I know this is a little off-topic but how many blades can you run on 1 ras pi?
We have been running 4 off one Pi with our firmware with no issues.
Think we maybe able to get to 6 or 8 …. working on it.
Thanks.
Topic: GRIDSEED G-BLADE Overclocking 7Mh/s, improvements and repair - page 31. (Read 74130 times)
We have been running 4 off one Pi with our firmware with no issues.
Think we maybe able to get to 6 or 8 …. working on it.
Thanks.
I know this is a little off-topic but how many blades can you run on 1 ras pi?
We have been running 4 off one Pi with our firmware with no issues.
Think we maybe able to get to 6 or 8 …. working on it.
Yo, they've done infrared tests on the card, just like you wished for
Have a look! http://cryptomining-blog.com/
Wolfey
Nice.
we can see that the 16v capacitor is also hot, but the ferrite beads aren't that bad.
So maybe double power line to feed the board can be done quite easily by duplicating the power board and connecting it to its outputs.
We need some input point to connect it, it would be much more easier if the board had some through hole components, but some places could be easy to connect to.
I might try a proto board in a week or so, I already have most of the components at home from other projects.
what about replacing the mosfets?
the twin ones are IR5300, the 90° flipped one however is a 5301
i made a table on Digikey with ones matchig the package type, sorted down by amperage:
HEXFETs
but someone who knows better should filter it down further by all that stuff like D2S, Vgs, Qg, Ciss etc pp...
the twin ones are IR5300, the 90° flipped one however is a 5301
i made a table on Digikey with ones matchig the package type, sorted down by amperage:
HEXFETs
but someone who knows better should filter it down further by all that stuff like D2S, Vgs, Qg, Ciss etc pp...
Thanks for your help.
if we stick with PQFN package, this one should be fine with higher specs.
http://www.irf.com/product-info/datasheets/data/irfh7004pbf.pdf
@ GenTarkin , I still don't really understand how to make appropriate calculation on HW errors percentage.
here is tonight result
actually, 487 errors on 10 hours @ 7488kh/s with difficulty 2300
My math would be 487/(7488*1000*3600*10)*100 = 0.0018%
can someone please confirm my math is OK?
It should be ok if 1 HW error is actually 1 failed hash over thousands because of HW
It would be different if it was 1 failed good share because of HW, but how can you guess it is a good share if it failed?
Yo, they've done infrared tests on the card, just like you wished for
Have a look! http://cryptomining-blog.com/
Wolfey
what about replacing the mosfets?
the twin ones are IR5300, the 90° flipped one however is a 5301
i made a table on Digikey with ones matchig the package type, sorted down by amperage:
HEXFETs
but someone who knows better should filter it down further by all that stuff like D2S, Vgs, Qg, Ciss etc pp...
the twin ones are IR5300, the 90° flipped one however is a 5301
i made a table on Digikey with ones matchig the package type, sorted down by amperage:
HEXFETs
but someone who knows better should filter it down further by all that stuff like D2S, Vgs, Qg, Ciss etc pp...
Thanks for your help.
if we stick with PQFN package, this one should be fine with higher specs.
http://www.irf.com/product-info/datasheets/data/irfh7004pbf.pdf
@ GenTarkin , I still don't really understand how to make appropriate calculation on HW errors percentage.
here is tonight result
actually, 487 errors on 10 hours @ 7488kh/s with difficulty 2300
My math would be 487/(7488*1000*3600*10)*100 = 0.0018%
can someone please confirm my math is OK?
It should be ok if 1 HW error is actually 1 failed hash over thousands because of HW
It would be different if it was 1 failed good share because of HW, but how can you guess it is a good share if it failed?
what about replacing the mosfets?
the twin ones are IR5300, the 90° flipped one however is a 5301
i made a table on Digikey with ones matchig the package type, sorted down by amperage:
HEXFETs
but someone who knows better should filter it down further by all that stuff like D2S, Vgs, Qg, Ciss etc pp...
the twin ones are IR5300, the 90° flipped one however is a 5301
i made a table on Digikey with ones matchig the package type, sorted down by amperage:
HEXFETs
but someone who knows better should filter it down further by all that stuff like D2S, Vgs, Qg, Ciss etc pp...
Question, the HW shown, is that HW count of diff1 shares or vardiff shares?
last screencap for today.
1 hour @ 7.2Mh/s with 9 HW errors in the first 3 minutes, time for the board to warm up.
Looks stable but needs active cooling on the powerboards
1 hour @ 7.2Mh/s with 9 HW errors in the first 3 minutes, time for the board to warm up.
Looks stable but needs active cooling on the powerboards
Cool! I mean, HOT!
So are you thinking of stopping there?
Do you think all blades will react the same way if standardized? Or is it too early to tell?
I'd love to get rid of the loud fan noise! I hear it's pretty ridiculous. Any chance for a down-volt mod on the fan? Of course, one can always find a quiet-er alternative but I'd rather not blow the money on a silly cooling fan in the name of a most efficient ROI period.
How much extra coin do you think this current mod will run on average?
I think it sucks that Gridseed didn't come out with dedicated Scrypt Asics instead of dual with more than half the chips processing power intentionally going unused! We'd have like 18MHs out of those 80 chips then!
Anyway.....onward!
Wolfey2014
last screencap for today.
1 hour @ 7.2Mh/s with 9 HW errors in the first 3 minutes, time for the board to warm up.
Looks stable but needs active cooling on the powerboards
1 hour @ 7.2Mh/s with 9 HW errors in the first 3 minutes, time for the board to warm up.
Looks stable but needs active cooling on the powerboards
42K resistor gives way less HW errors at 1063Mhz
7.3Mh/s seems to be the limit with the actual components on the power board.
Many components are getting really hot and need active cooling to work.
I'll try some improvements on the power board later, as suggested by wolfey.
For now, I'm happy with the results, but it's really a shame that Gridseed haven't designed a better power board to get the full potential from the chips.
7.3Mh/s seems to be the limit with the actual components on the power board.
Many components are getting really hot and need active cooling to work.
I'll try some improvements on the power board later, as suggested by wolfey.
For now, I'm happy with the results, but it's really a shame that Gridseed haven't designed a better power board to get the full potential from the chips.
Thanks J4!
You are getting great results in spite of the lack of more robust driver components IMO!
But, now we know there is a bit more room for improvement as at 7.3MHs, it's still less than 100KHs per chip.
91.25KHs per chip to be exact. I am pretty sure we can hit 8MHs! Awesome!
My seeds chips are stably running at 100KHs each! So we need to see about opening up the throttles just a tad more and we'll have it!
On with higher rating key components then?
Again, excellent work J4bberwock!
42K resistor gives way less HW errors at 1063Mhz
7.3Mh/s seems to be the limit with the actual components on the powerboard.
Many components are getting really hot and need active cooling to work.
I'll try some improvements on the powerboard later, as suggested by wolfey.
For now, I'm happy with the results, but it's really a shame that Gridseed haven't designed a better powerboard to get the full potential from the chips.
7.3Mh/s seems to be the limit with the actual components on the powerboard.
Many components are getting really hot and need active cooling to work.
I'll try some improvements on the powerboard later, as suggested by wolfey.
For now, I'm happy with the results, but it's really a shame that Gridseed haven't designed a better powerboard to get the full potential from the chips.
I have a question the 7mh-s is after the voltmod right?
Its a little disappointing I got mine tomorrow so I will follow this tread closely
Thanks
Its a little disappointing I got mine tomorrow so I will follow this tread closely
Thanks
Can you guess which one has the 42k resistor and which one still have the 39k?
The schematics for the blade is the 8 chip one from github.
At least, it's really close if it's not the good one.
Unfortunately, it seeems they have kept the same power board for the 40 chips blade that was supposed to be used for 8 chips, so not much room for overclocking.
The schematics for the blade is the 8 chip one from github.
At least, it's really close if it's not the good one.
Unfortunately, it seeems they have kept the same power board for the 40 chips blade that was supposed to be used for 8 chips, so not much room for overclocking.
Yo J4bberwock
I don't understand your question quite clearly enough re: 42k vs 39k ?
Alright, thanks. I have the 8 chip schematic on hand.
Right, I read that about the weak links on the driver portion of the cards. That's why I suggest larger value components which will translate to somewhat larger discreet components. But they can all be made to fit nicely on the card, I am sure.
Then we can really sock it to it! Right? Those figures of 3618 to 3622 are excellent! Do you think that is maximum speed? Or perhaps lowering the clock to something slightly less is already optimum?
Good work!
Can you guess which one has the 42k resistor and which one still have the 39k?
The schematics for the blade is the 8 chip one from github.
At least, it's really close if it's not the good one.
Unfortunately, it seeems they have kept the same power board for the 40 chips blade that was supposed to be used for 8 chips, so not much room for overclocking.
The schematics for the blade is the 8 chip one from github.
At least, it's really close if it's not the good one.
Unfortunately, it seeems they have kept the same power board for the 40 chips blade that was supposed to be used for 8 chips, so not much room for overclocking.
I believe to run more than one blade per raspberry pi you need a powered USB strip and then I think I saw that it could really only handle 5 blade units. aka 10 boards. The question I have though is could I buy small copper heatsinks and possible longer standoffs. Assemble all this and have more overclocking headroom? Or is the volt mod required to hit those higher clocks?
You won't gain anything with better cooling, the fan blowing air over the chips seems to be enough, even when overclocked, they are warm at most.
Voltmod is required for real overclocking.
Hello J4! Nice work!
Have you read the latest over volt report at http://cryptomining-blog.com/ rel volt modding the new gridseed g-blade scrypt asic miner?
They are saying that the voltage regulators overheat easily at even 39k ohms but can be compensated with additional heat sink and fan. Still, the overcurrents will possibly cause them to release magic black smoke!
My first thought is to replace them with TO-220 type with separate fin type heat sinks to help them catch the wind of the main cooling fan, which also sounds like it can be under-volted to make it run quieter since it provide well more than enough air flow to begin with.
