

He may be running close but under the PSU's max. Seasonics are tank PSU's.
Assuming your PSU is 90% effecient at that sort of powerdraw(which its likely less) that roughly a 90watt overhead from the wall. 915 - 90, leaves actual load near 810watts. His seasonic should have no prob with that.

Oh I agree SeaSonics are good.  I have an older one that mined a lot on GPUs.

What I was pointing out was 915W is @ 325mhz.  From the post above mine that reported draws at different speeds, I think its safe to say 950W-975W appears to be the "wall" draw @ 350mhz.  That puts him right at 860-885W PSU rating, which while it may work now WILL reduce lifespan of the PSU.  Just think about it, there is a reason Seasonic rated it 850 NOT 900 

I think the one benefit of being a former GPU miner is I personally feel I have a mini-bachelor in PSU performance....  Multi Rail, Single Rail, efficiency, ripple, brand, ratings... all plays into the mix.  At home today I have SilverStone, Seasonic, Thermaltake, PC Power & Cooling, Enermax,  CoolerMaster running today all 700-1KW.

But I digress, as long as it works, and you have a spare, go ahead and run that bad boy hard!

Worst case scenario is my 850W burns up and I am out $160.   Probable scenario is it continues to work for a few months or for years and I do not have to shell out more $$$ for a 1000-1300w PSU.  OR It does burn out and I sent it back for replacement (7 year warranty).

At anyrate, I have to cut my costs.  spending another $130-$250 on a PSU is a lot of money I really do not want to spend.

[cryptograd]

Batch#: 2
Modules:3
PowerSupply:750w Corsair Gold
Extra Cooling Setup: Temp controlled room set to 70F hottest module runs at 48C
Chip Frequency(Default: Advance):  325
More Options: Fans on constant 100%
HashRate: 82GH/s is average; Max is 92GH/s for random temporary spikes.

[crazyearner]

Batch#: 2
Modules:3

PowerSupply:860w Corsair Platinimum Digital removed the one that come with it in this picture and mods done at time new pics to be uploaded as more mods been done.

Extra Cooling Setup: Custom mods done to control and force airflow and more directed.
Chip Frequency manually set:  355-370 depending on temps in room and also hw errors
More Options: Fans on constant 100%
HashRate: 85GH/s is average; Max is 90+GH/s not fully tested this due to too many HW errors.

Avalon before pictures







Avalon after air flow mods.
As you can see in first pic using the cable to hold them in place as plenty of ribbon to use to make a little fold to hold plastic sheeting in place and where the molex connect theirs about a 0.5 to 1.5ml gap to allow to create a curve for the plastic to be molded into shape to allow to curve around and make the mod fit and old in place.


On this picture you will see where at the top this is still covering the gaps and also where air flows so I needed to push future to allow air flow to flow through the heat sink as I was a at first blocking this so had to gently guide it through and make marks on it where to mold curves in for cornering.




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

[Ytterbium]

It sounds like it's the voltage, rather then the heat that's causing problems at higher speeds. And, since the voltage isn't adjustable the only way to speed it up would be to replace the regulators (I guess?) which would be pretty invasive.

Someone actually built a water block but they still got around 82-84Gh/s.   

[lastbit]

AFAIK from reading and looking at design schematics, the main regulator or w/e it is is only capable of around 20A output. We may be gettin pretty close to that on just stock overclocks...
So, overvolting safely may be out of the question.

Why you are saying is capable of only 20A?

[crazyearner]

It sounds like it's the voltage, rather then the heat that's causing problems at higher speeds. And, since the voltage isn't adjustable the only way to speed it up would be to replace the regulators (I guess?) which would be pretty invasive.

Someone actually built a water block but they still got around 82-84Gh/s.   

Would need a voltage mod and thats not possible unless something is invented to work along side it to provide the extra power it needs but still  who knows what will happen in the next batch and gen 2 tech will have to wait.

Yeah I seen that thread for the water cooled unit guess needs better tanks and liquid for the unit. Am tempted at doing an mineral system and auction it off as am still considering to sell unit next week but will see how things pan out

[GenTarkin]

AFAIK from reading and looking at design schematics, the main regulator or w/e it is is only capable of around 20A output. We may be gettin pretty close to that on just stock overclocks...
So, overvolting safely may be out of the question.

Why you are saying is capable of only 20A?

Grr, I cant remember where yifu posted the design schematics. But the one that shows the layout for each miner board, shows in the VRM section. The main controller chip .. is it a TI or something? Look it up and its max output was like 20amps

[aneutronic]

Schematics are here: https://github.com/BitSyncom/avalon-ref/tree/master/PDF/SCH

The reg is a TPS40193.

Code:

The range of short circuit current thresholds that can be expected is shown in Equation 2 and Equation 3.

I SCP(max)=V ILIM(max)/RDS(on)min

I SCP(min)=V ILIM(min)/RDS(on)max

where
•
•
•
ISCP is the short circuit current
VILIM is the short circuit threshold for the low-side MOSFET
RDS(on) is the channel resistance of the low-side MOSFET
(3)
Note that due to blanking time considerations, overcurrent threshold accuracy may fall off for duty cycle greater
than 75% with the TPS40192, or 88% with the TPS40193. The reason for this is that the over current comparator
will have only a very short time to sample the SW pin voltage under these conditions and may not have time to
respond to voltages very near the threshold.
The short circuit protection threshold for the high-side MOSFET is fixed at 550 mV typical, 400 mV minimum.
This threshold is in place to provide a maximum current output using pulse by pulse current limit in the case of a
fault. The pulse will be terminated when the voltage drop acros the high side FET exceeds the short circuit
threshold. The maximum amount of current that can be specified to be sourced from a converter is found by
Equation 4.

I OUT(max)=V ILIM(min)/RDS(on)max

where
•
•
•
IOUT(max) is the maximum current that the converter is specified to source
VILIMH(min) is the short circuit threshold for the high-side MOSFET (400 mV)
RDS(on)max is the maximum resistance of the high-side MOSFET
(4)
If the required current from the converter is greater than the calculated IOUT(max) , a lower resistance high-side
MOSFET must be chosen. Both the high side and low side thresholds use temperature compensation to
approximate the change in resistance for a typical power MOSFET. This will help couneract shifts in overcurrent
thresholds as temperature increases. For this to be effective, the MOSFETs and the device must be well coupled
thermally.

(edit: only one FET on the board)
The high side FET is a PSMN2R2-25YLC which have an IDS of 100A and  an RDS(on)max of 2.4mohm so 400mv/2.4mohm=166.66A

Again I'm not a power supply expert but it looks like the limiting factor of this circuit is the inductor.
The schem says 1uh, I believe someone mentioned he had 1.5uh chokes in his unit so either ~32A or ~27A depending. I'll have to check mine the next time I have it open.