Yeah I am planning on building some 2 board m10's down the road as that gear slowly falls apart.
an s17 with an apw5 type psu designed like the old s7ln would be nice.
have it do 26th / 1100 watts in loud mode
and 20th/770 watts in quiet mode
my s17 does about 40th and 1540 watts in low speed.
so a 20th pulling 770 watts could be done.
they would need a psu that has a larger fan so it would not be real loud.
Right now I have
canaan avalon a1041 quietest of all at 31th setting pulling 2065 watts low speed
bitmain s17 decent sound at 40th setting pulling 1555 watts low speed
bitmain t17 one speed kind of loud
innosilicon t3 t39 Many settings but fans always run 100% = shit firmware and pretty loud
whatsminer m20s one speed 68-69 th loudest miner I ever owned
Topic: Next jump foward (Read 291 times)
I think the smartest move would be to make a miner as efficient as the S17 but half hashrate, thus half consumption and easier ventilation.
Two "halfS17"s that consume half and make half are as good as one. And that one is already taking two power plugs and 4 fans anyway...
But i guess Bitmain was starting to look puny next to the M20S concentrated power house so they went that way.
Efficiency, yes, but there is no need to concentrate that much power into a single unit.
I still think something like an S17 with just two or one board would be interesting to try, perhaps with a scrolling fan...
Two "halfS17"s that consume half and make half are as good as one. And that one is already taking two power plugs and 4 fans anyway...
But i guess Bitmain was starting to look puny next to the M20S concentrated power house so they went that way.
Efficiency, yes, but there is no need to concentrate that much power into a single unit.
I still think something like an S17 with just two or one board would be interesting to try, perhaps with a scrolling fan...
Due to the cost of 7nm nodes or lower,these process will be reserved for GPUs and CPUs for a while. Consider the number of chips required for an ASIC is in the hundreds.
You can put more transistors per chip, but then you run into thermal issues. We would need to move from air cooling to liquid cooling.
I think we are bottle-necking in terms of efficiency gains. Hopefully I wont have to buy a new miner for a while.
You can put more transistors per chip, but then you run into thermal issues. We would need to move from air cooling to liquid cooling.
I think we are bottle-necking in terms of efficiency gains. Hopefully I wont have to buy a new miner for a while.
If I'm reading things right, the S15 uses a 72-chip single string per board. That's why the PSU runs around 19V but would only need to source ~25A per phase (assuming it uses separate rails per board, which is best for per-board tuning but also unlikely given Bitmain's cheap-out design philosophy).
You can still do a board-level regulator with narrow strings and get decent efficiency (95-97% conversion) depending on the Vin/Vout ratio and most importantly the current draw.
S9 with a 3-wide string was asking close to 50A from a single phase, and resistive losses in FETs and copper are proportional to current squared, so by shifting to a two-phase design one could halve those losses. But Bitmain likes to build cheap and sell high and cares less about longevity so that never happened.
Dedicated PSU can give you a 5-10% efficiency boost depending on what implementation you're replacing, but it also comes at a cost - you're now locked in to their hardware and their replacement mechanism.
You can still do a board-level regulator with narrow strings and get decent efficiency (95-97% conversion) depending on the Vin/Vout ratio and most importantly the current draw.
S9 with a 3-wide string was asking close to 50A from a single phase, and resistive losses in FETs and copper are proportional to current squared, so by shifting to a two-phase design one could halve those losses. But Bitmain likes to build cheap and sell high and cares less about longevity so that never happened.
Dedicated PSU can give you a 5-10% efficiency boost depending on what implementation you're replacing, but it also comes at a cost - you're now locked in to their hardware and their replacement mechanism.
This is key. I have a t17 and a s17 either one can use the psu included. But those are dead if the psu dies and they do not sell replacement psu’s so if you burn the psu you lose 50th for about a month.
I have been getting Whatsminer gear more then bitmain because I was able to buy psu’s as spares.
What good is the most efficient gear in the world (s17?)
If you can not get spare parts for it?
Now project to a
H P H
H P H
S21. Say 25 watts a th and 120th. If the controller or the psu dies. You have 120Th dead in the water.
Not to mention shipping it would be hard as it would weigh quite a bit.
The s17 is a good design.
They simply need to sell spare parts for it.
If I could get 10 s17 units with 3 spare psu and 2 spare controller I would get them.
But they do not sell like that.
If I'm reading things right, the S15 uses a 72-chip single string per board. That's why the PSU runs around 19V but would only need to source ~25A per phase (assuming it uses separate rails per board, which is best for per-board tuning but also unlikely given Bitmain's cheap-out design philosophy).
You can still do a board-level regulator with narrow strings and get decent efficiency (95-97% conversion) depending on the Vin/Vout ratio and most importantly the current draw.
S9 with a 3-wide string was asking close to 50A from a single phase, and resistive losses in FETs and copper are proportional to current squared, so by shifting to a two-phase design one could halve those losses. But Bitmain likes to build cheap and sell high and cares less about longevity so that never happened.
Dedicated PSU can give you a 5-10% efficiency boost depending on what implementation you're replacing, but it also comes at a cost - you're now locked in to their hardware and their replacement mechanism.
You can still do a board-level regulator with narrow strings and get decent efficiency (95-97% conversion) depending on the Vin/Vout ratio and most importantly the current draw.
S9 with a 3-wide string was asking close to 50A from a single phase, and resistive losses in FETs and copper are proportional to current squared, so by shifting to a two-phase design one could halve those losses. But Bitmain likes to build cheap and sell high and cares less about longevity so that never happened.
Dedicated PSU can give you a 5-10% efficiency boost depending on what implementation you're replacing, but it also comes at a cost - you're now locked in to their hardware and their replacement mechanism.
Ya got a couple right:
The 2 biggest reasons are the dedicated PSU which eliminates the 2nd stage Vcore regulators saving around 10% or more on power and Competition which led to the publicly released improvements in Firmware. That said, the FW improvements should have been there all along...
Samsung has been building chips for decades including 14nm ones for miners back when BM and others were still at 22 & 28nm. Yeah BM got their 7nm chip right but at still horrible yields and Canaan proved with their 7nm chip that smaller is certainly NOT always better. More chips per string and dedicated PSU's goes back to early Bitfury days, the PCB gerber files they released to Open Source a few years ago were for 24 & 48-chip strings. 1st mass produced FinFET's were 90nm chips (DRAM) from Samsung in 2004.
The 2 biggest reasons are the dedicated PSU which eliminates the 2nd stage Vcore regulators saving around 10% or more on power and Competition which led to the publicly released improvements in Firmware. That said, the FW improvements should have been there all along...
Samsung has been building chips for decades including 14nm ones for miners back when BM and others were still at 22 & 28nm. Yeah BM got their 7nm chip right but at still horrible yields and Canaan proved with their 7nm chip that smaller is certainly NOT always better. More chips per string and dedicated PSU's goes back to early Bitfury days, the PCB gerber files they released to Open Source a few years ago were for 24 & 48-chip strings. 1st mass produced FinFET's were 90nm chips (DRAM) from Samsung in 2004.
I am of the opinion we won’t get much better then the s17pro.
On low power setting miner does 36 or 37 watts a th.
The dual tube design is easier to cool.
The dual plug psu pulling 2000 to 3000 watts is possible improve on.
Note using c20 receptacles the psu could pull 4400 watts.
So maybe a four tube setup
H -p - H
H -p - H
And have the psu supply 24dc
It could be wider using 120mm fans or 100mm fans
The hash boards could use 140mm fans.
The downside is heavy heavy gear.
And hard to ship it.
And dead psu means four dead tubes vs 1 dead tube. S9 or 2 dead tube s17.
On low power setting miner does 36 or 37 watts a th.
The dual tube design is easier to cool.
The dual plug psu pulling 2000 to 3000 watts is possible improve on.
Note using c20 receptacles the psu could pull 4400 watts.
So maybe a four tube setup
H -p - H
H -p - H
And have the psu supply 24dc
It could be wider using 120mm fans or 100mm fans
The hash boards could use 140mm fans.
The downside is heavy heavy gear.
And hard to ship it.
And dead psu means four dead tubes vs 1 dead tube. S9 or 2 dead tube s17.
We can say that around 40-50w/TH is now the norm for efficiency for new mining equipment.
This is double the efficiency of the S9, released 3 years ago. The efficiency improvements have been brought on by (Correct me if i'm wrong):
Increased competition
Dedicated PSUs
10nm and 7nm chip design
Samsung building chips
FinFET
Dual tube design
More chips per string
Better firmware options include different efficiency modes.
Higher total power draw, less losses. (2kw+ per miner)
ASIC Boost
I'm curious what other improvements can be made for the next generation of miners to bring efficiency to double again to 20-25w/th in another 3 years. I don't see much more gains in chip technology. 7nm has been problematic; and this is expensive and can take a long time with diminishing returns.
I could see a 347v or 277v miner in the future. Perhaps adding more tubes, 4 tube design for example. Not sure what else can be improved upon.
This is double the efficiency of the S9, released 3 years ago. The efficiency improvements have been brought on by (Correct me if i'm wrong):
Increased competition
Dedicated PSUs
10nm and 7nm chip design
Samsung building chips
FinFET
Dual tube design
More chips per string
Better firmware options include different efficiency modes.
Higher total power draw, less losses. (2kw+ per miner)
ASIC Boost
I'm curious what other improvements can be made for the next generation of miners to bring efficiency to double again to 20-25w/th in another 3 years. I don't see much more gains in chip technology. 7nm has been problematic; and this is expensive and can take a long time with diminishing returns.
I could see a 347v or 277v miner in the future. Perhaps adding more tubes, 4 tube design for example. Not sure what else can be improved upon.
Jump to: