GridSeed 5-chip USB miner voltage mod

[wolfey2014]

- Enable VID1: Short R336 (original VMOD)

- Increase DVDD to 1.4V: Cut R46 (0R jumper) and replace with 2.94k

- Increase VDD_PLL to 1.2V: Add 27.4k on top of R212 (27.4k || 10k = 7326R)

For reference, the default VDD_PLL is 0.8 x (R211 + R212) / R212, or 1.092V.

Not getting HW errors at 1000 MHz, but I haven't tested long term because I only got my units yesterday. I'm sure 1100+ could be achieved by increasing DVDD using the above method.

It's currently stable (no HW errors, short term) at 1250 MHz (529 kH/s) at 1.821V, with R46 being 14.7k.

Here are my R46 values and measured DVDD:

Resistor Value      Resulting DVDD Voltage
      4.32k                       1.450V
      5.90k                       1.509V
      7.79k                       1.581V
      12.1k                       1.734V
      14.7k                       1.821V

I think I'll stop there before I damage anything

Good. Thank you!
I edited it a bit for my own clarity. Hope you don't mind. 
What is your pool side hash rate - current hash rate and - average hash rate < most important....
Wolfey2014

[1714827325]

1714827325

[1714827325]

1714827325

[1714827325]

1714827325

[1714827325]

1714827325

[Reggie0]

So, you're saying to put a 120k resistor across C32 in parallel with it, right?

If one chooses the single resistor method of 54.54k instead, it is to replace R52 instead, right?

No longer any need for the 4 previous mods i.e. 2 jumpers, 39k and pencil mod, correct?

The ultimate option / mod is:
Just a single resistor (54.54k) solution and that's it?

If so,,,,
How sweet it is!  

Wolfey2014

Exactly.

28W power consumption

28W?? how?

There are an equalation for cmos logic IC power consumption:
P=c*f*V^2+Pleakage, where c is a constant.

Gridseed uses 45nm technology, so Pleakage is high, but you see P~V^2.
If you divide it with the nominal values, you will get P/Pn=1200/850*2^2/1.2^2 -> P=4*Pn plus the leakage, but it is not easy to calculate.

BTC: 1gqdzx8iSwUGt3vaoEaPCjvrWo7zKn7PK

[Reggie0]

Here is my resistor vs voltage table. + mean serial connected resistors, x mean parallel connected resistors

32k+1.8k   1.37V
32k+3k      1.42V
32k+4.7k   1.48V
32k+7.5k   1.58V
32k+10k    1.67V
32k+13k    1.78V
50k           1.81V
47k+5.1k   1.87V
100kx120k 1.97V

So, you're saying to put a 120k resistor across C32 in parallel with it, right?

If one chooses the single resistor method of 54.54k instead, it is to replace R52 instead, right?

No longer any need for the 4 previous mods i.e. 2 jumpers, 39k and pencil mod, correct?

The ultimate option / mod is:
Just a single resistor (54.54k) solution and that's it?

If so,,,,
How sweet it is!  

Wolfey2014

Exactly.

28W power consumption

28W?? how?

There are an equalation for cmos logic IC power consumption:
P=c*f*V^2+Pleakage, where c is a constant.

Gridseed uses 45nm technology, so Pleakage is high, but you see P~V^2.
If you divide it with the nominal values, you will get P/Pn=1200/850*2^2/1.2^2 -> P=4*Pn plus the leakage, but it is not easy to calculate.

There is obviously a point where overclocking no longer yields positive results at the pool, not to mention making magic black smoke . Seeing positive results locally 'client side' means nothing if it isn't reflected pool side!

Yes, there are a point after the real sweet spot which have the highest local hashrate and almost no HW error. It looks like it is a sweet spot, but it is the opposite(lets call it "black sweet spot"). As you know(or no:D), there are two kind of HW error.
 - The first is when the calculation error gives a false positive result and the share submitted to cgminer. Then cgminer does an error checking, and recognises it is a false result, and increases the HW error counter.
 - The second is when the calculation error gives a false negative result and the gs doesn't submit a share to cgminer. You will never get an error report in this situation because it is theoretically undetectable, only the WU decreases.
I experienced that, there are almost always a black sweet spot after a real sweet spot.

You keep saying that. However so far the reports from the modified miners don't back that up.

I can overclock my unmodified miner to get 500 kh/s, but that doesn't mean I'm getting 500 kh/s worth of completed work. Due to rejects/HW errors I'd be getting a lot less than that. You posted a screenshot showing 500kh/s but you had 429 HW errors over the course of about two days. For the same period of time, I have 9 HW errors. Those HW errors hurt and lower your effective hashrate.

It doesn't matter what hashrate your miner is reporting. What matters is how much work the pool sees you are doing. If, as we've seen, your miner is reporting 450 kh/s but the pool is reporting a 24 hour average of 350 kh/s then something is off.

We need OBJECTIVE tests to verify how effective the mods are. Testing shouldn't be run against pools or solomining anyway. There's too much variance. A real test would be run against test blocks. Remove the pools and coins from it entirely. Send the same workunit over and over again. Take the stats from that and calculate the effective hashrate. I think cgminer has a test mode that does something like this.

Yes i had 429 error and 43531 shares(including the errors). It means there are 429/43531*100%=0.9855% error rate, so the real hashrate is 510*(1-0.009855)=504.97 kH/s. I've tested it on some pools(ghash.io,clevermining,middlecoin,wemineltc) and it is fine.

BTC: 1gqdzx8iSwUGt3vaoEaPCjvrWo7zKn7PK

[poopypants]

- Enable VID1: Short R336 (original VMOD)
- Increase DVDD to 1.4V: Cut R46 (0R jumper) and replace with 2.94k
- Increase VDD_PLL to 1.2V: Add 27.4k on top of R212 (27.4k || 10k = 7326R)

For reference, the default VDD_PLL is 0.8 x (R211 + R212) / R212, or 1.092V.

Not getting HW errors at 1000 MHz, but I haven't tested long term because I only got my units yesterday. I'm sure 1100+ could be achieved by increasing DVDD using the above method.

It's currently stable (no HW errors, short term) at 1250 MHz (529 kH/s) at 1821 mV, with R46 being 14.7k.

Here are my R46 values and measured DVDD:

Resistance (R)	Voltage (mV)
4.32k	1450
5.90k	1509
7.79k	1581
12.1k	1734
14.7k	1821

I think I'll stop there before I damage anything

Edit: here is a microscope shot of the mods
Blue: R336
Green: R46
Purple: R212

If I wanted to run mine at 1150 stable, which resistor values should I use on R46 and R212?  I appreciate your picture and the risk you put in finding this mod.

[nemercry]

- Enable VID1: Short R336 (original VMOD)
- Increase DVDD to 1.4V: Cut R46 (0R jumper) and replace with 2.94k
- Increase VDD_PLL to 1.2V: Add 27.4k on top of R212 (27.4k || 10k = 7326R)

For reference, the default VDD_PLL is 0.8 x (R211 + R212) / R212, or 1.092V.

Not getting HW errors at 1000 MHz, but I haven't tested long term because I only got my units yesterday. I'm sure 1100+ could be achieved by increasing DVDD using the above method.

It's currently stable (no HW errors, short term) at 1250 MHz (529 kH/s) at 1821 mV, with R46 being 14.7k.

Here are my R46 values and measured DVDD:

Resistance (R)	Voltage (mV)
4.32k	1450
5.90k	1509
7.79k	1581
12.1k	1734
14.7k	1821

I think I'll stop there before I damage anything

Edit: here is a microscope shot of the mods
Blue: R336
Green: R46
Purple: R212

@Mixdio:
Could you explain me which benefit i'll get when i give it a higher PLL voltage ?
Doesnt make sense for me to higher the voltage there ?
Would be nice if you can help me on that here.

[twittmann]

Some tuning news on Gridseeds (Scrypt Mining)

1.) Tuning of Gridseed Hardware (Soldering required) upto 436 KH/s per Miner
2.) Fork from CGMiner for Gridseed Hardware (Frequency per Miner, Easy Overclocking for more stable Miners)

http://gridseed-blog.com (contains link to Originalpost, BLOG about Gridseed Mining is currently created)

[Reggie0]

So that's what's been mystifying us all - lost 'Completed Work' which is caused by a False Negative, of which the GS isn't programmed to detect and flag or display in the print out?
It just goes un-noticed in real time locally - then one sees the result pool side which ends up as Stale Shares or just No Shares of any kind reported at all.
Lost Completed Work = Lost Money = Throwing Coin Away!....Pssssshhhhhhhhhhhhhhhhh!
Is this correct?

Still, I see my Reward$ caufers filling up nearly twice as fast as before I made the mods.
Something still isn't being caught or explained here. Maybe I'm just hallucinating!
Better explanation. But the complete 'why' still eludes, I fear.

I wonder if an oscilloscope connected to the right spots on the card will give a better if not the indication of what's going on and even tell one when they've hit the real 'sweet spot' and not go any further. That would be some serious fine tuning. I happen to have a CRO. I guess I'm going to have to start probing and see what I can find out.

Or, is it all calculable (math) and the end result of the equations ARE the REAL answer and THAT"S THAT no matter what hardware changes are made or probing is done?

False Positive's SUCK!  

But....

The fact is, the original 4 mods have caused hardly any increase in power consumption while having reduced if not negated and stabilized an otherwise very unstable 'red noncey' stock miner at 850MHz or higher clock rate. So at this point, I still believe the original 4 mods are beneficial. I still want to circumvent the pencil mod without losing its positive results and I may have achieved this already. Still testing.

The thing is to get the clock and resulting counting / calculations in the chips to be stable and positive at higher clock rates so as to produce more valid shares i.e. $Moolah$! at the pool. I still believe this is achievable and we're on the right track.

All in the name of progress....right?  

Wolfey2014

Maybe, but I thin pencil mod is unnecessary, because PLL is stable.
There are no reasonable way to detect false negative results, because the only way if you check the job in full range on a trusted hardware or software. In this case, you do the job twice and it is an inadmissible overhead.

Awesome! I'll gonna try to beat that !

My opinion is it's not a good idea, because
- every 100mV gives more less extra frequency and efficiency will be very bad.
- i tried to add more 280mV and it caused temperature runaway in idle state.

Solder bridges + resistor + pencil ?

Tell me your secret!

Nope, you need to change resistors only. Replace R52 with 100k, and place 120k to C32. In parallel it is 54.54kOhm, and VDD will be 1.97V.
If you would like to use only one resistor there is a 54.9kOhm in E96 series, but VDD will be more than 1.97V maybe 2V.

If it is working pls. for you, pls support my risky experiments:)
BTC: 1gqdzx8iSwUGt3vaoEaPCjvrWo7zKn7PK

I am giving out 0.6BTC to the first to get his Gridseed miner stable at 1100 MHz (<10 HW error in 24h) and post the steps to mod the miner. So far I have managed to get it stable at 1013 MHz, but I feel we can push it further.

Disclaimer: as always, I am not responsible for damage to your miner, do this at your own risk!

So i won 0.6BTC  

Sorry to disappoint, but that's not yet stable by my standards, which is less than 10 HW error in 24h. But obviously the highest stable frequency wins, so it will be you if you manage to get it stable...
Now that, folks, is called pushing it to the limit, great work mate.  

Is it enought?



BTC: 1gqdzx8iSwUGt3vaoEaPCjvrWo7zKn7PK

[ZiG]

Some tuning news on Gridseeds (Scrypt Mining)

1.) Tuning of Gridseed Hardware (Soldering required) upto 436 KH/s per Miner
2.) Fork from CGMiner for Gridseed Hardware (Frequency per Miner, Easy Overclocking for more stable Miners)

http://gridseed-blog.com (contains link to Originalpost, BLOG about Gridseed Mining is currently created)

Nothing new...imho...and spam/pay-per-click redirect...I'll avoid...for now...

[mixdio]

If I wanted to run mine at 1150 stable, which resistor values should I use on R46 and R212?  I appreciate your picture and the risk you put in finding this mod.

I think I was able to get 1100-1150 at 1.581V (7.79k) - I make absolutely no guarantees.

@Mixdio:
Could you explain me which benefit i'll get when i give it a higher PLL voltage ?
Doesnt make sense for me to higher the voltage there ?
Would be nice if you can help me on that here.

Same concept as overclocking the core - increase voltage to increase stability. I tried decreasing VPLL to 0.91 V which resulted in hardware errors at 950 MHz.

To follow up with a 12 hour test:
1.821 V at 1250 MHz (~548 kH/s) - 0 HW errors
1.734 V at 1200 MHz (~526 kH/s) - 0 HW errors

The remaining 8 devices ran at 950 MHz with only the VID1 mod. There were 3 with HW errors.

The test was run with bfgminer compiled from nwoolls' feature/gridseed-support-new branch. girnyau's cgminer fork was segfaulting with clock speeds over 1200 MHz. I needed to be able to set individual clock rates because my devices have different mods. bfgminer allows this with the set-device parameter. bfgminer also appears to report hashrates a bit higher than cgminer.

Before drawing any conclusions from the data presented by myself and others, keep in mind that there are a lot of variables and the sample size is incredibly small.

[TrollboxChamp]

looks like you can use liquid solder and get an extra 40kh per unit. Nothing like an extra 100kh but still something to consider

There is a thread about it in this sub-forum

So kudos to whomever guessed it may be a viable option. Alas not the most effective it is still an option.

I personally am not comfortable micro soldering. Micro painting i can handle tho

[gbyg]

i believe R336, R46, R212 are all 0402? also it is very hard to find 14.7k and 27.4k where i live. closest i could find is 15k and 27K. will the performance will very much?

[mixdio]

i believe R336, R46, R212 are all 0402? also it is very hard to find 14.7k and 27.4k where i live. closest i could find is 15k and 27K. will the performance will very much?

The resistors are all 0402 (1005 metric) 1% tolerance.

I'm sure anything close will work, but again, there are a lot of variables, YMMV, etc.

[jamieb81]

is this it?

http://www.ebay.fr/itm/VISHAY-DRALORIC-CRCW04022K70FKED-RESISTOR-0402-2K7-1-Price-For-50-/171138323876?pt=UK_BOI_Electrical_Components_Supplies_ET&hash=item27d8a395a4

[wolfey2014]

Okay everyone, let's settle down. Take a deep breath. Start over.
Got it on spamming on here. I'm a bit new to forum-ing and all.
I HATE SPAM TOO! I get what it means on here now.

Oh yeah, no more signing all the time. It takes too much time

I do have some improved 'evidence' to post sometime over the next few to several hours 'with pertinent notes' I'm studying longer to figure out what all the variables are at the pool and tune these babies up.
I'm still tuning mine based on the original and current discoveries by others on here who deserve thanks for their contributions.

I am duplicating and applying data and reporting the results as I go. I'll keep it more concise as I already like to be as precise as possible. I am not afraid to be wrong either. After all, it's part of the game of being right - as it goes in life.

I'm always interested in contributing and making a positive difference in everyone's lives.

Please feel free to PM me if you have any questions, comments, challenges or concerns.
I want to stay excited and looking forward to the progress we all make as individuals while contributing to the group so that everyone has a fair chance to improve performance and profit$

If you can't do these mods yourself, I'll be happy to do them for you at a fair price! And, I guarantee my work to be 100% stable and permanent! If by some chance, I brick your pod, I'll replace it with one of my own. PM me for details.

[volum4]

please don't post this hashra shit here! it's just another spam.. it has nothing to do with voltage mod.

[wolfey2014]

please don't post this hashra shit here! it's just another spam.. it has nothing to do with voltage mod.

It's posted as 'vindication' for being accused of spreading BS about pricing....
I guess I should have said that.
But, like I said, no more spam.

[Cablez]

I have done the 3 mods on 2 of my pods and seen some benefit locally and pool side.  I do find that the choice of pool for testing is important as these multi-coin pools really have poor stratum results.  They toss all different kinds of rejected shares upon switching coins.  I would suggest for testing mod stability to use a sole coin pool as that has given me better numbers. 

Currently running at 1050Mhz stable and I do not have the PLL down to 0.94 yet. Looks like there is more room in these units.

[XScrypt]

I have done the 3 mods on 2 of my pods and seen some benefit locally and pool side.  I do find that the choice of pool for testing is important as these multi-coin pools really have poor stratum results.  They toss all different kinds of rejected shares upon switching coins.  I would suggest for testing mod stability to use a sole coin pool as that has given me better numbers. 

Currently running at 1050Mhz stable and I do not have the PLL down to 0.94 yet. Looks like there is more room in these units.

If you're using cgminer, use --benchmark to see how well the unit is performing. This hashes the same work unit over and over again while recording stats (hashrate, errors, etc.). It's probably the most objective measure you can get.

[kosch]

Stable 510kH/s@1200MHz mod of Reggie0, _only_ by replacing R52 with 100k Resistor and C32 with 120k? On the first view this sounds good!
@Reggie0: You've said GS takes 28W, do you have measured only the GS with fan or including the power supply?
Additionally I'm a bit confused. Here you talk about 510kH/s:
https://bitcointalk.org/index.php?topic=519112.msg5954771#msg5954771
But later, when it comes to proof stability you screenshot talks about 473kH/s
https://bitcointalk.org/index.php?topic=519112.msg5971887#msg5971887
Are we talking about different mods?

@nemercy: What are your modifications to get 489kH/s@1150MHz stable? It consumes 21W (with fan and PSU) you wrote, so I think its maybe the better choice.

[nemercry]

I have done the 3 mods on 2 of my pods and seen some benefit locally and pool side.  I do find that the choice of pool for testing is important as these multi-coin pools really have poor stratum results.  They toss all different kinds of rejected shares upon switching coins.  I would suggest for testing mod stability to use a sole coin pool as that has given me better numbers.  

Currently running at 1050Mhz stable and I do not have the PLL down to 0.94 yet. Looks like there is more room in these units.

PLEASE READ SO YOU DONT BRICK YOUR DEVICE

I just recall everyone: Dont play with PLL Voltage at all if you want to overclock. I did it once and it worked, but i found out that it was just because of the low frequency. As higher i got the higher the Errors did go.
So i can advise you to: whatever you do, dont change the PLL voltage. This week i'll gonna post the mod i am doing right now, to show everything its not hard to reach or complicated.
But pls dont resolder anything at your PLLs. Also did i calculated some values for the PLL resistors: dont use them, they are wrong!

For example: Even on my 1250mhz unit it just gets from 0 HW Errors to more when increasing the PLL.