Topic: Block Erupter USB - Overclocking/ hacking ? - page 24. (Read 168756 times)

Correct If you see the circuit you can see where they had initially cleaned the power up. If you do it just a little more and get the power stable enough the amount of HW errors will go drastically down as mine have.

It would be interesting to examine each ends of the communication protocol, I think bfgminer/cgminer are using the icarus protocol http://en.qi-hardware.com/wiki/Icarus#Communication_protocol_V3

The main reason I would like to know is because I would like to know how error handling of transmissions over rs232 is corrected. It would be embarrassing to find that the hardware errors were nothing to do with the asic, but actually due to data transmission errors.

I've noticed a stock erupter plugged directly into usb port gets about 1/2 the HW errors than when plugged in via extension cable.

Edit: ...And BTW I realize that noise over the usb transmission lines would not be a problem, rather noise over the usb power lines would effect the serial between the CP2102 (USB 2 UART) and the T2313A Microcontroller

Be aware factory spec of 1.8v is the breakdown voltage of die not operating voltage.


A bulk power cap is only necessary if your supplying power over thin cables and long distances. I have no such problem with the way I have been testing it, I have a stock Block Erupter that I run at the same time for comparison and hardware rates are comparable. I would not recommend supplying power over the usb cables especially cheap ones as you will have a higher voltage drop (and more noise).

The TEC works great the issue I ran into was after ~5min water would condense on parts of the board and cause errors and eventually lockup and timeouts. I placed the board directly on the TEC with only Arctic Silver between them as the outside of the TEC is ceramic there is no chance of shorting. The asic is cooled well in this setup however the buck converter did get warm. I found some pin compatible 5amp regulators made by Alpha & Omega that have a ground pad on the bottom that may be able to help with cooling however they are not easy to get. I post more pictures and info later in the week as I get my orders in. Everyone going off the theoretical maximums I mentioned needs to read the original thread by Fried Cat. These were early in the design. Personally from what I've seen I believe these will choke at about ~750. I don't think these will be able to dissipate the heat generated @ the voltages required. I would like to see some pictures of people running these overclocked so we can get a baseline of voltages/error rates and frequency's to be expected. Also post your cooling setups and temps.

Clean power certainly will help on your quest but I would recommend TEC cooling if you're gonna be approaching 1GHz+ speeds, smear it up with petroleum jelly and insulate it good, get that silicon as cold as you can for top clocks

Watch

Nice Thread

Yes to the parts. I suppose I will tell you. To reduce my HW errors I used a 10 uF capacitor on the power. Be sure the power is your problem by running an oscilloscope through it. I have a massive heatsink that will be going onto the board before it is fit into the box along with a fan.  To reduce a little more noise you can use something like silicon to cover the sides of the oscillator to keep any outside static from interfering with the cycles. When I have time I will post videos on my progress with the 1.4 Gh/s mod.

50MHz would give 1.4Ghash, that would be quite impressive.
Although, on 2.5V the chip would be on the edge of burning, i guess.
Where do you leave the heat? I guess that's far more than 10W, likely 14 or so... Liquid Nitrogen?

+1 on the Power.
I guess you replace the coil and caps with some better parts building a pi and use the chip SW-Pins (AZ1021 removed) as input to the stick?

if you don't tell, don't tease

Right now I am working on getting these up to their theoretical 1 GH/s
The hardware errors are caused by bad power. When the power is not clean (Has static) (which it is not if you look at it with an oscilloscope you will see that it has a lot of static) it will give you more hardware errors. That is why a longer cable produces more errors. So what is the best way to "clean" our power? Well I would tell you but then that would give away my secret. I am throwing a 50 Mhz oscillator on the board along with removing the resistors and replacing them with something a little lower. Factory specifications say max of 1.8V if I am correct however I am going up to 2.5V

All of which will be enclosed in its own case with a usb cable sticking out. Oh yeah you can't forget about the external power supply that will be required.

I just glued the usb sticks on a old and cheap P4 heat sink with ordinary double sided tape.

room temperature 20°C (68 °F)
heat sink 30°C (86 °F)
chip 45°C (113 °F) - everything fine!


Pictures: http://imgur.com/a/3QSWg#0

I've ordered some 15A external buck converters - I'm going to try and remove the coil off the board which will disconnect the onboard power but leave the filtering caps in place. This will give me better control of the voltage.  After hopefully getting 16mhz stable I'll try a 20mhz crystal.  Also got some big coolers coming.

I think it was haxtormoogle with the variable oscillator?

I was hoping that Bluestreak66 would test it since if I understood it correctly he has a test system with easy to change settings running...

You'd need to find a crystal that drives the hash chip at that rate.  12mhz = 333MH/s, 16mhz = 448MH/s - if you can find a compatible crystal in between then a lower voltage may be stable.

If I would like to go safer. 400 and 1,2V will that work out of USB and what do I need to do?

Bluestreak66, thank you and good on you for trying this out and publishing your technique and results.  Amigaman, you're now on my ignore list.  

I hope to contribute more once the crystals and heatsinks arrive. I've got 10 block erupters here and they are chugging along but about to get steamrollered by the new ASICs that are on the way, so I hope to make them last a bit longer.  I'd love to get 5GHs out of the 10 BEs, and maybe more if practical.

@Bluestreak66:
Exactly what Lucko said is my point.
Of course you can run your old '66 Volkswagen Beetle with Nitro, but how long would it withstand that?
You pumped up the clock to 16M, that's quite impressive, i must admit.
But you did not
- improve cooling professionally. Rubber bands will not suffice for longer periods of time.
- check for power consumption. Yeah it runs, but on the long terms?

If you designed SPS'es before and if you know the requirements you know i am right. It is hard to replace a given chip with another without changing all other parts, it is harder to find a suitable chip when the pcb is fixed, it is often almost impossible to get some of the chips you found.
Also, if you did not measure the current consumption (a task not so easy for itself) you really don't know where to go.
Do you have a datasheet of the asic? How much Volts and Amperes can it withstand, sufficient cooling provided?
How much can it withstand on the stick board?
What's the max temp?

I really hear you screaming "yeah dudes i got 450 out of it", but you did not provide any information if that is professionally repeatable and long term stable.

Nobody really knows but it should be fine as long as its kept cool. This is really unknown territory 448 @1.48 volts is as far as any body has push one as far as I know.

What is your suggestion? How can we increase the the current supplied to the asic without over stressing parts? I don't really see you contributing any ideas here, just criticizing. You said there was no way it would run with a 16Mhz oscillator and its clearly been running that way for 2 days now. I've designed switch mode supplies before I am well aware of the design requirements. Either contribute or get off my @ss

Just wondering how long chip would live like this... Isn't highest voltage on a blade 1,2 and about 400MH?