[BitBet] BitFury's ASIC WILL WORK WITH POWER < 1 W / GH/s

[Sine(X)]

Bit of a weird thread to bump, but I'll believe it when I see it. The last 3 chips have never been independently verified or even seen outside of a super farm.

Considering they make up nearly 1/3rd of the network alone .... thats evidence enough.

Of power claims? Its only evidence that they have money, and $60M buys you a lot of miners and a lot of PR.

Dogie, shame on you Bitmain pays you also for PR
I remember times when Bitmain and Spondoolies were competitors on B2C market and Dogie pushed tons of shit on Spondoolies, Guy Corem answered Dogie.. and there was again and again.. What was it? PR
Regarding "buys a lot of miners" - do you have some inside? Did BitFury bought their miners? Whom? Bitmain?? It would be a sensasion 

[Sine(X)]

Bit of a weird thread to bump, but I'll believe it when I see it. The last 3 chips have never been independently verified or even seen outside of a super farm.

Considering they make up nearly 1/3rd of the network alone .... thats evidence enough.

Of power claims? Its only evidence that they have money, and $60M buys you a lot of miners and a lot of PR.

Dogie is bang on the money here, this announcement by Bitfury is simply an exercise in misdirection and PR bullshit. They have had several attempts to make a really high efficiency chip and obviously don't know the recipe. It's easy to say 'my chip can do xJ/GH/sec' without revealing key factors like how big it is (very important), what voltage it runs at and the clock speed at the rated efficiency. In all fairness to Bitmain with their BM1385 they have fullfiled most of these conditions, but no mention of the die size as yet.

To believe that Bitfury can go from an alleged 0.2 J/GH/sec in 28nm to 0.05 in 16nm is nonsense, unless they have very specific measurement points that bear no resemblance to actual operating conditions in practice. 16nm is not a magic bullet - at best you will get half of the power consumption of an existing 28nm design and maybe 2.5 - 3 x the transistor density, very much depending on your design and memory content. KNC tried this same trick with their solar chip, presumably to try to dissuade their competitors and this is more of the same.

The only company that has ever been totally transparent in what they claim to be able to achieve in terms of power efficiency AND tell you how they actually achieved it (eg domino logic) is Spondoolies, and that's largely due to the collective industrial experience of their people and their attention to detail. I'm pretty damn sure that if they were still selling to the public they could come out with all kinds of data about what they could achieve and how they could do it, but they have wisely stayed out of this game and left it to the bullshitters.

By the way, anyone every looked at the size of Bitfurys various 'boards' - they have an awful lot of people to pay......what on earth do they all do? Is this some kind of weird vanity project?

brontosaurus, I always like to read your posts.
A little tidbit I can share (not under any kind of NDA):
BitFury (Valery Nebesny) attempted to do async, domino logic 55nm 5 engine scrypt chip and failed.
Here is the chip floor plan:


I really liked George BS statement: https://twitter.com/BitfuryGeorge/status/639064051612524544

There are two all important parameters in Bitcoin ASICs:
GH/mm^2 - Capex and J/GH - Opex

With extreme design techniques, sub 0.1 J/GH is possible also on 28nm processes, the question is what will be the GH/mm^2 - miner cost.

Whenever BitFury gives efficiency numbers (J/GH), they never give the all important GH/mm^2 numbers.

When they're selling hosted hash-rate of their 28nm machines to their customers, they claim the machine efficiency is 0.35 J/GH

Interesting, I didn't hear about BitFury's scrypt miners.. Were there any announcements? I remember epic fail with scrypt miners from Bitmain, when guys had pre-orders on their site and then said "ooops" 

[Guy Corem]

...
Interesting, I didn't hear about BitFury's scrypt miners.. Were there any announcements? I remember epic fail with scrypt miners from Bitmain, when guys had pre-orders on their site and then said "ooops" 

No, as I wrote it was a failed ASIC.

[dogie]

Bit of a weird thread to bump, but I'll believe it when I see it. The last 3 chips have never been independently verified or even seen outside of a super farm.

Considering they make up nearly 1/3rd of the network alone .... thats evidence enough.

Of power claims? Its only evidence that they have money, and $60M buys you a lot of miners and a lot of PR.

Dogie, shame on you Bitmain pays you also for PR
I remember times when Bitmain and Spondoolies were competitors on B2C market and Dogie pushed tons of shit on Spondoolies, Guy Corem answered Dogie.. and there was again and again.. What was it? PR
Regarding "buys a lot of miners" - do you have some inside? Did BitFury bought their miners? Whom? Bitmain?? It would be a sensasion  

I haven't been contracted by Bitmain since May, so no they do not "pay me for PR". I've also generally been a supporter of Bitfury, who had the top spot and a near perfect score in my guide for a long long time. Guy claimed that the rating criteria specifically targeted them when it just didn't as it was applied on BFL first, and they still had 88/100 score. The rating criteria are designed to protect buyers, and their claims that they'd be able to finance the next generation without preorders or investment did indeed turn out to be false.

Back on topic, all I said about Bitfury is that no one has actually ever tested their miners who is able to speak about it, and so its hard to really take their repeated claims seriously. If you think I'm paid for PR, why don't you talk to Bitfury's actual PR company or their Head of Global PR.

[tonycamp]

was not bit fury a thing of the cex site too they where using its technology no? i mean they use doges ltc btc and usd and cards and bank accounts and do cloud but also had a section for fury

[punin]

Back on topic, all I said about Bitfury is that no one has actually ever tested their miners who is able to speak about it, and so its hard to really take their repeated claims seriously. If you think I'm paid for PR, why don't you talk to Bitfury's actual PR company or their Head of Global PR.

Well, I actually did watch the whole testing process very closely and can confirm that 28nm figure to be empirically derived   (I hope my word still has some value on these forums)

[Sine(X)]

Bit of a weird thread to bump, but I'll believe it when I see it. The last 3 chips have never been independently verified or even seen outside of a super farm.

Considering they make up nearly 1/3rd of the network alone .... thats evidence enough.

Of power claims? Its only evidence that they have money, and $60M buys you a lot of miners and a lot of PR.

Dogie, shame on you Bitmain pays you also for PR
I remember times when Bitmain and Spondoolies were competitors on B2C market and Dogie pushed tons of shit on Spondoolies, Guy Corem answered Dogie.. and there was again and again.. What was it? PR
Regarding "buys a lot of miners" - do you have some inside? Did BitFury bought their miners? Whom? Bitmain?? It would be a sensasion  

I haven't been contracted by Bitmain since May, so no they do not "pay me for PR". I've also generally been a supporter of Bitfury, who had the top spot and a near perfect score in my guide for a long long time. Guy claimed that the rating criteria specifically targeted them when it just didn't as it was applied on BFL first, and they still had 88/100 score. The rating criteria are designed to protect buyers, and their claims that they'd be able to finance the next generation without preorders or investment did indeed turn out to be false.

Back on topic, all I said about Bitfury is that no one has actually ever tested their miners who is able to speak about it, and so its hard to really take their repeated claims seriously. If you think I'm paid for PR, why don't you talk to Bitfury's actual PR company or their Head of Global PR.

Dogie, each huge company has PR manager. Would you said Bitmain doesn't have?
Your rating, as I wrote earlier, is rather candid - I don't have any remarks.
You are good at communication, indeed Your arguments are convincing, but they are about other things. Last your post: "all I said about Bitfury is that no one has actually ever tested their miners who is able to speak about it" - and you are right. But one post before you had written absolutely another thing: "Its only evidence that they have money, and $60M buys you a lot of miners and a lot of PR".
Each company have some reputation and you can't check yourself each detail. You may just believe or not. Sometimes companies with PR and money don't have good reputation, because there are some more important things in this world.

[punin]

BitFury (Valery Nebesny) attempted to do async, domino logic 55nm 5 engine scrypt chip and failed.

That's actually incorrect info. You should sack the Mossad agent working for you

There are two all important parameters in Bitcoin ASICs:
GH/mm^2 - Capex and J/GH - Opex

Glad you brought that up! GH/mm^2 + J/GH is exactly the game we've been winning since the FPGA era.

With extreme design techniques, sub 0.1 J/GH is possible also on 28nm processes, the question is what will be the GH/mm^2 - miner cost.

What also counts is how much money you need to put down to have that chip hashing for you. I believe we're the leader in that as well.

Whenever BitFury gives efficiency numbers (J/GH), they never give the all important GH/mm^2 numbers.

Why would we give such revealing information to our competitors? And I don't really understand why the end user even needs to know about this? He cares for $/GH and J/GH and delivery times. GH/mm^2 is engineering porn. Our figures are NSFW

When they're selling hosted hash-rate of their 28nm machines to their customers, they claim the machine efficiency is 0.35 J/GH

Source?

[dogie]

Whenever BitFury gives efficiency numbers (J/GH), they never give the all important GH/mm^2 numbers.

Why would we give such revealing information to our competitors? And I don't really understand why the end user even needs to know about this? He cares for $/GH and J/GH and delivery times. GH/mm^2 is engineering porn. Our figures are NSFW

That's the type of transparency which is good for everyone. This shouldn't become an industry which relies on "this tiny 20g portion of cake is only 100 calories!" style marketing. If you're going to claim a power efficiency (J/GH) it should be:

• The power efficiency of a marketable machine at the wall (J/GH)
• Or the real world power efficiency of a chip test at the predicted machine voltage (J/GH)
• Or a full voltage / power efficiency curve similar to Bitmain (J/GH graph)
• Or nothing at all

Currently it appears you've got a graph of J/GH and have simply quoted the lowest J/GH at the lowest voltage possible, even though you know you'll never ever run a machine that slow. We understand you have VCs to appease, but that doesn't mean you can't still play the gentlemanly game. Everyone loses when manufacturers revert to unsavoury marketing, such as the latest bitcoin miner I designed which achieves a record breaking 0.000 J/GH! You can preorder it today and get free staples along with your stapler miner! VCs hit me up, you'll want to fund the market leader over here!

Does that explain why others are often skeptical about your claims? Even if they're true, its impossible to know if its 10% or 90% marketing fluff.

[TheRealSteve]

full custom, so is our 3rd gen.

Speaking of which, is that still Pickaxe (80/20 custom/auto), or do I need to be adding a line to the wiki?

On the BitFury end.. as 16nm is claimed to have taped out, is there any chance we can prise some more solid info on the 28nm chip?

[cniht]

full custom, so is our 3rd gen.

Speaking of which, is that still Pickaxe (80/20 custom/auto), or do I need to be adding a line to the wiki?

On the BitFury end.. as 16nm is claimed to have taped out, is there any chance we can prise some more solid info on the 28nm chip?

I don't trust any claim from any company until a product is brought to market.  In late June I think or early July KnC announced a 3D 16nm chip without any specifications or backup.

Lots of running of mouths to the press little in the way of public demonstrations.  This is something the community needs to get rough on with any and all manufacturers.  These companies need to be transparent in their home mines as well as the products they 'pronounce' or 'propose' to sell in the future.  It's bad enough the are moving full steam ahead with consolidated mining operations.  It's just like Central banks with a new spin, which wasn't the point of crypto-currency and BTC in the first place.

[Guy Corem]

BitFury (Valery Nebesny) attempted to do async, domino logic 55nm 5 engine scrypt chip and failed.

That's actually incorrect info. You should sack the Mossad agent working for you

There are two all important parameters in Bitcoin ASICs:
GH/mm^2 - Capex and J/GH - Opex

Glad you brought that up! GH/mm^2 + J/GH is exactly the game we've been winning since the FPGA era.

With extreme design techniques, sub 0.1 J/GH is possible also on 28nm processes, the question is what will be the GH/mm^2 - miner cost.

What also counts is how much money you need to put down to have that chip hashing for you. I believe we're the leader in that as well.

Whenever BitFury gives efficiency numbers (J/GH), they never give the all important GH/mm^2 numbers.

Why would we give such revealing information to our competitors? And I don't really understand why the end user even needs to know about this? He cares for $/GH and J/GH and delivery times. GH/mm^2 is engineering porn. Our figures are NSFW

When they're selling hosted hash-rate of their 28nm machines to their customers, they claim the machine efficiency is 0.35 J/GH

Source?

Pleasure to chat with you.
- LightFury floorplan came from Valery himself.
- The fact that it failed and the reason for its failure came from a designer known and respected by both parties.
- The 0.35 J/GH number came from two different bulk customers. You have them signed on draconian NDA so I can't reveal their identity.

Regarding GH/mm^2 (or $/GH/s):

The moderate hash-rate increase due to your 28nm deployment might be due to two different reasons:

- You reserve funds for your 16nm deployment
- Your number are not that good

[Guy Corem]

full custom, so is our 3rd gen.

Speaking of which, is that still Pickaxe (80/20 custom/auto), or do I need to be adding a line to the wiki?

On the BitFury end.. as 16nm is claimed to have taped out, is there any chance we can prise some more solid info on the 28nm chip?

It's pipelined (unrolled) full custom.

BitFury 16nm was indeed tapedout.
I don't believe they'll release any info on their 28nm, seems they're following "The Art of War"

[brontosaurus]

Back on topic, all I said about Bitfury is that no one has actually ever tested their miners who is able to speak about it, and so its hard to really take their repeated claims seriously. If you think I'm paid for PR, why don't you talk to Bitfury's actual PR company or their Head of Global PR.

Well, I actually did watch the whole testing process very closely and can confirm that 28nm figure to be empirically derived   (I hope my word still has some value on these forums)

Empirically derived? What on earth does that mean in the real world of engineering?

If you were an engineer at CERN working with subatomic particles with ultra short lifetimes I might understand using this expression, but in our macro world we have voltmeters, ammeters and oscilloscopes that negate the need for 'derivation'.

Are you trying to say that that put some figures into a simulator and uses some trial and error to get the result you want? Naturally, you are not answerable to anyone on this forum or indeed any paying customers, so whatever you claim doesn't need to be defended in any way.

[punin]

Well, I actually did watch the whole testing process very closely and can confirm that 28nm figure to be empirically derived   (I hope my word still has some value on these forums)

Empirically derived? What on earth does that mean in the real world of engineering?

If you were an engineer at CERN working with subatomic particles with ultra short lifetimes I might understand using this expression, but in our macro world we have voltmeters, ammeters and oscilloscopes that negate the need for 'derivation'.

Are you trying to say that that put some figures into a simulator and uses some trial and error to get the result you want? Naturally, you are not answerable to anyone on this forum or indeed any paying customers, so whatever you claim doesn't need to be defended in any way.

What I'm saying is we have tested and profiled the 28nm chips extensively and this figure is a real world measured figure, not based on simulation. Hence the wording.

[brontosaurus]

BitFury (Valery Nebesny) attempted to do async, domino logic 55nm 5 engine scrypt chip and failed.

That's actually incorrect info. You should sack the Mossad agent working for you

There are two all important parameters in Bitcoin ASICs:
GH/mm^2 - Capex and J/GH - Opex

Glad you brought that up! GH/mm^2 + J/GH is exactly the game we've been winning since the FPGA era.

With extreme design techniques, sub 0.1 J/GH is possible also on 28nm processes, the question is what will be the GH/mm^2 - miner cost.

What also counts is how much money you need to put down to have that chip hashing for you. I believe we're the leader in that as well.

Whenever BitFury gives efficiency numbers (J/GH), they never give the all important GH/mm^2 numbers.

Why would we give such revealing information to our competitors? And I don't really understand why the end user even needs to know about this? He cares for $/GH and J/GH and delivery times. GH/mm^2 is engineering porn. Our figures are NSFW

When they're selling hosted hash-rate of their 28nm machines to their customers, they claim the machine efficiency is 0.35 J/GH

Source?

It's baffling why you don't see the significance of GH/mm2 in the context of this discussion. Your company surely understands the advantages of running chips at low voltage and if you have a hashing engine running at the minimum energy point then it's possible to get some truly amazing efficiency. Trouble is, you then need multiple engines to get any speed.

That's why Guy's GH/mm2 metric is significant. You might also like to read this excellent paper by Tran and Baas that demonstrates very clearly the concept of minimum energy points although I'm guessing it's probably old news to you. Yes, I know it's for a 32 bit adder but the principle is still valid for SHA256 (which in any case uses 32 bit adders in the main pipeline and word generator)

http://web.ece.ucdavis.edu/~anhttran/files/papers/atran_icce10_adder.pdf


As for engineering porn, what you are currently offering isn't porn, it's more like burlesque where it' all tease. I don't think anyone believes the figures you released other than a crude attempt to worry your competitors and maybe dissuade them from building more capacity. Just like the statement one of your engineers made about everyone else being 18 months behind you, it's insulting their intelligence and
demeaning their abilities.

Personally, I would 100 times more trust what previous Intel engineers say can be done than a bunch of enthusiastic amateurs - unless the latter can actually prove how big their balls are.  

[Sine(X)]

BM1385 is full custom, so is our 3rd gen.

So currently you are in the same team with Dogie, aren't you? (Bitmain team)

[Guy Corem]

BM1385 is full custom, so is our 3rd gen.

So currently you are in the same team with Dogie, aren't you? (Bitmain team)

No.

btw:
It seems that KNC finally started to deploy it's 16nm
Interesting times ahead.