heat is wasted power, at least for us, so you should be right.

with these heatsinks and at least a fan blowing on them or a cover to direct the flow of the stock fan, you could try 43k resistor too.
Higher than 43k, and it will stop sending shares after a few minutes or seconds on mine, even after extra cooling.

Thanks for your feedback on ferrite beads removal and direct soldering for less heat

Btw, where did you get the heatsinks from?
the bigger ones look like they were made for CPU.
The small blue ones are nice and I like blue.

A2 is 12.5x the price of a blade for basically 13x the hashing power, so not that bad.
but in a month or so, the A2 will cost around 5000$ if not less.
the race has begun.
bye bye ROI.

I'm still curious if A2 can be hardware modded to 50% boost, but they are too pricey for now. I can't afford to pay one to play with it.

I'll stick with the zeusminer blizzard when it will hit 100$ in 1 or 2 weeks as my next project.

Too bad knc and innosilicon don't have entry level miners to test for cheap in the 150-200$ range.

For now, the blade was fun to mod, I'll keep working on my homemade powerboard, just as a proof of concept.
Without heavy changes to the powerboard, we will be limited around 7.3Mh/s, but the chips can provide 8Mh/s with 47k or 49.9k resistor on one blade and I want to reach it.

1HW error per minute on the first one looks high to me.

Can you check if by chance it is the top pcb that is giving more errors ?
Slight overheat of the chip on the upper board gives higher error count for me.

33k replaced with 43k

you can see small dissipators on top of Q6 Q7 Q8 on the upper right corner of the third picture.
plus larger ones on the back side of the pcb under the mosfets and choke

as promised, some more pics

The blade in action with the prototype cover




cover removed




the cooling of the chips.




closer view of both pcb modded
I used the secondary location on one of the boards because my 0805 resistor was keeping shorting with the nearest empty pad.
still needs some cleaning of the pads on this one




on the second board, the resistor is just where the 0603 33K was




and in case you want to try the bridge voltmod, here are the places to connect for pin 12 and pin 13 of up1509.
As usual, you do it at your own expense.
Beware you can fry your blade.

sorry, IRFH5302 on my board

resistors are as follow from left to right on my picture
33k (the one I replaced for the first mod)
27k
33k
36k <- should use this one when bridging



it's been discussed on this thread.
they are 2x IRF5300 and one IRFH5302

I ordered IRFH7004 to try as a direct replacement and IRFB7437PBF for an homemade powerboard

here is the bridge voltmod for the blade. still untested, I'll have to physically check on the board for voltage =1 to be sure

I checked on the schematics of the pod, so, basically, you are bridging R333, and then you enable voltage=1 all the time by bridging R337.
They are connected to pin 13 and 12 of the up1509, so quite easy to find.
 I'll have a look after breakfast.

I'll check on the PCB, but not sure I'll mod this one. I managed to get 2 hw errors on 8 hours @ 7.2MH/s with my custom cooling.
I'll buy a second one to try.

[2014-05-04 08:14:07] Runtime: 8 hrs : 48 mins : 59 secs
[2014-05-04 08:14:07] Average hashrate: 7.2 Megahash/s
[2014-05-04 08:14:07] Solved blocks: 0
[2014-05-04 08:14:07] Best share difficulty: 1.5M
[2014-05-04 08:14:07] Share submissions: 2352
[2014-05-04 08:14:07] Accepted shares: 2337
[2014-05-04 08:14:07] Rejected shares: 15
[2014-05-04 08:14:07] Accepted difficulty shares: 3346304
[2014-05-04 08:14:07] Rejected difficulty shares: 20864
[2014-05-04 08:14:07] Reject ratio: 0.6%
[2014-05-04 08:14:07] Hardware errors: 2
[2014-05-04 08:14:07] Utility (accepted shares / min): 4.42/min

if it's the one on the left of R52 (R139), you shouldn't need it, it's only used when switching the pod to different modes by software.
Just try the pod without it.
BTW, it should be R63, so 30k. you can replace it with your old R52(R139) that is 33k. The pod should work a little faster than 350Kh/s if you switch to this predefined mode.
the real R43 isn't connected anyway.

picture would have been better, but after checking the actual PCB, you are looking for R63, so 30K, or your old 33K or you can even leave it without the resistor, it should work

for some reasons, item restricted by ebay... can't see it

I used the ones from this auction.

http://cgi.ebay.fr/ws/eBayISAPI.dll?ViewItem&item=251251434021&ssPageName=STRK:MEWNX:IT&_trksid=p3984.m1439.l2649
1mm too small, but they are doing the job.
higher fins would have been better too

heatsinks on each mining chip. after overclocking with 43K resistor, they are getting hot. actually, the body of the blade is around 40°C. warm to touch.
heatsink under the power chips helps too, but I only have 1 that fits at the moment.

To stay on topic, since I'm back home, I made some improvements to the cooling.
Looks interesting for now.
heatsinks in each chips, forced air flow on the overheating parts with custom cover.
a few pics will be posted soon.

I also tried dual 25V 1000uF capacitors I had at hand, but they don't improve anything.
So 220uF 25V should be enough and a little smaller.

BTW, I managed to melt a power plug from my cables by trying back a 46K resistor. Simply too much current for it to handle if you use a single plug.

all the overheating parts will be doubled, including L2.
The up1509 will be powered by a doubled channel because the actual limiting factor is the capacity of the channel.

It should work since the other components aren't stressed that much, even when I overclock @ 7.5MH/s
We might be able to achieve around 8MH/s, maybe a little more.
On my pods, the chips are running at least at 104kh/s, so my math says 8.3Mh/s is doable with the good power board.

Hi,

you are close to what I want to do. I want to double the part of the board that feeds the up1509. this way, the chips will still have the same regulated power supply.
The idea is to try to bridge 2 pods before going on the blade.

I already identified easy connection points on the board, yellow/orange circles

red circles are our hot spots to cool down.
Funny thing is the initial design used 4 vrm (Q5 to Q8) for 8 chips and they decided to cut it down to 3 as on the 5chips pod, but for 40 chips. Should be related to the LTC only mode on the blades.

+5V will be collected after bat54C on R10, 12V at C12 and the return will be at R14, R117, R15, EC3, R31, R36 plus ground. only 9 connection points, should be ok.

 

A simple copy/paste of the actual board with same or upgraded components connected in parallel should work.
That's what I'll try.

@xcooling,
I replaced the 16v cap with 2 parallel 220uF 25V I had.
It will be replaced by 2 parallel 470uF 35V since I'm back at home.

best results so far
https://bitcointalk.org/index.php?topic=576784.msg6386770#msg6386770

1100Mhz with some HW errors 7.5Mh/s
or 1083MHz with almost no HW errors 7.2Mh/s

More tests soon since I'm back at home, but I'm still awaiting some heatsinks ordered 2 weeks ago.
We should be able to get 7.5Mh/s with almost no HW errors by cooling the chips.
Errors are appearing after some time and they are happening faster when the board gets hotter.

Next step will probably be an homemade improved power board that will be tried on one 5 chips pod before I try it on the blade.

more pics will be posted when I'll upgrade the PCB with cooling.

27MHz arrived at home, so I'll test it on one pod if it still hashes, I'll go for 33Mhz or close
I'm wondering where we can see the difference since the pod only uses the oscillator/crystal as a clock reference.
1000Mhz setting will actually be 1080, but where/how to check?

I'm using 43 +/- 5%

One of them is 42, the second one is 42.7.
Both PCB works correctly but very hot so active cooling .
Doubled power plugs, upgraded capacitors on one PCB and 0.75mm² wires to power the blades.