Block Erupter Prisma (>=1.4 T/device, 0.75-0.78 W/G, <1 BTC/T, October Shipping)

[Chris_Sabian]

Hy, i want to share some opinions i have discussed with intron regarding the undervolting of the prismas.

What he told me is that being a one string design meaning that each row has 16 chips @12V that is 0.75V.

In order to undervolt you would have to come down from 12V to 9...10V that would give 0.65V to each chip.
Also those D15NF10 mosfets act as a shunt when a chip in the string goes bad, alowing the string to keep going,
but here comes the bad news, if a shunt is activated there are 15 chips in the string, that equals to 0.8v/chip...more amps being drawn...

From the pictures and calculations and after doing some numbers, those mosfet are working @20~22A, maximum amperage is 23A
If im right then in all the cases a board caught fire was because of a bad chip going wrong, shunt working and the mosfet exploding because it was to its limit.

Its a good design but poorely equiped.

A question to the ones the board caught fire: Were they powered from Server PSUs?
Server PSUs are known to be more stable on the 12V line, normal ATX would go down to 11.8v or so no resulting in such a bad avalanche heating disasters.

What could work is modding a server PSU to 9...10.5V and powering one blade to see what is the maximum clock it would go and power consumption.
It will make the blade safer and power consumption will go down.
OR
A rectifier 30A diode with a 2.6...2.8V voltage drop on it in series with the whole blade in order for the blade to work undervolted with a standard 12V.
The goal is to make them safer in the rectifier diode situation. Power consumption will not go down so much as the rectifier will dissipate about 50W of heat.


All i said is theory, until my blades arrive and i can send one to intron or test myself.
None of this has been tested or we could be dead wrong about the design.
Please test at your own risk.

I think that you are about 10 months late with this post.

[adib]

Well i didnt have a prisma until recently and never saw a post that gave some undeclocking/undervolting solutions.
The tube is easy to undervolt....Prisma..not so much.

Thought i should share the info, i think there are enough in the wild to be saved
Ill be presenting the Prisma tower when it arrives and if i succeed in building it.

If anyoane has or knows about a server PSU capable of voltage adjusment to 10.5v let me know

[sidehack]

One of my projects once I find time in the next few weeks is undervolting Prismas. I have about a dozen retired units laying around I was going to see what I could do with, if only as a test.

[adib]

If you plan to get rid of the blades(without heatsinks), let me know
Also let me know what power supply youll be using

For me these come to mind
DPS-1570AB(BB)
NPS-700AB
They are voltage adjustable and hoping they go all the way down to 10V

[sidehack]

I'm not sure what undervolting has been done on Tubes (or BE200 in general) but I was just playing with my Tube test rig a bit and had trouble getting it to see chips reliably below 730mV with a clock at 20MHz. Not sure where the limitation is and I'll keep playing with it until I have a Prisma setup to mess with.

EDIT - of course, it's also possible that it got set to 200MHz instead of 20MHz, given I was seeing 10x% expected hashrate... Let's see where 100MHz starts dropping out.

[adib]

I'm not sure what undervolting has been done on Tubes (or BE200 in general) but I was just playing with my Tube test rig a bit and had trouble getting it to see chips reliably below 730mV with a clock at 20MHz. Not sure where the limitation is and I'll keep playing with it until I have a Prisma setup to mess with.

EDIT - of course, it's also possible that it got set to 200MHz instead of 20MHz, given I was seeing 10x% expected hashrate... Let's see where 100MHz starts dropping out.

you have to change the driver-blockerupter.c on line 193
from if (!opt_bet_clk || opt_bet_clk< 19 || opt_bet_clk > 31)
to if (!opt_bet_clk || opt_bet_clk< 9 || opt_bet_clk > 31)
or even 0 if you want to be able to go lower than 20(200Mhz) on the chip.

as described in the driver info->clock = (opt_bet_clk + 1) * 10;
so 20 is 200 mhz on the chip

[sidehack]

I was running off the standalone controller. Currently I have 'em set at 180MHz and am getting expected hashrate. Chips still didn't like to light up below 730.

When I have some time, maybe in a few weeks, I'll take a look again. What version of cgminer are you pulling from? Nothing I had handy reported HW errors correctly so was kinda useless for testing.

[adib]

Latest build of cgminer, ran them from 100 to 270.

Even at 270 the TPS VRMs have a hard time keeping up @250~260W, also i added a top side heatsink
The blade should be run @200W
Seems like these are build on the limit of everything...chips...vrms...

730mV is still better than 800mV, what is you power consumption @wall? and hashrate?
You could try swaping the 20mhz oscillator with a 10mhz one from a prisma maybe it will help lowering the start of the chips...i dunno.

Ill have more time in the weekend to tinker with the resistors and frequencies.

I dont think error reporting is even in the code.

[sidehack]

I did a half-board power measure (four VRMs disabled, four with adjustable voltage) during initial testing and the board-level power consumption dropped to a hair below 0.8W/GH board-level according to my measurements, compared to about 1W/GH unmodified. With my power that's still viable for a while yet, especially since the shop leaks air like a seive and it still got balls cold with 40KW of miners running last winter so I need the heat.

Those TPS were definitely stressed. Every other BE200 design I've seen had two ASICs per, but they were also clocked higher. I have had to replace a few blown schfifty-threes. One of my test Tubes is down a node from a roasted buck chip, and I think Novak's space heater Tube is short one as well.

If I have time coming up I might try a modified cgminer build and see what happens. For now I'm satisfied with their performance. Anything I was going to do with Prismas is sidelined for a bit for more pressing projects.