CoinTerra announces its first ASIC - Hash-Rate greater than 500 GH/s

[Xian01]

Cartels wont work for bitcoin miners.

 How can ghash.io / Bitfury's recent turn be described as anything other than approaching behavior similar to a cartel ?

[Puppet]

did i say 12 PH/s?

No. Did I say you said it? No.
Please read carefully.

Oh, yes i do think that the pace will pick up. Who doesnt? Nevertheless its still bullshit to assume "hundreds of PH/s coming online next 6 months".

Sigh.

I never assumed 100s of PH wil be coming online in 6 months, please look up the definition of a hypothetical.
Moreover, the hypothetical didnt state that 100's of PH would be online in 6 months, merely that during the next 6 months we would see them being sold and being deployed. Note the present continuous.
Finally I am saying the hypothetical is at least plausible, and more plausible than the other hypothetical I gave of only a few dozen PH. Considering you agree at the current rate without anyone known to be shipping 28nm gear, we could already be on track to add 50PH its hard to see how you are arguing against the plausibility of that number multiplying considering about 10 vendors are expected to start shipping latest gen miners in the next months and currently there are none (known).

[VolanicEruptor]

http://cointerra.com/engineering-update-asic-bringup-diagnostic-testing/

new update for Jan 7th...

[Minor Miner]

http://cointerra.com/engineering-update-asic-bringup-diagnostic-testing/
new update for Jan 7th...

Very exciting.   Even more exciting that an asic company is updating people in an effective manner.   Great Job CT and looking forward to more progress reports.

[VolanicEruptor]

"Our first round of tests were focused on characterizing one of the four dies on a single GoldStrike I ASIC. The engineers slowly increased the number of active hashing pipelines and monitored their performance. At the end of this round of tests we achieved our target of 132GH/s on a single die.

The next steps will be to characterize the remaining dies on the ASIC and test multiple samples to verify consistent performance and efficiency. We will continue to post engineering updates reporting progress so check back with us soon."

So by the sounds of it, the chips are capable of reaching their targeted hash rate, and now they're going to start testing the entire batch?  

[willphase]

Interesting update.  It seems they have gone from using just three of their dies to assuming they will use all four.

In the FPGA video from last year - http://www.youtube.com/watch?v=jSpNZMYunls  it was stated:

"120 cores per die for a total of 360 cores.  Nominal 500 Gh/sec up to 720 Gh/sec" and "1.4Ghz" (implied top speed for 720 Gh/sec of 2Ghz)

so that was 3 x 120 x 1400 = 504,000 Gh per chip, with four chips = 2.016 Th/sec which meets their specification, back last year.

Now, it seems they are running at only 1.1 Ghz:

120 x 1100 = 132 Gh/s on each die (as said on their post) x 4 = 528 Gh/s per chip, with four chips = 2.1 Th/sec

It's interesting to see their plans from using just three dies have changed to underclocking from their nominal 1.4Ghz to 1.1Ghz and using four dies.  Possible their yield from the first wafer was higher than they expected (they expected to only be able to use three dies), or it's possible that the power profile of running at a lower clock speed were better.

Either way, this is probably good news for cointerra customers, given the potential to clock each of the four cores at 1.4Ghz would give a projected yield of each TerraMiner of 1400 * 4 * 4 * 120 = 2.68Th - and if they are able to clock each die at the 2 Ghz they said in their FPGA video, the power systems can support it, and the yield on the dies is good, then a possible top speed of 2000 * 4 * 4 * 120 = 3.84 Th/sec

All this is speculation, of course.

Will

[VolanicEruptor]

Thanks Will, you made me smile a bit  

Perhaps they have known this for some time, and that could be why they aren't lowering their prices on the may batch from $6k

[Carlton Banks]

It's interesting to see their plans from using just three dies have changed to underclocking from their nominal 1.4Ghz to 1.1Ghz and using four dies.  Possible their yield from the first wafer was higher than they expected (they expected to only be able to use three dies), or it's possible that the power profile of running at a lower clock speed were better.

Either way, this is probably good news for cointerra customers, given the potential to clock each of the four cores at 1.4Ghz would give a projected yield of each TerraMiner of 1400 * 4 * 4 * 120 = 2.68Th - and if they are able to clock each die at the 2 Ghz they said in their FPGA video, the power systems can support it, and the yield on the dies is good, then a possible top speed of 2000 * 4 * 4 * 120 = 3.84 Th/sec

All this is speculation, of course.

Will

Judging from what's said to have been ordered, I'm thinking the power throughput might be a problem there (for 3.84 TH/s). But that's not to say there won't be some nice overclocking headroom.

[Minor Miner]

the yield on the dies is good, then a possible top speed of 2000 * 4 * 4 * 120 = 3.84 Th/sec

Theoretically, a porsche GT3 can do the "cannonball run" in 14 hours and 45 minutes.   But it hasn't.   But it could.   I will be happy cruising along with my unit at 2TH/s.   Very happy.

[VolanicEruptor]

the yield on the dies is good, then a possible top speed of 2000 * 4 * 4 * 120 = 3.84 Th/sec

Theoretically, a porsche GT3 can do the "cannonball run" in 14 hours and 45 minutes.   But it hasn't.   But it could.   I will be happy cruising along with my unit at 2TH/s.   Very happy.

An engine has moving pistons, constantly moving against the cylinder walls..
I think if you can overclock anything, and its stable, its not going to wear out the same as a car..

[Epoch]

An engine has moving pistons, constantly moving against the cylinder walls..
I think if you can overclock anything, and its stable, its not going to wear out the same as a car..

I guess you're not familiar with electromigration, then.

[willphase]

Judging from what's said to have been ordered, I'm thinking the power throughput might be a problem there (for 3.84 TH/s). But that's not to say there won't be some nice overclocking headroom.

it's also distinctly possible that for each chip they will tune them to run at different Ghz depending on how many of the hashing cores and dies are actually working - having a range of 1.1 Ghz to 1.4 Ghz (and potentially even up to 2 Ghz) gives the Cointerra engineers considerable latitude to have their firmware automatically tweak each chip to get to the target 2 Th for the entire unit.

Will

[VolanicEruptor]

An engine has moving pistons, constantly moving against the cylinder walls..
I think if you can overclock anything, and its stable, its not going to wear out the same as a car..

I guess you're not familiar with electromigration, then.

I still think an engine would wear out a lot faster, but I'm definitely not an expert

[plato14]

anyone entertain me as to how long testing will take?

[Epoch]

An engine has moving pistons, constantly moving against the cylinder walls..
I think if you can overclock anything, and its stable, its not going to wear out the same as a car..

I guess you're not familiar with electromigration, then.

I still think an engine would wear out a lot faster, but I'm definitely not an expert

That makes 2 of us.

[Praeconium]

Hey guys, could anyone give advice on buying March delivery batch @9btcs in terms of whether its even considerable or should I move on?

If you're contemplating then look at this chart that I have kept up.

It compares available hardware vs possible network increase.

Included is an exponential network growth at 20% and 30%.

However as far as network difficulty forecast is concerned anything greater then 2 - 3 months out is highly speculative.

Make a copy for yourself and key in your own details and existing setup.

Here are the return results if the network increases an average of 20% every 12 days.
Or by end of March 4380085631 difficulty.

https://docs.google.com/spreadsheet/ccc?key=0AmeuPljmUNHCdEpqX2RmMDFwemJyLURVUWFtZ3J3aGc&usp=sharing

Wow, awesome table!Thanks

Hey guys, could anyone give advice on buying March delivery batch @9btcs in terms of whether its even considerable or should I move on?

The difficulty is getting obliterated.  Every coin is getting more and more precious.  I personally wouldn't risk it on these delayed companies.

Ok, thanks

[newguy05]

wish they showed a screenshot of the laptop hashing at the stated rate. It's nice and all to see those photos of the hardware, but the key is seeing photo of the laptop screen running at the stated rate and how long the duration is.

[JoseSan]

wish they showed a screenshot of the laptop hashing at the stated rate. It's nice and all to see those photos of the hardware, but the key is seeing photo of the laptop screen running at the stated rate and how long the duration is.

I don't see why you can't trust them to state their testing hash rate truthfully. It's still text on a screen whether it's in a blog post or a cgminer screenshot; either can be faked. Notice they're forgoing the DC-DC units? Functional operation of the chips is fantastic, but it's a symphony of parts that need to be tested, and I suspect the final concert will give a more formal presentation of things like cgminer grabs. Quite frankly working, packaged silicon is really encouraging news.

[testerx]

wish they showed a screenshot of the laptop hashing at the stated rate. It's nice and all to see those photos of the hardware, but the key is seeing photo of the laptop screen running at the stated rate and how long the duration is.

I highly doubt that cointerra would lie about the chip bringup so far, and what they've reported is obviously very conservative so it's not something that would make sense to lie about lol.

[testerx]

Judging from what's said to have been ordered, I'm thinking the power throughput might be a problem there (for 3.84 TH/s). But that's not to say there won't be some nice overclocking headroom.

Power and heat, as it is these are chips that will be putting out massive amounts of heat.  It would be pretty interesting to get these going under phase or something like that to see just how much you could squeeze out of them but the amount of heat they're going to be putting out is going to be a nightmare to cool with anything other than liquid.  Still...if someone were to make a beefy custom waterblock for chilled liquid cooling with some kind of inert liquid and a custom loop it'd certainly be very interesting.  Would be very hard to run anything like that in a racked/data center type setup though so it's not really practical if you have more than one lol.