Topic: BFL Labs 60 GH/s Single Tweaking Help for Monkey-brained Stupid People (Read 7418 times)
September 15, 2013, 02:45:09 AM
good luck babi
September 12, 2013, 03:09:49 PM
watching
August 13, 2013, 06:42:13 AM
hope ck releases a single firmware and maybe updates jalapeno to 1.2.6
July 22, 2013, 11:47:07 PM
Agreed! I might give it a try - thanks for experimenting and releasing your firmware to the public though!
Yeah thanks. I figure there's got to be some other ways to bump up performance. I just know very little about the processing architecture or these hash calculations. If anyone knows anything about the firmware or wants to find a place to experiment:
in JOBPIPE_MODULE.h
#define PIPE_MAX_BUFFER_DEPTH 40
#define PIPE_JOB_BUFFER_OK 0
#define PIPE_JOB_BUFFER_FULL 1
#define PIPE_JOB_BUFFER_EMPTY 2
#define MAX_RESULTS_TO_SEND_AT_A_TIME_FROM_BUFFER 16
Any reason why it has to be 40 for buffer depth? What about 16 at a time?
July 22, 2013, 10:46:09 PM
Correct me if I am wrong, but the HW error rate of 3.2% means your effective hashrate is actually 60GH/s, not 62GH/s? So the increase is truly only 2GH/s?
I had an error rate of around 0.4% to begin with. I'll also admit I wasn't keeping meticulous records of my pre-modded hash rates. I suppose it may be "only" 2GH/s, but such a mod will cost you a lot less than buying USB ASICs, though with the exponential difficulty 2GH/s isn't going to mean that much soon. It's certainly better than a kick in the pants that's for sure.
July 22, 2013, 10:28:01 PM
Correct me if I am wrong, but the HW error rate of 3.2% means your effective hashrate is actually 60GH/s, not 62GH/s? So the increase is truly only 2GH/s?
I had an error rate of around 0.4% to begin with. I'll also admit I wasn't keeping meticulous records of my pre-modded hash rates. I suppose it may be "only" 2GH/s, but such a mod will cost you a lot less than buying USB ASICs, though with the exponential difficulty 2GH/s isn't going to mean that much soon. It's certainly better than a kick in the pants that's for sure.
July 22, 2013, 07:12:43 PM
Ok cool. So I set the frequency to 9 in std-def.h, disabled ""#define __RUN_HEAVY_DIAGNOSTICS_ON_EACH_ENGINE" and changed "#define __TOTAL_DIAGNOSTICS_RUN 10" down to 1 in std-def.h. (Thanks, danattacker)
Also took off the case and froze the unit for about 30 minutes.
Used to be averaging about 58 GH/s...now:
5s:62.57 avg:62.02 u:63.48 Gh/s | A:114 R:20+0(.40%) HW:127/3.2%
HW error rate is a little high, but it's stable. So ~5-10% gain depending on the average, which isn't anything to sneeze at.
So I've uploaded the compiled Singles firmware .elf here, with the source code with my edits should you wish to tweak it further:
https://www.dropbox.com/sh/2v6151n50g4hb0s/CdGbzmnDiV
I would expect this should work for all non-little singles/jalapeno units. I've also included an untested version that should work with the minirigs as well.
I also have included two pictures of the board with the jtag connectors visible, to the left of the power ports. There's no need to disassemble the fans or heat sinks.
If you use this and it works for you, a donation of one day's worth of your increased earnings (~.05-0.1 BTC at current rates for a singles, ~5-10BTC for a minirig) is suggested and appreciated.
Donations: 1QGGpaa8YX1eC2Kp99yxCPfrMoRuMUoudD
Correct me if I am wrong, but the HW error rate of 3.2% means your effective hashrate is actually 60GH/s, not 62GH/s? So the increase is truly only 2GH/s? Also took off the case and froze the unit for about 30 minutes.
Used to be averaging about 58 GH/s...now:
5s:62.57 avg:62.02 u:63.48 Gh/s | A:114 R:20+0(.40%) HW:127/3.2%
HW error rate is a little high, but it's stable. So ~5-10% gain depending on the average, which isn't anything to sneeze at.
So I've uploaded the compiled Singles firmware .elf here, with the source code with my edits should you wish to tweak it further:
https://www.dropbox.com/sh/2v6151n50g4hb0s/CdGbzmnDiV
I would expect this should work for all non-little singles/jalapeno units. I've also included an untested version that should work with the minirigs as well.
I also have included two pictures of the board with the jtag connectors visible, to the left of the power ports. There's no need to disassemble the fans or heat sinks.
If you use this and it works for you, a donation of one day's worth of your increased earnings (~.05-0.1 BTC at current rates for a singles, ~5-10BTC for a minirig) is suggested and appreciated.
Donations: 1QGGpaa8YX1eC2Kp99yxCPfrMoRuMUoudD
July 22, 2013, 11:57:32 AM
Quote
The second trick is to put the Single in a freezer for a half hour before you power it on. Don't run it in the freezer, but just get the chips nice and freezing cold before you turn it on.
That sounds like a great way to get condescation and/or liquid into your box :p
I double bagged my unit before putting it in the freezer. No noticeable condensation on the unit after 30 minutes or so.
July 22, 2013, 11:26:16 AM
Quote
The second trick is to put the Single in a freezer for a half hour before you power it on. Don't run it in the freezer, but just get the chips nice and freezing cold before you turn it on.
That sounds like a great way to get condescation and/or liquid into your box :p
July 22, 2013, 05:46:44 AM
For this performance you need C or less. Unless there is a different performance of chips in a Single
July 21, 2013, 05:43:24 PM
The most you would get out of a Single with 16 Grade A chips would be 64Gh
You have probably 16 Grade B chips (give or take)
You have probably 16 Grade B chips (give or take)
July 21, 2013, 12:21:42 PM
Ok cool. So I set the frequency to 9 in std-def.h, disabled ""#define __RUN_HEAVY_DIAGNOSTICS_ON_EACH_ENGINE" and changed "#define __TOTAL_DIAGNOSTICS_RUN 10" down to 1 in std-def.h. (Thanks, danattacker)
Also took off the case and froze the unit for about 30 minutes.
Used to be averaging about 58 GH/s...now:
5s:62.57 avg:62.02 u:63.48 Gh/s | A:114 R:20+0(.40%) HW:127/3.2%
HW error rate is a little high, but it's stable. So ~5-10% gain depending on the average, which isn't anything to sneeze at.
So I've uploaded the compiled Singles firmware .elf here, with the source code with my edits should you wish to tweak it further:
https://www.dropbox.com/sh/2v6151n50g4hb0s/CdGbzmnDiV
I would expect this should work for all non-little singles/jalapeno units. I've also included an untested version that should work with the minirigs as well.
I also have included two pictures of the board with the jtag connectors visible, to the left of the power ports. There's no need to disassemble the fans or heat sinks.
If you use this and it works for you, a donation of one day's worth of your increased earnings (~.05-0.1 BTC at current rates for a singles, ~5-10BTC for a minirig) is suggested and appreciated.
Donations: 1QGGpaa8YX1eC2Kp99yxCPfrMoRuMUoudD
Also took off the case and froze the unit for about 30 minutes.
Used to be averaging about 58 GH/s...now:
5s:62.57 avg:62.02 u:63.48 Gh/s | A:114 R:20+0(.40%) HW:127/3.2%
HW error rate is a little high, but it's stable. So ~5-10% gain depending on the average, which isn't anything to sneeze at.
So I've uploaded the compiled Singles firmware .elf here, with the source code with my edits should you wish to tweak it further:
https://www.dropbox.com/sh/2v6151n50g4hb0s/CdGbzmnDiV
I would expect this should work for all non-little singles/jalapeno units. I've also included an untested version that should work with the minirigs as well.
I also have included two pictures of the board with the jtag connectors visible, to the left of the power ports. There's no need to disassemble the fans or heat sinks.
If you use this and it works for you, a donation of one day's worth of your increased earnings (~.05-0.1 BTC at current rates for a singles, ~5-10BTC for a minirig) is suggested and appreciated.
Donations: 1QGGpaa8YX1eC2Kp99yxCPfrMoRuMUoudD
July 21, 2013, 12:56:09 AM
Isn't firmware realised by BFL only for 8 chips not 16? If it is it will not work...
Different PSU helps because unit do a self diagnostic at start-up. Finds safe clock speeds and working engine. More stable voltage helps...
EDIT: Stop reading before the end to give you this warning but I see I was too late... But now it looks like you fixed it to 16 chips...
Any chance you would took some detail pictures of the board... We are making BFL boards for chips and we would like to know ASAP what was the problem with shortboard used at first and still used in jalapenos.
Different PSU helps because unit do a self diagnostic at start-up. Finds safe clock speeds and working engine. More stable voltage helps...
EDIT: Stop reading before the end to give you this warning but I see I was too late... But now it looks like you fixed it to 16 chips...
Any chance you would took some detail pictures of the board... We are making BFL boards for chips and we would like to know ASAP what was the problem with shortboard used at first and still used in jalapenos.
July 20, 2013, 05:16:56 PM
Ok, thanks!
It's extremely easy to get at the JTAG port-unplug everything, undo four screws on each side, remove the grates over the fans, then push the bottom portion out of the case a few inches to the left with the power ports facing you. No need to disassemble heat sinks or anything like that.
I tried flashing the 1.2.5 firmware, only to find my performance had been halved to 30 GH/s (eep!). I went through the code and realized that the firmware available for download, though for all models, has already pre-defined the Jalapeno as the product and it consequently was doing things like reporting only half as many engines and throttling down the fans. A quick change to std_defs.h such that "__PRODUCT_MODEL_SINGLE__" was defined and "Jalapeno" was commented fixed that. I'll definitely take a look at those firmware mods and do some tweaking when I have the time.
It's extremely easy to get at the JTAG port-unplug everything, undo four screws on each side, remove the grates over the fans, then push the bottom portion out of the case a few inches to the left with the power ports facing you. No need to disassemble heat sinks or anything like that.
I tried flashing the 1.2.5 firmware, only to find my performance had been halved to 30 GH/s (eep!). I went through the code and realized that the firmware available for download, though for all models, has already pre-defined the Jalapeno as the product and it consequently was doing things like reporting only half as many engines and throttling down the fans. A quick change to std_defs.h such that "__PRODUCT_MODEL_SINGLE__" was defined and "Jalapeno" was commented fixed that. I'll definitely take a look at those firmware mods and do some tweaking when I have the time.
July 20, 2013, 03:57:55 PM
When you put it in the freezer, doesnt it condensate after being brought back out? I don't think id recommend that
July 20, 2013, 03:53:21 PM
They run the SC Single firmware. They should ship with version 1.2.5, which is the newest.
The only tweaking I've seen were done on Jalapenos, where some tricks transfer over to the SC Single, and some don't. For the Jalapeno, there have been 2 big ways to increase hashrate: Flash a newer firmware, and then put the Single in a freezer before you start it up.
The BFL ASICs do a sort of self-diagnostic every time you power it on. It's got 16 chips, and each ship has 16 cores. Most chips only enable 14-15 of it's cores, and some might work at higher speeds than others. To increase your hashrate, you want to first of all increase the number of cores per chip you're running, and then increase the hashrate they're running at. You might get more HW errors with both of these improvements, but the overall hashrate will more than compensate for that.
For Jalapenos, they come with Firmware version 1.0.0, which limits you to 5GH/s. Flashing them to 1.2.5 gives some substantial performance boosts, enabling all available chips/cores at a safe frequency. The Singles come with a stock version 1.2.5, but there's still some room for improvement. You could try flashing Ckolivas' modified 1.2.5, which should be a little better at enabling questionable cores and/or running them at higher speeds. I don't know of anyone who's done this on a Single yet, so who knows?! You might brick your device, I'm not sure.
The second trick is to put the Single in a freezer for a half hour before you power it on. Don't run it in the freezer, but just get the chips nice and freezing cold before you turn it on. When it does that self-diagnostic test, you might get better results if the chips are 0C rather than 25C. Of course these lower temps won't last long once it start mining, but it will be mining at a higher clock than it was, and it will give you more hashes. Again, you'll prolly get more HW errors, but it's probably worth it.
I sold my Single, and the guy said he went from 57-62GH/s just by using a different PSU. I'm not sure why that is, as it was getting the same 57GH/s with the BFL PSU, my Seasonic X-650, and his Antec 900W one, and finally when using it with a CX430, it went to 62GH/s. No clue why, except for many the self-test decided to play nice and give him good results.
TL:DR You could try flashing a modified firmware, but that's quite risky. For low-risk, put it in the freezer for an hour before you power it up and mine with it.
The only tweaking I've seen were done on Jalapenos, where some tricks transfer over to the SC Single, and some don't. For the Jalapeno, there have been 2 big ways to increase hashrate: Flash a newer firmware, and then put the Single in a freezer before you start it up.
The BFL ASICs do a sort of self-diagnostic every time you power it on. It's got 16 chips, and each ship has 16 cores. Most chips only enable 14-15 of it's cores, and some might work at higher speeds than others. To increase your hashrate, you want to first of all increase the number of cores per chip you're running, and then increase the hashrate they're running at. You might get more HW errors with both of these improvements, but the overall hashrate will more than compensate for that.
For Jalapenos, they come with Firmware version 1.0.0, which limits you to 5GH/s. Flashing them to 1.2.5 gives some substantial performance boosts, enabling all available chips/cores at a safe frequency. The Singles come with a stock version 1.2.5, but there's still some room for improvement. You could try flashing Ckolivas' modified 1.2.5, which should be a little better at enabling questionable cores and/or running them at higher speeds. I don't know of anyone who's done this on a Single yet, so who knows?! You might brick your device, I'm not sure.
The second trick is to put the Single in a freezer for a half hour before you power it on. Don't run it in the freezer, but just get the chips nice and freezing cold before you turn it on. When it does that self-diagnostic test, you might get better results if the chips are 0C rather than 25C. Of course these lower temps won't last long once it start mining, but it will be mining at a higher clock than it was, and it will give you more hashes. Again, you'll prolly get more HW errors, but it's probably worth it.
I sold my Single, and the guy said he went from 57-62GH/s just by using a different PSU. I'm not sure why that is, as it was getting the same 57GH/s with the BFL PSU, my Seasonic X-650, and his Antec 900W one, and finally when using it with a CX430, it went to 62GH/s. No clue why, except for many the self-test decided to play nice and give him good results.
TL:DR You could try flashing a modified firmware, but that's quite risky. For low-risk, put it in the freezer for an hour before you power it up and mine with it.
July 18, 2013, 04:59:18 PM
No. At the time you can't register at BFL forum.. You'll get an activation link via E-Mail and then you have a "Administrator error".
Good luck with tweaking your SC!
Good luck with tweaking your SC!
July 18, 2013, 12:50:43 PM
I'm trying to tweak my BFL Labs 60 GH/s ASIC single and am having a god-awful time getting any documentation or facts straight on it. I am a monkey-brained stupid person and need your help. For whatever reason, I cannot register on Butterfly Labs forums.
PRESENT: Running at an average 59 GH/s over an often-on AC vent at ~50-54 degrees C (61-63 C with the AC off).
I feel like this is probably a pretty decent chip considering the range I've seen other users reporting and feel like it could get at least a 10% more through some sort of overclocking.
Questions:
Does this ASIC 60 GH/s single use the "singles" firmware listed on www.butterflylabs.com/drivers or the BitForce SC firmware https://forums.butterflylabs.com/announcements/3282-bitforce-sc-firmware-version-1-2-5-a.html or something else?
What firmware version did these 60 GH/s units ship with?
When I run Easyminer v1.1.20130703.02.40, I cannot autotune the unit. Right clicking in Easyminer to autotune here per the instructions here https://bitcointalk.org/index.php?topic=90426.0;all does not show a dropdown menu. Is the Autotune only for other units? My unit reports the incorrect GH/s to Easyminer- does it need a firmware upgrade to function?
Can these Single's firmware be upgraded like the Jalapenos with an Atmel AVR Dragon? Will this produce a similar 30-50% increase in performance as seen with the Jalapenos?
Is there any other method of improving performance I should try?
PRESENT: Running at an average 59 GH/s over an often-on AC vent at ~50-54 degrees C (61-63 C with the AC off).
I feel like this is probably a pretty decent chip considering the range I've seen other users reporting and feel like it could get at least a 10% more through some sort of overclocking.
Questions:
Does this ASIC 60 GH/s single use the "singles" firmware listed on www.butterflylabs.com/drivers or the BitForce SC firmware https://forums.butterflylabs.com/announcements/3282-bitforce-sc-firmware-version-1-2-5-a.html or something else?
What firmware version did these 60 GH/s units ship with?
When I run Easyminer v1.1.20130703.02.40, I cannot autotune the unit. Right clicking in Easyminer to autotune here per the instructions here https://bitcointalk.org/index.php?topic=90426.0;all does not show a dropdown menu. Is the Autotune only for other units? My unit reports the incorrect GH/s to Easyminer- does it need a firmware upgrade to function?
Can these Single's firmware be upgraded like the Jalapenos with an Atmel AVR Dragon? Will this produce a similar 30-50% increase in performance as seen with the Jalapenos?
Is there any other method of improving performance I should try?
Jump to: