Pages:
Author

Topic: Hacking The KNC Firmware: Overclocking - page 6. (Read 144314 times)

sr. member
Activity: 266
Merit: 250
any details how we can used novec 7000?
i must put each asic chip at his own bottle of novec?Huh

Novec 7000 would be a massive cooling headache; 7100 is much more feasible.

Depends how big your bottle is; you could put a hundred of them in one tank if your tank was big enough and your cooling could keep up.

My experience says no they will not tolerate 50+
Must qualify that statement.
I can have a couple pushin 50.1 or so.
If I see 51.x it might not have puked yet but it is likely.
Some quit before 50A.
Seems like 49A all day works.
Way past recommended with temp(for most) and current.
I lowered all volt 1 selection and it helped everything.
-0.0439
HW error dropped from >0.8% to <0.7%
My results could be unique.
If you try it please report results.
Different PSU/wires combo might produce different results.
YMMV
Smiley

-0.0439 means you're pulling a higher voltage across the chip or not?
sr. member
Activity: 386
Merit: 250
Do we think that the VRM's will tolerate 50A+ if they are properly cooled?
u27

My experience says no they will not tolerate 50+
Must qualify that statement.
I can have a couple pushin 50.1 or so.
If I see 51.x it might not have puked yet but it is likely.
Some quit before 50A.
Seems like 49A all day works.
Way past recommended with temp(for most) and current.

I lowered all volt 1 selection and it helped everything.
-0.0439
HW error dropped from >0.8% to <0.7%

My results could be unique.
If you try it please report results.
Different PSU/wires combo might produce different results.

YMMV
Smiley
legendary
Activity: 2408
Merit: 1004
any details how we can used novec 7000?
i must put each asic chip at his own bottle of novec?Huh
sr. member
Activity: 266
Merit: 250
52a is max for the vrm before OCP kicks in n kills the vrm...so with that limitation you don't have much oc'in room...i'd say you'll be lucky to get 550mhz if that outta these units unless you super cool the vrms lmao...but that wouldn't be worth it

What is OCP? (please excuse my ignorance)

How super cool are we talking? Pretty sure the Novec would still keep them at 65 if you were running 100A through them/

EDIT:
Over Current Protection right?
hero member
Activity: 784
Merit: 504
Dream become broken often
Did you have Status or Advanced web page running?
ssh connection?
miner.php using the API?
I have stopped using the gui completely when possible.
Not even running 'screen -r'
I'll see how long going native lasts.
I get hashrate with
'api-cgminer devs'
Get VRM details and clock volt temp stuff same as Advanced
'waas -g all-asic-info'
This also identifies choices for filtering.
To filter out Frequency
'waas -g all-asic-info | grep Freq'
or a different way to see clock only
'cat /var/volatile/run/.waas_curfreq'
The teaser
Code:
root@Neptune:~# cat /var/run/.waas_curfreq
-1 -1 -1 -1 500 500 550 500 500 500 525 525 500 500 550 500 475 500 525 500 475 525 500 525
Find sleeping VRM
'waas -g all-asic-info | grep Iout'
Change things manually instead of Advanced page
'vi /config/advanced.conf'
apply button replacement, use after manual edit of /config/advanced.conf
'waas'
WARNING this may result just like clicking on Apply in the advanced page with some VRM off.
It's random same as Advanced.
YMMV
Smiley
EDIT
I found a way to wake up sleeping VRM
WITHOUT POWER CYCLE!
it's faster!
stop cgminer with
'/etc/init.d/cgminer.sh stop'
if Advanced page is open wait for numbers to drop (20 seconds?)
To just wake up sleepers no changes
'waas'
If you need to apply changes from Advanced do it with cgminer stopped
If you manually edited /config/advanced.conf, run
'waas'
Wait 20-30 seconds for waas to finish.
restart cgminer with
'/etc/init.d/cgminer.sh restart'
look to see if VRM are happy!
Smiley

This is all looking very promising; custom firmware with a "kick sleeping VRM's" button would be really sweet.

