Bitmain Announces Ethereum ASIC Miner, Shipping Mid-July

Latest Videos FromTom's Hardware
Contributor

Kevin Carbotte is a contributing writer for Tom's Hardware who primarily covers VR and AR hardware. He has been writing for us for more than four years.

  • takeshi7
    I'm going to laugh when all of Bitmain's R&D money goes down the toilet after Ethereum changes their mining algorithm.

    https://cointelegraph.com/news/ethereum-community-considers-hard-fork-to-fight-asic-miners
    Reply
  • tslot05qsljgo9ed
    Anyone who buys this is buying a lump of sand as these will be useless.

    New EIP Suggests Ethereum Hard Fork For ASIC Resistance
    https://www.ethnews.com/new-eip-suggests-ethereum-hard-fork-for-asic-resistance

    The price of $800 is a sure giveaway that this will be useless when it arrives.
    Reply
  • teodorehh
    You only need six RX580 with 1050W power consumption. And

    (a) if ethereum changes the algorithm, you just switch to new software while ASICs are uselss
    (b) When Ethereum goes POS you sell the GPUs while you throw to the garbage can the ASIC.

    ..and btw, ethereum WILL change the algorithm if many of those get sold.
    Reply
  • InvalidError
    20853527 said:
    ..and btw, ethereum WILL change the algorithm if many of those get sold.
    I bet Bitmain has had thousands of those in its own mining farms for months already, just like it has used Bitcoin miners internally for months prior to making them available for sale.
    Reply
  • 10tacle
    Jumping into cryptocurrency mining with BC or ETH nowadays is like going back in time to the late 1800s heading westward for gold with a horse and buggy. You just aren't going to get rich after those who got there years prior already beat you to it.
    Reply
  • annabelle75
    Hey guys, been lurking @Tom's for a long time, love the site! :-)

    That said, here's a summary of a 13-gpu miner built around two 1000w PSUs:
    miner3 10.0.0.13:3333 4 days, 06:38 364.61 MH/s 33110/1 (0.00%) 70C/72% 64C:72% 57C:72% 52C:72%
    63C:72% 57C:72% 63C:72% 64C:72% 51C:72% 68C:72% 64C:72% 52C:72% 55C:72% us2.ethermine.org:4444
    10.6 - ETH THIRTEEN

    So it's 364 MH/s and pulling about 1900W at the wall. A bit under the efficiency of the Bitmain offering, but close. The difference is the GPU build way at this low of a power draw and stable, high hashrate, requires an understanding of the unix kernel and how to undervolt the PCI bus and underclock the GPUs via the kernel's amdgpu driver. And the GPU's need to have their bioses modified. Which, if you don't understand what you're doing, can leave you with "bricked" GPUs that are a PITA to recover. So quite a bit of work to build.

    Bitmain's ASICs have a nice web interface. Pointy-clicky and done, you're mining.

    And post #1 above isn't really accurate. The E3s won't be bricks if you know how to modify the software running them. Which is kinda like building a ROM for Android, not really that hard... just update the bundled mining software with the new algo, build, flash, poof. Already done for Bitmain D3s, google "BlissZ D3 firmware".

    So the E3s are, IMHO, a good thing. They're not drastically faster than six AMD GPUs of similar power, but they are available to hobbyists that just want to mine. And they draw less power, which is always a good thing...

    And I doubt PiRL will fork so you can always mine that, or ETC/ELLA/MVP/Something ETHASH-based.
    Reply
  • Giroro
    An ASIC can be anything that you want it to be. You could even build a GPU into an ASIC - you could actually even make the argument that all GPUs are technically ASICs. They solved the "Fast Memory" problem by simply designing an ASIC that can use fast memory, and then providing that memory. The only real challenge is that an ETH ASIC probably operates very differently, so they would have been starting from scratch with their design. It can take about 2 years (and a lot of money) to design and produce an ASIC from scratch, and ETH became profitable about 2 years ago - so it didn't even take them an unreasonable amount of time.
    A hard fork or a changed algo on ETH won't necessarily even break an ASIC, if they designed the ASIC to be re-programmable - but I imagine you would lose a lot of speed and efficiency if you did that.
    FPGAs on the other hand can easily be reconfigured, but those devices are where you actually run into the memory problem. For the most part only FPGAs made for AI are designed for that kind of memory access, and you would have a hard time building a box around a single one of those for $800.
    Even if they need to build a new chip for a new algorithm, the turnaround for that will be much faster this time. So ETH fans are going to find themselves in an arms race where there is a hard fork every 6 months, until investors get tired of it and the currency loses so much value that an updated ASIC is not worthwhile.

    Sure, there are not many people in the world with the knowledge and motivation (and absurd bankroll) needed to design and build an ASIC, but they didn't exactly pull off some technical miracle here. Anybody who claims their currency is ASIC-resistant doesn't actually understand what an ASIC is, so you should not trust their opinion on such things. All the current ASIC-resistant currencies are simply FPGA-resistant (granted ASIC designs are often iterated and tested on FPGAs first, so that could slow ASIC development).
    Reply
  • annabelle75
    20854404 said:
    An ASIC can be anything that you want it to be....

    Interesting clarification, thanks!

    My sum total of knowledge about GPU hardware: You can carefully cut the x16 slot down to x8 or x4 to fit in old slots on, say, 1U or 2U servers. A figurative hacksaw job to make an elegant AI solution for colocation. And, when replacing fans on non-blower models, the chip underneath says "ASIC" on it. Meaning a GPU is a less-specific "ASIC" than what carries the label of "ASIC"?

    Further I don't understand the difference between a programmable CPU feeding a GPU's ASIC Vs. what's considered an FPGA? Haven't gate arrays fallen out of favor/use with the advent of CPU/GPU hardware and Cuda/OpenCL/OpenACC?

    And you wimped out on proffering an opinion if the E3 is a good or bad thing ;-)
    Reply
  • InvalidError
    20854404 said:
    An ASIC can be anything that you want it to be. You could even build a GPU into an ASIC - you could actually even make the argument that all GPUs are technically ASICs.
    Practically anything that isn't a basic generic function technically qualifies as an ASIC. I'd argue that even CPUs qualify as they are integrated circuits designed for the specific purpose of executing the ISA it was designed for.
    Reply
  • Xenocrates
    20854454 said:
    20854404 said:
    An ASIC can be anything that you want it to be....

    Interesting clarification, thanks!

    My sum total of knowledge about GPU hardware: You can carefully cut the x16 slot down to x8 or x4 to fit in old slots on, say, 1U or 2U servers. A figurative hacksaw job to make an elegant AI solution for colocation. And, when replacing fans on non-blower models, the chip underneath says "ASIC" on it. Meaning a GPU is a less-specific "ASIC" than what carries the label of "ASIC"?

    Further I don't understand the difference between a programmable CPU feeding a GPU's ASIC Vs. what's considered an FPGA? Haven't gate arrays fallen out of favor/use with the advent of CPU/GPU hardware and Cuda/OpenCL/OpenACC?

    And you wimped out on proffering an opinion if the E3 is a good or bad thing ;-)

    Why on earth would you cut down the card? It's far easier to cut the back off the PCIe slot, so that a full length card will fit in a 4x slot. That way the cards are undamaged, and can operate at full speed, and you only lose warranty on one component (if you're using risers, it's cheap), if the MB or riser doesn't already have the slot open at the back.

    As far as FPGA's, yes and no. FPGA's are for development work, or for niche algorythm's that need to be fast anyways. IIRC, Nvidia was simulating new architectures on FPGA hardware at one point. Once you have a good FPGA design, you could almost as easily send it to be fabbed for ASIC production, and likely get better speeds at cheaper unit prices, once you overcome the cost of taping it out for lithography anyways.
    Reply