Swedish ASIC miner company kncminer.com

[edgar]

exactly

[Searing]

Hi

Now after 24 hours my jup is going with 552ghs rock solid but I figured out something interesting.

If all my vrm output above 0.7 V the pool shows great 550-560Ghs, if just only 1 or 2 vrm output goes down under 0.7 V (for example 0.687V) the pool hashrate dorps down to 510-540. In cgminer still 552.

Any ideas?

KNC's cgminer includes hardware errors in the reported hashing rate. Something not done by the other versions of cgminer. From the maker:
https://asktom.cf/index.php?topic=319533.0

Please compute your hashing rate from cgminer using:
2^32 * Accepted / Elapsed


What are your temps and HW in both circumstances?

using above formula mine is 7.15% I'm running 565ss 555ave on .95 temps 27c 40.5c and two at 33.5c so a bit higher then the 5.4% I got just using HW values

Searing

[jelin1984]

Which is these formula?

What I must write at cgminer at Jupiter to have it solid run??

[tunctioncloud]

Hopefully KNC latest drivers are always best.

Because it is hard to follow this discussion to have the best hashrate possible

[timmmers]

Hopefully KNC latest drivers are always best.

Because it is hard to follow this discussion to have the best hashrate possible

Have been for me. Not had a single problem yet .97 from the day after it was released (just to make sure if the latest version wasn't homicidal).
I don't get this .97 doesn't work stuff, how can that be the firmware for one rig and not for others?

[FeedbackLoop]

Which is these formula?

These formulae were posted by Kano in a KNC thread, they merely allow you to compute your hashrate properly:

https://asktom.cf/index.php?topic=306969.msg3309728#msg3309728

ElGabo has:

11213824*2^32/(25*3600+2.5*60)
   5.3425410e+11
534 GH/s in accepted shares.

420356*2^32/(25*3600+2.5*60)
   2.0026792e+10
20 GH/s in hardware errors for less than 5% HW

72191.*2^32/(25.*3600.+2.5*60)
   3.4393565e+09
3.4 GH/s in rejected shares (538 GH/s full hashrate).


Which are outstanding values compared with what many people here, including me, have. Unfortunately it's hard to figure out what factors affect the real hashrates as most people do not post the source of their numbers and some others, like edgar, will prefer to cover the thread in flame instead of making some effort to search the threads, understand and cooperate.

My other question was if temperatures affect the output in your machine ElGabo, and if that affects HW (positively or negatively) but your HW is so low that your machine doesn't seem good to test this. Unless when the temperature is high (seeing that your screenshot shows very low temperatures) your hardware error rate goes much higher  (which would be the opposite of what people including Phoenix and DigginDeep just above seem to be seeing). Then we could see if that VRM output correlates. Is that the case?


tunctioncloud: 0.97 is horrible for my machine and many others.

[timmmers]

Which is these formula?

These formulae were posted by Kano in a KNC thread, they merely allow you to compute your hashrate properly:

https://asktom.cf/index.php?topic=306969.msg3309728#msg3309728

ElGabo has:

11213824*2^32/(25*3600+2.5*60)
   5.3425410e+11
534 GH/s in accepted shares.

420356*2^32/(25*3600+2.5*60)
   2.0026792e+10
20 GH/s in hardware errors for less than 5% HW

72191.*2^32/(25.*3600.+2.5*60)
   3.4393565e+09
3.4 GH/s in rejected shares (538 GH/s full hashrate).


Which are outstanding values compared with what many people here, including me, have. Unfortunately it's hard to figure out what factors affect the real hashrates as most people do not post the source of their numbers and some others, like edgar, will prefer to cover the thread in flame instead of making some effort to search the threads, understand and cooperate.

My other question was if temperatures affect the output in your machine ElGabo, and if that affects HW (positively or negatively) but your HW is so low that your machine doesn't seem good to test this. Unless when the temperature is high (seeing that your screenshot shows very low temperatures) your hardware error rate goes much higher  (which would be the opposite of what people including Phoenix and DigginDeep just above seem to be seeing). Then we could see if that VRM output correlates. Is that the case?


tunctioncloud: 0.97 is horrible for my machine and many others.

It's s shame that there seems to be such a variance in build quality between rigs, things like slack screws make me think whoever assembled these things probably hadn't much experience. The temperature may just be causing better contact and compensating for these things and on two well made rigs may not have any effect at all. I've been lucky it seems, but if I had one of the rigs with problems I'd have it to bits and rebuild it and make sure things like thermal paste and contacts were all tickety boo. Like they should have when assembling them , which is not racket science really is it?

[fubly]

TEST_ASIC_BOARD_FOUND=FAIL

type only in your console when logged in:

cat /././var/log/initc.log


root@Jupiter-fuckoff:/var/volatile/log# cat initc.log
Return code = -11
TEST_READ_BBB_SN=OK
TEST_I2C_CONNECT=OK
TEST_LM75_TEST=OK
TEST_TPS65217_TEST=OK
TEST_TPS65217_CONFIG=OK
TEST_FPGA_CONFIG=OK
TEST_FPGA_TEST=OK
TEST_ONBOARD_EEPROM_EMPTY=OK
TEST_GET_SERIAL_NUMBER=OK
TEST_ONBOARD_EEPROM_WRITE=OK
TEST_MMC_REFLASH=OK
TEST_I2C_MUX_PROBE=OK
  TEST_ASIC_BOARD_FOUND=FAIL  
TEST_READ_IOBOARD_SN=OK
TEST_IOBOARD_SN_GOOD=OK
TEST_GET_CGMINER_CONFIG=OK
TEST_WRITE_CGMINER_CONFIG=OK

what can i do?

[Puppet]

It's s shame that there seems to be such a variance in build quality between rigs, things like slack screws make me think whoever assembled these things probably hadn't much experience.

I doubt its the assembly. Its almost certainly because they chose not to test and bin the chips. All wafers have defunct chips and chips of varying quality, thats a given. Almost all asic manufacturers test their chips on the wafer, and then again once they are packaged, tossing out the broken ones and binning the others according to how well they perform (power consumption, clock speed, if applicable, number of working cores) and use different bins for different products (think various AMD or Intel chips with different clockspeeds), or combine good and not so good chips to get a more predictable end product (afaik, thats what BFL does).

KnC seems to solder all their chips on a module, and they will do a final assembly functional test, but whether you get 4 excellent or 4 mediocre chips or a mix of both is just a matter of luck.

[jelin1984]

11213824*2^32/(25*3600+2.5*60)
   5.3425410e+11
534 GH/s in accepted shares.

420356*2^32/(25*3600+2.5*60)
   2.0026792e+10
20 GH/s in hardware errors for less than 5% HW

72191.*2^32/(25.*3600.+2.5*60)
   3.4393565e+09
3.4 GH/s in rejected shares (538 GH/s full hashrate).



Where I must write that ??

[crumbs]

It's s shame that there seems to be such a variance in build quality between rigs, things like slack screws make me think whoever assembled these things probably hadn't much experience.

I doubt its the assembly. Its almost certainly because they chose not to test and bin the chips. All wafers have defunct chips and chips of varying quality, thats a given. Almost all asic manufacturers test their chips on the wafer, and then again once they are packaged, tossing out the broken ones and binning the others according to how well they perform (power consumption, clock speed, if applicable, number of working cores) and use different bins for different products (think various AMD or Intel chips with different clockspeeds), or combine good and not so good chips to get a more predictable end product (afaik, thats what BFL does).

KnC seems to solder all their chips on a module, and they will do a final assembly functional test, but whether you get 4 excellent or 4 mediocre chips or a mix of both is just a matter of luck.

Sort-a like buying any CPU/GPU -- some take clock better than others, no cause for complaints as long as all manage the datasheet specs.
If it becomes profitable in the future, KNC will sell you "Deluxe 1337 Black Edition," with 100% good cores guaranteed

[FiatKiller]

Hotter is better! I had a supplemental house fan & turned it off. ASICs fans are unplugged. Only
left the tilted case fans on. FW 9.7 and Enablecores afterwards. No Bertmod yet. HW error rate
was 19% before. Hashrate was mid-250s before. WU was 3900ish before.
Considering putting the case back on also...
Edit: power is now only 298 watts. I just removed the fans entirely and put a long piece of closed cell foam
across the tops of both heatsinks. Will see what happens.
Credit goes to user FoolPartedWithMoney for pointing-out first that more heat may be the key!

[timmmers]

It's s shame that there seems to be such a variance in build quality between rigs, things like slack screws make me think whoever assembled these things probably hadn't much experience.

I doubt its the assembly. Its almost certainly because they chose not to test and bin the chips. All wafers have defunct chips and chips of varying quality, thats a given. Almost all asic manufacturers test their chips on the wafer, and then again once they are packaged, tossing out the broken ones and binning the others according to how well they perform (power consumption, clock speed, if applicable, number of working cores) and use different bins for different products (think various AMD or Intel chips with different clockspeeds), or combine good and not so good chips to get a more predictable end product (afaik, thats what BFL does).

KnC seems to solder all their chips on a module, and they will do a final assembly functional test, but whether you get 4 excellent or 4 mediocre chips or a mix of both is just a matter of luck.

Sort-a like buying any CPU/GPU -- some take clock better than others, no cause for complaints as long as all manage the datasheet specs.
If it becomes profitable in the future, KNC will sell you "Deluxe 1337 Black Edition," with 100% good cores guaranteed

Not what I meant really, the temperatures seem to have an effect on some machines but not others, they seem to all have a top end of about 140Gh per module but some take some help getting that.
True enough they rushed and didn't test the chips, but I don't think they assembled some rigs at all well at a basic level, and that's only what we can see. Leaving the sucking plastic on them doesn't show much attention to detail, so if they didn't do that simple thing what's the odds on thermal paste being well done? A few bent boards caused by the assembly or other minor defects could be why the heat makes such a change for some as they expand maybe I was thinking.

"Spoling the sheep for a hapeth of tar" my granny would say.

[Searing]

It's s shame that there seems to be such a variance in build quality between rigs, things like slack screws make me think whoever assembled these things probably hadn't much experience.

I doubt its the assembly. Its almost certainly because they chose not to test and bin the chips. All wafers have defunct chips and chips of varying quality, thats a given. Almost all asic manufacturers test their chips on the wafer, and then again once they are packaged, tossing out the broken ones and binning the others according to how well they perform (power consumption, clock speed, if applicable, number of working cores) and use different bins for different products (think various AMD or Intel chips with different clockspeeds), or combine good and not so good chips to get a more predictable end product (afaik, thats what BFL does).

KnC seems to solder all their chips on a module, and they will do a final assembly functional test, but whether you get 4 excellent or 4 mediocre chips or a mix of both is just a matter of luck.

well they likely will do this same method in the future..they thought they'd get 400gh they got up 550gh or more...so the plan to skip that step seemed to work in speeding up the delivery I guess...sloppy thou it maybe be....24hrs is the devil on rollerskates....in speed if you get the stuff off the plane via courier and into a working box

http://www.youtube.com/watch?v=ygE01sOhzz0

ie they went "plaid"

[FeedbackLoop]

Also getting some better stats, 10 more cores, by just disabling the fan on just my problematic board.   Still no numbers on actual average long-term performance but I guess/hope it will follow  Phoenix's path.

Temperature does not seem to change significantly matching FiatKillers pics. Perhaps the sensors are indeed not in chip itself like someone mentioned?  Sorry if that was already posted. Can't find it.

The sink does get crazy hot to the touch. Switching the bad board fan with a worse one. No fan at all feels too risky.

[sbfree]

Hotter is better! I had a supplemental house fan & turned it off. ASICs fans are unplugged. Only
left the tilted case fans on. FW 9.7 and Enablecores afterwards. No Bertmod yet. HW error rate
was 19% before. Hashrate was mid-250s before. WU was 3900ish before.
Considering putting the case back on also...

fiatkiller, you are bold w/ turning off heat sink fans....but i have to agree....

miner_knc   250.75 GH/s   44,156k (99.83%)

my stats above on btcguild....AFTER TURNING OFF THE A/C....asic temps went up, but so did the HASHING as shown above...

something that would normally range around 220GH/s is now a solid 250+GH/s,

and yes, since error rate went from around 20% to now under 15%, that difference caused less wasted shares and more good share= higher hash rate at server. YAY!

wish I had one that was super like phoenix1969's, who has 3 saturns all above 270GH/s, but I am now happy with my unit though and no a/c....

ps asic temps went from low to mid 40'c, to now upper 40's to low 50's celsius...

[DimensionsOfHell]

Well I still am getting the "Running (Connect to CGMiner API failed)" error from the web GUI.
Yes, I have cleared the cash on the browser, AND used a different browser.

More importantly I can tell it isn't doing anything because the pool isn't getting much work.

Looks like 0.96 isn't up to prime time for my Mercury.  Going back to 0.95 because at least that was hashing something.  Unless there is something else to do to get 0.96 to work?

Thanks,

_theJestre

I have similar problems with my Saturn. Download bertmod and check the stats of your VRMs. You might have a bad unit or two. I have 3 bad VRMs out of my two boards. My guess is that with BAD VRMs you cannot update to the higher fw, probably because they are more well adjusted to work with the VRMs, and since some VRMs are not functioning [properly, or at all, like in my case], its causing the later fw to not work. The only fw I can get running stable is fw 0.9.4. Out of my Saturn, I am averaging 150gh/s.

Check with the bertmod patch, you might need to RMA the board. Or you could try all the tricks the good folks here have discussed. I've tried them all (other then disabling fans), and nothing fixed it.

Bertmod: http://forum.kncminer.com/forum/main-category/main-forum/6183-bertmod-0-2-unofficial-firmware-mod-feedback-thread

[sbfree]

Well I still am getting the "Running (Connect to CGMiner API failed)" error from the web GUI.
Yes, I have cleared the cash on the browser, AND used a different browser.

More importantly I can tell it isn't doing anything because the pool isn't getting much work.

Looks like 0.96 isn't up to prime time for my Mercury.  Going back to 0.95 because at least that was hashing something.  Unless there is something else to do to get 0.96 to work?

Thanks,

_theJestre

I have similar problems with my Saturn. Download bertmod and check the stats of your VRMs. You might have a bad unit or two. I have 3 bad VRMs out of my two boards. My guess is that with BAD VRMs you cannot update to the higher fw, probably because they are more well adjusted to work with the VRMs, and since some VRMs are not functioning [properly, or at all, like in my case], its causing the later fw to not work. The only fw I can get running stable is fw 0.9.4. Out of my Saturn, I am averaging 150gh/s.

Check with the bertmod patch, you might need to RMA the board. Or you could try all the tricks the good folks here have discussed. I've tried them all (other then disabling fans), and nothing fixed it.

Bertmod: http://forum.kncminer.com/forum/main-category/main-forum/6183-bertmod-0-2-unofficial-firmware-mod-feedback-thread

I don't know if your problem is related, but I am now starting to suspect that a little extra heat (rather than cooling) and now as of recently, tightening down (slightly) on the heat sink crossmember screws (WHICH MOST LIKELY PUSHES THE CHIP DOWN AS WELL) has improved performance ON SOME machines....I suspect on yours, and I don't know DIDDLY (other than Bo), that you will probably find some of the screws on the crossmember for the heatsink can use a little tightening.....which in turn will probably make the chip have better contact to the board, then you will notice change in VRM stats...HOPEFULLY.

Disclosure...I KNOW NOTHING.
good luck.

[helmax]

eligius pool anyone can connect?

i see errors in my cgminer

[RHA]

I was right when I signaled two weeks ago at #kncminer channel that the Saturn seems to work better in higher temperature.
Now many people confirm it.
Going from 40-42 C to 48.5-50 C moves avg from 267 to 274 (at cgminer AND at the pool) and WU from 3850-3900 to 4010.