Let me know what you think.
Wolfey
Hey Wolfey,
The post wasn't there a few hours ago, but they are facing the exact same limitations as me, but they haven't noticed the ferrite beads overheating. Thermal picture wold be nice to identify where are the hot/weak spots.
I managed to get the blade stable for 25 minutes at 1125Mhz, giving 3800Kh/s for the panel with 47K resistor, but after that, it's overheating, even with a fan blowing directly on it and stops sending shares
It's actually running with very few hw errors at 1063Mhz with 42k resistor.
TO-220 package might be a good idea to replace the 09N03. Datasheet says 25V 50A, but power dissipation might be too low I'll have to check if I have an acceptable replacement when back home next week.
Which components exactly are the FET's or Voltage Regulators on the card. I don't have one so I can't find out for myself.
You can parallel FET's to achieve maximum 'overhead' current capacity. You can increase capacitor value/voltage by adding more caps, like they do on motherboards for cleaner power and good of not excellent quiescent current characteristics.
Darn things! They make them cheaply by cutting corners forcing us to spend more doe and engineering time to make them what they can/should be in the first place! Oh well, that's part of the profit vs modding game. They are helping us too in a way, I guess
Can you provide me with the data sheet on the blades? I'd love to see it and follow along.
You're making progress! Excellent!
I believe to run more than one blade per raspberry pi you need a powered USB strip and then I think I saw that it could really only handle 5 blade units. aka 10 boards. The question I have though is could I buy small copper heatsinks and possible longer standoffs. Assemble all this and have more overclocking headroom? Or is the volt mod required to hit those higher clocks?
You won't gain anything with better cooling, the fan blowing air over the chips seems to be enough, even when overclocked, they are warm at most.
Voltmod is required for real overclocking.
Hello J4! Nice work!
Have you read the latest over volt report at http://cryptomining-blog.com/ rel volt modding the new gridseed g-blade scrypt asic miner?
They are saying that the voltage regulators overheat easily at even 39k ohms but can be compensated with additional heat sink and fan. Still, the overcurrents will possibly cause them to release magic black smoke!
My first thought is to replace them with TO-220 type with separate fin type heat sinks to help them catch the wind of the main cooling fan, which also sounds like it can be under-volted to make it run quieter since it provide well more than enough air flow to begin with.
Let me know what you think.
Wolfey
Hey Wolfey,
The post wasn't there a few hours ago, but they are facing the exact same limitations as me, but they haven't noticed the ferrite beads overheating. Thermal picture wold be nice to identify where are the hot/weak spots.
I managed to get the blade stable for 25 minutes at 1125Mhz, giving 3800Kh/s for the panel with 47K resistor, but after that, it's overheating, even with a fan blowing directly on it and stops sending shares
It's actually running with very few hw errors at 1063Mhz with 42k resistor.
TO-220 package might be a good idea to replace the 09N03. Datasheet says 25V 50A, but power dissipation might be too low I'll have to check if I have an acceptable replacement when back home next week.
I believe to run more than one blade per raspberry pi you need a powered USB strip and then I think I saw that it could really only handle 5 blade units. aka 10 boards. The question I have though is could I buy small copper heatsinks and possible longer standoffs. Assemble all this and have more overclocking headroom? Or is the volt mod required to hit those higher clocks?
You won't gain anything with better cooling, the fan blowing air over the chips seems to be enough, even when overclocked, they are warm at most.
Voltmod is required for real overclocking.
Hello J4! Nice work!
Have you read the latest over volt report at http://cryptomining-blog.com/ rel volt modding the new gridseed g-blade scrypt asic miner?
They are saying that the voltage regulators overheat easily at even 39k ohms but can be compensated with additional heat sink and fan. Still, the overcurrents will possibly cause them to release magic black smoke!
My first thought is to replace them with TO-220 type with separate fin type heat sinks to help them catch the wind of the main cooling fan, which also sounds like it can be under-volted to make it run quieter since it provide well more than enough air flow to begin with.
Let me know what you think.
Wolfey
So what I'm hearing is that these chips are nowhere near their maximum speed, and if the blade was engineered just slightly better they could be pushed way faster? Well dam. Makes me wanna try and mod the heck out of the power board..
I believe to run more than one blade per raspberry pi you need a powered USB strip and then I think I saw that it could really only handle 5 blade units. aka 10 boards. The question I have though is could I buy small copper heatsinks and possible longer standoffs. Assemble all this and have more overclocking headroom? Or is the volt mod required to hit those higher clocks?
You won't gain anything with better cooling, the fan blowing air over the chips seems to be enough, even when overclocked, they are warm at most.
Voltmod is required for real overclocking.
I believe to run more than one blade per raspberry pi you need a powered USB strip and then I think I saw that it could really only handle 5 blade units. aka 10 boards. The question I have though is could I buy small copper heatsinks and possible longer standoffs. Assemble all this and have more overclocking headroom? Or is the volt mod required to hit those higher clocks?
Jump to: