GridSeed 5-chip USB miner voltage mod

[Reggie0]

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.

28W is only GS's power consumption without fan.
The second picture is for win sandor's 0.6 BTC reward. You can see less than 10 HW errors in 24 hour. It is the same mod, only freq was reduced.

BTC: 1gqdzx8iSwUGt3vaoEaPCjvrWo7zKn7PK

[nemercry]

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.

28W is only GS's power consumption without fan.
The second picture is for win sandor's 0.6 BTC reward. You can see less than 10 HW errors in 24 hour. It is the same mod, only freq was reduced.

BTC: 1gqdzx8iSwUGt3vaoEaPCjvrWo7zKn7PK

Give me a little more time and ill give you a run of my 1250mhz and of my 1150mhz units. 24h each. I was able to setup a headless machine now and can let them run 24/7 now.
Probably i'll get a higher stable rate. Lets see

[Reggie0]

Give me a little more time and ill give you a run of my 1250mhz and of my 1150mhz units. 24h each. I was able to setup a headless machine now and can let them run 24/7 now.
Probably i'll get a higher stable rate. Lets see

OK. Be careful, output voltage is limited by organic polymer capacitor. Max. 2.5V allowed(but keep a small safety range).

[wolfey2014]

Okay, my 'non-pencil-trick' mod is:
Replace C36 10nf with a 100nf capacitor = no more pencil tricks! 'yay'
This includes the other 3 mods i.e. the two jumpers and resistor change out on page 14 (Sandor's mods) of this thread.
I have run over 36 hours with ZERO hardware errors - 1000MHz (stable) at 568KH/s peak - 378KH/s average pool side .07% stale shares out of 587,114 valid shares. Comm port FIFO's were set at 14 Receive and 16 Transmit.
It's still running strong and stable. So I believe this mod is a pass!

I"m looking forward to trying out nemercry's mods when he finally releases the particulars. His results sound fantastic!

As usual, if you do these mods, you do them at your own risk!

[nemercry]

It's currently still running but i wanted to share some results.(yet only 6h runn)
first batch (red)
second batch (dark gold)
third batch (white gold)

I modded one of the newest Gridseed units (those with smdleds on it, batch 3) and an older one,batch2.
For some reason the newer version of the gridseeds isnt stable like the older ones.
After the run is finished i will measure the voltages between some parts, to see if there are any differences...
I am not sure if this is just variance or if they did change transistors, which causes this now.

[Andareed]

Okay, my 'non-pencil-trick' mod is:
Replace C36 10nf with a 100nf capacitor = no more pencil tricks! 'yay'
This includes the other 3 mods i.e. the two jumpers and resistor change out on page 14 (Sandor's mods) of this thread.
I have run over 36 hours with ZERO hardware errors - 1000MHz (stable) at 568KH/s peak - 378KH/s average pool side .07% stale shares out of 587,114 valid shares. Comm port FIFO's were set at 14 Receive and 16 Transmit.
It's still running strong and stable. So I believe this mod is a pass!

I"m looking forward to trying out nemercry's mods when he finally releases the particulars. His results sound fantastic!

As usual, if you do these mods, you do them at your own risk!

378 kh/s is lower than you should expect at 1000 MHz. Reported hardware error counts only tell half the story. A reported hardware represents a nonce that your miner thought was valid, but really isn't. It's way more likely that your miner would be corrupting and missing valid nonces, which are unfortunately not detectable.

[wolfey2014]

Okay, my 'non-pencil-trick' mod is:
Replace C36 10nf with a 100nf capacitor = no more pencil tricks! 'yay'
This includes the other 3 mods i.e. the two jumpers and resistor change out on page 14 (Sandor's mods) of this thread.
I have run over 36 hours with ZERO hardware errors - 1000MHz (stable) at 568KH/s peak - 378KH/s average pool side .07% stale shares out of 587,114 valid shares. Comm port FIFO's were set at 14 Receive and 16 Transmit.
It's still running strong and stable. So I believe this mod is a pass!

I"m looking forward to trying out nemercry's mods when he finally releases the particulars. His results sound fantastic!

As usual, if you do these mods, you do them at your own risk!

378 kh/s is lower than you should expect at 1000 MHz. Reported hardware error counts only tell half the story. A reported hardware represents a nonce that your miner thought was valid, but really isn't. It's way more likely that your miner would be corrupting and missing valid nonces, which are unfortunately not detectable.

And....let's not forget. Pool side results are inaccurate to the degree of and methods by which they are averaged. A different pool can see the same results 'differently' - substantially differently, from my own experience as well as others on here.
1000MHz @ 378KH/s pool side vs 950MHz @ 283
That's what I'm seeing over here.

I'm just sayin

[amix]

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.

are you surpised? PLL Phase Locked Loop. basically a feedback loop to keep  frequency in check stop it from sprilaling out of control. the output is feed back into the input.

[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.

are you surpised? PLL Phase Locked Loop. basically a feedback loop to keep  frequency in check stop it from sprilaling out of control. the output is feed back into the input.

I just got a lot of requests from people asking me which PLL voltage they should set.
And as for my person, i dont mind soldering different resistors for testing purposed, but i do not want people to try something what brings them no benefit at all and probably more work to correct it again.

[Xer0]

anyone tried to build a test bench,

soldering precision spindle trimmers instead of resistors?

[nemercry]

24 hours run completed:
Cgminer:

Pool:


I am still not sure why the third batch unit (the one on the top) throws that much hw errors but i will investigate and report.

[SVK]

24 hours run completed:



I am still not sure why the third batch unit (the one on the top) throws that much hw errors but i will investigate and report.

Nice results 
Can you post a link to a Cgminer you are using ?
I have RPi so it has to work on it.

[nemercry]

@SVK:
Thank you.
It should be this repository: https://github.com/girnyau/cgminer-gc3355 and i found it in this thread: https://bitcointalk.org/index.php?topic=477709.msg5752544#msg5752544

If anyone interested I've modified Andareed's version of cgminer a bit to let you better finetune the gridseed units. There's a big variance in the resistors they used as voltage references so every unit has a different maximum frequency they can run at.

https://github.com/girnyau/cgminer-gc3355

New frequency table: 800, 813, 825, 838, 850, 863, 875, 888, 900, 913, 925, 938, 950, 963, 975, 988, 1000, 1013, 1025, 1038, 1050, 1063, 1075, 1088, 1100, 1113, 1125, 1138, 1150, 1163, 1175, 1188, 1200

You won't need the ghz ones however unless highly overvolted

You can also define a base frequency as usual, then set per device frequencies by the units' serial number if you want:

--gridseed-options freq=888 --gridseed-freq 8D74488A4949=900,8D96227B5449=900,8D7357785355=875

Factory units running since ~48 hours without any hw mods but the fans desoldered:
...

[ajareselde]

24 hours run completed:
Cgminer:

Pool:


I am still not sure why the third batch unit (the one on the top) throws that much hw errors but i will investigate and report.

with what mods is this done in the end ?








.

[ZiG]

anyone tried to build a test bench,

soldering precision spindle trimmers instead of resistors?

Yep...I am experimenting with mine Gridseed...2 trim pods...VDD voltage ...and PLL voltage...

[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.

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.

are you surpised? PLL Phase Locked Loop. basically a feedback loop to keep  frequency in check stop it from sprilaling out of control. the output is feed back into the input.

I just got a lot of requests from people asking me which PLL voltage they should set.
And as for my person, i dont mind soldering different resistors for testing purposed, but i do not want people to try something what brings them no benefit at all and probably more work to correct it again.

So you are saying that the PLL modification is not really necessary for stability?  I can try to back it to default and see if the results change.

[amix]

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.

are you surpised? PLL Phase Locked Loop. basically a feedback loop to keep  frequency in check stop it from sprilaling out of control. the output is feed back into the input.

I just got a lot of requests from people asking me which PLL voltage they should set.
And as for my person, i dont mind soldering different resistors for testing purposed, but i do not want people to try something what brings them no benefit at all and probably more work to correct it again.

So you are saying that the PLL modification is not really necessary for stability?  I can try to back it to default and see if the results change.

