Gridseed GC3355 -Hybrid Scrypt/SHA256 ASIC - page 19.

tacotime

legendary

Activity: 1484

Merit: 1005

Quote from: senseless on December 09, 2013, 09:43:44 PM

It appears what they've done to keep the chip small is to implement 4x 256Kbit on-die memories for hashing as I haven't seen anything mentioned about an off chip memory. At 55nm 1mbit of sram is about 7.63mm2. At 28nm 1mbit of sram is 3.88mm2. The max size for QFN-48 (which is a 7x7mm package) silicon is around 4x4mm so have roughly 16mm2 of space. They couldn't have implemented 4x 1mbit as it would be to large (with an exception for the 28nm which would be 15.52mm2, but even then there wouldn't be enough room for all those bitcoin hashers). The speed and stats are certainly within the realm of possibility.

Looks like their site is up. Didn't work the first time.

55nm chip, so definitely within the realm of possibility.

That's interesting. If they did implement the SRAM on chip, I guess the could maximally fit 14 scratchpads (14 x 128 KB). If with 8 they could reach 84 KH/s at let's say a theoretical 0.750 W, we'd max out on a single chip at 147 KH/s and 1.3 W. Not too shabby if these only cost about $5-10 each to produce on 55 nm QFN-48.

I'd guess they aren't using straight up scratch pads, though, but rather the TMTO desynchronization trick from Solar Designer. Regardless, the physical limitations as above should still be the same.

Egon

member

Activity: 85

Merit: 10

Anyone try to sign up for the Weibo feed?

senseless

hero member

Activity: 1118

Merit: 541

Translated auto-reply from sales address:

Quote

Important Note: We currently do not have any place in the pre-sale, Taobao are fake, the official website information please prevail.

Follow weibo.com/gridseed That is our official microblogging.

Relating to the purchase, agents, price issues,
Please wait for our official website www.gridseed.com announced
If you need to apply the agent, please specify in the message header "application proxy" and described their own information, and other advantages, if only one sentence is difficult to our attention that, thank you.

For technical discussion, please www.cybtc.com gridchip zone discussion, where maintenance by our technical team.

Documents to download, go to the open source project: github.com/gridseed

If we are unable to respond to you request, please understand, because the volume of mail than we expected, but we will read every message.

I found more pictures on some chinese site:
http://www.cybtc.com/thread-3005-1-1.html

tacotime

legendary

Activity: 1484

Merit: 1005

Quote from: Gator-hex on December 09, 2013, 09:33:59 PM

Scrypt requires high speed access to memory so I'm gonna say this is bull until there is some evidence it works. Wink

I'm sure someone will reverse engineer these sooner rather than later, and then we'll be looking at a lot of potential scrypt ASICs hitting the market. SEMs aren't that costly these days.

senseless

hero member

Activity: 1118

Merit: 541

Quote from: greaterninja on December 09, 2013, 01:55:45 AM

It is possible to do scrypt with a fpga, so that implies asic is possible too. I have seen a POC. With that said I am extremely skeptical if it can hit above 60k/s rate. Let alone even 10k/s scrypt hashing rate.

My FPGA is hashing at 60Kh/s, granted, it's a 5000$ fpga, but it is possible.

Litecoin will work fine on asic with on-die sram. You can even calculate the hash rate based on known factors.

Code:

200 / (2048 * 8) * Ncores = scrypt hashrate in Mh/s based on 200Mhz clock speed

It appears what they've done to keep the chip small is to implement 4x 256Kbit on-die memories for hashing as I haven't seen anything mentioned about an off chip memory. At 55nm 1mbit of sram is about 7.63mm2. At 28nm 1mbit of sram is 3.88mm2. The max size for QFN-48 (which is a 7x7mm package) silicon is around 4x4mm so have roughly 16mm2 of space. They couldn't have implemented 4x 1mbit as it would be to large (with an exception for the 28nm which would be 15.52mm2, but even then there wouldn't be enough room for all those bitcoin hashers). The speed and stats are certainly within the realm of possibility.

Looks like their site is up. Didn't work the first time.

55nm chip, so definitely within the realm of possibility.

Gator-hex

hero member

Activity: 490

Merit: 500

Scrypt requires high speed access to memory so I'm gonna say this is bull until there is some evidence it works. Wink

Tomatocage

legendary

Activity: 1554

Merit: 1222

brb keeping up with the Kardashians

When will they be available and how can I buy one?

Sy

legendary

Activity: 1484

Merit: 1003

Bounty Detective

Yeah but i bet they don't even think about it, its alot of work to put 1 million chips into working devices - why even bother if you can just sell them at much higher instant profit Grin

just sayin, that would only double the current diff, raising the ltc diff to 10k and above is not an easy task xD

Sy

legendary

Activity: 1484

Merit: 1003

Bounty Detective

Quote from: ar88 on December 09, 2013, 06:47:59 AM

Quote from: Sy on December 09, 2013, 06:28:30 AM

Power efficiency is nice but the number doesn't change that much, you need 10 chips (600$?) to outmatch a 280x (300$?), yes power is the key but this is not "a game changer" just a shift in power consumption, the kh numbers are pretty much the same.

Du(al)-Coin mode up to 1.75G/s BTC Hash Rate + 60K/s LTC Hash Rate, or up to 2.25G/s BTC Hash
Rate + 38K LTC Hash Rate

comes with 1.75GH/s of BTC hash rate as well..when running 60Kh/S LTC, doubt one 280x can do that? Grin

Yep that's a neat feature, doesn't concern the ltc network at all though, ppl react like this will really hit the ltc network hard but in fact, it doesn't change much at all, availability will suck as usual - you need 1,058,750 chips to just double the current diff and bring it up to 4200...

Don't get me wrong, i'll most likely buy some of these since it's really really nice to have but it won't change things as much as most here tend to think.

ar88

newbie

Activity: 24

Merit: 0

Quote from: Sy on December 09, 2013, 06:28:30 AM

Power efficiency is nice but the number doesn't change that much, you need 10 chips (600$?) to outmatch a 280x (300$?), yes power is the key but this is not "a game changer" just a shift in power consumption, the kh numbers are pretty much the same.

Du(al)-Coin mode up to 1.75G/s BTC Hash Rate + 60K/s LTC Hash Rate, or up to 2.25G/s BTC Hash
Rate + 38K LTC Hash Rate

comes with 1.75GH/s of BTC hash rate as well..when running 60Kh/S LTC, doubt one 280x can do that? Grin

Sy

legendary

Activity: 1484

Merit: 1003

Bounty Detective

Power efficiency is nice but the number doesn't change that much, you need 10 chips (600$?) to outmatch a 280x (300$?), yes power is the key but this is not "a game changer" just a shift in power consumption, the kh numbers are pretty much the same.

jayeeyee

full member

Activity: 191

Merit: 110

You lost me at LTC. When you say 60K/s.. does that equate the same to 60KH/s?

greaterninja

hero member

Activity: 924

Merit: 1000

Quote from: dvdrewritable on December 02, 2013, 07:28:59 PM

Quote

1 GENERAL DESCRIPTION
GC3355 is high performance and low power SHA256 processor designed by GRIDCHIP. With
advanced technology and highly integrated design, GC3355 target to provide multiple function and low
cost solution in SHA256 application fields.

Key feature:
 160 BTC Units
 4 LTC Units
 BTC mode up to 2.25G/s BTC Hash Rate, with 2.4W/GHash
 LTC mode up to 60K/s LTC Hash Rate
 Due-Coin mode up to 1.75G/s BTC Hash Rate + 60K/s LTC Hash Rate, or up to 2.25G/s BTC Hash
Rate + 38K LTC Hash Rate
 Highly integrated with PLL and Pre-Calculation Engine
 Support dual configuration and report interface, UART and Custom-Defined 2-Wires Bus
 Support Crystal and Oscillator
 Fully adjustable clock frequency

http://gridseed.com
https://github.com/gridseed/gc3355-doc
https://github.com/gridseed/cpuminer

It is possible to do scrypt with a fpga, so that implies asic is possible too. I have seen a POC. With that said I am extremely skeptical if it can hit above 60k/s rate. Let alone even 10k/s scrypt hashing rate.

dan99

sr. member

Activity: 406

Merit: 250

getting very interesting.

Sitarow

legendary

Activity: 1792

Merit: 1047

Quote from: tacotime on December 08, 2013, 08:45:23 PM

Chips and dev boards for now

Pictures confirm the release event and functionality. One person was able to get a single chip to hash at 84 KH/s.

This could be a game changer.. at least as hardware is concerned.

papamoi

sr. member

Activity: 378

Merit: 250

hi

so who is producing it?

tacotime

legendary

Activity: 1484

Merit: 1005

Chips and dev boards for now

Pictures confirm the release event and functionality. One person was able to get a single chip to hash at 84 KH/s.

Vagnavs

legendary

Activity: 1121

Merit: 1003

just chips or a miner?

fractal02

hero member

Activity: 504

Merit: 500

$Profile photo of fractal02$

Very interesting !

LTC & BTC chips...i buy ! Grin

Cheshyr

full member

Activity: 168

Merit: 100

First chips are getting into developer hands today.

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

Code:

Scheduled at 10:00 on December 8, 2013, in West Street, Haidian, Beijing, 70 3W coffee (Haidian Book City, south of the membership across the sea floor), held Gridchip GC3355 chip product launches and mining machines. In addition to the chip and mine-site machine product launches, and BTC, LTC demonstration outside mining, but also on-site sales of 50 sets of single-chip development board and a few chips.

Development board: each 500 yuan 
chip: 400 yuan per piece, per pack of 1600 yuan four chips 

This price is only preliminary offers developers, not as a formal reference to the product price. Each person can purchase two development boards, chip purchase of two packs per person.

$65 a chip for the early pieces.

Topic: Gridseed GC3355 -Hybrid Scrypt/SHA256 ASIC - page 19. (Read 107978 times)