Topic: Swedish ASIC miner company kncminer.com - page 382. (Read 3050076 times)

unfortunately you are likely correct on the above....man knc had quite the rapid slide to the other side so to speak...$$$ talk we are irrelevant now off to do the BFL way...find 'more' newbies..hope they don't read bitcoin talk or google before they buy knc stuff

Searing
 

My new worthless prediction (I'm batting .000, lol)...

After all of the 5 piece preorders are delivered (or maybe before), KNC will introduce a new product that uses more chips but is much more efficient. They're going to have to do this to compete with Spondoolies.

IMO, they are only shipping 5 board Neptunes because this was the cheapest configuration to get to the 3THs target. But my guess it isn't what they had originally planned nor what is optimal for the new ASIC.

There is some very compelling evidence that supports this hypothesis. But the most glaringly obvious is the single PCI power connector. They're not so stupid to design it to handle 420w. But they were compelled to squeeze as much hash out of each chip that they threw away common sense.

Remember "under promise, over deliver"? Yeah, not to so much anymore. The new motto... "We deliver dangerous crap because we're cheap bastards."

Ofc there is since the industry never stands still, only Intel (however their 14nm launches has been delayed) is currently moving over anything major to 14nm and they are usually 12-18 months ahead of the rest of the industry when it comes to production. Usually when foundries starts talking about a node you can be sure that their mass production is still 12 months+ away. Samsung will probably start their 14nm production next year, good luck getting access to their fabs for something like btc mining however. 20/16nm at TSMC/Globalfoundries will be as good as it's get for 2014/2015. The only opening I can see in that time frame iirc is that GF struck some kind of deal with Samsung on 14nm tech utilization, so they could be converting sooner rather than later. However converting a fab is not something that you do overnight exactly.

Also doesn't change the fact that for every new process node the costs have increased exponentially for RND while the gains in power reduction/performance increases have been decreasing. It used to be that you should jump on a new process node asap since you could offer a product with better performance metrics at a lower price, these days however it makes more sense to wait until the new tech has matured unless performance/density are your only concerns.

The whole argument has been about why you should go 20nm instead of 28nm when the NRE costs are higher while the cost  for performance x with efficiency y is the same on both processes. Currently it costs more to develop for 20nm than 28 while you gain almost nothing in terms of performance/$, this will change with time however when the process matures and price comes down for manufacturing.

However if you like KNC simply make someone else pay for your NRE costs it made a lot more sense to jump directly to 20 vs 28nm. They could have offered the same kind of performance at the same price with a new better 28nm design and delivered it sooner, they wouldn't have filled up their orders as easily however since "OMG 20NM SO GUD MUST HAVE" mentality. In the long term KNC will be the winners since costs of 20nm wafers will go down faster than the cost of 28nm so they will be able to fill their datorhall(s) with higher density chips paying less/GH, and we paid for it pretty much.

My new worthless prediction (I'm batting .000, lol)...

After all of the 5 piece preorders are delivered (or maybe before), KNC will introduce a new product that uses more chips but is much more efficient. They're going to have to do this to compete with Spondoolies.

IMO, they are only shipping 5 board Neptunes because this was the cheapest configuration to get to the 3THs target. But my guess it isn't what they had originally planned nor what is optimal for the new ASIC.

There is some very compelling evidence that supports this hypothesis. But the most glaringly obvious is the single PCI power connector. They're not so stupid to design it to handle 420w. But they were compelled to squeeze as much hash out of each chip that they threw away common sense.

Remember "under promise, over deliver"? Yeah, not to so much anymore. The new motto... "We deliver dangerous crap because we're cheap bastards."

Without knowing how many millions they will spend on wafers, thats not a conclusion you can draw. What you can probably safely conclude, is that they intend to order a shitload of wafers. Lets make some guesses. Lets say the mask set cost $5M and lets say per TH cost scales linearly with node/transistor size. 20²/28²~=0.5. So if they intend to order >$5M worth of wafers, it  makes sense excluding even the improved power efficiency.

Does anyone know if they can tell if they really do have 20nm chips as opposed to 28nm if they are going to ship you 5 boxes? What if they shipped Nov jups at 700G underclocked to conserve power and put them in 5 shiny boxes that would equal 3T.

That still doesn't justifies throwing a couple of millions $ for the R&D and for the mask.

Stop drinking Inaba's coolaid. 20nm structured asics (altera hardcopy etc) dont even exist yet.

Now given how new/experimental 20nm still is, I am sure its a standard cell implementation which will never be optimally efficient. But keep in mind that smaller process nodes offer 2 advantages; first is power efficiency, the second is transistor density which directly translates in to cost per TH.

NRE might have been sky high, but unless yields are in the toilet, 20nm chips should be cheaper to produce than 28nm chips per TH.

[The fact that KNC are running these chips at 0,7W/GH could mean a number of things.
1. Another hard copy rather than custom design (worse density/effiency)

Also, 0.7W/GH on 20nm is a waste of Silicone, you can get that on 40nm or even 55nm.... It just makes for unreasonably high NRE.

Be sure to stack dry kindling on and around those PCI cables to keep them cool. If you don't have any kindling on hand, oily rags will do in a pinch. Also be sure to have a full gas can venting into the surrounding area.

Isn´t KNC late on their shipment?

Didn´t they say the "customer appreciation" batch would ship one month before end of Q2?

Also, 0.7W/GH on 20nm is a waste of Silicone, you can get that on 40nm or even 55nm.... It just makes for unreasonably high NRE.