The Chili – 30+GH/s BFL based Bitcoin Miner Assembly

[joeventura]

I have a wide range of hardware error rates.. I am wondering if they are related to cooling, luck or random chance...  So before I go tearing down my cards and reseating them how do these numbers sound?

1 card at 3.5% error rate, 7 cards around 8% error rate 2 cards at 10% (and change).

So should I be worried with 10% error rates or is this one of those random chance things?

Thanks in advance.

No you are good!!

I have one with a 14% error rate and one with a 7%

[1714742690]

1714742690

[1714742690]

1714742690

[1714742690]

1714742690

[daemonfox]

I have a wide range of hardware error rates.. I am wondering if they are related to cooling, luck or random chance...  So before I go tearing down my cards and reseating them how do these numbers sound?

1 card at 3.5% error rate, 7 cards around 8% error rate 2 cards at 10% (and change).

So should I be worried with 10% error rates or is this one of those random chance things?

Thanks in advance.

It is definitely a factor of chip cooling and airflow over the mosfets.

I started with an Accelero II mounted in line like you would on a GPU and the heatsink fins were over the mosfets... I averaged about 8% errors.

I rotated it 90 degrees and an 80mm fan right on top of the mosfets and gained a couple GH/s and down to 5% errors.

I have now moved on to an H60 Corsair water cooler and the 80mm on the mosfets but am now back at 8% errors (i believe due to the fail hardware included in the refurb box that wouldn't mount properly...) I am getting some long bolts and nuts to use a much better backplate tomorrow and expect the change to fix my issue and increase hashes.

UPDATE: Bolts, washers and nuts used to reseat and mount the water block... 33 GH/s and 4.4% HW errors.

[daemonfox]

Question about the H60 from Corsair... pump speed can be adjusted through the BIOS or firmware or in Windows when this is plugged into a motherboard for an actual CPU.

Is there something I can add to my batch files that will tell the 3 pin header on the Chili to lower the power so it lowers the pump rate? Others have mentioned this helps increase the hashrate due to the fluid picking up more heat and spending more time in the radiator to dissipate it.

[lightfoot]

Question about the H60 from Corsair... pump speed can be adjusted through the BIOS or firmware or in Windows when this is plugged into a motherboard for an actual CPU.

Is there something I can add to my batch files that will tell the 3 pin header on the Chili to lower the power so it lowers the pump rate? Others have mentioned this helps increase the hashrate due to the fluid picking up more heat and spending more time in the radiator to dissipate it.

I do not know about the CPU speed ability, to be honest I cheated. My Noctura 92 fan came with a wonderful assortment of cables including one with an in-line resistor to slow the fan down. Works perfectly on the pump line, I use the splitter cable to power the fan at full speed, and the pump at reduced speed.

If you want to buy a resistor, you just need to use the r=e/i answer to find the right size. The question is how many watts does the pump pull? Then watts=i*e (current*voltage), volts is 12, solve for current, then put 2/current in the top equation to get the resistor size in ohms. (if you want to drop the voltage by 2 volts or 20ish%)

C

[Hands]

Deamonfox are you running thermal pad or thermal grease?

[daemonfox]

Deamonfox are you running thermal pad or thermal grease?

Both... pads on chips AND the compound that was on the block when it arrived. Earlier in this thread others had noted using the TIM with grease on top helped make an even better contact since some 1st gen boards had chips that were not level.

[MrTeal]

I can partially confirm what PhDminer was talking about earlier with the low temperatures. I took a unit offline to do a firmware update and after letting it sit for a couple minutes I had a hell of a time restarting it. It is a good performing (39GH/s @ 1.1V) unit, but it kept on restarting while it was powering up as it never got time to warm up. I ended up getting it started by removing the fan blowing cold air on it (it's -29C outside right now) and letting it get started and warmed up. I'll have to look into it deeper in the next couple days why it's doing this.

[MrTeal]

Question about the H60 from Corsair... pump speed can be adjusted through the BIOS or firmware or in Windows when this is plugged into a motherboard for an actual CPU.

Is there something I can add to my batch files that will tell the 3 pin header on the Chili to lower the power so it lowers the pump rate? Others have mentioned this helps increase the hashrate due to the fluid picking up more heat and spending more time in the radiator to dissipate it.

I'm not sure on the H60, but my Water 2.0 Extreme has a USB plug that you plug into one of the USB headers on a motherboard, and you can control the fan speed separate of the pump speed. Is the H60 like that? If so, the Chili can't control it. All it does is PWM the fan, with a 5V minimum so the tach operates properly. I don't believe the cgminer fan control will work though, the fan is just controlled based on temperature at this point.

[daemonfox]

Question about the H60 from Corsair... pump speed can be adjusted through the BIOS or firmware or in Windows when this is plugged into a motherboard for an actual CPU.

Is there something I can add to my batch files that will tell the 3 pin header on the Chili to lower the power so it lowers the pump rate? Others have mentioned this helps increase the hashrate due to the fluid picking up more heat and spending more time in the radiator to dissipate it.

I'm not sure on the H60, but my Water 2.0 Extreme has a USB plug that you plug into one of the USB headers on a motherboard, and you can control the fan speed separate of the pump speed. Is the H60 like that? If so, the Chili can't control it. All it does is PWM the fan, with a 5V minimum so the tach operates properly. I don't believe the cgminer fan control will work though, the fan is just controlled based on temperature at this point.

The fan on the radiator has its own three pin (albeit that it is a 4 pin head it has the guides and works fine on three pins) that is plugged into the first set of pins on the Chili (the set closer to the USB port) and the pump is plugged into the three pin that is closest to the power input. What I read on the Corsair forums was that in BIOS it sees it as any other device on the fan three pin... a fan that it can scale down.

So, if you are saying the that 3 pin fan header is only controlled by what temp is read... there is no way to just set it at a stable target?

[Keefe]

I might have just bricked my Chili. The flash utility crashed mid-flash, and now no lights come on after a power cycle, except the power light next to the socket. The flash utility no longer displays stats and it hangs as soon as I attempt to flash again. What can I do to diagnose this?

I checked the 3.3v and 5v test points and get correct results. The chip core voltage regulators are putting out 0v but I guess those don't come on until the firmware loads?

WinXP does recognize the Chili's presence just like before, so I guess the FTDI chip is ok? But with corrupt firmware, the ARM can't accept a flash over USB? Do I need to use JTAG now?

[MrTeal]

I might have just bricked my Chili. The flash utility crashed mid-flash, and now no lights come on after a power cycle, except the power light next to the socket. The flash utility no longer displays stats and it hangs as soon as I attempt to flash again. What can I do to diagnose this?

I checked the 3.3v and 5v test points and get correct results. The chip core voltage regulators are putting out 0v but I guess those don't come on until the firmware loads?

WinXP does recognize the Chili's presence just like before, so I guess the FTDI chip is ok? But with corrupt firmware, the ARM can't accept a flash over USB? Do I need to use JTAG now?

It's possible. The chip doesn't change the startup vector between the banks of flash until the very end so there is a very short window where you could actually brick the unit, so it should be very resilient to bricking. I did a few dozen trials of pulling the USB mid-flash, pulling the power to the board, and killing the ChiliFlash process but I was never able to get one to brick.
It does sound like you were able to though. Fire me off an email, and if you don't have a JTAG programmer we can probably reflash it for you.

[MrTeal]

Question about the H60 from Corsair... pump speed can be adjusted through the BIOS or firmware or in Windows when this is plugged into a motherboard for an actual CPU.

Is there something I can add to my batch files that will tell the 3 pin header on the Chili to lower the power so it lowers the pump rate? Others have mentioned this helps increase the hashrate due to the fluid picking up more heat and spending more time in the radiator to dissipate it.

I'm not sure on the H60, but my Water 2.0 Extreme has a USB plug that you plug into one of the USB headers on a motherboard, and you can control the fan speed separate of the pump speed. Is the H60 like that? If so, the Chili can't control it. All it does is PWM the fan, with a 5V minimum so the tach operates properly. I don't believe the cgminer fan control will work though, the fan is just controlled based on temperature at this point.

The fan on the radiator has its own three pin (albeit that it is a 4 pin head it has the guides and works fine on three pins) that is plugged into the first set of pins on the Chili (the set closer to the USB port) and the pump is plugged into the three pin that is closest to the power input. What I read on the Corsair forums was that in BIOS it sees it as any other device on the fan three pin... a fan that it can scale down.

So, if you are saying the that 3 pin fan header is only controlled by what temp is read... there is no way to just set it at a stable target?

Not right now, no.

[PhDMiner]

I can partially confirm what PhDminer was talking about earlier with the low temperatures. I took a unit offline to do a firmware update and after letting it sit for a couple minutes I had a hell of a time restarting it. It is a good performing (39GH/s @ 1.1V) unit, but it kept on restarting while it was powering up as it never got time to warm up. I ended up getting it started by removing the fan blowing cold air on it (it's -29C outside right now) and letting it get started and warmed up. I'll have to look into it deeper in the next couple days why it's doing this.

Thanks, Man!!

[lightfoot]

Side question: What is the theoretical *maximum* a Chili can go if fully cooled? See I have a water block on this unit and it's hashing 39/37. But I could dunk the radiator into a 50 gallon ran barrel outside and suspend the Chili over it to bring the water temps down to 32 degrees. Can the chips go faster, or is there a theoretical maximum (the fact that the BFL chips are clocking at 5gh each *now* is without a doubt creepy.)

[MrTeal]

Side question: What is the theoretical *maximum* a Chili can go if fully cooled? See I have a water block on this unit and it's hashing 39/37. But I could dunk the radiator into a 50 gallon ran barrel outside and suspend the Chili over it to bring the water temps down to 32 degrees. Can the chips go faster, or is there a theoretical maximum (the fact that the BFL chips are clocking at 5gh each *now* is without a doubt creepy.)

There's really not that much more room on them. The hashrate just seems to hit a wall around 5GH/s per chip.

[i3luefire]

Side question: What is the theoretical *maximum* a Chili can go if fully cooled? See I have a water block on this unit and it's hashing 39/37. But I could dunk the radiator into a 50 gallon ran barrel outside and suspend the Chili over it to bring the water temps down to 32 degrees. Can the chips go faster, or is there a theoretical maximum (the fact that the BFL chips are clocking at 5gh each *now* is without a doubt creepy.)

it seems like to me that the limit is 39ish GH/s because of the current limit on the power supply FETs.
 and

The hashrate just seems to hit a wall around 5GH/s per chip.

[xselam1988]

A double Chili please...

how do you do this?
where can I buy these?

[MrTeal]

Side question: What is the theoretical *maximum* a Chili can go if fully cooled? See I have a water block on this unit and it's hashing 39/37. But I could dunk the radiator into a 50 gallon ran barrel outside and suspend the Chili over it to bring the water temps down to 32 degrees. Can the chips go faster, or is there a theoretical maximum (the fact that the BFL chips are clocking at 5gh each *now* is without a doubt creepy.)

it seems like to me that the limit is 39ish GH/s because of the current limit on the power supply FETs.

Not really. I have a 4 chip board that I play with that is similarly limited, even if I only run 2 chips on it. It's more a function of the BFL chips being broken, the highest clock divisor doesn't work so you reach a maximum frequency setpoint and then can only vary the frequency using the voltage. There's a limit to how far you can push the voltage though.

[i3luefire]

Side question: What is the theoretical *maximum* a Chili can go if fully cooled? See I have a water block on this unit and it's hashing 39/37. But I could dunk the radiator into a 50 gallon ran barrel outside and suspend the Chili over it to bring the water temps down to 32 degrees. Can the chips go faster, or is there a theoretical maximum (the fact that the BFL chips are clocking at 5gh each *now* is without a doubt creepy.)

it seems like to me that the limit is 39ish GH/s because of the current limit on the power supply FETs.

Not really. I have a 4 chip board that I play with that is similarly limited, even if I only run 2 chips on it. It's more a function of the BFL chips being broken, the highest clock divisor doesn't work so you reach a maximum frequency setpoint and then can only vary the frequency using the voltage. There's a limit to how far you can push the voltage though.

So then we are meeting both limits at the same time on an 8 perfect chip board. Which means the boards were designed with minimal wasted overhead. Which is  why you were able to make these things for such a badass price and still max out the BFL chips. You guys really nailed it. I am very impressed by your work. The only thing I feel like we are missing is the z-link feature. It would have made my setup a little neater. But all in all these are excellent boards with an impressive FW capable of quickly auto-tuning for max speed.

Hope to see z-link soon.

[Hands]

I've got like 3 or 4 cards that all of the sudden are showing 250GH/100% hardware errors?

I have them off now and will let them cool completely down before turning them back on.. But reading on here that some people solved this with firmware reflash's.

Do these cards get in situations where they need to be re-flashed? Or can a reboot always get them "back"

I am currently running the 14e flash that has the 1.1v limit.