Pages:
Author

Topic: GRIDSEED G-BLADE Overclocking 7Mh/s, improvements and repair - page 24. (Read 74047 times)

hero member
Activity: 840
Merit: 1000
Did anyone check if these companies which sell those scrypt machines use the same chips or not?
I think they do. And if they can buy it, we can do it also.
 How hard is to made miner like that when you get chips, using directly raspberry pi on it.

you still need the firmware to communicate with the chips.
The electronic part can be done, but might not worth it if you only plan to do a few.

http://www.innosilicon.com/A2.htm

200$ each chip if you order 50 pieces
roughly 10000$ = the actual price for the complete miner with 48 chips

these chips are too expencive and I am soure you can get cheapper one.
And if more people takes it we get discount also.

We need to search cheapper chips Smiley

Gridseed GC3355 chips would cost nothing to buy now. Could worth the try, plus the firmware and the design to be improved already exists.
It all depends if they sell the chips and if they sell for really cheap.
They aren't that power hungry to move 28nm now. simply marketing.
sr. member
Activity: 412
Merit: 250
With what did you glue it on the backplate ?

I assume that i have to be very carefull, that the Glue is not electronically conductive.
Can someone post me an example of a glue used ?

Regards


I will glue it with
http://www.ebay.com/itm/Thermal-Conductive-Adhesive-Heatsink-Viscous-Plaster-Compound-Glue-For-GPU-IC-PC-/231001799393?pt=AU_Components&hash=item35c8c7bae1


There is no word on electronic non conductivity ?


I buyed this two
Details about  2Pcs Thermal Conductive Heatsink Plaster Viscous Compound Glue for PC GPU NI5L
which is the same, two because tube are really small.

Features:
Prevent overheating & extend the life of your computer parts, LED flashlights and many other devices
Does not dry in the tube
Lasts for many small projects
Very strong after it dries
Great heat transfer / thermal
Low electrical conductivity (highly insulating)
Can be used for thermal coupling of electrical device to heatsinks, especially for CPU, Northbridge, Southbridge and GPU cooler use

Specification:
Melting capacity: 0 (200 degree celsius/ 24Hours)
Evaporation: 0.001% (200 degree celsius/ 24Hours)
Thermal conductivity: > 0.671W/m-K Thermal
Impedance: <0.246
Clotting time: 3min (25 degree celsius)
Strength of connected buildings: 25Kg
Insulation coefficient: > 5.1
Dissipation coefficient: <0.005
Temperature resistance: 200 degree celsius
Size: Approx. 6 x 1.7 x 1.2cm (L*W*H)
Net Weight: About 5g
hero member
Activity: 826
Merit: 1000
°^°
Did anyone check if these companies which sell those scrypt machines use the same chips or not?
I think they do. And if they can buy it, we can do it also.
 How hard is to made miner like that when you get chips, using directly raspberry pi on it.

you still need the firmware to communicate with the chips.
The electronic part can be done, but might not worth it if you only plan to do a few.

http://www.innosilicon.com/A2.htm

200$ each chip if you order 50 pieces
roughly 10000$ = the actual price for the complete miner with 48 chips

these chips are too expencive and I am soure you can get cheapper one.
And if more people takes it we get discount also.

We need to search cheapper chips Smiley
they price the chips so high for purpose
you wont buy their complete miners if you could build yourself one cheaper
sr. member
Activity: 339
Merit: 250
Vice versa is not a meal.
With what did you glue it on the backplate ?

I assume that i have to be very carefull, that the Glue is not electronically conductive.
Can someone post me an example of a glue used ?

Regards


I will glue it with
http://www.ebay.com/itm/Thermal-Conductive-Adhesive-Heatsink-Viscous-Plaster-Compound-Glue-For-GPU-IC-PC-/231001799393?pt=AU_Components&hash=item35c8c7bae1


There is no word on electronic non conductivity ?

sr. member
Activity: 412
Merit: 250
Did anyone check if these companies which sell those scrypt machines use the same chips or not?
I think they do. And if they can buy it, we can do it also.
 How hard is to made miner like that when you get chips, using directly raspberry pi on it.

you still need the firmware to communicate with the chips.
The electronic part can be done, but might not worth it if you only plan to do a few.

http://www.innosilicon.com/A2.htm

200$ each chip if you order 50 pieces
roughly 10000$ = the actual price for the complete miner with 48 chips

these chips are too expencive and I am soure you can get cheapper one.
And if more people takes it we get discount also.

We need to search cheapper chips Smiley
sr. member
Activity: 412
Merit: 250
With what did you glue it on the backplate ?

I assume that i have to be very carefull, that the Glue is not electronically conductive.
Can someone post me an example of a glue used ?

Regards


I will glue it with
http://www.ebay.com/itm/Thermal-Conductive-Adhesive-Heatsink-Viscous-Plaster-Compound-Glue-For-GPU-IC-PC-/231001799393?pt=AU_Components&hash=item35c8c7bae1
sr. member
Activity: 339
Merit: 250
Vice versa is not a meal.
With what did you glue it on the backplate ?

I assume that i have to be very carefull, that the Glue is not electronically conductive.
Can someone post me an example of a glue used ?

