GRIDSEED G-BLADE Overclocking 7Mh/s, improvements and repair

[xcooling]

initial overclocking and modding



completed unit

[1714742780]

1714742780

[1714742780]

1714742780

[1714742780]

1714742780

[J4bberwock]

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!

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.

[J4bberwock]

initial overclocking and modding

completed unit

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.

[xcooling]

@J4bberwock , my thoughts exactly.


The heatsinks on the rear were from an old amd athlon cpu cooler, cut in half. (half per a board)

the small blue heatsinks i salvaged from old graphics cards, however you can make some if you cut up a small chipset heatsink.

[xcooling]

Ive tested up to 50kOhm, and had 3 thermal probes attached to various components.

however by killing the ferrite beads the voltage/current seems to be increased for the chips.

Right now im waiting for the chips to "burn in" and then ill increase the frequency to 1000+

[J4bberwock]

however by killing the ferrite beads the voltage/current seems to be increased for the chips.

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

[wolfey2014]

Radical! Just Radical!
Think there's any chance of getting to more than 8MHs?
Those heat sinks are humongous!
Also, removing the FB's will probably add noise to the supported circuit and may cause more HW errors than it's worth. Hope not. I know that under overclock conditions, those little things are like tiny fire crackers just waiting to be lit off! 
Nice work guys! Nice workmanship!

[xcooling]

@wofly..

the blades were designed for those crap industrial 12v switching powersupplies.

Im using a High quality atx powersupply, using the 12v pcie rails. Not going to be that amount of noise.

Hardware error wise.. 0.1% .. is acceptable in my book. Stock unmodifed i had 5%

Btw..

[xcooling]

Ill push my clocks higher, once ive finished the 24hour stress test.

Current Ambiant air temp is 35deg C (95 deg F), the miners highest temp is 55 deg C (131 deg F)

[Janezki]

xcooling: You running at 963Mhz?

Modded mine with 93kOhm resistors, heatsinks & "widebody kit" with 120mm fan.
Stock power connectors with 14AWG wires straight from PCIE of Corsair TX850.





About those ferrite beads; if removed, do the pads (where the FB's were soldered at) have to be bridged?
I was planning to direct solder the power wires, maybe next weekend.

[Xer0]

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...

[GenTarkin]

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

[Janezki]

GenTarkin: Vardiff on mine.

[J4bberwock]

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

[natbyte]

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.

[ZiG]

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...?

[Blisk]

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?

[natbyte]

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.

[xcooling]

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

[Blisk]

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