Let me know if I can do anything to help; my Novec 7100 just arrived but the tank is not here yet. I'm going to leave it well alone until KfC sort out my dead die issues and then throw one of them in the Novec for consistent 65 degree running across the board. Do we think that the VRM's will tolerate 50A+ if they are properly cooled?

u27

52a is max for the vrm before OCP kicks in n kills the vrm...so with that limitation you don't have much oc'in room...i'd say you'll be lucky to get 550mhz if that outta these units unless you super cool the vrms lmao...but that wouldn't be worth it
sr. member
Activity: 266
Merit: 250
Did you have Status or Advanced web page running?
ssh connection?
miner.php using the API?
I have stopped using the gui completely when possible.
Not even running 'screen -r'
I'll see how long going native lasts.
I get hashrate with
'api-cgminer devs'
Get VRM details and clock volt temp stuff same as Advanced
'waas -g all-asic-info'
This also identifies choices for filtering.
To filter out Frequency
'waas -g all-asic-info | grep Freq'
or a different way to see clock only
'cat /var/volatile/run/.waas_curfreq'
The teaser
Code:
root@Neptune:~# cat /var/run/.waas_curfreq
-1 -1 -1 -1 500 500 550 500 500 500 525 525 500 500 550 500 475 500 525 500 475 525 500 525
Find sleeping VRM
'waas -g all-asic-info | grep Iout'
Change things manually instead of Advanced page
'vi /config/advanced.conf'
apply button replacement, use after manual edit of /config/advanced.conf
'waas'
WARNING this may result just like clicking on Apply in the advanced page with some VRM off.
It's random same as Advanced.
YMMV
Smiley
EDIT
I found a way to wake up sleeping VRM
WITHOUT POWER CYCLE!
it's faster!
stop cgminer with
'/etc/init.d/cgminer.sh stop'
if Advanced page is open wait for numbers to drop (20 seconds?)
To just wake up sleepers no changes
'waas'
If you need to apply changes from Advanced do it with cgminer stopped
If you manually edited /config/advanced.conf, run
'waas'
Wait 20-30 seconds for waas to finish.
restart cgminer with
'/etc/init.d/cgminer.sh restart'
look to see if VRM are happy!
Smiley

This is all looking very promising; custom firmware with a "kick sleeping VRM's" button would be really sweet.

Let me know if I can do anything to help; my Novec 7100 just arrived but the tank is not here yet. I'm going to leave it well alone until KfC sort out my dead die issues and then throw one of them in the Novec for consistent 65 degree running across the board. Do we think that the VRM's will tolerate 50A+ if they are properly cooled?

u27
sr. member
Activity: 386
Merit: 250

I just woke up, and saw that cgminer 4.4.1 dumped again (and restarted automatically)...  looks that there is some bug in fw-1.0 or cgminer_4.4.1 (I didn't have crashes with rc9 and 4.4.0, while hardware and Volt-tuning are all the same).
Next week, I hope to have some more time and will check that log-file (thanks for the info).


Did you have Status or Advanced web page running?
ssh connection?
miner.php using the API?

I have stopped using the gui completely when possible.
Not even running 'screen -r'
I'll see how long going native lasts.

I get hashrate with
'api-cgminer devs'

Get VRM details and clock volt temp stuff same as Advanced
'waas -g all-asic-info'
This also identifies choices for filtering.

To filter out Frequency
'waas -g all-asic-info | grep Freq'

or a different way to see clock only
'cat /var/volatile/run/.waas_curfreq'

The teaser
Code:
root@Neptune:~# cat /var/run/.waas_curfreq
-1 -1 -1 -1 500 500 550 500 500 500 525 525 500 500 550 500 475 500 525 500 475 525 500 525

Find sleeping VRM
'waas -g all-asic-info | grep Iout'

Change things manually instead of Advanced page
'vi /config/advanced.conf'

apply button replacement, use after manual edit of /config/advanced.conf
'waas'
WARNING this may result just like clicking on Apply in the advanced page with some VRM off.
It's random same as Advanced.

YMMV
Smiley

EDIT
I found a way to wake up sleeping VRM
WITHOUT POWER CYCLE!
it's faster!

stop cgminer with
'/etc/init.d/cgminer.sh stop'
if Advanced page is open wait for numbers to drop (20 seconds?)
To just wake up sleepers no changes
'waas'
If you need to apply changes from Advanced do it with cgminer stopped
If you manually edited /config/advanced.conf, run
'waas'
Wait 20-30 seconds for waas to finish.
restart cgminer with
'/etc/init.d/cgminer.sh restart'
look to see if VRM are happy!
Smiley
full member
Activity: 203
Merit: 100

I got a cgminer dump and auto-restart as well (it ran very nicely, for 9 hours, at 3.585 Ths and 50080 WU).

I'm going to assume it was not possibly VRM temp related in your case.

I still have a couple that reach low 90 on occasion.
Those are still just sitting in some wind.

I mention that because it is certainly 'possible' that they got warmer I doubt it.

Some kind of clue is output from
cat /var/log/monitordcdc.log

It shows throttling of all die(on one ASIC) and documents cgminer restarts indirectly.
The timestamps have big jumps at 'events'.

Been runnin clean for a few hours now.
Pool thinks it did 3.7T for an hour! Smiley
3.600 would be perfect @ 500 and I had 2 die(or 6???) slower at the time.
It's just a calculating average wierdness when block times change during period averaged.
Still pool is not indicating anything to get alarmed about.

I'm just lettin it ride. It recoveres quickly.

YMMV
Smiley

I just woke up, and saw that cgminer 4.4.1 dumped again (and restarted automatically)...  looks that there is some bug in fw-1.0 or cgminer_4.4.1 (I didn't have crashes with rc9 and 4.4.0, while hardware and Volt-tuning are all the same).
Next week, I hope to have some more time and will check that log-file (thanks for the info).
sr. member
Activity: 386
Merit: 250

I got a cgminer dump and auto-restart as well (it ran very nicely, for 9 hours, at 3.585 Ths and 50080 WU).

I'm going to assume it was not possibly VRM temp related in your case.

I still have a couple that reach low 90 on occasion.
Those are still just sitting in some wind.

I mention that because it is certainly 'possible' that they got warmer I doubt it.

Some kind of clue is output from
cat /var/log/monitordcdc.log

It shows throttling of all die(on one ASIC) and documents cgminer restarts indirectly.
The timestamps have big jumps at 'events'.

Been runnin clean for a few hours now.
Pool thinks it did 3.7T for an hour! Smiley
3.600 would be perfect @ 500 and I had 2 die(or 6???) slower at the time.
It's just a calculating average wierdness when block times change during period averaged.
Still pool is not indicating anything to get alarmed about.

I'm just lettin it ride. It recoveres quickly.

YMMV
Smiley
member
Activity: 67
Merit: 10
Im @ 3.53 with new FW 1.0

You found the reason behind the performance gain tolip.. I have two Neptunes, one with its controller card and two cubes currently stuck in a slower than communicated RMA process.

I repurposed a Jupiter controller card to run the orphaned cubes and this 3-cube machine has seen zero change on the new firmware. I believe that supports your pegged CPU observation quite nicely Smiley That 2% was probably getting lost on the wire or flushed from a full buffer.
full member
Activity: 203
Merit: 100
Im @ 3.53 with new FW 1.0

No voltage adjustment
I have all but 2 die @ 500
2 die @ 475 to keep VRM alive
(still nekkid VRMs here)

After sleeping I noticed uptime is only 4 hours.
It must have restarted itself at some point.

WHILE TYPING THIS cgminer crashed.
Not the first time, I was typing 'I must have had a storm pass by..'
when it happened.

Now I blame new FW not nature.
cgminer restarts on its own, not a critical issue.
Shit happens, we are lucky to be allowed to find some of the bugs! Wink
Thx KnCMiner.


YMMV
Smiley

I got a cgminer dump and auto-restart as well (it ran very nicely, for 9 hours, at 3.585 Ths and 50080 WU).
sr. member
Activity: 386
Merit: 250
Im @ 3.53 with new FW 1.0

No voltage adjustment
I have all but 2 die @ 500
2 die @ 475 to keep VRM alive
(still nekkid VRMs here)

After sleeping I noticed uptime is only 4 hours.
It must have restarted itself at some point.

WHILE TYPING THIS cgminer crashed.
Not the first time, I was typing 'I must have had a storm pass by..'
when it happened.

Now I blame new FW not nature.
cgminer restarts on its own, not a critical issue.
Shit happens, we are lucky to be allowed to find some of the bugs! Wink
Thx KnCMiner.


YMMV
Smiley
sr. member
Activity: 386
Merit: 250
Regarding Neptune, I have asked KnC the following questions but got no positive reply. I raise them here to ask for your advice:

1. How to view VRM and ASIC temperature in command line? Now the only way to view temperature is via the web controlling page, but using GUI will cause high hardware error. Is it possible to view temperature info from command line/terminal?

2. KnC uses their customized ARM linux, and there is no package management tool like yum or aptitude. How to install other software package on KnC linux? Now I need python to run some scripts, but I could not install it...

THIS WILL ONLY LAST UNTILL REBOOT

Below is the script Luke-Jr gave us to build bfgminer.
I commented out the part after creating the build environment.

I DO SOMETHING WIERD TO IT
I make a folder to save downloaded stuff for faster repeat builds

'mkdir /config/opkg.cache'

I changed every instance of
install
to
--cache /config/opkg.cache install

if you don't like this remove 4 instances of
--cache /config/opkg.cache
below

running this will make the STATUS wepbage lie and say STOPPED
it is the web interface that is stopped not the hashing

Code:
cat >/etc/opkg/feeds.conf <<\EOF
src/gz noarch http://feeds.angstrom-distribution.org/feeds/v2013.06/ipk/eglibc/all/
src/gz base http://feeds.angstrom-distribution.org/feeds/v2013.06/ipk/eglibc/armv7ahf-vfp-neon/base/
src/gz beaglebone http://feeds.angstrom-distribution.org/feeds/v2013.06/ipk/eglibc/armv7ahf-vfp-neon/machine/beaglebone/
EOF

opkg update
opkg --cache /config/opkg.cache install angstrom-feed-configs
rm /etc/opkg/feeds.conf
opkg update

opkg --cache /config/opkg.cache install update-alternatives
opkg --cache /config/opkg.cache install automake autoconf make gcc cpp binutils git less pkgconfig-dev ncurses-dev libtool nano bash i2c-tools-dev
while ! opkg --cache /config/opkg.cache install libcurl-dev; do true; done

###curl http://www.digip.org/jansson/releases/jansson-2.0.1.tar.bz2 | tar -xjvp
###cd jansson-2.0.1
###./configure --prefix=/usr CC=arm-angstrom-linux-gnueabi-gcc --disable-static NM=arm-angstrom-linux-gnueabi-nm
###make install && ldconfig
###cd ..

###git clone git://github.com/luke-jr/bfgminer
###cd bfgminer
###./autogen.sh
###git clone git://github.com/troydhanson/uthash
###./configure --host=arm-angstrom-linux-gnueabi --enable-knc --disable-other-drivers CFLAGS="-I$PWD/uthash/src"
###make AR=arm-angstrom-linux-gnueabi-ar

###/etc/init.d/cgminer.sh stop
###./bfgminer -S knc:auto -c /config/cgminer.conf


This will create a build environment, install compiler etc.
It also loads up the opkg info so you can install stuff with opkg
opkg is similar to apt-get etc.

Put stuff you want available after reboot in a folder below /config
I use things like
 
/config/bfg
/config/opkg.cache
/config/other

etc.
You should save the sctipt above in a similar folder.
Mine is in /config/bfg where I start when I compile bfgminer
(bfgminer is not ready for Neptune yet)

IT IS NOT A HUGE DISKDRIVE
use 'du' to keep an idea on what is left for the miner to use

Many thanks to Luke-Jr for bfgminer and the build toys installed easily


YMMV
Smiley
sr. member
Activity: 476
Merit: 250
Regarding Neptune, I have asked KnC the following questions but got no positive reply. I raise them here to ask for your advice:

1. How to view VRM and ASIC temperature in command line? Now the only way to view temperature is via the web controlling page, but using GUI will cause high hardware error. Is it possible to view temperature info from command line/terminal?

2. KnC uses their customized ARM linux, and there is no package management tool like yum or aptitude. How to install other software package on KnC linux? Now I need python to run some scripts, but I could not install it...
sr. member
Activity: 476
Merit: 250
Yes but some of your vrms are running too hot.

You need to tweak it carefully.

What is the level, you guys have experience of, giving trouble ?  Some are already at 104 to 106 C

I would not let it run that high.

I know it spits an error at around 95C.

Ericsson says the limit is 120C but I would not dare to let it run more than 95C.

Anyone else to share experiences?

so far my neppy has the temperature on the VRM between 70-95 degree C. Is it safe? although KnC tells me that the temperature can reach up to 120 degree.
full member
Activity: 203
Merit: 100
After 2.5 days running with Volt-tuned 500 Mhz, this is what currently is possible: 3.500 Th/s

That's nice Smiley I'm a little lower as I can't dial everthing up to 500 just yet, but I have picked up about 100Gh/s over the last 8-10 hours with today's firmware release.

Firmware 1.0 is 2% better and much more stable (no jumping 5seconds rate anymore).  But still the browser-pages cause errors visible in cgminer (just don't use the browser for best results).

legendary
Activity: 2408
Merit: 1004
HOW I CAN UNDERVOLT JUPITER?HuhHuhHuh??
member
Activity: 67
Merit: 10
After 2.5 days running with Volt-tuned 500 Mhz, this is what currently is possible: 3.500 Th/s



That's nice Smiley I'm a little lower as I can't dial everthing up to 500 just yet, but I have picked up about 100Gh/s over the last 8-10 hours with today's firmware release.
sr. member
Activity: 386
Merit: 250

Always very good comments tolip Smiley

I have done most of my homework.. chlorine does attack copper but pH is the big issue. Pool heater heat exchangers are often copper. I did buy cheapie WBs from HK with side entry Smiley Smiley so they are probably low on copper content. There should be minimum airflow distruption and I plan to reorganize fans and heatsinks for better VRM cooling. Maybe experiment with some piping and JB weld on a Jupiter board for lower risk to earnings. I will have a pressure relief value to 'short' the loop and will have the setup after the filter on a 'side' loop from the main run and expect the pressure should stay quite low.

The radiator in the pool is a good idea.. I thought about this and my concern at the time was that I have a pretty long run (would add ~50-60 feet to the loop) and about 10-12ft of vertical going into and coming out of the attic space.

Freezing in the attic isn't a big concern here in central Florida, but with the heat these put out we can engineer it so it cannot freeze in any climate. If it did get cold enough I could even skip the radiator Wink

Anyway... still thinking it all through and I do very much appreciate your input!

Thx for the positive feedback,

I can see you are already all in so I'll provide a bit more OPINION for you to ponder.

Pitstop, pitstop, pitstop.
Build the whole loop one time not 5 times if possible for shorter pitstops.
I added the ASIC one at a time to loop during short as possible pitstops.

If you mount PCB/waterblock vertical, height and direction become considerations.
If the flow us UP it is complimentary to filling/burping system.
I use a low input and a high output and I can mostly fill/burp without pressure just siphon.
My saturns each have a PCB upside down.
If I get a big bubble on an upside down waterblock I am running uncooled possibly.
Flow should cure, but things have a way of going wonky in a chain reaction.
2 pumps in series(each can be anywhere in loop),  running off seperate supplies, is my protection.

Hose length:
You have a pool pump ignore.
What is your backup?
5 eggs are in the same basket not 1 fan to replace.

I have 2 800L/hour 4m lift pumps in series, one is enough but 2 assures faster flow.
I have all in single strand, #5 in the line has 1-4's heat pass thru it.
Some imaginary ideal flow rate is 'still cools nearly equally but leaks won't squirt all over'.
The reason I mention all this is higher pressure can create issues that you might/can ignore with lower pressures.
WIth a small resivour at lower pressure I might notice water missing and not know where it went, some pinhole somewhere. My Saturns have a pinhole of a few cc a week. Can go months before concern. Not so with my Nep setup.
The leak that stopped instantly with pump off was quite noticable drippage with Nep pumps goin.

Rise and fall can be ignored with one opening in loop.
Will that opening be the pool?
Shutoffs?

My system has different considerations, I can open close resivour/sight at will while running.
If you want to run loop at lower pressure than the pump you have to have open unrestricted return to pool or it won't flow or won't be lower pressure.

My 20nm ASIC are high in the middle.
The cheapo WB were high in the middle too.
Some laborer crankin down to git er done lowered the corners.

Scrape existing goo with straightedge to get an idea on ASIC flatness.

If the WB have about 40mm^2 flat and lowered corners you're golden.
Stock heatsink has ~35mm^2 if I'm optimistic.

Goo will get U by, have plenty handy.
I had grit in the goo came with Nep WBs.
No idea what the package says.
I dug into my saved supply of artic silver that comes with some heatsinks.
I never use a whole 3 grams?(whatever size they are) on 2 heatsinks.
My thinkin is a shiny new WB should get decent goo.
According to Toms Hardware the Artic Silver is OK stuff so I saved it Smiley

Not if but WHEN IT LEAKS, don't panic, execute PLAN B, pump first.
Vacuume off the water and motor on.
Use air pressure with caution drying, dried & filtered else sandblast PCB that stays wet.
I had VRMs dis-appear till they buddies baked em dry.

I'm taking a look at refurb truck radiators for my semi-permanent setup.
You are starting on the way other end of that spectrum with cooling power.

Keep us posted!

YMMV
Smiley
member
Activity: 67
Merit: 10
Pages:
Jump to: