Fury/Blizzard tuning and mods

[nst6563]

I'm starting this thread mostly because there isn't one for it already with the aim to get more from our Fury's/Blizzards (all based on the Zeus chips).

So here's some links to get started:
A WIKI!  http://wiki.zeusminer.info/index.php/Main_Page  : Thanks LongAndShort!
Link to Silab's for Windows USB driver:  http://www.silabs.com/products/mcu/Pages/USBtoUARTBridgeVCPDrivers.aspx

There are two hardware versions of the USB devices.

The first versions used USB DRIVER CP201X http://www.silabs.com/products/mcu/pages/usbtouartbridgevcpdrivers.aspx
Batch two uses FT232R USB UART http://www.ftdichip.com/Drivers/CDM/CDM%20v2.10.00%20WHQL%20Certified.exe

*Above was taken from http://wiki.zeusminer.info/index.php/Software

Documentation as well as user experimentation has shown that Zeus hardware prefers higher difficulties as compared to Gridseed.  Using difficulty values of 1024 or 2048 are not unheard of for use on a single Fury/Blizzard and offer good results.  
Link to thread that has a growing list of pools which support the manual setting of difficulty:  https://bitcointalk.org/index.php?topic=645532.0 (Thanks jekv2!)

Zeusminer page with links to cgminer binaries, sources, windows drivers, and rpi images:
http://zeusminer.com/user-manual-ver-1-0/

Builds of cgminer with Zeus support:
The zeus version of cgminer base on cgminer 3.11 https://github.com/zeusminer/cgminer_zeus (Thanks happydaze for the link)

GAWMiners support page for cgminer source and binaries for the Zeus based hardware:
https://gawminers.zendesk.com/hc/en-us/articles/202184744-cgminer-source-and-config

Updated build(s) of bfgminer with Zeus support:
  Link to file:  https://github.com/Darkwinde/bfgminer
  Link to thread: https://litecointalk.org/index.php?topic=20477.0 (Thanks worldlybedouin for updated link)

  Link to file (Windows binary):  http://sgminer.builders/
  Link to Sources:  https://www.dropbox.com/sh/fygluxm446cjctn/AAAFUAJDeIEYiWnuEefimKjWa
  Link to thread: https://hashtrader.com/t/bfgminer-fork-supports-for-zeus-miners/354/4
  Link to litecointalk thread:  https://litecointalk.org/index.php?topic=16301.5760
 
  Link to Nwoolls' bfgminer: https://github.com/nwoolls/bfgminer/tree/feature/zeusminer-support-cleaner
  Link to Luke-jr's bfgminer: https://github.com/luke-jr/bfgminer

To start mining, run:

Code:

Code:

bfgminer --scrypt -S zus:all -o host:port -u username -p password

If you have more than 6 chips, use:

Code:

Code:

bfgminer --scrypt -S zus:all -o host:port -u username -p password --set zus:chips=6

To set the clock, use:

Code:

Code:

bfgminer --scrypt -S zus:all -o host:port -u username -p password --set zus:clock=340

Finally, there are a few ZeusMiners with buggy firmware that don't properly respond to detection. For those you may need to use:

Code:

Code:

bfgminer --scrypt -S zus:all -o host:port -u username -p password --set zus:ignore_golden_nonce=1

No CPUMiner support yet.  

Raspberry Pi images:
Hashra Lunar Lander :  
* Support Page:  http://hashra.com/support/
* Direct link to image: https://mega.co.nz/#!D1QFmBpL!scLfSE3782eFOv4At6KAcZv-x0rr_Vgxe-1pD8T8P0o
*    Default Login Username: hashra
*    Default Login Password: hashra
*    Root access to firmware:
*    User: pi
*    Password: controla

Starminer:
* Litecointalk thread:  https://litecointalk.org/index.php?topic=19129.0
* Direct link to image download page: http://sourceforge.net/projects/starminer/files/
* There are two user accounts for StarMiner, root and pi. Both have the password of live.  (Listed on the Sourceforge Wiki "Getting Started" for Starminer)

Minepeon
Uses bfgminer (which is making rapid progress with the zeus based miners)
* Link to "main page": http://minepeon.com/index.php/Main_Page
* Username:    minepeon
* Password:    peon

MinerEU (ScriptA)
* Download + instructions: https://github.com/MinerEU/scripta_a2/tree/zeus

Zenminer:
* Here's the webpage:  https://zenminer.com/
Info below provided by thebotnet - Thanks!