Regards
hero member
Activity: 840
Merit: 1000
Did anyone check if these companies which sell those scrypt machines use the same chips or not?
I think they do. And if they can buy it, we can do it also.
 How hard is to made miner like that when you get chips, using directly raspberry pi on it.

you still need the firmware to communicate with the chips.
The electronic part can be done, but might not worth it if you only plan to do a few.

http://www.innosilicon.com/A2.htm

200$ each chip if you order 50 pieces
roughly 10000$ = the actual price for the complete miner with 48 chips
sr. member
Activity: 412
Merit: 250
Did anyone check if these companies which sell those scrypt machines use the same chips or not?
I think they do. And if they can buy it, we can do it also.
 How hard is to made miner like that when you get chips, using directly raspberry pi on it.
hero member
Activity: 840
Merit: 1000
From the two blades I am getting about 13.8 mh/s with an average of about 1.6 h/w errors per hour over the pair of blades, I will be increasing the clock once I've a 24 hour run under my belt, I suspect a combined total 14.5 mh/s maybe possible across the two units.
I am soure you can get more if you change thosw two Q7 and Q8

my replacement ones are on their way.
I hope to tell you it works in a few days
sr. member
Activity: 412
Merit: 250
From the two blades I am getting about 13.8 mh/s with an average of about 1.6 h/w errors per hour over the pair of blades, I will be increasing the clock once I've a 24 hour run under my belt, I suspect a combined total 14.5 mh/s maybe possible across the two units.
I am soure you can get more if you change thosw two Q7 and Q8
member
Activity: 145
Merit: 10
Ferrite bead removal, once they are removed you need to bridge them. Make sure the joint is 2-3mm thick between the 2 pads.

39kohm : running 963mhz @ 6.6 Mh/s
member
Activity: 70
Merit: 10
From the two blades I am getting about 13.8 mh/s with an average of about 1.6 h/w errors per hour over the pair of blades, I will be increasing the clock once I've a 24 hour run under my belt, I suspect a combined total 14.5 mh/s maybe possible across the two units.
sr. member
Activity: 412
Merit: 250
So these units are now on sale everywhere for about $1,100.. I'm wondering why.. something has got to be driving that cut..

I just heard the Blades are going for around $960 each and the pods are going for $65...
The new A2 is being sold for $12,000...
Man, what a racket!
It's like the olden days when I was Mr. cutting edge! Expensive game and right now, the way things are changing so fast, hardly profitable! Sheeeeiiiiittttt!
Still to expencive for what you get out and question is, is it possible to overclock it.
We have already expected too much from blades

natbyte you did exactly what I plan to do, just waiting for that heatsinks with fans.
I will also add these, for security on 3 places. Because if some fan dies, blade will be fried in seconds, this will turn it off.
http://www.ebay.com/itm/10PCS-Bimetal-90-Celsius-NC-Temperature-Control-Switch-Senser-Thermostat-KSD9700-/181226456499?pt=LH_DefaultDomain_0&hash=item2a31f045b3

What hashrate you have?
ZiG
sr. member
Activity: 406
Merit: 250
I have modded two complete blade units so far, I found rotating the whole blades orientation gives less hardware errors over the long term, allowing for better heat dissipation, I just fitted sticky pads to the metal heat sink, so you have the foam sheets out to the side rather than on the top or bottom.

I fitted the following a 35v 220uf capacitor, 43k ohm resistor 0.1 +/-, 2 Northbridge heat sinks and fans attached on the rear of each power board, passive heatsinks on the FETs and choke.

Running very stable at 1013mhz, with virtually no hardware errors at all.




What is your hash rate after the mods...?
member
Activity: 70
Merit: 10
I have modded two complete blade units so far, I found rotating the whole blades orientation gives less hardware errors over the long term, allowing for better heat dissipation, I just fitted sticky pads to the metal heat sink, so you have the foam sheets out to the side rather than on the top or bottom.

I fitted the following a 35v 220uf capacitor, 43k ohm resistor 0.1 +/-, 2 Northbridge heat sinks and fans attached on the rear of each power board, passive heatsinks on the FETs and choke.

Running very stable at 1013mhz, with virtually no hardware errors at all.



hero member
Activity: 840
Merit: 1000
Im still waiting for someone to confirm if HW count is diff1 or vardiff
From my experimentation it seems to be vardiff, because when my vardiff is lower at a pool, I get a lot more HW ... when my vardiff is higher, the HW count decreases proportionally.

If thats indeed the case, then its HW count * vardiff / (diffA+diffR+diffS) * 100 = HW error rate
If ur miner supports reporting of diff1work then its HW count * vardiff / diff1work * 100 = HW error rate

in cgminer, should I do HW count *100 / (submitted shares + rejected shares + HW count) using the summary?

I can see the exact same behaviour as you regarding vardif on my blade
member
Activity: 85
Merit: 10
GenTarkin: Vardiff on mine.
legendary
Activity: 2450
Merit: 1002
Im still waiting for someone to confirm if HW count is diff1 or vardiff
From my experimentation it seems to be vardiff, because when my vardiff is lower at a pool, I get a lot more HW ... when my vardiff is higher, the HW count decreases proportionally.

If thats indeed the case, then its HW count * vardiff / (diffA+diffR+diffS) * 100 = HW error rate
If ur miner supports reporting of diff1work then its HW count * vardiff / diff1work * 100 = HW error rate
hero member
Activity: 826
Merit: 1000
°^°
When i got the blade it would do a 800mhz and no more.. with an average hash rate of 5.2.. and a 5% hardware error rate.

1.3 mh/s for 30mins of soldering.. which is a 25% increase in hash rate...

which is worth a minimum of 0.2 btc extra per a month... you decide if it was worth it.
sounds like your blade is faulty
peeps reporting to run them up to 6mh without mods...
Pages:
Jump to: