Topic: ZTEX voltmod (Read 5074 times)

I think of water cooling my boards as well

I have a simple NB cooler from Watercool called "HEat killer - Micro" here that would fit.

I tried the voltmod today on my 1.15y. Works like a charm. Went from 216 to 232 after raising the voltage to 1.32V. Also tried 1.4V for a short time, which gave me

001-0: ztex_ufm1_15y1-04A36E42D1-4: f=248.00MHz,  errorRate=0.00%,  maxErrorRate=0.00%,  hashRate=248.0MH/s,  submitted 13 new nonces,  luckFactor=0.66

Now I am thinking about watercooling, also because of the noise reduction and about recompiling BTCMiner to allow higher frequencies ;-)

@ rupy

working ~65 hrs a week atm and building a house too...not much time for "btc-work".

2-ztex modded today...

Why not? Is there something your not satisfied with?

I'm really considering doing this since it's not much left of 50 BTC blocks... can just revert to lower voltage after!

Been runin at 1.4v for weeks now... 

@ rupy: not yet...

@ Andows: Im on 1,34-1,36v - no temperature issues, no stability problems, ~240-248MHZ

What voltages have you guys ended up using on long term? Have you experienced any hardware failures or performance degradation?

Xilinx suggests (in the Spartan-6 FPGA Data Sheet: DC and Switching Characteristics -document) using 1.14-1.26V for V_CCINT and lists the 1.32V as absolute maximum.
I suspect that 1.26V would probably be the safest maximum for long-term use, but I would love to hear more about your experiences.

@nyana Did you get around to mod all your FPGAs?

Not directly. But when push comes to shove people do this by building some sort of multivibrator circuit involving the I/O pads and external analog components. Then the temperature can be measured indirectly by counting up the time it takes for some capacitor to charge/discharge through some internal resistance. Calibration can be difficult for such circuits.

As far as i know they dont have an Inbuild temp Sensor.

is it possible to messure the temperature of the chip?

You can, but having the ability to fine-tune and change voltage on the fly is very handy.
And by using a big multiturn VR in parallel with the existing R, you can tune the voltage very precisely. (I also have off/on switch).


Prolly yes, depends on the ambient temp, HS-chip contact, etc...


I started with 50k VR, but then I broke one pin and changed it to 25k VR + 3k R. I'd recommend at least 25k or higher.

What are the specs of the trimmer ?

Edit: Found above

Can't you just unsolder the R13 and replace it instead of all these ghetto cables?

How much warmer does the chip become? Can you bump it to 1.320v without adding heatsinks etc.?

+1 

Nice work, nice pics

Hey all,

the Voltmod works for me very well too.

I measured a deafult voltage of 1,23v on my "lazyest" ztex Module.

The Resistor R13 (2,4kohm) was original @ 885 ohms (dont know why and i dont know much about elecronics). after soldering the Trimmer i "regulated" it to 818 Ohms and a Voltage of 1,37.

The Results on 18 degree Celsius

Stock:
1,230v / 204mhz / 885 Ohm@R13 / 0,10% Error

Modded:
1,375v / 240mhz / 818 Ohm@R13 / 0,00% Error

Im very happy with these results, after a few days of testing i will mod all my 22 ztex Modules!

attached some Pics from "work"


By daki123 at 2012-07-28

By daki123 at 2012-07-28

By daki123 at 2012-07-28

By daki123 at 2012-07-28

Anny extremcooling attempts planed. Like dry ice or ln2 ?

My tex doesn't look that ghetto anymore after seeing this :p




A little update on the tex. Ran 276 mhash at 1.48v with TML (X:276 C:182,182,188) today with minimal invalids. Next step is to mount a waterblock on in and try to go past 1.5v.

And Xilinx told me that 45nm performance was not strongly voltage-sensitive..

-rph

Thanks

I just love me some ghetto modding!!!  Congrats mate.

This will be my last update or at least for a while, I have achieved maximum clock 248mhz at 1.41V, which is the ztex bitstream limit.
Still waiting for new releases of ET's TML release - maybe we will see some interesting numbers there.

Some changes: redone the soldering with a better equipment, much nicer now,
also added a off/on switch for vmod. Off being default and on starts at 1.3v now due to changed trimmer. Also added HS on some micro hotspots.













Remember kids, ambient temperature is very important, no matter how much air you throw at the chip.

 

Ran 1.355v@236mhash over night - everything is in check. Ambient temp in the room has risen about 10°C in the meanwhile (from 22°C to 32°C), so now an invalid poped up from time to time but ztex didn't lower clock.

I got a hold of some thermal tape and attached a bottom HS. Testing now.

http://shrani.si/f/2D/11A/3b8KJP16/img9220.jpg

http://shrani.si/f/3J/Jg/10zWNYA7/img9222.jpg

http://shrani.si/f/2d/6o/3U6cKOP9/img9224.jpg

Correction, its already working - we are just testing 2 new release candidates as I type.

Also I have x6500 at hand and Icarus, maybe maybe they are ready for some modifications as well, but we need better bitstreams there.

After reading about the difficulties in the tricone mining bitstream I was wondering when someone was going to bump the voltages a bit to solve the instability issues. Perhaps you should chat with eldentyrell and beta test that?

Curious how the MH/w efficiency will be affected with higher voltages and core clocks. FPGA overvolting might make watercooling something to consider. More likely though would be an immersed solution in fluorinert(which can be found for cheap on eBay from time to time).


edit: my mistake, you're one of the guys trying to get eldentyrell's bitstream working on a ztex single.

Nice to see that this is possible.

Now u have to test them on LN2 plz

So it turns up its not really hard to do....

Here are ztex schematics http://www.ztex.de/downloads/usb-fpga-1.15x.pdf

We need to fiddle with R12 and/or R13 to bump the voltage.
Formula to get the voltage is 0.8*(1+R12/R13)

On schema or (high res pic1, high res pic2) we can see:
R13 = 2k4
R12 = 1K3 , pcb print says 1K2 in reality R12 is 1370 ohm, which means (by the formula) default voltage should be 1.256V

VCCINT can be checked here:



and it showed me 1.264V on my multimeter by default.

I decided to go with 50k parallel multi-turn trimmer that would result at 2290ohm instead 2400 at 50k. That should bring us to about 1.28V.

I know its ugly but it works (i redid some of the soldering after seeing this pic). Also it doest matter if the R12 and R13 are shorted because they are anyway via pcb.





The readout on the red and black wire straight after - 1.286V.
This increased my stable hashrate from 216 to 220, ztex miner did try to push higher but it went down to 220.

Time to crank it up, some sources say ztex tested up to 1.4v and for bitfury we know he runs 1.5v in OC mode, Xilinx says it's okay to go up to 1.32v.

Also, I am not responsible for your fried boards.


EDIT:

Results w latest ztex sw/fw, 0% invalid:
1.264V 216mhash //default
1.286V 220mhash
1.300V 228mhash
1.320v 232mhash //Xilinx approved
1.350v 236mhash
1.370v 240mhash
1.390v 244mhash
1.410v 248mhash // tested with extensive cooling,.. this is the max that ztex bitstream can go.