in theory and in practice really go hand in hand. From looking the diagrams and brushing op abit. the PLL voltage is a feedback loop to help stablise the oscillator frequency. feed the output back into itself - i believe in an inverted format - which goes without saying with produce an offset if the frequency is out. and result in no feedback when the output frequency matches its input - ie frequency stable. Oscillators can vary drastically depending on how well they are made with what not to mention the normal factors of voltage current, temperature etc. so the PLL is to keep a consent frequency, when thats its designed use anyway. But as always when overclocking other things need to be tweaked, up the frequency , to get it stable up the voltage, change timings etc. anyone whos every overclocked there pc and memory knows. So offsetting the PLL voltage SLIGHTLY could/should help to stabilizing it WHEN its already been overclocked, but nbot as a direct overclocking resource. It runs the risk of producing an amplified feedback loop and youll probably find that its starts off fine at a higher frequency then starts generating hardware errors slowly, then more so left to its own id say its only a matter of time before its generating 100% hardware errors - unless its offset elsewhere. But i could be wrong i seem to be having trouble finding the datasheet for the chips used. Personally id say a tweaking of the PLL voltage (ie pencil mod) to fine tune it. just as each gridseed chip can produce different results so can each oscillator if not produced in a controlled environment, and these are made in china, so whos know the quality of the parts that have been used. Comparing my first gridseed to the second there is vast difference in quality and finish - particular on the boring of the heatsink, not to mention the heat channels on the actually gridseed chip.
 

[wolfey2014]

The mods I made to one of my pods using a slightly modified version of Sandor's idea on page 6 are still working and stable at 1000MHz.
No pencil mod. No PLL voltage adjustment. It has been left at the factory setting of 1.091V
The change I made was to change out C36 from 10nf to 100nf. This also yields virtually no HW errors at 1000MHz.
The other 3 mods, resistor change out to 38k 5% from 36k and the two jumpers on page 14 are still in place.
I haven't heard of anyone bricking their pod 'yet' by adjusting PLL voltage or core voltage either, for that matter and they have been set all over the place + and -! Hopefully no one will, either!
Still, so far, so good.
I am very curious and excited to see what mods nemercry has come up with to achieve the results he has been talking about.

[amix]

The mods I made to one of my pods using a slightly modified version of Sandor's idea on page 6 are still working and stable at 1000MHz.
No pencil mod. No PLL voltage adjustment. It has been left at the factory setting of 1.091V
The change I made was to change out C34 from 10nf to 100nf. This also yields virtually no HW errors at 1000MHz.
The other 3 mods, resistor change out to 38k 5% from 36k and the two jumpers on page 14 are still in place.
I haven't heard of anyone bricking their pod 'yet' by adjusting PLL voltage or core voltage either, for that matter and they have been set all over the place + and -! Hopefully no one will, either!
Still, so far, so good.
I am very curious and excited to see what mods nemercry has come up with to achieve the results he has been talking about.

oh yeah i forgot to add that PLL voltage has a rated maxmium of 1.1 . So be extra warned, you run the risk of frying something ( according to gridseed datasheet). So it depends on which PLL voltage your modding as well. erm cooked grid

[nemercry]

A new run with 2 more modded.
The two from the last run are also included.


I clocked down the bad units, which where havin to much hardware errors, to see where they stabilize.
It looks like as it has nothing to do with the new units itself, as the HW Errors also occured at another unit(one of the dark gold units).

In my oppinion the benefit of tweaking the PLL Voltage, if its even is possible to achieve positive results by doing this, wont be significantly better.
Trying to regulate a system, while you do not even know in which direction, cant bring you any further. It is just like playing lotto and thats something i wont bring in here.
There is not even one guys here, who actually can proof, that his change in PLL voltage, actually brought him any advantage.
It's just ppl. saying, that they assume, that it must be done, if you have done the resistor change. Thats not science, thats not reliable it is an assumption.
All PLL tweakings i have done so far, brought me bad resulst. So for now on i just call it bullshit and advice everyone not to play with PLL voltage.

Some user wrote me that he thinks, that the mod is not stable, because it has to much HW Errors.
I gave him credit for the fact that he recognized correctly, that the Hashrate(e.g. mhz) which the device was running on wasnt "stable".
And said yes: There are units where you need to go down with the frequency. till its "stable" again.
But that has nothing to do with the mod itself.
It's like the 950mhz mod. There are units which are running fine with 950mhz and there are units which are only able to run at 900-913mhz.
But still are there a lot of ppl. doing the mod because it still brings you more hashrate than 850mhz or 888mhz.
Also if the unit is a bad one and is only stable at 900mhz (with the mod) it must be that it throws also hws at the stock of 850mhz. All what the mod does is increase the range in where your units operate in.
A unit which runs bad even at stock clocks will also run bad on a higher clock compared to good units and thats nothing you can change.
But lets say you clock a bad unit from 850mhz to 1100mhz and the HWs stay the same, it will be still an increase of 30%.
Independant if its a good or a bad unit.