Hash rate is coming back up. I'll let it run overnight at 342 clk. 340 clk might actually be the number for this unit until I get some thermal paste. Your numbers look really good.
Topic: Fury/Blizzard tuning and mods - page 43. (Read 115281 times)
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.
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.
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.
great post ! 
I order mine today and try to share with you some tweaks
I order mine today and try to share with you some tweaks
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.
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.
348 clk sent the HW errors to 11% so I'll try 342.
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.
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!
I'll try for 1400 kh/s too. Can't let you get too far ahead!
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.
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.
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
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.
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
Quote
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.
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
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
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.
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.
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.
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.
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.
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
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:
caution is advised
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
Quote
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:
Quote
..... but please make sure you are doing it in cool ambient environment because over heating WILL damage the hardware.
caution is advised
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.
Hopefully we can tweak these Fury/Blizzard units to get more out of them like the gridseeds.
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
nst6563's photo shows 1.2v printed, I wonder if the voltage is already upped a bit?
and nst6563 had this info:
here's the post re: Zeus voltage https://litecointalk.org/index.php?topic=16301.msg177074#msg177074
Quote
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.
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:
Quote
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
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
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:
I'm currently running 340 clk and poolside it was 1,374 kh/s after 24 hours.
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:
Quote
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.
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.
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.
*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:
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.
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:
Quote
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.
-------------------------------------------------------------------------------------------------------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.
Jump to: