Overclocked Avalon?
Magnificent.
Water cool that suxor!
Topic: My Impression of Avalon Machine (cooling info updated) - page 4. (Read 15188 times)
3 HW Error about 0.55% ~ 0.65% @ 282. ~0.75% @ overclocking to 300. hashrate increase by 5% if overclocked.
Dude,
The review is just great!
Please, can you confirm what options are you using for overclocking
--avalon-options 115200:24:10:45:300? Particularity i am interested in time-out value --avalon-options 115200:24:10:40-45??:300. Have you tried with something smaller than 45 (default)
Best
I'll try 42:300 soon.
I am very curious if that will increase the hash rate a little bit more:)
Best
I'm really happy of this, let's hope that they find a way to make the process smoother and respect the deadlines.
3 HW Error about 0.55% ~ 0.65% @ 282. ~0.75% @ overclocking to 300. hashrate increase by 5% if overclocked.
Dude,
The review is just great!
Please, can you confirm what options are you using for overclocking
--avalon-options 115200:24:10:45:300? Particularity i am interested in time-out value --avalon-options 115200:24:10:40-45??:300. Have you tried with something smaller than 45 (default)
Best
I'll try 42:300 soon.
3 HW Error about 0.55% ~ 0.65% @ 282. ~0.75% @ overclocking to 300. hashrate increase by 5% if overclocked.
Dude,
The review is just great!
Please, can you confirm what options are you using for overclocking
--avalon-options 115200:24:10:45:300? Particularity i am interested in time-out value --avalon-options 115200:24:10:40-45??:300. Have you tried with something smaller than 45 (default)
Best
Thank you for your review.
I'm a first batch Avalon buyer in China, purchased via taobao / alipay (& paid RMB).
Avalon sent the machine on Feb 18 & I got my first machine (*: my order is 3 machines) on Feb 20 and it hashes well on HHTT.
The first buyer in China got his machine afternoon or evening on Feb 18 (+0800) I think. Mine delayed for 2 days.
Currently (Feb 22) even someone in China get their Avalon (5 is confirmed including mine, no more than 10) in their hands, they each has only ONE machine delivered no matter how many are ordered. Avalon planned to ship the remaining at the end of Feb.
Have not heard any new confirmed shipped machine in China as Feb 25, so the previous paragraph has NOT been invalidated.
Now I have all the 3 machines ...
Here is some unordered summaries & notices that may help you prepare.
You may ask other related questions & I'll try hard to answer.
Tips will be appreciated: 1HFhtvzNEiy9eooLSJRMyZ1XB2FXgYGvCH
You may want to see xiangfu's config guide as well:
http://en.qi-hardware.com/wiki/Avalon
And ngzhang's advices:
https://bitcointalksearch.org/topic/m.1604081
About the system (openwrt):
1 the embedded wr703n is MEM modded.
original 703n is 32M mem, it has been modded to 64M. but no flash mod. flash size is 4M, the same as original.
those who would like to have ipv6 may become disappointed. may have to use another machine & connect via USB.
(if you prepare to do so, longer USB A-to-B type cable or USB extension cable & teardown is required, see Update 2 blow)
Update: in the first sample machine, the cable is short & it would be nice if you have a replacement.
make sure the connector is fastened, "no_matching_work" in cgminer "stats" api should be always 0.
the newly sent machines have much better cable & protection, should be ok.
Update2: ngzhang says "it will not work if there is no hub between the 703N and controller board"
2 the miner is cgminer 2.10.5, should be the same as the git tree. it should have no memory leak bug (for stratum).
Update: please update to a newer firmware from http://downloads.qi-hardware.com/people/xiangfu/avalon/
3 fan control is a little buggy, should be fixed by xiangfu soon (already in tree).
Update: upgrade firmware should fix.
4 the jffs2 partition is so small (320k) & can only hold config files.
that means cant upgrade cgminer via jffs2 unless done firmware repackaging. cgminer is installed in squashfs.
5 default IP is 192.168.0.100/24, gateway 192.168.0.1 (you may have known)
6 there is some monitoring logic in cron, but it does NOT prevent the machine to get wonky (jgrazik & me both have experienced, crying~~).
Here is a snippet might be useful.
(default port is 4028. I'm not responsible for any loss if the snippet doesnt work as expected. you should know a little bash scripting)
IP=xxx.xxx.xxx.xxx
PORT=4028
RESTART_MONITOR="(echo -n 'devs' | nc ${IP} ${PORT} | grep 'MHS 5s=0.0') && (echo -n 'restart' | nc ${IP} ${PORT})"
echo $RESTART_MONITOR
while true; do date; sleep 10; sh -c "$RESTART_MONITOR"; done
7 version: openwrt rev r35097, flash chip info (should be the same as newer rev wr703n/wr720n):
[ 0.480000] m25p80 spi0.0: found en25q32b, expected m25p80
[ 0.480000] m25p80 spi0.0: en25q32b (4096 Kbytes)
About the packaging & machine:
1 package include a power cable, an wifi antenna (SMA connector, gain is 5dBi I think) and the machine it self.
2 no instruction included (hopefully yours may include one).
3 to teardown the box, use Hex Key Wrench of size M4.0. your fingers may get hurt, put on gloves first.
4 the USB cable may get dropped from the board during shipping, make sure there is a blue light on the 703n board when powered on, or you will be in trouble...
5 there is an 4 port USB hub included, unopened 3M double-sided tape pasted on the back of the hub.
6 the PSU is Antec EA-650 Green (80P Bronze, not the Platinum one)
7 There is apower board(?) power distribution board inside, on the board there are 2 * 6pin PCI-E power connectors remaining.
so if you want to replace a better PSU, make sure it have 2 more 6pin pci-e connectors MORE THAN EA-650.
I made a mistake!
the PDU board layout by ngzhang:
10pin (out to control board)
24pin (ATX input)
12pin | 12pin | 12pin | 12pin (out, each to power one module, one empty for expansion. it really looks like dual 6pin pci-e)
8pin | 8pin | 4pin | 4pin (in, 8pin ones are pci-e 8pin, 4pin ones are cpu 4pin)
so any PSU with 24pin ATX + 2 * 8pin pci-e + 2 * 4pin CPU would be ok.
if your PSU have 3 or more ways of 12V output, you may want to consider how to merge them.
I guess 650w should be ok for 4 modules?
8 the control board
10pin power input from PDU board
one B-type USB socket (provides power to USB hub reversely, then 703n get powered), so it is appeared as USB serial device.
4 * IDE-like sockets (for the data to each module, one empty for expansion)
3 * small 3pin temp sensor socket (1 empty, not sure what type of temp sensor. one at the smaller 3pin fan for air temp, one installed in the front, on the heat sink of 3rd / hottest module for module temp)
5 * 4pin PWM fan headers (the same size of cpu/chassis fans on motherboards. 2 fans are 4pin, 1 fan is 3pin, 2 headers empty, so you are able to install 2 fans at front. make sure the wire should be long)
Note: the speed of the two 4pin fans are controlled by the controller (cgminer driver call adjust_fan/temp), it is unknown that the exact relation between the command and RPM of EACH fan.
9 there is no thermal grease between the bottom of the heat sink (of the modules) and the upside of bottom aluminium board (& it is a little hot).
you may wish to mod them and do some work outside the machine to cool down the downside of bottom board.
