Bitcoin Forum

The spec sheet is here:

http://www.cybtc.com/uc_server/data/avatar/000/00/22/13_avatar_middle.jpg
https://github.com/gridseed/gc3355-doc/blob/master/GC3355_DataSheet.pdf

These are hybrid miners that utilize the BTC hashing core to do a few of the SHA256 calculations for LTC apparently, they are claimed to be able to simultaneously hash BTC at 1.75 GH/s and LTC at 59.6 KH/s while pulling 4.56 W.  Hashing LTC alone, they are claimed to pull 59.6 KH/s at 0.44 W.


http://www.cybtc.com/thread-3226-1-1.html

Well damn...

If it's true I'll be cranking my GPUs hard for the new few months until these come out -- ASIC introduction into Bitcoin caused a 100-fold increase in valuation, I would be surprised if we don't see a big increase with Litecoin too.

This is 0.00738 W/KH/s versus 0.300 W/KH/s for a GPU -- a 40.7 fold increase in efficiency.  This is about the same increase in efficiency as compared to BTC when ASICMINER first introduced its chips; I'm curious to see how they did it.

Added release date

I believe this when i see it ... No preorder for me at least!

Hopefully it doesn't happen soon. I want to put some more GPUs to better use.

If it happens it happens... maybe that primecoin GPU miner will finally get off the ground.

Some threads with photos:

http://www.cybtc.com/thread-3370-1-1.html
http://www.cybtc.com/thread-3379-1-1.html
http://www.cybtc.com/thread-3355-1-1.html

So it is confirmed I guess... the second thread shows a single chip hashing at 84 KH/s.

Well perfect time for people to invest and switch to more asic resistance coins.

lol , ASIC are a good thing and strengthen the network .

Interesting, something like this was always more likely to come from left-field rather than the scammers shouting "we're developing multi-MH/s LTC ASIC" to the rooftops.

The documentation is a little light on specifics (such as the internal RAM, TMTO or threading), but at around 60kH/s in four LTC units its not too dissimilar to the open source engine, so we're probably looking at a few MB of RAM at most.

I'm most intrigued by the register specification which has 8 words of midstate and (the usual) 19 words of data. I wasn't aware that midstate was applicable to the scrypt algorithm (or at least, it only applies to the very initial stages of the PBKDF_128). I'll need to look at this in more detail as I may have missed an optimization here.

Anyway, good luck to them.

what about the cost?

ok 4x the cost of a single 7950

I'm no expert at this but if I'm reading this datasheet right, claims to hash btc and ltc at the same time ???

I wonder if this is true. Where can we order?

Also interested in getting some chips...but not sure I know where to click... my chinese definitely not great!

Who do I throw my money at and when?

It depends on price

Is there anyone that speaks chinese and understand how to order a couple of board and some chips?

Are there any readymade samples? What is their speed and price? I would buy them.

so like $1200USD for a single chip, for 60KH/s?  I guess if youre gambling that the price of power will skyrocket this may be a good deal.

Or I could be understanding something wrong.

i made a mistake it is only x2 the cost still not worth imho, cuz you are basically paying the bills in advance

Nobody ever said there wouldn't be a scrypt ASIC, just that it probably won't be cost effective for some time.

Is that not the case here? Looks like we have a rough price for just the chip, no other supporting hardware like RAM?

No external off-chip RAM, a small quantity of SRAM is on-die in each scrypt core.  The only external support components are 8 bypass caps on the supply rails, an external clock reference to drive the PLL, and a pulldown or pullup resistor on the clock output pin to select the interface mode.

You can get brief technical specs here at their website (English version): http://www.gridseed.com/main.php

Orders are not accepted yet, but the release date is expected to be announced soon (they said they would sell the actual items directly, not pre-orders).

They sold some engineering samples at the release conference, but the price seemed pretty high (400 RMB per chips, or roughly $65 per chip -- according to: http://www.cybtc.com/thread-3226-1-1.html ).

Sort of.  It looks like if they were to make a scrypt only ASIC, the power consumption for 330 KH/s would be about 2.5 W.  This was similar to the original 130 nm ASIC miners for SHA256, but this time we're on a 55 nm process node.  130 nm should use about three fold the amount of power in theory, so scrypt is slightly less efficiency in this ASIC implementation, but not by much.

Interesting i wonder how long before it's really out on the market. Will be good for scrypt but the same pre order back order nightmare will happen regardless of what they say, i doubt they can manufacture to meet demand.  .

with just a tiny bit of effort you could deliver even more hopeless and despair.  BTC-Market, thats what I call a DEAL.

In summary, if i understand right, they put 4 (four) LTC cores alongside a 160 BTC cores.  

The question is begged, why?  I can imagine its proof of concept, but they've lost die space that could have been better used for BTC, shirley?

Another question, if they are placing ram on the die, will that scale?  i.e. if they had 160 LTC cores what would that look like.  I reckon 2400 K/s for 17W by crude calcs, which seems decent.  That's equivalent to 4x 7950 for about 15% the power.  So, why didn't they just push out an all LTC device?  

Or is there less than meets the eye?


oh... and at 2.5GH/s aren't they waaaay behind the curve? 

Senseless wrote a useful summary on the other thread https://bitcointalk.org/index.php?topic=355268.msg3899931#msg3899931

I agree that its proof of concept, but the RAM takes up a significant amount of the die area so its quite an expensive one. Who knows what the rationale was for this device? Perhaps they had a slot available at the foundry and decided to use this as a test/development vehicle for new hashing architectures? If so then perhaps the next version will be scrypt-only, but I doubt it will scale to 160 LTC cores (see senseless's comment on the RAM I linked above).

GPUs can be made more efficient though. A 7870 undervolted to 0.95V, hashing at ~410 Kh/s and pulling 5A from the 12V rail, consumes 0.145 W/Kh.

+1