Quantum effects have been an issue for semiconductor design for over 20 years now - first started getting noticeable in the 1990s.
Every "smaller" node they have been MORE of an issue.
No, the boards in the T9 are NOT the "bad S9 boards" - they have fewer chips, they are a different BOARD.
I do suspect Bitmain does part-level testing to decide which chips go in which boards though - that's been standard practice in the industry for DECADES, to match performance as closely as possible of the parts on a specific board when those parts have to work together.
I was doing that on MRF 455 amplifier transistors in the early 1980s for use in push-pull amps, and it was a widespread old technique THEN.
Topic: Bitmains Plan to sell anything? Am I wrong? (Read 607 times)
yep, they mix chips in 14-16nm GPUs too and some software even indicates the %.
Silicon lottery.
Silicon lottery.
Not to argue but here goes.
the 14/16 chip size has issues even with intel.
Basic reason is an electron has a width set by God or the universe or whatever you believe.
Therefore the smaller a chip gets the tighter it is for the electrons to be squeezed into behaving correctly. Ie the etchs in the chip are so freaking small you get crossover or leaking or bleeding of electrons into the wrong spots.
so 14/16 is getting close to the smallest size possible for electrons to fit.
yeah 10 then 7 then done. (we need to be able to go beyond electrons)
So it is not that chips are poorly made it is that it is hard to make chips this small.
That said The shit chip pile may be:
30% fully dead
30% so so
30% good
10% great
this is for 1387 chips
maybe 1385 chips were
10% dead
30% so so
30% good
30% great.
So bitmain is trying to come up with ways to use the 14/16 nm chips
Just wait for 3 years when the 10nm come out they will be worse then these.
the 14/16 chip size has issues even with intel.
Basic reason is an electron has a width set by God or the universe or whatever you believe.
Therefore the smaller a chip gets the tighter it is for the electrons to be squeezed into behaving correctly. Ie the etchs in the chip are so freaking small you get crossover or leaking or bleeding of electrons into the wrong spots.
so 14/16 is getting close to the smallest size possible for electrons to fit.
yeah 10 then 7 then done. (we need to be able to go beyond electrons)
So it is not that chips are poorly made it is that it is hard to make chips this small.
That said The shit chip pile may be:
30% fully dead
30% so so
30% good
10% great
this is for 1387 chips
maybe 1385 chips were
10% dead
30% so so
30% good
30% great.
So bitmain is trying to come up with ways to use the 14/16 nm chips
Just wait for 3 years when the 10nm come out they will be worse then these.
Could be wrong...
I don't see much difference in comparison with prior (S7) story.
Initial batches of S7 were 0.25 J/H, less watts for more hash, later batches were 0.285
In T9/S9 case the difference is even larger in %, hence the name change.
as far as stability, it remains to be seen.
Obviously, T9 are operating at higher freqency, which can theoretically make them more stable, not less, but it is an open question.
I wouldn't buy T9 because it is not power-efficient enough for me and i don't want to buy more PSUs ; some S9 work fine on EVGA 1300 or two per 2880 PSU and I want to keep it this way.
I don't see much difference in comparison with prior (S7) story.
Initial batches of S7 were 0.25 J/H, less watts for more hash, later batches were 0.285
In T9/S9 case the difference is even larger in %, hence the name change.
as far as stability, it remains to be seen.
Obviously, T9 are operating at higher freqency, which can theoretically make them more stable, not less, but it is an open question.
I wouldn't buy T9 because it is not power-efficient enough for me and i don't want to buy more PSUs ; some S9 work fine on EVGA 1300 or two per 2880 PSU and I want to keep it this way.
Okay so looking to see if I'm nuts.
But I think I understand how bitmain is doing things... S9 T9 anyway
So they launch the s9 and it's OKAY..
Then at some point they start making boards and I'm assuming they then bench test the boards and catagorize them.
As I've seen no real design difference in the different batches I think this is what they are doing.
Seems the chip manufacturing is garbage. Good design, cheap build.
So they get 3 blades running at peak performance, slam them together and launch a 14TH batch.
Taking the lesser blades and sticking them together to put out 11 and 12th batches etc..
This would explain why they have freq locked the miners, the lower TH ones are only stable where they are set.
Now it would seem they put the worst of their boards into the T9. And threw those at people.
It screams crap quality and the failure rates at 15% or greater proves this.
Why are people scrambling to buy these T9s...
Am I the only one thinking this? anyone else noticed this pattern?
But I think I understand how bitmain is doing things... S9 T9 anyway
So they launch the s9 and it's OKAY..
Then at some point they start making boards and I'm assuming they then bench test the boards and catagorize them.
As I've seen no real design difference in the different batches I think this is what they are doing.
Seems the chip manufacturing is garbage. Good design, cheap build.
So they get 3 blades running at peak performance, slam them together and launch a 14TH batch.
Taking the lesser blades and sticking them together to put out 11 and 12th batches etc..
This would explain why they have freq locked the miners, the lower TH ones are only stable where they are set.
Now it would seem they put the worst of their boards into the T9. And threw those at people.
It screams crap quality and the failure rates at 15% or greater proves this.
Why are people scrambling to buy these T9s...
Am I the only one thinking this? anyone else noticed this pattern?
Jump to: