Topic: got problem with 5 x 5870 (Read 2887 times)
My apologies to the original poster. Did not intend to hi-jack your post. But this is "desperation time" with these GA-990FXA-UD3 boards
Thanks for the link to the M/board page, been looking for that for weeks. Mentions GA-990FXA-UD3 but does not clarify which Version. That is what I am starting to think is an issue here.
So as not to disrupt this thread anymore I did originally post a question regarding these damn Gigabyte boards in Newbies. So if anybody has any startling revelations regarding making these POS work with more than 3 GPU's please post there.
https://bitcointalksearch.org/topic/four-5850s-on-ga-990fxa-ud5-how-86995
Thanks
I see from your first post that you have 5 Cards (5870's)on one M/board using extenders. Would you please be kind enough to tell me exactly what make and model the M/board is and are the extenders 1 - 16 or 16 - 16 or a combo of the two types.
Currently I have two Gigabyte GA-990FXA-UD5 boards that totally refuse to run more than 3 cards (5850), whatever you do or try. They also refuse to reliably boot from USB. They have cost me days and days of "down time" as I fought with them, not to mention the intense frustration!!!.
I need to replace them ASAP but need to know which M/board is a 'proven performer" with 4+ cards.
Thanks
There's a thread for that, check it out:
https://bitcointalksearch.org/topic/motherboards-what-works-wfour-or-more-gpus-list-them-37880
I use 2 mobos..
ASUS crosshairV
MSI gd70
the gd70 will do 6 cards, the crosshairV will do 5 cards.
ASUS crosshairV
MSI gd70
the gd70 will do 6 cards, the crosshairV will do 5 cards.
I see from your first post that you have 5 Cards (5870's)on one M/board using extenders. Would you please be kind enough to tell me exactly what make and model the M/board is and are the extenders 1 - 16 or 16 - 16 or a combo of the two types.
Currently I have two Gigabyte GA-990FXA-UD5 boards that totally refuse to run more than 3 cards (5850), whatever you do or try. They also refuse to reliably boot from USB. They have cost me days and days of "down time" as I fought with them, not to mention the intense frustration!!!.
I need to replace them ASAP but need to know which M/board is a 'proven performer" with 4+ cards.
Thanks
yes,I am using riser for all of them
Are you using risers/extenders?
With my 5970 if I plug it into the board I can clock the engine to 875 without problems but if I put it on a riser I can only clock it to 850. Higher clocks on risers cause issues for me.
With my 5970 if I plug it into the board I can clock the engine to 875 without problems but if I put it on a riser I can only clock it to 850. Higher clocks on risers cause issues for me.
solved loss network problem by following this link:
http://www.bitcointrading.com/forum/index.php?topic=416.msg1235#msg1235
still testing
http://www.bitcointrading.com/forum/index.php?topic=416.msg1235#msg1235
still testing
I set the WORKSIZE=128 today,and I got 373 per card right now
I got another problem,I am using powerline ,but after couple hours,the network disconnected, and it won't reconnect,I have to do it manually,anybody has anycode can check the connection and can reconnect automatically?
thanks everyone!
I got another problem,I am using powerline ,but after couple hours,the network disconnected, and it won't reconnect,I have to do it manually,anybody has anycode can check the connection and can reconnect automatically?
thanks everyone!
Anywhere from 1.125 to 1.175, i.e. stock. If you are undervolting, those probably won't work.
Interesting , I have 3 5870's.
The 2 reference design cards, I am running at 875 mhz core, 150 mhz mem @ 1.025v and getting around around 401 Mh/s for each one.
Then I have the big boy with the arctic cooling set up, 1000 mhz core, 150 mhz mem @ 1.080v and getting around 450 mh/s....
Using Cgminer 2.4.1 with a 128 work size, 2 vectors..phatk kernel with 11.11 catalyst drivers.
Probably is the worksize. Note that I said 160, not 150. For a negligible power bump, you get a few more hashes with that. That is assuming a core of 850 to 900 or so.
+1 for truth. Mind, memory frequencies are approximate. It's hard to explain without rambling, but if you have a bunch of 5830's (different VGA BIOS -may- cause a difference due to possibly different memory latencies coded in. Untested) at a certain core frequency, they should all have a the same "favorite" memory frequency. If you have a 5870 at the same core frequency though, it won't have the same "favorite" as the 5830's as I've found. My 5870's prefer a slightly lower memory at a certain core compared to my 5830's.
Probably is the worksize. Note that I said 160, not 150. For a negligible power bump, you get a few more hashes with that. That is assuming a core of 850 to 900 or so.
just can get about 333mhash when I set it to 850/150(I can get 410+before when I set to 900/300,big difference)
still testing if it is stable!
thanks for all the suggestions!thanks
wtf 333 is wrong, you must have wrong vectors or something. Sorry I don't have rigs running right now so I can't check.still testing if it is stable!
thanks for all the suggestions!thanks
Add WORKSIZE=128 to your flags. Right now, you are using the default (256), which is probably too large for mem=150. If that doesn't help, go back to mem=300 and try both 128 and 256 as your worksize. You might even try 64 for your worksize - I think some people get best results with that, but 128 is best for me.
I also noticed a weird thing with BAMT where mem=250 (and I think worksize 256) gave me a boost in reported hash rate. But as far as I could tell it actually throttled my share submission rate, so I didn't stick with it.
just can get about 333mhash when I set it to 850/150(I can get 410+before when I set to 900/300,big difference)
still testing if it is stable!
thanks for all the suggestions!thanks
wtf 333 is wrong, you must have wrong vectors or something. Sorry I don't have rigs running right now so I can't check.still testing if it is stable!
thanks for all the suggestions!thanks
that sounds like he is using a worksize of 256, which likes 300+ memory and hates 150ish memory. 150ish is only good for 128 worksize.
just can get about 333mhash when I set it to 850/150(I can get 410+before when I set to 900/300,big difference)
still testing if it is stable!
thanks for all the suggestions!thanks
wtf 333 is wrong, you must have wrong vectors or something. Sorry I don't have rigs running right now so I can't check. still testing if it is stable!
thanks for all the suggestions!thanks
just can get about 333mhash when I set it to 850/150(I can get 410+before when I set to 900/300,big difference)
still testing if it is stable!
thanks for all the suggestions!thanks
still testing if it is stable!
thanks for all the suggestions!thanks
Anywhere from 1.125 to 1.175, i.e. stock. If you are undervolting, those probably won't work.
850/160 works great on 5870s
your core voltage and mhash?
850/160 works great on 5870s
Probably so, but the OP needs to troubleshoot here, so it would be better to start with something really modest to find out if the issue is the clock settings or something else entirely. I have 6 5850s, and all of them are fine with going to 150 memory setting, except one, which gets unhappy much below 450. So it may be that in general 5870s are fine with 850/160 @1.05 volts, but it doesn't follow that all the OP's 5870s are happy with this.
850/160 works great on 5870s
Looking at your config file, the first thing to try is backing way off of 900 as your core speed. You are are at 1.05 volts, so you are undervolting, and at that voltage 900 is *really* optimistic. Drop down to 600 to start, and I would also raise your memory clock to at least 450 - I have one 5850 that doesn't like to clock much below that. If you can get all 5 working with these moderate settings, then you can start raising the core/lowering the memory to find out where your cards are getting unstable. If you still can't get all 5 working that way, then try one card at a time by disabling the other four to see if it is just one card giving you trouble. Also, what motherboard are you using? I have read that for some motherboards you have to go into bios and change the latency timer to get 5 cards working. The needed change is to go from 64 --> 96. This didn't seem to matter for my biostar, but I read that gigabyte boards need this setting changed.
Haven't you been having trouble with this for a while? It's little hard to advise w/o reviewing what you have tried so far, but here are some basic suggestions. I got 5x5850 going on BAMT, but it was pretty unstable at first. You have probably already tried this, but if you have made *any* hardware configuration changes, even if they seem trivial, boot from a fresh copy of BAMT. At one point I swapped one 5850 for another, so the two were in different slots, and BAMT completely freaked out about this. I had to start over with a fresh copy. Also, you might try being uber anal about that fresh copy, as in write zeros to the usb drive to completely wipe it and then use dd in linux as opposed to win32diskimager to copy the image. Another thing to try is to get the cards going manually in phoenix - i.e. cd /opt/miners/phoenix - and start each card manually. If you can get all 5 going this way, this will at least tell you that the problem is specific to the process of starting automatically from the configuration script. Also, try starting your GPUs with *REALLY* moderate settings (like 600/500 core/clock at stock voltage). If they work this way, you know you are pushing the clocks of one or more GPUs too much. If they still don't work, lower the voltage - then if they work you know your PSU can't handle 5 cards at stock voltage.
I think BAMT is awesome, but for me it did a lot of really strange things. I know it's supposed to be essentially plug and play, but in my experience it was a bit less friendly than this. I've switched now to cgminer on a flavor of ubuntu 11.04 (Peppermint Two), and I'm liking this better. I definitely recommend cgminer if you can't get BAMT working.
I think BAMT is awesome, but for me it did a lot of really strange things. I know it's supposed to be essentially plug and play, but in my experience it was a bit less friendly than this. I've switched now to cgminer on a flavor of ubuntu 11.04 (Peppermint Two), and I'm liking this better. I definitely recommend cgminer if you can't get BAMT working.
thanks for your suggestion,I will try
here 's my setting
Code:
# Big A Miner Thing configuration
# /etc/bamt/bamt.conf
#
# When done editing, restart mining to make changes here take effect
# If you're just setting up BAMT, be sure to edit pools as well.
#
# As of 0.5, many less common options have been removed from this
# default config to make it easier on the noobs.
#
# You will find a complete example with all options described
# at /opt/bamt/examples/bamt.conf
#
# Note: This is a YAML file. Indentation is significant.
#
# If you break things badly, a backup of this file in it's original
# form can be found at /opt/bamt/examples/bamt.conf.simple
#
---
settings:
# miner_id is used to set hostname and identify this rig to munin and
# other processes
# IMPORTANT! miner_id must be a valid hostname: all lower case, no weird
# characters, underscores, periods, or spaces. Dash and numbers are ok.
miner_id: bamt-miner
# loc is any additional string you wish to use to identify or group this rig
miner_loc: unspecified location
# should we monitor the cards? 1 = yes, 0 = no
do_monitor: 0
# if monitoring, how do we send email alerts?
# these are required..
# smtp_host: smtp.gmail.com
# smtp_to: [email protected]
# these are optional..
# smtp_from: somefrom@address
# smtp_subject: I wanted a custom subject in my alerts...
# smtp_port: 587
# smtp_auth_user: user@where
# smtp_auth_pass: secret
# smtp_tls: 1
# smtp_ssl: 1
# GPU Configurtation
# You'll need one gpuX: section per GPU installed. Not sure what's installed,
# or the order? Open a root shell and type: aticonfig --list-adapters
#
# You can delete gpu sections if you don't have that many cards.. or add more
# if you find a way to run them
gpu0:
# remove disabled: or set it to 0 to actually use this card..
disabled: 0
# overclocking.. all are optional
# optional command to execute prior to overclocking
# any valid command (or the path to a script) is fine
# pre_oc_cmd:
# (core/mem)_X = the value to set for this profile (0-2)
# Profile 2 is used while mining, but setting values
# in profile 2 that are *lower* than the values in the other
# profiles is not allowed on some GPUs, so you have to adjust
# the other profiles as well.
# For instance, to get 300Mhz mem clock on my 5830s, I have
# to set all 3 profiles to 300.
# core_speed_0: 300
# core_speed_1: 800
# core_speed_2: 980
core_speed_2: 900
# mem_speed_0: 300
# mem_speed_1: 300
# mem_speed_2: 300
mem_speed_2: 300
# core_voltage_0: 1.125
# core_voltage_1: 1.125
# core_voltage_2: 1.125000
core_voltage_2: 1.05
# recommend to set and forget fan, preferably at 100
fan_speed: 40
# optional command to execute after overclocking
# post_oc_cmd:
# Phoenix mining setup (see full example for how to use cgminer instead)
# We strongly recommend you use phoenix at, least until you are familiar
# with BAMT
# kernel to use.. one of: phatk phatk2 poclbm
# if in doubt, use phatk2
kernel: phatk2
# phoenix style kernel params. *do not* include DEVICE=x, otherwise
# exactly same as any phoenix setup.
# if in doubt: BFI_INT VECTORS AGGRESSION=11
# (though 9 on GPU0 often makes using the GUI more pleasant)
kernel_params: BFI_INT VECTORS FASTLOOP=false AGGRESSION=9
# Pool file
# IMPORTANT: pool_file must contain a list of pool URLs.. one at least, more if
# you like. You can share the same file for all GPUs, or have one for each.
# Miner will start with first URL in file, however if no shares are accepted
# for any reason in (pool_timeout) seconds, the miner will move to next URL
# in the file. Wraps around to first URL when last one fails.
# You can also set pool specific values, see the example pools file.
# if in doubt: /etc/bamt/pools
pool_file: /etc/bamt/pools
# max time to allow phoenix to run without finding a share..
# exceed this and we kill phoenix, start a new instance with
# next line in pools file
pool_timeout: 180
# monitoring limits, used by email alerts and gpumon
# an alert is generated if the current value goes below a 'lo'
# or above a 'hi'.
# GPU temp sensor reading (in C)
monitor_temp_lo: 45
monitor_temp_hi: 80
# GPU load (in %)
monitor_load_lo: 80
# average hashrate (in mh/s)
monitor_hash_lo: 125
# fam rpm (in rpm)
monitor_fan_lo: 2000
# effective reject rate (in % of total)
monitor_reject_hi: 2
# OK.. hopefully you get the idea. Stubs for additional GPUs below.
# There are many additional paramters to help you deal with various
# scenarios, see /opt/bamt/examples and the docs (if we write some)
gpu1:
disabled: 0
# core_speed_0: 300
# core_speed_1: 800
# core_speed_2: 1060
core_speed_2: 900
# mem_speed_0: 300
# mem_speed_1: 300
# mem_speed_2: 1120
mem_speed_2: 300
# core_voltage_0: 1.125
# core_voltage_1: 1.125
# core_voltage_2: 1.125000
core_voltage_2: 1.05
fan_speed: 30
kernel: phatk2
kernel_params: BFI_INT VECTORS FASTLOOP=false AGGRESSION=11
pool_file: /etc/bamt/pools
pool_timeout: 180
monitor_temp_lo: 45
monitor_temp_hi: 80
monitor_load_lo: 80
monitor_hash_lo: 125
monitor_fan_lo: 2000
monitor_reject_hi: 1
gpu2:
disabled: 0
# core_speed_0: 300
# core_speed_1: 800
# core_speed_2: 980
core_speed_2: 900
# mem_speed_0: 300
# mem_speed_1: 300
# mem_speed_2: 300
mem_speed_2: 300
# core_voltage_0: 1.125
# core_voltage_1: 1.125
# core_voltage_2: 1.125000
core_voltage_2: 1.05
fan_speed: 30
kernel: phatk2
kernel_params: BFI_INT VECTORS FASTLOOP=false AGGRESSION=11
pool_file: /etc/bamt/pools
pool_timeout: 180
monitor_temp_lo: 45
monitor_temp_hi: 80
monitor_load_lo: 80
monitor_hash_lo: 125
monitor_fan_lo: 2000
monitor_reject_hi: 1
gpu3:
disabled: 0
# core_speed_0: 300
# core_speed_1: 800
# core_speed_2: 980
core_speed_2: 900
# mem_speed_0: 300
# mem_speed_1: 300
# mem_speed_2: 300
mem_speed_2: 300
# core_voltage_0: 1.125
# core_voltage_1: 1.125
# core_voltage_2: 1.125000
core_voltage_2: 1.05
fan_speed: 35
kernel: phatk2
kernel_params: BFI_INT VECTORS FASTLOOP=false AGGRESSION=11
pool_file: /etc/bamt/pools
pool_timeout: 180
monitor_temp_lo: 45
monitor_temp_hi: 80
monitor_load_lo: 80
monitor_hash_lo: 125
monitor_fan_lo: 2000
monitor_reject_hi: 1
gpu4:
disabled: 0
# core_speed_0: 300
# core_speed_1: 800
# core_speed_2: 980
core_speed_2: 900
# mem_speed_0: 300
# mem_speed_1: 300
# mem_speed_2: 300
mem_speed_2: 300
# core_voltage_0: 1.125
# core_voltage_1: 1.125
# core_voltage_2: 1.125000
core_voltage_2: 1.05
fan_speed: 40
kernel: phatk2
kernel_params: BFI_INT VECTORS FASTLOOP=false AGGRESSION=11
pool_file: /etc/bamt/pools
pool_timeout: 180
monitor_temp_lo: 45
monitor_temp_hi: 80
monitor_load_lo: 80
monitor_hash_lo: 125
monitor_shares_lo: 1
gpu5:
disabled: 1
# core_speed_0: 300
# core_speed_1: 800
# core_speed_2: 980
core_speed_2: 900
# mem_speed_0: 300
# mem_speed_1: 300
# mem_speed_2: 300
mem_speed_2: 300
# core_voltage_0: 1.125
# core_voltage_1: 1.125
# core_voltage_2: 1.125000
core_voltage_2: 1.05
fan_speed: 50
kernel: phatk2
kernel_params: BFI_INT VECTORS FASTLOOP=false AGGRESSION=11
pool_file: /etc/bamt/pools
pool_timeout: 180
monitor_temp_lo: 45
monitor_temp_hi: 80
monitor_load_lo: 80
monitor_hash_lo: 125
monitor_shares_lo: 1
gpu6:
disabled: 1
# core_speed_0: 300
# core_speed_1: 800
# core_speed_2: 980
# mem_speed_0: 300
# mem_speed_1: 300
# mem_speed_2: 300
# core_voltage_0: 1.125
# core_voltage_1: 1.125
# core_voltage_2: 1.125000
fan_speed: 100
kernel: phatk2
kernel_params: BFI_INT VECTORS FASTLOOP=false AGGRESSION=11
pool_file: /etc/bamt/pools
pool_timeout: 180
monitor_temp_lo: 45
monitor_temp_hi: 80
monitor_load_lo: 80
monitor_hash_lo: 125
monitor_shares_lo: 1
gpu7:
disabled: 1
# core_speed_0: 300
# core_speed_1: 800
# core_speed_2: 980
# mem_speed_0: 300
# mem_speed_1: 300
# mem_speed_2: 300
# core_voltage_0: 1.125
# core_voltage_1: 1.125
# core_voltage_2: 1.125000
fan_speed: 100
kernel: phatk2
kernel_params: BFI_INT VECTORS FASTLOOP=false AGGRESSION=11
pool_file: /etc/bamt/pools
pool_timeout: 180
monitor_temp_lo: 45
monitor_temp_hi: 80
monitor_load_lo: 80
monitor_hash_lo: 125
monitor_shares_lo: 1
Haven't you been having trouble with this for a while? It's little hard to advise w/o reviewing what you have tried so far, but here are some basic suggestions. I got 5x5850 going on BAMT, but it was pretty unstable at first. You have probably already tried this, but if you have made *any* hardware configuration changes, even if they seem trivial, boot from a fresh copy of BAMT. At one point I swapped one 5850 for another, so the two were in different slots, and BAMT completely freaked out about this. I had to start over with a fresh copy. Also, you might try being uber anal about that fresh copy, as in write zeros to the usb drive to completely wipe it and then use dd in linux as opposed to win32diskimager to copy the image. Another thing to try is to get the cards going manually in phoenix - i.e. cd /opt/miners/phoenix - and start each card manually. If you can get all 5 going this way, this will at least tell you that the problem is specific to the process of starting automatically from the configuration script. Also, try starting your GPUs with *REALLY* moderate settings (like 600/500 core/clock at stock voltage). If they work this way, you know you are pushing the clocks of one or more GPUs too much. If they still don't work, lower the voltage - then if they work you know your PSU can't handle 5 cards at stock voltage.
I think BAMT is awesome, but for me it did a lot of really strange things. I know it's supposed to be essentially plug and play, but in my experience it was a bit less friendly than this. I've switched now to cgminer on a flavor of ubuntu 11.04 (Peppermint Two), and I'm liking this better. I definitely recommend cgminer if you can't get BAMT working.
I think BAMT is awesome, but for me it did a lot of really strange things. I know it's supposed to be essentially plug and play, but in my experience it was a bit less friendly than this. I've switched now to cgminer on a flavor of ubuntu 11.04 (Peppermint Two), and I'm liking this better. I definitely recommend cgminer if you can't get BAMT working.
I tried to run 5 5870 with BAMT,but after couple minutes,BAMT crashed(mouse doesn't move,keyboard doesn't work,stop mining,no network),have to restart again,
anybody has any ideas?
anybody has any ideas?
You set your setting up incorrectly.
but it works in the first five minutes,
I tried to run 5 5870 with BAMT,but after couple minutes,BAMT crashed(mouse doesn't move,keyboard doesn't work,stop mining,no network),have to restart again,
anybody has any ideas?
anybody has any ideas?
You set your setting up incorrectly.
I tried to run 5 5870 with BAMT,but after couple minutes,BAMT crashed(mouse doesn't move,keyboard doesn't work,stop mining,no network),have to restart again,
anybody has any ideas?
anybody has any ideas?
Jump to: