We are Dead! - ETH ASIC are out!

adasebadaseb Member Posts: 1,043 ✭✭✭
http://www.lightminers.com

I really hope its a scam or else its time to dump GPUs on eBay
«1

Comments

  • HelioxHeliox Member, Moderator Posts: 633 mod
    Seriously.. no way this is real.. chill out..
  • dlehenkydlehenky Member Posts: 2,249 ✭✭✭✭
    How on earth could you have 80 asic chips sharing a 3GB VRAM? That's not possible. The chips may be capable of 2.5 MH/s without a memory bottleneck, but what difference does that make? The only other explanation would be they have 3 GB of VRAM per chip, which would be .24 TB of VRAM. And, of course, they give no price.
  • Chris31Chris31 Member Posts: 21
    The dimension on their USB miner (Dimensions: 100mm(W) x 35mm(L) x 45mm(H)) is already wrong

    It is no way that long according to the picture.
  • ThatDudeThereThatDudeThere Member Posts: 134
    Click on "more info" and it gives a price. $1699
    http://www.lightminers.com/z02.php
  • agent412agent412 Member, Moderator Posts: 278 mod
    i call bullshit
  • dlehenkydlehenky Member Posts: 2,249 ✭✭✭✭
    I just don't get this kind of blatant fraud. At 25 MH/s, you're using ~200+ GB/sec of VRAM bandwidth. That's not a theory, it's a fact. So, this 200 MH/s Z02 would need 1.6 TB/sec of memory throughput to a single 3 GB GDDR VRAM memory. That is not possible with today's memory technology, or tomorrow's either. That doesn't even consider 80 ASICs trying to share the address and data bus to the same VRAM. This thing is so far from reality that "fairy tale" is a radical understatement. The saddest thing is, some folks will actually believe this crap and buy it.
  • rdnkjdirdnkjdi Member Posts: 135 ✭✭
    Scam incoming. . .
  • HashingTurtleHashingTurtle Member Posts: 107
    edited April 2016
    Let's assume for a moment this is a valid product and you are willing toss a chunk of money to get some of the usb's or one of the boxes. What are you going to mine when ethereum goes pos? There are not many options. Has anyone found 5 or 6 other mineable coins for this specific product (2 or 3 might not be enough for me) with some kind of a business plan that would create value for their coin and draw the mining community?

    Large up front cost
    Not shipping today
    Limited mining options
    Coming out right before AMD Polaris (hashrate unknown)
  • workwork Member Posts: 2,084 ✭✭✭✭
    dlehenky said:

    How on earth could you have 80 asic chips sharing a 3GB VRAM? That's not possible. The chips may be capable of 2.5 MH/s without a memory bottleneck, but what difference does that make? The only other explanation would be they have 3 GB of VRAM per chip, which would be .24 TB of VRAM. And, of course, they give no price.

    This was actually discussed in critisism's for dagger's design
    VB didn't make it sequentially memory hard, so shared memory was deemed a significant potential advantage for asic hardware.

    That said, I doubt these "ASIC" chips are real.
  • dlehenkydlehenky Member Posts: 2,249 ✭✭✭✭
    @work Since all the memory accesses are pseudo-random, that's a lot of addresses on the memory bus, non-cacheable, so how does that play to shared memory?
  • workwork Member Posts: 2,084 ✭✭✭✭
    edited April 2016
    @dlehenky one of the original criticisms: https://bitslog.wordpress.com/2014/01/17/ethereum-dagger-pow-is-flawed/

    The way I see it, the solution is simple. (To put it oversimplified), you don't store and access the dag, you store and access the cache, and then each asic chip on-the-fly computes the full dag, grabbing the needed components for a given nonce and discarding the buffer.

    Claymore's miner doesn't need DAG files now because the GPU can compute the DAG in under 5 seconds. Create an asic designed to do exactly that, and you might get that down to a handful of clock cycles.

    This is why I have somewhat regularly pointed out (well, at least a few times), that ethash isn't really memory-hard, it is memory-capacity-hard. And requiring memory-capacity certainly doesn't make an algorith sufficiently memory hard.
  • dlehenkydlehenky Member Posts: 2,249 ✭✭✭✭
    @work So, you're computing the DAG hashes (2 sequential) you need for the loop iteration based on their position (index) in the current "virtual" DAG?
  • bbcoinbbcoin Member Posts: 376 ✭✭✭
    edited April 2016
    I registered using junk mail to see if I can snoop something out.

    No notifications/emails from them about registration or my order = they don't want to leave a trace.

    Website was registered with Go-daddy and it's based on a server in the US.

    Payments are accepted to bitcoin addresses generated at random.
    That way unless you will send them money yourself you will never know where the money is going.
    (most likely straight into a mixer)

    I will make a call this hasn't got a thing to do with China. I'd suspect it's a US citizen trying to scam ETH noobs.
  • ThatDudeThereThatDudeThere Member Posts: 134
    I hate scammers
  • HelioxHeliox Member, Moderator Posts: 633 mod

    I hate scammers

    Bitcoin world/scene is full of it. Another reason why Bitcoin will have a serious hard time to make it to the general public.. :(
  • workwork Member Posts: 2,084 ✭✭✭✭
    edited April 2016
    Heliox said:

    Bitcoin world/scene is full of it. Another reason why Bitcoin will have a serious hard time to make it to the general public.. :(

    Blockchain-based currencies trade network security for personal security. From a network perspective, a blockchain is significantly more secure then a bank; but, from a user perspective, it is the opposite. Comparison bitcoin to classic fintech, risk moves from the network to the user, and security moves from the user to the network.

    It's really just another way of thinking that demands people be more careful with their money. This is hardly a bad thing, but it's definitely a major change from the "reverse-anything" banking world we have now.
  • adasebadaseb Member Posts: 1,043 ✭✭✭
    If it was a scam why didn't they just say they made a 1Gh/s miner using 1000Watts for $500.

    The price is high and everything and makes it seem like its plausible.
  • dlehenkydlehenky Member Posts: 2,249 ✭✭✭✭
    @adaseb Yep, they COUNT on folks like you! Disregard all technical feasibility and solid evidence on the 'net that it's bogus, but sooo wanting it to be true that you swallow it hook, line, and sinker. Go for it! Buy the limit! We'd all like to hear about your experience with it, really.
  • AetsenAetsen Member Posts: 30
    It's a scam look at the images on their about page. Stolen from various other sites across the web. OP if this is legit contact me and send me a Z01 to test and report on.

    http://www.oz-chill.com/wp-content/uploads/2014/02/China-Office-C.png

    https://www.shopify.com/blog/5103612-5-tips-to-help-entrepreneurs-work-from-home

    http://www.reuters.com/article/us-china-pmi-hsbc-idUSTRE7AM08P20111123
  • jstefanopjstefanop Member Posts: 43
    dlehenky said:

    @work So, you're computing the DAG hashes (2 sequential) you need for the loop iteration based on their position (index) in the current "virtual" DAG?

    Im pretty sure claymore miner just sends over the 16MB cache to the GPU, and he just creates the data set locally on the GPU instead of the CPU. Its still brilliant, as there is like a 100x speedup in DAG creation, and eliminates the need to send the whole data set from CPU RAM over the bus to GPU VRAM, and no need to store it since the GPU can generate it fast enough.
  • dlehenkydlehenky Member Posts: 2,249 ✭✭✭✭
    edited April 2016
    @jstefanop Yes, I understood that part :) And it is indeed a slick improvement to the current implementation. I was asking @work about the ASIC "on the fly" DAG calculation in the kernel inner loop that I *think* he was putting forward as an ASIC alternative to doing random memory accesses to fetch the DAG bits from the whole DAG stored in GPU memory. Sorry for the long sentence. :)
  • downhillfromheredownhillfromhere Member Posts: 50
    work said:

    @dlehenky one of the original criticisms: https://bitslog.wordpress.com/2014/01/17/ethereum-dagger-pow-is-flawed/

    The way I see it, the solution is simple. (To put it oversimplified), you don't store and access the dag, you store and access the cache, and then each asic chip on-the-fly computes the full dag, grabbing the needed components for a given nonce and discarding the buffer.

    Claymore's miner doesn't need DAG files now because the GPU can compute the DAG in under 5 seconds. Create an asic designed to do exactly that, and you might get that down to a handful of clock cycles.

    This is why I have somewhat regularly pointed out (well, at least a few times), that ethash isn't really memory-hard, it is memory-capacity-hard. And requiring memory-capacity certainly doesn't make an algorith sufficiently memory hard.

    Brilliant explanation work, now I comprehend why Claymore's miner works so well. If I ever get appointed to one of Vitalik's recently metioned "Ethereum courts" any virtual violations will be dismissed automatically.
  • jstefanopjstefanop Member Posts: 43
    Obviously a lot more research would have to go into it, but I find it hard to believe that grabbing the dag bits via an "on the fly" asic method would be any faster than random memory accesses on a DAG stored in memory.

    The dag creation core on the asic would have to compute the random bits on the fly before the next main loop is finished on the hashing core.

    I see the hashing core waiting for a "long" time before it can get the next loops random bits, so probably the same thing as the current memory access bottleneck in the GPU implementation.

    What this would do though is save ALOT of cost in memory. Since essentially all you would need is 16MB of cache.

    Interesting idea either way.
  • wirelessnet2wirelessnet2 Member Posts: 368 ✭✭✭
    oh shit... that'd better be a scam. Seems like they stole the photos...
  • JukeboxJukebox Member Posts: 529 ✭✭✭
    adaseb said:

    If it was a scam why didn't they just say they made a 1Gh/s miner using 1000Watts for $500.

    The price is high and everything and makes it seem like its plausible.

    http://www.lightminers.com/z01.php
    SCAM in every letter.

    USB sticker-size miner have VRAM: 3GB Hynix GDDR VRAM

    What a f*cking f%ck? :D

    Ok, let's call Capt. Obvious.

    3 GB GDDR5 it's at least SIX memory chips for now. It's impossible to fit them in that USB stick board, but you also need a memory controller and computing chip to place there.

    Each GDDR5 chip on high clocks consumes at least 1 watt, so we have 6 watt consumption only for memory,
    and unknown amount for memory controller and computing chip. So it's impossible to power this device from USB port.

    Other weird things:
    1) why 3 GB? Not 2, not 4 - but 3?
    2) why VRAM? Are there some video?
  • Chris31Chris31 Member Posts: 21
    Jukebox said:

    adaseb said:

    If it was a scam why didn't they just say they made a 1Gh/s miner using 1000Watts for $500.

    The price is high and everything and makes it seem like its plausible.

    http://www.lightminers.com/z01.php
    SCAM in every letter.

    USB sticker-size miner have VRAM: 3GB Hynix GDDR VRAM

    What a f*cking f%ck? :D

    Ok, let's call Capt. Obvious.

    3 GB GDDR5 it's at least SIX memory chips for now. It's impossible to fit them in that USB stick board, but you also need a memory controller and computing chip to place there.

    Each GDDR5 chip on high clocks consumes at least 1 watt, so we have 6 watt consumption only for memory,
    and unknown amount for memory controller and computing chip. So it's impossible to power this device from USB port.

    Other weird things:
    1) why 3 GB? Not 2, not 4 - but 3?
    2) why VRAM? Are there some video?
    They appear to be using QFP 64 package for their ASICs and there are 6 of them on each USB dongle

    Orginal version
    https://upload.wikimedia.org/wikipedia/commons/7/79/SSDTR-ASIC_technology.jpg

    Lightminer version
    http://www.lightminers.com/images/eth01-2.png
  • kotariuskotarius Member Posts: 329 ✭✭✭
This discussion has been closed.