ASICminer Blades running slower than expected?

[aznatama]

20 Blades v2
2 backpanes
2 HP 1000w power supplies

At first, I used a seasonic 1250w PSU, and w/ 10 blades, it would trip the psu internal breaker and the PSU would turn off. I ran it using connections from 3 separate rails.  Could be I didn't use enough wiring for the amp pull.  Voltage at the blade measured only 11.7v.  I added another Seasonic 480w PSU to the setup, and that allowed it all to run just fine.  Voltage still at 11.7 though. (ATX PSU temp solution for th eweekend til HP PSU's arrive)

I now have the backpane HP server power supply, and voltage at the board is measuring a stable 12.7v

Both instances, my hash rate via the web interface shows only ~9mh/s
Efficiency is showing only 90-93% at the web interface

My pool will report a hash rate that fluctuates between 9~11 per board (180~220gh/s)
I also have a rather high (to me) rejected/stale share rate of several %

Settings:
Worker difficulty set at pool: 128
Backpane 1 -> Netgear 16p switch
Backpane 2 -> Netgear 16p switch
2x Netgear 16p switched -> Netgear Tomato wireless router
BFGminer proxy run on computer w/ wireless connection to router
Wireless network traffic very minimal, but proxy host is also running GPU scrypt mining

Could the network connection be the issue here?  I have a 75ft cat6 cable coming in soon, which I will use to connect my computer directly to one of the hubs.  Should I get a 24p managed hub instead?  traffic seems low enough that this setup won't be an issue.  Should I daily-chain the hubs and have only 1 connection go to the router?

Thanks

[sidehack]

Some power supplies in the 1000+ range aren't rated to output full power unless provided with 220V input. You might check on the input requirements for that PSU - not really a problem now you have the backplane supply.

What's your cooling setup? Blades don't like heat, and the stock heatsinks aren't always very effective. Might try smoothing out the mating surface on a heatsink, possibly adding some thermal paste to increase heat transfer, and see if that gets the speed up. Fans are also your friend.

[aznatama]

I'm using a floor fan to cool the blades, which are placed on a wire shelf.  The center 5-6 blades are cool to the touch, and the ones on the edges are just somewhat warm.  the ASIC chips themselves are ~75C max (the chip farthest from the fans on the outside blades), but most chips are under 70C.

Seems like after I plugged my computer directly into the same switch as the blades, their speeds have gotten better, average is higher according to bfgminer..

[bitminer9000]

I have several of these units and I would say power/cooling/internet quality all take a part in the ghash output.
You want to make sure your using quality cables with a thick enough gauge. Using molex cables that have small wires will make the unit run poorly, not at all, or start a electrical fire/burn. I even received some of these units with very crappy molex adapter. So dont just use anything!

I'v had the best luck wiring the blades directly to the cpu 8 pin plug and/or the PCI-E 8 pin plug. Cut the plug and use the shortest cable length possible.
Use a cpu 8 pin extender cable pci-e 8 pin extender cable if you dont want to chop up your psu cpu 8 pin plug.

It is ok to use the regular molex plugs to power it up but I dont think that is as efficient as doing a direct run to the cpu 8pin plug wires or the pci-e 8pin plug wires.

Make sure to use a rail that has 12amps minimal IMO, This unit needs 80 watts not including overhead which is prolly 84-88watts or so.
The guide for these says 10Amps (IIRC) but that the absolute minimal IMO. Running off a rail that can do 12 amps or more will keep the psu nice and happy.

Make sure to keep them cool, my units love cold air, They all run at 10900 or higher, here is output of some of my workers

worker4
Total MHS: 10915                                     
Received: 0001297816                               
Accepted: 0001288884                               
Per Minute: 149.15                                   
Efficiency: 099.31%                                   
         
 
worker5
Total MHS: 10921                                     
Received: 0001298768                               
Accepted: 0001290503                               
Per Minute: 149.23                                   
Efficiency: 099.36%                                   
 
 
worker6
Total MHS: 11050                                     
Received: 0000060784                               
Accepted: 0000061176                               
Per Minute: 151.09                                   
Efficiency: 100.64%

[bitminer9000]

Ooops,  I see that you now have a backplane,

I am not really sure what issue you are having then as I run all my units in their own 1u chassis. Maybe your units are just getting to hot being so close.

[vm1990]

its the wifi. wifi isnt very stable at the best of times, its great for internet browsing but it misses packets and takes longer to go through the air than a wire so all these small details add up to you loosing money

wired is the best
homeplugs are 2nd best
wifi is bad for mining full stop

[aznatama]

I have it hard wired to my computer w/ CAT6 cabling, which is plugged into the same switch as Blades 1~10 (Blades 11~20 are daisy chained to the first hub).  The hubs are then connected to the Router.

My ping/jitter for my cable internet connections all test out really well, and the voltage output is measuring 12.7v, suggesting that the current is sufficient.

The lowest efficientcy is at ~ 92%, highest at ~98%  Reported gh/s on bfgminer is right under the 10.7/blade mark, but poos speed is lower.  Also has a seemingly high rejected/stale share rate...

Any way to lower these stale shares and get the blades to mine faster (up to spec at least)?

Heat isn't a problem, 18/20 of the heatsinks are cool to the touch, while 2/20 are just warm.  Nothing like my USB eruptors which were pretty hot (yet still hashed at over 333)

Any other thoughts? Should I get a 24p managed switch instead of using 2 16p switches?  I'm using the netgear 116? switch right now.

[vm1990]

the only other thing i can suggest is look on your pool to see if it has a backup port. and try that (most pools have 2 or 3 ports but 99%of people use the same pool port)

the switches dont make a difference aslong as there good quality switches.

if you can some pools alow you to change the difficulty lowering it or highering it might make the difference for blades most people stick the difficulty at around 12

[aznatama]

12 difficulty for 200+gh rate??  I read that it should be about 128 for this hashrate...

I'll try setting my diff to 64 and see what happens

[sidehack]

Vardiff about 1 per GH is typical I believe, so two backplanes would be 200-250. Some pools automatically adjust this from the server.

Also, a heatsink that's cool to the touch doesn't always mean much on these, as thermal contact between the chips and heatsink can be pretty crappy on stock units - I'd especially be wary of the one reading out 92%. With good cooling (might mean slight alterations to stock heatsink and/or chipsinks) and a good network you should easily see 99+% across the board.

[aznatama]

The ASIC chip temps for the blades I can read (farthest out on each backpane) are all reading under 70C. 

I'll check the low efficiency one, and try ordering some chipsinks on amazon

Thanks for the suggestion.