They claim this, but the claim seems kind of absurd. The whole point of covert is that it's cover, it's nearly impossible to detect.

Consider though: at great expense they developed an significant performance optimization and baked it into their chips, huge numbers of which they ran themselves ... and then kept it secret but ... just didn't use it?  Why? because they like giving money away to the power company?

There is evidence that bitmain was asicboosting: among other things (Edit: e.g. as frodo mentions, the huge number of empty blocks, which have more or less magically gone away), they mined a small number of blocks that were invalid because transactions were out of dependency order. This would be a pretty difficult mistake to make-- unless you were grinding blocks by swapping around transaction order.  It's just not conclusive proof, unfortunately short of some massive internal document leak conclusive proof is likely impossible.

[Edit: The empty block part is really strong because unlike any other case you can compute the collisions for the empty blocks arbitrarily far in advance, so even if takes you tens of seconds to find one on the FPGA you can keep mining until one is found.]

Of course, there is also no reason to think that bitmain was the only booster.  I think its very interesting that the bitmain funded developers of bcash pushed surprisingly hard for their "canonical transaction order" change shortly after bitmain published overt AB support for their existing hardware... Might well be that once covert boosting wasn't an advantage it was best for overt supporters kill it on bcash in order to disrupt any miners whos hardware was less flexible than Bitmain's. (Because Bitmain used a rather overpowered FPGA for control they could switch between covert/overt/no-ab with just a firmware change... it would have been more power efficient though to have a separate asic for collision finding, but such a design might result in devices that lose half their hashrate when used with a mandatory transaction order or otherwise have to only mine empty blocks...)


Wow. No, I completely hadn't seen that-- I mean it was buried deep in a thread about a vendor who's hardware I didn't have at the time, and didn't mention asicboost in the post... all the details were in a pdf... I wouldn't be surprised if almost no one ever read it.  Too bad, because that was a pretty big revelation. Kudos to you.   It would have saved me an astonishing amount of time if I'd been aware of it.  (in particular figuring out the how to use the multi-midstate with no actual docs on the chip but simply picking through the software and bus captures with unreliable information extracted from the chip design was a big chore).  

As an aside, given that you knew this-- how come when I announced it and so many people were saying that I was lying about it being there, you didn't step up to point it out?