Swedish ASIC miner company kncminer.com

[joae1975]

Notice the page says, "ship to you in October."  Meaning you probably won't get it until Nov, unless you live close.  Cheaper units offered for Nov. shipping probably won't be received until Dec.  I've accepted that I probably won't receive my day 2 Sept unit until Oct.  Day 2 of what?  It's certainly not Sept. 2nd.  It's the 2nd day of shipping.  That could be Sept. 27th.  Who knows?  Curious how this will play out.

[1713500056]

1713500056

[1713500056]

1713500056

[1713500056]

1713500056

[kingcoin]

5) Extra-credit question for people who aren't engineers but have experience mining: how many mining devices/rigs you had that booted and started mining correctly, but kept failing after several hours or only in specific circumstances? Were you willing to consider them completely faulty and throw them away or were you willing to delve in and debug the problem? How many hours of "testing" was your cutoff time before you decided to throw that device away?

5b) Also how would you feel after spending X thousand dollars on your miner to discover that when you receive it it only has 100-Y% of the cores operating, while some other guy (also payed X K$) on the forum received his miner which has 100% of the cores operating? This after you have spent numerous of hours trying to restart your miner, upgrading software, power cycle, swapped pools, replaced the PSU you bought because initially because you thought it could not supply enough power to the miner, etc. Of course the hash rate is within the late announced hash rate given by the vendor, so you can't ship it back for a replacement.

6) Extra-extra-credit question for engineers: What do you think about other engineers that have an obsesive-compulsive disorder about some testing methodology but have no practical experience running an actual bitmine?

KnC is skipping a testing methodology used by the vast majority of the ASIC industry: chip level testing to make sure they detect faults at the chip tester and not in the assembled product, where they might not even know if the cause is a defective ASIC. Rather than stopping or sorting the bad chips at the fab, they pass them on to their customers. Again, which hopefully will receive miners above the later announced rate.

As I've mentioned earlier I would have expected the hash rate to be given by the architecture and static timing analysis (assuming they have designed the rest of the miner so the operating conditions of the ASIC timing model is not violated). I was not aware that KnC was skipping this common testing methodology, but it might explain why they are so uncertain about the actual final performance of their miners: It will depend heavily on the yield of ASIC's as they will be mounting defective ASIC's into the miners as they don't know in advance which chips are defective or not.

BTW: Just keep your rude personal attacks and wrong assumptions of what I do and have or have not done coming. I simply ignore them, but they seem to be good for your inflated ego.

[Anenome5]

KnC is skipping a testing methodology used by the vast majority of the ASIC industry: chip level testing to make sure they detect faults at the chip tester and not in the assembled product, where they might not even know if the cause is a defective ASIC. Rather than stopping or sorting the bad chips at the fab, they pass them on to their customers. Again, which hopefully will receive miners above the later announced rate.

Once they receive the completed machines they'll go a burn-in and anything not within specs will simply be set aside and they'll move to the next device and send you that one. It's not a big deal yo.

[kingcoin]

KnC is skipping a testing methodology used by the vast majority of the ASIC industry: chip level testing to make sure they detect faults at the chip tester and not in the assembled product, where they might not even know if the cause is a defective ASIC. Rather than stopping or sorting the bad chips at the fab, they pass them on to their customers. Again, which hopefully will receive miners above the later announced rate.

Once they receive the completed machines they'll go a burn-in and anything not within specs will simply be set aside and they'll move to the next device and send you that one. It's not a big deal yo.

Then throwing away a full PCB containing working chips or go through the process of re-balling, re-flowing the BGA's? That's expensive, especially if the ASIC yield is low. I can't see why they are skipping this common simple test methodology for the price of possibly throwing away almost complete miners. But what do I know, perhaps is related to some requirement or incomplete procedure given by the undisclosed mysterious ASIC vendor.

[J35st3r]

Oh poo. The thread went chip-design while I was asleep. I missed all the fun 

I'll just point out that KNC have had plenty of time since tape-out to refine their testing procedures. Yes is is insane to omit the wafer test on production devices, but these are prototypes (for all they say about commercial product, it will take many more months to shakedown a volume-scale design, and quite right too, time is very much of the essence with bitcoin mining).

Provided they have sorted out a test socket/interface for the packaged chips, so they can weed out the DOAs (dead shorts between power rails do not make it onto PCB's), and (rare) controller failures, they will be fine.

[RHA]

Isn't it a bit hard for the factory to test properly a chip with such a high power requirements? Without a radiator? How many test beds can deliver ~200 W to the chip?

Each chip is mounted on its own PCB. If the failure level of the chip is high (many engines are faulty), they simply replace the PCB with the chip.

[Rampion]

I clearly remember how Knc stated they would have a working prototype by August. What do you say Bitcoinorama, will you be worried if this milestone is missed or not?

[timmmers]

Do you guys think it is sound that kncminer should offer a option to buy miners for November delivery, at a discount. For them to offer that now instead of when October is sold out. I would like to buy a miner, but I dont want it to be so close to November that I was a few days off from a cheaper price. I rather have a firm delivery date, like early November, than Late October with a higher price.

They didn't offer November delivery. They said a Nov price drop.
A firm delivery date. Good luck with that. Anywhere

I think all the Oct scheduled ones have a good chance to be early Nov...a 2 week delay is Usain Bolt performance  in this business.

[kingcoin]

Oh poo. The thread went chip-design while I was asleep. I missed all the fun 

I'll just point out that KNC have had plenty of time since tape-out to refine their testing procedures. Yes is is insane to omit the wafer test on production devices, but these are prototypes (for all they say about commercial product, it will take many more months to shakedown a volume-scale design, and quite right too, time is very much of the essence with bitcoin mining).

Running scan insertion and inserting signature analysis on the hash core will require adding gates. More than what can be handled by a simple metal fix. Hence it would require a number of masks being produced at a high cost and more time and resources.  But why?

Provided they have sorted out a test socket/interface for the packaged chips, so they can weed out the DOAs (dead shorts between power rails do not make it onto PCB's), and (rare) controller failures, they will be fine.

BGA fixtures are expensive and not always so easy to deal with. It not impossible even though the package here is probably not the cheapest one around. But why?

[kingcoin]

Anyway its the same problem CPU and GPU manufacturers have to deal with, so the solutions will be out there (I defer to 2112 et al, my chip industry insider knowledge is two decades out of date now).

They use scan insertion and test methodologies and don't expect the motherboard/graphics board manufactures to take care of the testing and yield problems.

[J35st3r]

I'll probably get out of my depth here and show my ignorance, but WTF, its a quiet day so ...

Running scan insertion and inserting signature analysis on the hash core will require adding gates. More than what can be handled by a simple metal fix. Hence it would require a number of masks being produced at a high cost and more time and resources.  But why?

But (as has been pointed out upthread), this is totally unneccessary as its already built in - a bitcoin hasher is almost the ultimate in a self-testing design. Probably better than most of the self test bolted onto typical commodity ASICs.

BGA fixtures are expensive and not always so easy to deal with. It not impossible even though the package here is probably not the cheapest one around. But why?

So how do all the manufacturers of CPU, GPU, FPGA and all the other high power devices manage then? Of course the test fixtures are expensive, have a long lead time, and need specific expertise to program  (my first job, 30 years back was writing test patterns a TekTest for the 5 micron ULA devices that were just coming onto the market). ORSoc will have outsourced this to the experts (one hopes).

[Phoenix1969]

KingCoin...Ever thought that maybe Markus knows something you/we don't?

If it were me wanting to know the answers to the techno-garble, I would call them.

My thinking is... either you trust them, or you don't; it's as simple as that.

[YipYip]

Check it out 2 to 1 payout for October delivery on BitBet

http://bitbet.us/bet/472/kncminer-will-deliver-asic-devices-before-october-1st/

Get in on the action...KNC are sure to deliver ...

(I may have a vested intrest in this blatant ad ..lolz)

[Ytterbium]

I'll probably get out of my depth here and show my ignorance, but WTF, its a quiet day so ...

Running scan insertion and inserting signature analysis on the hash core will require adding gates. More than what can be handled by a simple metal fix. Hence it would require a number of masks being produced at a high cost and more time and resources.  But why?

But (as has been pointed out upthread), this is totally unneccessary as its already built in - a bitcoin hasher is almost the ultimate in a self-testing design. Probably better than most of the self test bolted onto typical commodity ASICs.

Another way to put it - if a Hasher has errors, and those errors are undetectable, then those errors don't matter.  The system will mine away without any problems.

Although I did think of something - suppose a hasher had an error that caused it to miscount the number of trailing zeros, but only if they were above a certain threshold. in that case your miner would appear to be working working just fine - but it would be ignoring hashes above a certain difficulty.

It would mean that, if that difficulty were reached, everything would appear to be working perfectly and in fact you'd be getting full credit at a mining pool, but you would never find a block and the effect would be slightly reduced luck for the pool.

But that said, they're likely including known high diff block headers in their tests.

[J35st3r]

Another way to put it - if a Hasher has errors, and those errors are undetectable, then those errors don't matter.  The system will mine away without any problems.

Although I did think of something - suppose a hasher had an error that caused it to miscount the number of trailing zeros, but only if they were above a certain threshold. in that case your miner would appear to be working working just fine - but it would be ignoring hashes above a certain difficulty.

It would mean that, if that difficulty were reached, everything would appear to be working perfectly and in fact you'd be getting full credit at a mining pool, but you would never find a block and the effect would be slightly reduced luck for the pool.

But that said, they're likely including known high diff block headers in their tests.

I don't think that's an issue. The pool will check the submitted share and reject it if the hash is invalid for the worker difficulty level (target). For that matter cgminer et all will reject it as a hardware error before it even reaches the pool.

The issue of the miner mis-calculating the hash against the target just means it will fail to report a valid hash, ie just another duff core fault. I'm not sure about current ASICs, but the original fpgaminer (and ngzhang's icarus and the ztex's) all work to a target of 00000000 (difficulty one). It would make sense for next-gen asics to accept a variable target to reduce the comms bandwidth to cgminer, but that requires a driver coding change. I guess cklovas or kano could confirm this.

[EDIT] Having thought about it, you are correct. If a hasher is incapable of reporting a high difficulty share, but OK for difficulty 1 shares. then it will act as a leecher, submitting valid shares but never solving a block. Its a pretty unlikely failure mode though, though could be designed-in at the development stage (evil grin  ). That's the trouble with proprietary systems, no knowledge of exactly what is going on inside the black box (cue conspiracy theorists  )

[klee]

KnC is skipping a testing methodology used by the vast majority of the ASIC industry: chip level testing to make sure they detect faults at the chip tester and not in the assembled product, where they might not even know if the cause is a defective ASIC. Rather than stopping or sorting the bad chips at the fab, they pass them on to their customers. Again, which hopefully will receive miners above the later announced rate.

Once they receive the completed machines they'll go a burn-in and anything not within specs will simply be set aside and they'll move to the next device and send you that one. It's not a big deal yo.

Would be great if they sold these too cheaper!

[DPoS]

You guys forget, these guys have a modular design were we probably get 50% more hashing power thrown in the case to be safe... They are in fast mode and don't look like they are trying to maximize profits on design testing and guessing, just make profits on how fast they send us stuff above what we paided for... If we get 600Gh on a Jupiter I don't think they care they gave too much

[timmmers]

You guys forget, these guys have a modular design were we probably get 50% more hashing power thrown in the case to be safe... They are in fast mode and don't look like they are trying to maximize profits on design testing and guessing, just make profits on how fast they send us stuff above what we paided for... If we get 600Gh on a Jupiter I don't think they care they gave too much

How does that work? The design I saw had one module for the Mercury and 4 for a Jupiter. 100 G/hash each?
Unless they built in some extra as recently mentioned and it wasn't such a surprise as we may think?

[Bitcoinorama]

In case anyone's interested, interview with Austin and Becky Craig on the 'Life on Bitcoin' film project;

http://www.coindesk.com/life-on-bitcoin-is-a-challenge-for-newlywed-documentarians/

[dwdoc]

In case anyone's interested, interview with Austin and Becky Craig on the 'Life on Bitcoin' film project;

http://www.coindesk.com/life-on-bitcoin-is-a-challenge-for-newlywed-documentarians/

Pretty cool but no mention of KNC.