If anyone would like to use the zen os on their pies then here is a quick guide.  

1. Download the zen os image from this guide put together by zenminer. * * *https://docs.google.com/document/d/1UilAJ3wMOgoHFqz9e1zhOvz79KQDascYaC6Aq3sbmqg

2. Write it to your raspberry pi's sd card using win32 disk imager (or your preferred method)

3. Boot the pi with Ethernet and the miners plugged in (or plug them in at a later time, up to you)

4. Go to https://zenminer.com/LogIn/ and click on "Don't have an Account?" then sign up and sign in

5. Click the add miner button and it should show up, it may take a while for it first to show up but it will (also try refreshing)

Remember this guide is for people that want to use their own raspberry pi hardware and dont have an activation code that came with the zenminers.

Minera (as of version 3.0.0)
* Official site
* https://github.com/michelem09/minera

Basic cgminer command (as if it's really that much different than what we're all used to anyway):
cgminer.exe -o stratum+tcp://us.clevermining.com:3333 -u 1FC7cRPFXUXpgGoduMy1EZcA86zbQAaPKW -p x --nocheck-golden --chips-count 6 --ltc-clk 328 -S //./COM4
Example json config file ( base config file is from starminer with minor additions).

Code:

{
"pools" : [
	{
		"url" : "stratum+tcp://stratum.nicehash.com:3333",
		"user" : "16LfWJ6pvica3PrfgR4HVSVrr2wMWBy2RL",
		"pass" : "d=32"
	},
	{
		"url" : "stratum+tcp://ny.clevermining.com:3333",
		"user" : "1FC7cRPFXUXpgGoduMy1EZcA86zbQAaPKW",
		"pass" : "x"
	}	
],
"api-listen" : true,
"api-port" : "4028",
"expiry" : "120",
"nocheck-golden" : true,
"chips-count" : "6",
"ltc-clk" : "328",
"log" : "5",
"no-pool-disable" : true,
"queue" : "9999",
"hotplug" : "5",
"scan-serial" : [
"/dev/ttyUSB0",
"/dev/ttyUSB1",
"/dev/ttyUSB2",
"/dev/ttyUSB3",
"/dev/ttyUSB4",
"/dev/ttyUSB5",
"/dev/ttyUSB6",
"/dev/ttyUSB7",
"/dev/ttyUSB8",
"/dev/ttyUSB9"
],
"scan-time" : "30",
"scrypt" : true,
"verbose" : true,
"shares" : "0",
"kernel-path" : "/usr/local/bin",
"api-allow" : "W:127.0.0.1",
"api-description" : "cgminer v3.1.1-zm"
}

I've seen it mentioned over on litecointalk forums that if you increase the clock speed to 336 and use a higher difficulty the error rate will go down.  I can understand the difficulty having an effect, but not sure about the clockspeed.  I'm still experimenting with that.
--------------------------------------------------------

As for mods....they're now in the thread and some technical details in the 2nd post
***DO NOT ATTEMPT IF YOU CANNOT ACCEPT RESPONSIBILITY SHOULD SOMETHING GO WRONG***
I noticed that while the Fury is hashing away, the heatsink wasn't THAT warm...but the tops of the chips were too hot to touch for very long at all.  So, I took it all apart.  
6 screws hold the circuitboard in place on the heatsink.  Another 4 hold the fan down.  10 screws and you're circuitboard has the freedom it needs.
First thing I noticed was that there was thermal paste on the back...but not much....at all....in fact...there wasn't even any on the heatsink after removing the circuitboard.  So that's #1 on the list.  
I applied thermal paste (Arctic Silver Ceramique - USE NON ELECTRICALLY CONDUCTIVE PASTE ONLY) to the backside of the circuitboard behind each of the ASIC chips as well as the mosfets, choke, and the other coupe IC's.  After a test run I noticed that the heatsink was noticably warmer...which was good since more heat is being transferred to the heatsink from the board.
Seeing as the ASIC chips were still REALLY hot to the touch, I cut up a larger heatsink and applied heatsinks to each IC on the board as well as all parts of the power circuitry.
Image of the cooling mod.  (forgive the look of the heatsinks...I had a hell of a time with it...but they do the job.)

Do note that there is only about 7mm-8mm of real clearance between the circuit board and the case.  The heatsink I had cut up was 5mm high.  I also bent the fins a little so they were spread farther apart and not just standing straight up.

