Topic: Swedish ASIC miner company kncminer.com - page 1211. (Read 3049524 times)

Thanks for the reply. Any good stories?

Learn a new thing everyday, thanks hypermega for that nice find.

I honestly think it's a VRM thing... not an asic thing..  
But the temperature inversion link says it all! wow...

I don't think so...  That phenomenon might make it possible to clock a chip faster at a higher temperature, but it should not effect the speed of a chip that is clocked at a constant speed.  Right?  I mean, if I am reading it right, the rise-times and fall-times on the edges will be faster, but I don't think these chips are clocked close to theoretical maximum where that might make a real difference.

After reverting back, the two fans at the front of the case do not function!!!!

They are in denial because it doesnt make sense,,, lol  But it's the truth. Mine are still rockin' solid now, no drastic drop overnight this time.
It may not be this way for everybody, but the higher temps cured both ailing saturns, point blank.

I also noticed that the coldest sat is the slowest although I'm tickled with the speed

It's probably a effect called temperature inversion. 28nm logic gets faster with higher temperatures.
http://siliconsaint.blogspot.com/2012/07/temperature-inversion-in-deep-sub.html?m=1

just tried .97 firware, got a nice hashrate and low HW rate, but the fans were not spinning   73c on one core. reverted back. WTF KNC?

Seems the Jupiter that I am in a Group Buy with (Aug 2nd order) just arrived last week and is hashing at 315Ghash/s. Looks like a chip is bad or the like.

Any good stories out there of KNC shipping out a replacement board or the like and not asking for the unit back with an RMA and such?

Thanks,
IAS

Is there a way to monitor clockspeed?

well guess you are another anomaly, welcome to the club.

just a guess, but maybe warmer temps allow the VRMs to do their jobs more effciently....look at your betmod stats to see any changes in
input/out current and output volt....

 

Thanks. I always hated white on black. Harder to read to me. I like dark font on lighter backgrounds.

Also, been playing around with making the ASICs warmer. Best results so far are with case fans tilted towards the BBB and ASIC fans unplugged, but I still have a small house fan on Low at the empty end of the Saturn case tilted down to provide cooling for the VRM chips mainly. Temps are 46 & 41. WU is 4006(best ever), avg hashrate is 257(same as best ever), error rate is still high at over 18%, but best ever. This is after over 300K shares. I may actually turn the house fan off next, but that makes me nervous. Glad to hear the max temp is around 105C for the ASICs.

Jupiter FW 0.95

3 x Replacement boards.

After 24 Hour Burn in

Average 550Gh/s CGMiner and 550Gh/s at BTC Guild.

Less than 1 % HW errors.

All cores on and 100%.

Temps Range across 4 Asics 59.5 up to 65.5.

The suggestion that the ASIC's ball joints could get damaged due to Shipment with heat-sinks attached may have some merit.

My Under performing Jupiter is now 100%, Not changed firmware since RMA of the Faulty Boards (Faulty due to Hash Rate between 80 and 100Gh/s and a VRM not detected on 1 board).

Only other difference is I have applied my own Thermal paste.

little that I know, you can only use firmware updates. I think I saw on irc some guys chatting away, or was it here, about having reversed engineered KNC's code.....there might be some real computer literates out there hacking away but nothing concrete yet....

Some tips to overclock?

thanks

nemo, I actually had added 2 high cfm 140mm computer case fans, 1 for each board/chip on my Saturn. And it cooled the chips...no noticeable gain, no noticeable decline in performance neither. But at the time I wasn't considering or thinking yet about hotter is better at the time, so I wasn't paying attention to that. All I noticed was that I added those 2 fans it did cool it, but no performance gain.

So my Saturn came with .95. I did all the suggested mods, not happy went to .96 and was somewhat satisfied, ended with .97 but has high error rate.

So I have gone direct with power supply to asic boards (lost the knc extenders). Supposedly less resistance blah blah....nothing. More cooling with a large fan over head facing down, nothing.

So then soy suggested .90 blah blah....I gave it a whirl (even though some early units had fried on this firmware) and whamo, LOW ERRORS

HIGH HASH RATE (both on cgminer and at the pool server)....I was happy...for like 20 minutes til it crashed.....worked my way all the way back to .97 (which at this point works the best for me) but hashes with a 20% error rate, which is bad.....

So I have kept a journal of all changes, times, stats (temp, wu, hw etc etc etc), and saw that .90 performed the best, and per bertmod, showed output volt slightly higher, and output current way higher (like 50A+, and yeah I was scared), and temps were definitely higher like 52c...

so then I think people on the board have been mentioning ball arrays, and shipping possibly harming chip and chips seating better with heat (put asic board in oven or something)  blah blah then a few post about heat making it better....

So all I did was unplug the large fan, the 2 added high cfm case fans and turn off the a/c unit in the office, and the number I noticed the most was less errors....slowly down to like 14%- 15%, which is still high but better than 20%. Slightly higher chip temps, and even slight higher watt usage.....throughout most of the stable mining on this saturn (usually .94 and up) 1 chip with 100% cores enabled and the "problem" chip  with 0 upto 10 cores disabled

I am sorry, you are not going to get a reflow effect at the temperatures that this chip is running at.  The max temp was stated at 105deg. but they hardly run anywhere near that. 

I do not doubt the empirical evidence seen, just the conclusion.

Can it be electrical interference from the fan's motor? They have quite a strong electromagnetic field when running.

I'm curious to know how orama could possibly know who Avenger is and find out whether he has or not an order. I also find it very much less than honest for the knc ppl to divulge their order book to non-employees and I find orama's witch hunt for Avenger puerile and fruitless.

I'm sorry you're having it raw, but you chose to believe in phonies with no (successful) track record (unless you still believe that knc is orsoc and that having experience in design is that same as having experience in manufacturing). Can't say I didn't warn you.

As for the cake, sure, I'll probably have some anyway. Maybe you should consider why I was asking for pictures?