20nm asic miner technology | Bitcointalksearch.org

Bicknellski

hero member

Activity: 924

Merit: 1000

Quote from: gmaxwell on January 02, 2014, 10:05:09 PM

It's not just the size that counts.

The numbers you should care about are delivery time, BTC/GH, and W/GH. Technical details are interesting, but if they don't improve those three numbers they're not really relevant.

$ per GH/s that is the real important number along with delivery date especially when you are making that miner.

Hippievogel

newbie

Activity: 57

Merit: 0

Quote from: ProfMac on January 04, 2014, 02:15:30 PM

Quote from: gmaxwell on January 02, 2014, 10:05:09 PM

It's not just the size that counts.

The numbers you should care about are delivery time, BTC/GH, and W/GH. Technical details are interesting, but if they don't improve those three numbers they're not really relevant.

The correct unit is J/GH you may also say W/(GH/s)

maybe thats the correct unit of measurement but it is shit to say because you dont pay joule on your Energy bills. the pricing is in kwh.
W/(GH/s) is better

crazyates

legendary

Activity: 952

Merit: 1000

Quote from: brontosaurus on January 04, 2014, 07:08:29 PM

What everyone forgets is that an efficient 28nm design will, for sha256 at least, outperform an inefficient 20nm one. KNC, for example, did a conversion from an FPGA for their 28nm design - soon to be ported to 20nm, and if you calculate back from their watts/GH metric, you can see that their pipelines must contain about 35 - 50% more switching gates than normal. So what's the advantage in paying twice as much NRE and probably three times as much for the silicon? This to be expected from a conversion.

Likewise, in certain circumstances a full custom 40nm design could easily outperform a 28nm (or even 20nm) design done at gate level, but it's an order of magnitude more difficult to implement unless you really know what you're doing, and not many asic designers nowadays have the necessary skills.

Mmm could you imagine a BitFury chip or something similar at 20nm?

brontosaurus

sr. member

Activity: 441

Merit: 250

What everyone forgets is that an efficient 28nm design will, for sha256 at least, outperform an inefficient 20nm one. KNC, for example, did a conversion from an FPGA for their 28nm design - soon to be ported to 20nm, and if you calculate back from their watts/GH metric, you can see that their pipelines must contain about 35 - 50% more switching gates than normal. So what's the advantage in paying twice as much NRE and probably three times as much for the silicon? This to be expected from a conversion.

Likewise, in certain circumstances a full custom 40nm design could easily outperform a 28nm (or even 20nm) design done at gate level, but it's an order of magnitude more difficult to implement unless you really know what you're doing, and not many asic designers nowadays have the necessary skills.

smoothrunnings

hero member

Activity: 616

Merit: 500

Quote from: gamersglory on January 02, 2014, 08:12:10 PM

Unless someone can get Intel to use there 14nm process then yes 20nm is as small as it gets right now Wink

Intel's Labs have a 10nm chip, but that doesn't mean we'll see anytime soon!

ProfMac

legendary

Activity: 1246

Merit: 1001

Quote from: gmaxwell on January 04, 2014, 04:34:37 PM

Quote from: ProfMac on January 04, 2014, 02:15:30 PM

The correct unit is J/GH you may also say W/(GH/s)

But the pedantry doesn't extend to correcting $/GH? Or do your miners only perform a fixed number of hash operations during their lives? Tongue

I think that $/GH or BTC/GH are well formed units.

gmaxwell

staff

Activity: 4172

Merit: 8419

Quote from: ProfMac on January 04, 2014, 02:15:30 PM

The correct unit is J/GH you may also say W/(GH/s)

But the pedantry doesn't extend to correcting $/GH? Or do your miners only perform a fixed number of hash operations during their lives? Tongue

ProfMac

legendary

Activity: 1246

Merit: 1001

Quote from: gmaxwell on January 02, 2014, 10:05:09 PM

It's not just the size that counts.

The numbers you should care about are delivery time, BTC/GH, and W/GH. Technical details are interesting, but if they don't improve those three numbers they're not really relevant.

The correct unit is J/GH you may also say W/(GH/s)

xselam1988

jr. member

Activity: 59

Merit: 10

Quote from: crazyates on January 02, 2014, 10:34:02 PM

Quote from: gmaxwell on January 02, 2014, 10:05:09 PM

It's not just the size that counts.

Yeah yeah. That's what all the guys say. Wink

Lol J/K, but you really are right. It always amazed me how KNC was a 28nm ASIC, and yet the BitFury chips are 55nm and consumed less power (per GHs). What's the advantage of a small 28nm process if a competitors 55nm chip is more efficient, and runs cooler?

(yes, I know the actual answer of how BitFury is more efficient, but I still think it's funny).

yea thats what I think about,LOLOLOL

crazyates

legendary

Activity: 952

Merit: 1000

Quote from: gmaxwell on January 04, 2014, 12:55:00 AM

Quote from: crazyates on January 04, 2014, 12:18:10 AM

Quote from: gmaxwell on January 03, 2014, 09:13:43 PM

Quote from: crazyates on January 02, 2014, 10:34:02 PM

Lol J/K, but you really are right. It always amazed me how KNC was a 28nm ASIC, and yet the BitFury chips are 55nm and consumed less power (per GHs). What's the advantage of a small 28nm process if a competitors 55nm chip is more efficient, and runs cooler?
(yes, I know the actual answer of how BitFury is more efficient, but I still think it's funny).

I facepalm just a little bit every time I see someone brag that KNC was the first to 28nm. (hell, if you're going to be that liberal: there were people running on 28nm _FPGAs_ a LONG time before KNC shipped Tongue

)

Where did I say that KNC was the first to use 28nm? I don't even own any KNC hardware. I was just saying that 28nm SHOULD be more power efficient, but in the case of KNC vs BitFury, they're not.

I also face-palm when I agree with someone and say 'yea right, and don't you also hate it when' and they think I'm saying they're doing it. I'm not. The fact that bitfury is more efficient at 55nm is a concrete example of why bragging about 28nm is silly.

Ah, I gotcha now. Sorry I was a bit tired and out of it last night, so when you quoted me, I was like huh? Wink

gamersglory

member

Activity: 114

Merit: 10

No more Crypto in this world

Yes most of the good chips are on a 28nm die. 20nm is going to be very overpriced sense its newer process. but the reason you go for the smaller process is you get a better yield on the wafer.

gmaxwell

staff

Activity: 4172

Merit: 8419

Quote from: crazyates on January 04, 2014, 12:18:10 AM

Quote from: gmaxwell on January 03, 2014, 09:13:43 PM

Quote from: crazyates on January 02, 2014, 10:34:02 PM

Lol J/K, but you really are right. It always amazed me how KNC was a 28nm ASIC, and yet the BitFury chips are 55nm and consumed less power (per GHs). What's the advantage of a small 28nm process if a competitors 55nm chip is more efficient, and runs cooler?
(yes, I know the actual answer of how BitFury is more efficient, but I still think it's funny).

I facepalm just a little bit every time I see someone brag that KNC was the first to 28nm. (hell, if you're going to be that liberal: there were people running on 28nm _FPGAs_ a LONG time before KNC shipped Tongue

)

Where did I say that KNC was the first to use 28nm? I don't even own any KNC hardware. I was just saying that 28nm SHOULD be more power efficient, but in the case of KNC vs BitFury, they're not.

I also face-palm when I agree with someone and say 'yea right, and don't you also hate it when' and they think I'm saying they're doing it. I'm not. The fact that bitfury is more efficient at 55nm is a concrete example of why bragging about 28nm is silly.

crazyates

legendary

Activity: 952

Merit: 1000

Quote from: gmaxwell on January 03, 2014, 09:13:43 PM

Quote from: crazyates on January 02, 2014, 10:34:02 PM

Lol J/K, but you really are right. It always amazed me how KNC was a 28nm ASIC, and yet the BitFury chips are 55nm and consumed less power (per GHs). What's the advantage of a small 28nm process if a competitors 55nm chip is more efficient, and runs cooler?
(yes, I know the actual answer of how BitFury is more efficient, but I still think it's funny).

I facepalm just a little bit every time I see someone brag that KNC was the first to 28nm. (hell, if you're going to be that liberal: there were people running on 28nm _FPGAs_ a LONG time before KNC shipped Tongue

)

Where did I say that KNC was the first to use 28nm? I don't even own any KNC hardware. I was just saying that 28nm SHOULD be more power efficient, but in the case of KNC vs BitFury, they're not.

gmaxwell

staff

Activity: 4172

Merit: 8419

Quote from: crazyates on January 02, 2014, 10:34:02 PM

Lol J/K, but you really are right. It always amazed me how KNC was a 28nm ASIC, and yet the BitFury chips are 55nm and consumed less power (per GHs). What's the advantage of a small 28nm process if a competitors 55nm chip is more efficient, and runs cooler?
(yes, I know the actual answer of how BitFury is more efficient, but I still think it's funny).

I facepalm just a little bit every time I see someone brag that KNC was the first to 28nm. (hell, if you're going to be that liberal: there were people running on 28nm _FPGAs_ a LONG time before KNC shipped Tongue

)

crazyates

legendary

Activity: 952

Merit: 1000

Quote from: gmaxwell on January 02, 2014, 10:05:09 PM

It's not just the size that counts.

Yeah yeah. That's what all the guys say. Wink

Lol J/K, but you really are right. It always amazed me how KNC was a 28nm ASIC, and yet the BitFury chips are 55nm and consumed less power (per GHs). What's the advantage of a small 28nm process if a competitors 55nm chip is more efficient, and runs cooler?

(yes, I know the actual answer of how BitFury is more efficient, but I still think it's funny).

gmaxwell

staff

Activity: 4172

Merit: 8419

It's not just the size that counts.

The numbers you should care about are delivery time, BTC/GH, and W/GH. Technical details are interesting, but if they don't improve those three numbers they're not really relevant.

gamersglory

member

Activity: 114

Merit: 10

No more Crypto in this world

Unless someone can get Intel to use there 14nm process then yes 20nm is as small as it gets right now Wink

Silv0r

full member

Activity: 204

Merit: 100

Not only KnC! Hashblaster releases a 20nm ASIC too! Grin

(...not)

https://www.hashblaster.com/

vpasic

sr. member

Activity: 266

Merit: 250

not yet.
KNC Neptune is only one announced and release date is unknown.

ayayay

sr. member

Activity: 260

Merit: 250

Am I correct to say this is the latest asic technology? any word of a newer more advanced asic coming down the road and if yes when?

Topic: 20nm asic miner technology (Read 3422 times)