:EDIT:  After 3 full days of testing with the application of Arctic Silver Ceramique and the added heatsinks to all IC's on the board, I can run stably at 350Mhz with a 4% error rate (screenshots are in the thread).

See the following posts for examples of fan mods... (Thanks Ktulu1 for compiling the list!)

https://bitcointalk.org/index.php?topic=644921.msg7304782#msg7304782
https://bitcointalk.org/index.php?topic=644921.msg7354916#msg7354916
https://bitcointalk.org/index.php?topic=644921.msg7371079#msg7371079
https://bitcointalk.org/index.php?topic=644921.msg7543334#msg7543334

Also note there are numerous test point locations with voltages silkscreened next to them.  I'm just making an assumption here...but the 1.2vc would indicate a test point for the Core voltage which should read 1.2v.  I measured this point and got 1.26v.

"Fan lubrication mod" (not Zeus hardware specific but good to keep things running smoothly for as long as possible... Thanks jekv2 for the suggestion!)
Link to thread describing mod: http://www.overclock.net/t/773256/prepping-a-sleeve-bearing-fan-for-work1
Link to oil mentioned:  http://acehardwaremaldives.com/product/auto/10705/


So....any other mods?  Suggestions?  Tuning tricks?

I'll keep the first post updated as best as I can with notable information that's contributed.

:Edit - added Zeusminer links

[1714467252]

1714467252

[1714467252]

1714467252

[1714467252]

1714467252

[1714467252]

1714467252

[nst6563]

*Reserved*

Technical info:
"Zeus scrypt chip’s software implementation" doc http://zeusminer.com/user-manual-ver-1-0/

From the doc we know the chips have 8 cores.
Also it indicates the hash rate is approx 70% of the clock rate if "--nocheck-golden" is used.

So 340 clk would equal 238kh/s per chip x 6 chips for the Fury = 1.4mh/s approx
328 clk would equal 230kh/s per chip x 6 chips for the Fury = 1.377mh/s approx

Readcount Timeout is explained as:

4.3   Readcount Timeout
After we send one command packet to the chained chips, that BIG chip will cover all the 32 bits nonce in a certain TIME. So we must send a new work to it before that, otherwise its hash power is wasted.

Cgminer’s readcount timeout is used for that. Every count means 0.1 second. We now set the readcount to be 3/8 of the covering time. It may be 3/4 later.

-------------------------------------------------------------------------------------------------------
As for IC's....
2R030
PEM
1349A6
4214
This looks like it may be it: http://www.nxp.com/documents/data_sheet/PSMN2R0-30YL.pdf

The other chip that looks to control the voltage is marked with:
0193
391
CCON
Looks like this may be the IC in question: http://www.ti.com/lit/ds/slus719e/slus719e.pdf
http://ca.mouser.com/ProductDetail/Texas-Instruments/TPS40193DRCR/?qs=kcKC6WcubkU/HLcVuupS9g==

Silicon Labs USB to UART bridge:
http://www.silabs.com/Support%20Documents/TechnicalDocs/CP2102-9.pdf
-------------------------------------------------------------------------------------------------------
------------------------------SCHEMATICS ...!!! (Thanks Zig!)-------------------------------------------
-------------------------------------------------------------------------------------------------------
http://zeusminer.com/wp-content/uploads/2014/06/BLIZZARD_X6Chips.pdf

alternate link:
https://www.dropbox.com/s/r9i4n8o2yhsqfh8/BLIZZARD_X6Chips.pdf

Blizzard and 32-chip Schematics  (Thanks happydaze!)
http://zeusminer.com/shcematics/    (ok....did anyone else realize that schematics is spelled wrong?)

Zeus ZMC230 datasheet!  (Thanks Zig!)
Direct Link: https://mega.co.nz/#!t54SgYIR!_WzDmy7SuwVurOdWZUIRnNh93C3lSJ_6bUld-BJdlsM
Also here at the very bottom of the page:  http://zeusminer.com/shcematics/

-------------------------------------------------------------------------------------------------------
------------------------------MOD INFO (Thanks J4bberwok!)-------------------------------------------
-------------------------------------------------------------------------------------------------------
In the formula from the datasheet, R7 is th resistor I have swapped, R8 is just under it on the PCB with 93A marking

R7=0.591xR8/(desired voltage-0.591)

5.6K gives 1.57v
6.2k gives 1.48v
6.8k gives 1.4v
7.5k gives 1.3v
8.2k gives 1.26v

I measured a bias of +5% on my board. I haven't measured the R8 out of the board, but if it's slightly out bu 4 or 5% it could be the reason.
1.65v

for undervolting, if you want less power draw
10k gives 1.14v
11k gives 1.09v
12k gives 1.05v
13k gives 1.01v

Changing the R8 will also work
don't touch the 9.1k one, replace the 93A labelled one

10k gives 1.24v
11k gives 1.30v
12k gives 1.37v
13k gives 1.44v
15k gives 1.57v

for undervolt
6.2k gives 0.99v
6.8k gives 1.03v
7.5k gives 1.08v
8.2k gives 1.12v


picture to clear things up


-------------------------------------------------------------------------------------------------------

The chips are 8mm x 8mm.

[suchmoon]

Good job, thank you. I was going to take the stock fans off, put 4 or 8 blizzards together and add a big fan. Should be ok in theory (8 blizzards is ~380W, same as e.g. AntMiner S1) but if there is an issue with thermal paste I might need to rethink this project.

[happydaze]

Thanks for starting this topic.  I think the Zeus Blizzard and the Gawminer Fury are the same.

This is a good link to Zeusminer info:  http://zeusminer.com/user-manual-ver-1-0/  There's the Zeus chip doc there for download.

From the doc we know the chips have 8 cores.
Also it indicates the hash rate is approx 70% of the clock rate if "--nocheck-golden" is used.

So 340 clk would equal 238kh/s per chip x 6 chips for the Fury = 1.4mh/s approx
328 clk would equal 230kh/s per chip x 6 chips for the Fury = 1.377mh/s approx

You can manipulate the Readcount Timeout by changing the "--chips-count" cmd parameter.  Doing so will distort the hash rate displayed by cgminer.  I played around with that but haven't come to a any conclusion if helps yet.  Setting 12 chips reduces the Readcount Timeout by more than half I think (on a 6 chip device).  I set 400 chips for a test.  It slowed the miner down a lot but it was fun seeing a spoofed 90mh/s hash rate in cgminer.

Readcount Timeout is explained as:

4.3   Readcount Timeout
After we send one command packet to the chained chips, that BIG chip will cover all the 32 bits nonce in a certain TIME. So we must send a new work to it before that, otherwise its hash power is wasted.

Cgminer’s readcount timeout is used for that. Every count means 0.1 second. We now set the readcount to be 3/8 of the covering time. It may be 3/4 later.

I'm currently running 340 clk and poolside it was 1,374 kh/s after 24 hours.

[happydaze]

Earlier we started talking about voltage so here's a copy:

here's the post re: Zeus voltage https://litecointalk.org/index.php?topic=16301.msg177074#msg177074

it's possible for a physical mod. if change some of the resistors to get it to work at 1.2 ish V, you save some more power there comparing to 1.3.

power consumption is related to voltage it's running on, which is tunable with hardware mods.

nst6563's photo shows 1.2v printed, I wonder if the voltage is already upped a bit?

and nst6563 had this info:

I measured the 1.2vc mark and it reads 1.26v.  So if the chips were originally spec'd at 1.3v, there's some room to play with. 

As for IC's....I don't have a camera other than the one on my htc....but around the choke are all the same type of mosfet with the markings:
2R030
PEM
1349A6
4214

The other chip that looks to control the voltage is marked with:
0193
391
CCON

Since this is a Fury, there are NO markings on any of the ASIC IC's themselves.  The only other IC with markings is the Silab's USB interface - which from my understanding can support up to 1Mbps transfer rate however the Fury/Blizzard have it limited to 115200Kbps (could there be some performance gained by upping this limit??  It appears that the chips themselves are daisy chained to a nonce solved by the last chip in the chain has to travel through each chip before it and then to the interface....it could be possible that more data is flowing through the port than 115200Kbps can effectively handle I suppose).

:EDIT:  So it looks like the 0193 chip is possibly this TPS40193:  http://www.ti.com/lit/ds/slus719e/slus719e.pdf 

[nst6563]

Thanks!  Updated the first post with the links, 2nd post with the technical information.

Hopefully we can tweak these Fury/Blizzard units to get more out of them like the gridseeds.

[happydaze]

A volt mod would be great.  Not too much, we don't want 100 watt glowing bricks    A nice stable 1.5mh/s or 1.6mh/s would be awesome.

another interesting quote from Terry from Zeus on Litecointalk forum:

https://litecointalk.org/index.php?topic=16301.msg180519#msg180519

we were okay (in a/c room) about 340~350ish at stable speed. but anything above that you would have a lot of errors

350 x .7 = 245 kh/s x 6 = 1.47mh/s  that could be our first goal, maybe obtained by keeping it nice and cool

but beware:

..... but please make sure you are doing it in cool ambient environment because over heating WILL damage the hardware.

caution is advised

[suchmoon]

As for volt mods, given the ungodly power consumption there might be interest in undervolting, not just overvolting. Maybe in a few months when the profitability will be outweighed by power costs.

[happydaze]

Yes, I think some guys with bigger miners wanted that.  Under volt it a bit. 

I'll run  --chips-count 8 for a test (if it continues to look stable) to see if a slightly faster Readcount Timeout helps or hurts.  It'll will be a 24hr test.

Note: if you set more chips than what you have your hashrate in the miner software will be wrong.  It will report 70% of --ltc-clk multiplied by the number of chips you choose. If you set less chips the hash rate will appear low.  If you set more chips the hash rate will appear high. The miner hash rate is not accurate.  Poolside hash rate is what matters.

[nst6563]

Well, I've added the thermal paste to the back, applied heatsinks to all IC's on the top, and there is a fan blowing air through the case over the heatsinks I applied.  Ambient temp is around 73f.  Temperature probe attached to the side of the case/heatsink reads 82f.

I hadn't known about the higher speed information...however....for kicks I started it running at 350 over 2 hours ago.  Right now it's sitting at 1.39Mhs and around 4% error.

[ZiG]

GOOD job, guys...

Let's start digging some info about schematics...

Otherwise should be a tough task...trials and errors...

I don't want anybody to burn anything...

ZiG

[happydaze]

nst6563 - yours may be the fastest Fury so far!

I've been running it more than 12 hrs with  --chips-count 8 and it hasn't hurt things and hasn't helped much either.  340 clk 1378kh/s poolside 24hr average.  That's up by 4kh/s and could be just normal variance.   I find the 24hr avg hash rate number on the pool to be the best gauge of stability and performance.

re: schematics, Zeus had this to say late May

we can open source all that

So I'll think they'll be made available or at least enough info for us to know how to raise or lower the voltage.

[nst6563]

If it's the fastest, it's likely only due to the cooling mods I've done to it by adding just a little more thermal paste to the back of the circuitboard and adding the heatsinks to the IC's on the top.  It does run cool now with the small fan blowing through the inside of the case...temp probe still reads 80f on the heatsink.  I may try to move it to an empty spot on the circuit board sometime today and see what it reads on the board itself.

Btw...Hashra only goes to 350Mhz clock speed.  Not sure if it can be manually specified higher than that or not....I was looking through some of the file last night but didn't really find anything.  If the temps on the fury really are only around 80f as read by the heatsink....I have no doubt we could probably push the speeds a little higher without even touching the voltages.

[happydaze]

The docs say 70% of clk for approx hash rate.  Looks like 67% for me so clk 348 should get close to 1400kh/s.  I think yours is similar with 350 clk and 1400kh/s.
I'll try for 1400 kh/s too.  Can't let you get too far ahead!

[nst6563]

I've got another fury coming this week so I'll take all kinds of measurements on it and post some pics with the values.  Hopefully this will help the collective until we find schematics.

[happydaze]

348 clk sent the HW errors to 11% so I'll try 342.

[nst6563]

Stock cooling or have you done anything to it? 

Here's a 4hr run on mine at the full 350Mhz that Hashra allows.  Still sitting around the 4% error range. 



If a better application of thermal paste and the addition of heatsinks on the top of the IC's can take the speed from 328/336 to 350+ on stock voltage while maintaining the same error rate I'd say that's a step in the right direction.

[daupheus]

great post ! 
I order mine today and try to share with you some tweaks

[happydaze]

Stock cooling or have you done anything to it? 

Just stock cooling....  I lost a bit of speed today.  1,365 kH/s avg now.  I tried too many things at once & HW errors were over 8%.  Time to go get some thermal grease.

[nst6563]

I would definitely recommend the thermal paste.  Anything that's non conductive should be better than what's on there now. 

My Fury has been running for over 10 hrs now, 3103 accepted shares, 125 hw errors.  Sitting right at 4%.  Heatsink temp probe is still at 80f.  I'm moving the probe to the circuitboard to see if there's any difference.