10 lsusb:
Bus 002 Device 008: ID 0403:6001 Future Technology Devices International, Ltd FT232 USB-Serial (UART) IC
Device Descriptor:
bLength 18
bDescriptorType 1
bcdUSB 2.00
bDeviceClass 0 (Defined at Interface level)
bDeviceSubClass 0
bDeviceProtocol 0
bMaxPacketSize0 8
idVendor 0x0403 Future Technology Devices International, Ltd
idProduct 0x6001 FT232 USB-Serial (UART) IC
bcdDevice 6.00
iManufacturer 1
iProduct 2
iSerial 3
bNumConfigurations 1
Configuration Descriptor:
bLength 9
bDescriptorType 2
wTotalLength 32
bNumInterfaces 1
bConfigurationValue 1
iConfiguration 0
bmAttributes 0xa0
(Bus Powered)
Remote Wakeup
MaxPower 90mA
Interface Descriptor:
bLength 9
bDescriptorType 4
bInterfaceNumber 0
bAlternateSetting 0
bNumEndpoints 2
bInterfaceClass 255 Vendor Specific Class
bInterfaceSubClass 255 Vendor Specific Subclass
bInterfaceProtocol 255 Vendor Specific Protocol
iInterface 2
Endpoint Descriptor:
bLength 7
bDescriptorType 5
bEndpointAddress 0x81 EP 1 IN
bmAttributes 2
Transfer Type Bulk
Synch Type None
Usage Type Data
wMaxPacketSize 0x0040 1x 64 bytes
bInterval 0
Endpoint Descriptor:
bLength 7
bDescriptorType 5
bEndpointAddress 0x02 EP 2 OUT
bmAttributes 2
Transfer Type Bulk
Synch Type None
Usage Type Data
wMaxPacketSize 0x0040 1x 64 bytes
bInterval 0
11 the two main fans (for modules) are "4715KL-04W-B56"
About running:
1 you may wish to move all the 3 modules to the left, can lower the temperature about 3 degree.
aware: this may increase the difficulty of installing the 4th module (if released in the future).
Update: 1 on the left & 2 on the right would be a better choice.
2 64300-64500 MHash/s measured by submitted work units, 67000 MHash/s displayed by default.
3 HW Error about 0.55% ~ 0.65% @ 282. ~0.75% @ overclocking to 300. hashrate increase by 5% if overclocked.
I recommend not touch the last 45 in the options.
when do overclocking, there are some duplicated shares appeared, not sure it is cgminer or HW's fault.
the overclocking may have reduced the total efficiency, so I guess we should use something between 282~300.
4 please DO preconfigure the diff of ALL the pools (including the backup ones), var diff may slow down the machine at first until the diff is adjusted.
5 fans must remain 2000+ RPM to keep the heat out.
with optimized fan control, fan2 is 2640, fan3 is 2280 while temp is 46 deg (incoming air is 28 deg measured by temp1, outsite temp is 23 deg).
the noise level is comparable to a multi GPU cluster with 60-70% fan. I dont have a measure device to get the exact number.
(I tried to use some apps, 54dB for 1 meter away, details see the reply #38)
currently there's a bug in fan controlling (if not fixed),
but if that bug is not fixed, the full power of fans may help to cool down the machine & make the uptime a little longer.
so the conclusion is:
IF YOU WANT TO DO MOD, GO AHEAD & IT WILL INCREASE STABILITY!
personally I would guess there will be more problems if the machines are stacked.
6 I dont think a tutorial is required if you use the web interface. if you have openwrt experiences...
just make sure you have a machine in the same subnet (192.168.0.0/24) and have a different IP, type in the default IP...
(I dont want to type the full steps, sorry... the only thing worth mentioning is, you can put http:// or stratum+tcp:// in the pool URL / address field.)
if you use telnet (first time to openwrt) / ssh (after set the root password), use vi / cat / sed to write or modify your config files.
please refer to openwrt docs to config /etc/config/network, and see the github repo below to modify /etc/config/cgminer.
you may get these file samples on https://github.com/BitSyncom/cgminer-openwrt-packages , as that is a source repo, so filename may different.
7 monitoring
monitoring is achieved via cgminer's api. https://github.com/ckolivas/cgminer/blob/master/API-README
there is a command-line utility called cgminer-api stayed inside the openwrt system, it would parse the result into a more human readable format. use nc (netcat) is also ok by echo -n "command" | nc IP PORT.
if you like to try cgminer-api, just compile the api-example.c in cgminer and rename XD
just type cgminer-api in telnet / ssh it should display "summary".
the command syntax is:
cgminer-api [command [IP [PORT]]]
temp in included in "devs" query (only temp3 as the "Temperature" field) and "stats" query has more including temp1~3 & fan1~3.
8 monitoring again...
fan1 is alway 0 (no connect, add a 4pin PWM fan to trigger), fan2 / fan3 should >= 2000.
temp2 is always -1 (no connect, unknown sensor type if want a replacement), temp1 is income air temp (or machine temp?), temp3 is module temp (should be 40+ after working for a while)
it would be strange if fan2 & fan3 are 0, MHS 5s from devs is 0.0, it may require hard power off & (maybe) wait several sec for the back light to be off, then power on again.
Note: I may update the main thread for updates.
Avalon sent the machine on Feb 18 & I got my first machine (*: my order is 3 machines) on Feb 20 and it hashes well on HHTT.
The first buyer in China got his machine afternoon or evening on Feb 18 (+0800) I think. Mine delayed for 2 days.
Currently (Feb 22) even someone in China get their Avalon (5 is confirmed including mine, no more than 10) in their hands, they each has only ONE machine delivered no matter how many are ordered. Avalon planned to ship the remaining at the end of Feb.
Have not heard any new confirmed shipped machine in China as Feb 25, so the previous paragraph has NOT been invalidated.
Now I have all the 3 machines ...
Here is some unordered summaries & notices that may help you prepare.
You may ask other related questions & I'll try hard to answer.
Tips will be appreciated: 1HFhtvzNEiy9eooLSJRMyZ1XB2FXgYGvCH
You may want to see xiangfu's config guide as well:
http://en.qi-hardware.com/wiki/Avalon
And ngzhang's advices:
https://bitcointalksearch.org/topic/m.1604081
About the system (openwrt):
1 the embedded wr703n is MEM modded.
original 703n is 32M mem, it has been modded to 64M. but no flash mod. flash size is 4M, the same as original.
those who would like to have ipv6 may become disappointed. may have to use another machine & connect via USB.
(if you prepare to do so, longer USB A-to-B type cable or USB extension cable & teardown is required, see Update 2 blow)
Update: in the first sample machine, the cable is short & it would be nice if you have a replacement.
make sure the connector is fastened, "no_matching_work" in cgminer "stats" api should be always 0.
the newly sent machines have much better cable & protection, should be ok.
Update2: ngzhang says "it will not work if there is no hub between the 703N and controller board"
2 the miner is cgminer 2.10.5, should be the same as the git tree. it should have no memory leak bug (for stratum).
Update: please update to a newer firmware from http://downloads.qi-hardware.com/people/xiangfu/avalon/
3 fan control is a little buggy, should be fixed by xiangfu soon (already in tree).
Update: upgrade firmware should fix.
4 the jffs2 partition is so small (320k) & can only hold config files.
that means cant upgrade cgminer via jffs2 unless done firmware repackaging. cgminer is installed in squashfs.
5 default IP is 192.168.0.100/24, gateway 192.168.0.1 (you may have known)
6 there is some monitoring logic in cron, but it does NOT prevent the machine to get wonky (jgrazik & me both have experienced, crying~~).
Here is a snippet might be useful.
(default port is 4028. I'm not responsible for any loss if the snippet doesnt work as expected. you should know a little bash scripting)
IP=xxx.xxx.xxx.xxx
PORT=4028
RESTART_MONITOR="(echo -n 'devs' | nc ${IP} ${PORT} | grep 'MHS 5s=0.0') && (echo -n 'restart' | nc ${IP} ${PORT})"
echo $RESTART_MONITOR
while true; do date; sleep 10; sh -c "$RESTART_MONITOR"; done
7 version: openwrt rev r35097, flash chip info (should be the same as newer rev wr703n/wr720n):
[ 0.480000] m25p80 spi0.0: found en25q32b, expected m25p80
[ 0.480000] m25p80 spi0.0: en25q32b (4096 Kbytes)
About the packaging & machine:
1 package include a power cable, an wifi antenna (SMA connector, gain is 5dBi I think) and the machine it self.
2 no instruction included (hopefully yours may include one).
3 to teardown the box, use Hex Key Wrench of size M4.0. your fingers may get hurt, put on gloves first.
4 the USB cable may get dropped from the board during shipping, make sure there is a blue light on the 703n board when powered on, or you will be in trouble...
5 there is an 4 port USB hub included, unopened 3M double-sided tape pasted on the back of the hub.
6 the PSU is Antec EA-650 Green (80P Bronze, not the Platinum one)
7 There is a
so if you want to replace a better PSU, make sure it have 2 more 6pin pci-e connectors MORE THAN EA-650.
I made a mistake!
the PDU board layout by ngzhang:
10pin (out to control board)
24pin (ATX input)
12pin | 12pin | 12pin | 12pin (out, each to power one module, one empty for expansion. it really looks like dual 6pin pci-e)
8pin | 8pin | 4pin | 4pin (in, 8pin ones are pci-e 8pin, 4pin ones are cpu 4pin)
so any PSU with 24pin ATX + 2 * 8pin pci-e + 2 * 4pin CPU would be ok.
if your PSU have 3 or more ways of 12V output, you may want to consider how to merge them.
I guess 650w should be ok for 4 modules?
8 the control board
10pin power input from PDU board
one B-type USB socket (provides power to USB hub reversely, then 703n get powered), so it is appeared as USB serial device.
4 * IDE-like sockets (for the data to each module, one empty for expansion)
3 * small 3pin temp sensor socket (1 empty, not sure what type of temp sensor. one at the smaller 3pin fan for air temp, one installed in the front, on the heat sink of 3rd / hottest module for module temp)
5 * 4pin PWM fan headers (the same size of cpu/chassis fans on motherboards. 2 fans are 4pin, 1 fan is 3pin, 2 headers empty, so you are able to install 2 fans at front. make sure the wire should be long)
Note: the speed of the two 4pin fans are controlled by the controller (cgminer driver call adjust_fan/temp), it is unknown that the exact relation between the command and RPM of EACH fan.
9 there is no thermal grease between the bottom of the heat sink (of the modules) and the upside of bottom aluminium board (& it is a little hot).
you may wish to mod them and do some work outside the machine to cool down the downside of bottom board.
10 lsusb:
Bus 002 Device 008: ID 0403:6001 Future Technology Devices International, Ltd FT232 USB-Serial (UART) IC
Device Descriptor:
bLength 18
bDescriptorType 1
bcdUSB 2.00
bDeviceClass 0 (Defined at Interface level)
bDeviceSubClass 0
bDeviceProtocol 0
bMaxPacketSize0 8
idVendor 0x0403 Future Technology Devices International, Ltd
idProduct 0x6001 FT232 USB-Serial (UART) IC
bcdDevice 6.00
iManufacturer 1
iProduct 2
iSerial 3
bNumConfigurations 1
Configuration Descriptor:
bLength 9
bDescriptorType 2
wTotalLength 32
bNumInterfaces 1
bConfigurationValue 1
iConfiguration 0
bmAttributes 0xa0
(Bus Powered)
Remote Wakeup
MaxPower 90mA
Interface Descriptor:
bLength 9
bDescriptorType 4
bInterfaceNumber 0
bAlternateSetting 0
bNumEndpoints 2
bInterfaceClass 255 Vendor Specific Class
bInterfaceSubClass 255 Vendor Specific Subclass
bInterfaceProtocol 255 Vendor Specific Protocol
iInterface 2
Endpoint Descriptor:
bLength 7
bDescriptorType 5
bEndpointAddress 0x81 EP 1 IN
bmAttributes 2
Transfer Type Bulk
Synch Type None
Usage Type Data
wMaxPacketSize 0x0040 1x 64 bytes
bInterval 0
Endpoint Descriptor:
bLength 7
bDescriptorType 5
bEndpointAddress 0x02 EP 2 OUT
bmAttributes 2
Transfer Type Bulk
Synch Type None
Usage Type Data
wMaxPacketSize 0x0040 1x 64 bytes
bInterval 0
11 the two main fans (for modules) are "4715KL-04W-B56"
About running:
1 you may wish to move all the 3 modules to the left, can lower the temperature about 3 degree.
aware: this may increase the difficulty of installing the 4th module (if released in the future).
Update: 1 on the left & 2 on the right would be a better choice.
2 64300-64500 MHash/s measured by submitted work units, 67000 MHash/s displayed by default.
3 HW Error about 0.55% ~ 0.65% @ 282. ~0.75% @ overclocking to 300. hashrate increase by 5% if overclocked.
I recommend not touch the last 45 in the options.
when do overclocking, there are some duplicated shares appeared, not sure it is cgminer or HW's fault.
the overclocking may have reduced the total efficiency, so I guess we should use something between 282~300.
4 please DO preconfigure the diff of ALL the pools (including the backup ones), var diff may slow down the machine at first until the diff is adjusted.
5 fans must remain 2000+ RPM to keep the heat out.
with optimized fan control, fan2 is 2640, fan3 is 2280 while temp is 46 deg (incoming air is 28 deg measured by temp1, outsite temp is 23 deg).
the noise level is comparable to a multi GPU cluster with 60-70% fan. I dont have a measure device to get the exact number.
(I tried to use some apps, 54dB for 1 meter away, details see the reply #38)
currently there's a bug in fan controlling (if not fixed),
but if that bug is not fixed, the full power of fans may help to cool down the machine & make the uptime a little longer.
so the conclusion is:
IF YOU WANT TO DO MOD, GO AHEAD & IT WILL INCREASE STABILITY!
personally I would guess there will be more problems if the machines are stacked.
6 I dont think a tutorial is required if you use the web interface. if you have openwrt experiences...
just make sure you have a machine in the same subnet (192.168.0.0/24) and have a different IP, type in the default IP...
(I dont want to type the full steps, sorry... the only thing worth mentioning is, you can put http:// or stratum+tcp:// in the pool URL / address field.)
if you use telnet (first time to openwrt) / ssh (after set the root password), use vi / cat / sed to write or modify your config files.
please refer to openwrt docs to config /etc/config/network, and see the github repo below to modify /etc/config/cgminer.
you may get these file samples on https://github.com/BitSyncom/cgminer-openwrt-packages , as that is a source repo, so filename may different.
7 monitoring
monitoring is achieved via cgminer's api. https://github.com/ckolivas/cgminer/blob/master/API-README
there is a command-line utility called cgminer-api stayed inside the openwrt system, it would parse the result into a more human readable format. use nc (netcat) is also ok by echo -n "command" | nc IP PORT.
if you like to try cgminer-api, just compile the api-example.c in cgminer and rename XD
just type cgminer-api in telnet / ssh it should display "summary".
the command syntax is:
cgminer-api [command [IP [PORT]]]
temp in included in "devs" query (only temp3 as the "Temperature" field) and "stats" query has more including temp1~3 & fan1~3.
8 monitoring again...
fan1 is alway 0 (no connect, add a 4pin PWM fan to trigger), fan2 / fan3 should >= 2000.
temp2 is always -1 (no connect, unknown sensor type if want a replacement), temp1 is income air temp (or machine temp?), temp3 is module temp (should be 40+ after working for a while)
it would be strange if fan2 & fan3 are 0, MHS 5s from devs is 0.0, it may require hard power off & (maybe) wait several sec for the back light to be off, then power on again.
Note: I may update the main thread for updates.
Jump to: