Pages:
Author

Topic: [LTC] Litecoin = ASIC resistant is a lie! (Read 3029 times)

newbie
Activity: 42
Merit: 0
June 12, 2013, 12:33:32 AM
#23
Just finished reading all the posts here and I'm left with one question: What was the point of this thread again? I'm pretty new here and this is not even the first time I've seen this subject posted in the last little while.

I can't see the point as well. It seems some people can't use search...

Exactly, and no one has ever denied the possibility or plausibility of such machines, just the practicality of the implementation of same. There's a ton of stuff that's 'possible' but not financially viable. For instance, I could go out and spend all of my cash and savings on lottery tickets and possibly win, and even have a better chance than if I never spent anything. That does not make it a financially viable decision though.

OF COURSE no hash is FPGA or even ASIC resistant. It's just impractical and costly with the current gen technology. There are several DYI and company projects working on SCRYPT FPGA right now, and sooner or later it will be available. The whole POINT of developing a new coins is that you do if on the premise of working and available technology. Hell, when bitcoin came out it was 'GPU Proof' until the first person realized how to do it, and do it well.

Ok and... I can't really see what you added to my previous post regarding that it was of course possible, just not practical at this current point to warrant anyone with serious funds putting a significant financial investment into.
legendary
Activity: 1632
Merit: 1010
Yes, Litecoin ASICs do not exist yet, but are not harder (or  costlier) to build than any Bitcoin ASIC. It's just a different hashing algorithm!

Scrypt has different parameters than SHA256, but those obstacles should be easily tackled.

really this old arguement again.  scrypt is sha256.  scrypt does a salsa20 8 times mix backwards and forwards, it makes the lookup tables difficult to predict.  you need to load it all into memory to hash it.  it requires a lot of on die memory to load the tables into.  there are no fpga boards on the market that have that much on die ram.  it would cost a fortune to develop.  with most scrypt coins having low value and low liquidity there is not incentive for an investor to deploy the several million it would take to build an fpga or asic just so that can make a few hundred grand.  they woould be competing with GPUs manufacturers which have already invented the wheel.

correction: there are no asic baords with memory enough for scrypt
fpga is right around the corner and the boards already exist
full member
Activity: 224
Merit: 100
Just finished reading all the posts here and I'm left with one question: What was the point of this thread again? I'm pretty new here and this is not even the first time I've seen this subject posted in the last little while.

I can't see the point as well. It seems some people can't use search...

Exactly, and no one has ever denied the possibility or plausibility of such machines, just the practicality of the implementation of same. There's a ton of stuff that's 'possible' but not financially viable. For instance, I could go out and spend all of my cash and savings on lottery tickets and possibly win, and even have a better chance than if I never spent anything. That does not make it a financially viable decision though.

OF COURSE no hash is FPGA or even ASIC resistant. It's just impractical and costly with the current gen technology. There are several DYI and company projects working on SCRYPT FPGA right now, and sooner or later it will be available. The whole POINT of developing a new coins is that you do if on the premise of working and available technology. Hell, when bitcoin came out it was 'GPU Proof' until the first person realized how to do it, and do it well.
newbie
Activity: 42
Merit: 0
Just finished reading all the posts here and I'm left with one question: What was the point of this thread again? I'm pretty new here and this is not even the first time I've seen this subject posted in the last little while.

I can't see the point as well. It seems some people can't use search...

Exactly, and no one has ever denied the possibility or plausibility of such machines, just the practicality of the implementation of same. There's a ton of stuff that's 'possible' but not financially viable. For instance, I could go out and spend all of my cash and savings on lottery tickets and possibly win, and even have a better chance than if I never spent anything. That does not make it a financially viable decision though.
full member
Activity: 182
Merit: 100
fml
Yes, Litecoin ASICs do not exist yet, but are not harder (or  costlier) to build than any Bitcoin ASIC. It's just a different hashing algorithm!

Scrypt has different parameters than SHA256, but those obstacles should be easily tackled.

really this old arguement again.  scrypt is sha256.  scrypt does a salsa20 8 times mix backwards and forwards, it makes the lookup tables difficult to predict.  you need to load it all into memory to hash it.  it requires a lot of on die memory to load the tables into.  there are no fpga boards on the market that have that much on die ram.  it would cost a fortune to develop.  with most scrypt coins having low value and low liquidity there is not incentive for an investor to deploy the several million it would take to build an fpga or asic just so that can make a few hundred grand.  they woould be competing with GPUs manufacturers which have already invented the wheel.

And, even if someone did do this, the diff would raise so fast, you'd never see a return on the investment, and LTC, would be dead.
hero member
Activity: 574
Merit: 523
Just finished reading all the posts here and I'm left with one question: What was the point of this thread again? I'm pretty new here and this is not even the first time I've seen this subject posted in the last little while.

I can't see the point as well. It seems some people can't use search...
newbie
Activity: 42
Merit: 0
Just finished reading all the posts here and I'm left with one question: What was the point of this thread again? I'm pretty new here and this is not even the first time I've seen this subject posted in the last little while.
hero member
Activity: 574
Merit: 523
Just a thought.

Let's imagine, one creates ASIC for LTC mining. It should perform approximately with the same ratio to sha256d as CPU/GPU implementations do  or even worse as the algorithm requires 1024 sequential write operations and 1024 randomly addressed but sequentially executed read operations to/from the memory of 1024-bit blocks (let's assume that the salsa-8 is unrolled completely so it takes just 1 clock cycle, i.e. is very small number compare to 1024x2). But the actual ratio is not really important for the sake of my question:

"What for to produce a chip that performs worse and produces things valued lower than the already produced chip that has the same production cost but performs much better in terms of $/chip/sec"?

The chip development costs are nearly the same in both cases, so I would select to implement sha256d if I were to start from the scratch.

The income that could generate early adopters of the new technology is also not a reason to develop such an ASIC - the same story as with bitcoin would happen to ltc - everyone would eventually switch to ASIC to mine LTC.

Then the second gen of LTC would be invented for these who owns GPU farms. And the story will take a next round: "Let's develop an ASIC implementation of our new even more complex algorithm"
member
Activity: 67
Merit: 10
A much cheaper and easier solution is to interface the memory on the PCB, no need to have the memory directly inside the die.

its a matter is distance.  its expensive to have to go out to the PCB.  updated my post above.

I don't think that is true.  Take a GPU card for instance.  A GPU card has it's memory on the PCB board which it accesses through an ASIC chip.
hero member
Activity: 798
Merit: 1000
‘Try to be nice’
It doesn't really matter if Scrypt is ASIC resistant or not.  BTC is the 900 pound gorilla in the room.

Nobody will take the alternate coins, including LTC, seriousy if all the GPU's mining those coins combined can't add 1% or 1/10'th of 1% to the bitcoin network.  That gross imbalance between hashing power will be reflected in the LTC/BTC ratio.  It is already starting to fall as ASIC gains traction

poor guy you're about to learn all about economic.
hero member
Activity: 1395
Merit: 505
It doesn't really matter if Scrypt is ASIC resistant or not.  BTC is the 900 pound gorilla in the room.

Nobody will take the alternate coins, including LTC, seriousy if all the GPU's mining those coins combined can't add 1% or 1/10'th of 1% to the bitcoin network.  That gross imbalance between hashing power will be reflected in the LTC/BTC ratio.  It is already starting to fall as ASIC gains traction
sr. member
Activity: 434
Merit: 250
Just kidding. I was pulling a prank on the community.
hero member
Activity: 714
Merit: 500
Eh, just saying it can be done much cheaper that way.

There are at least two companies/groups working on Scrypt FPGAs, you seriously think they're both spending a million+ for on die memory?

I believe they're using the method I suggested.

Wouldn't trust anyone saying they are working on FPGAs for Scrypt.  It is technically possible to build.  Hell they can just use system board RAM if they really wanted to.  It's a matter of cost (performance) and what the net result is.  Pound for pound they will realize they just invented a GPU with no net power efficiency improvement.
sr. member
Activity: 322
Merit: 250
Yes, Litecoin ASICs do not exist yet, but are not harder (or  costlier) to build than any Bitcoin ASIC. It's just a different hashing algorithm!

Scrypt has different parameters than SHA256, but those obstacles should be easily tackled.


So wrong all the way around.

It's clear you're an expert on this subject.


 Grin Grin Grin




Kelly Lee!  So nice to see you again.  How have you been?  Nice to see you back!
legendary
Activity: 1484
Merit: 1026
In Cryptocoins I Trust
Eh, just saying it can be done much cheaper that way.

There are at least two companies/groups working on Scrypt FPGAs, you seriously think they're both spending a million+ for on die memory?

I believe they're using the method I suggested.
hero member
Activity: 714
Merit: 500
A much cheaper and easier solution is to interface the memory on the PCB, no need to have the memory directly inside the die.

its a matter is distance.  its expensive to have to go out to the PCB.  updated my post above.
legendary
Activity: 1484
Merit: 1026
In Cryptocoins I Trust
A much cheaper and easier solution is to interface the memory on the PCB, no need to have the memory directly inside the die.
hero member
Activity: 714
Merit: 500
Yes, Litecoin ASICs do not exist yet, but are not harder (or  costlier) to build than any Bitcoin ASIC. It's just a different hashing algorithm!

Scrypt has different parameters than SHA256, but those obstacles should be easily tackled.

really this old arguement again.  scrypt is sha256.  scrypt does a salsa20 8 times mix backwards and forwards, it makes the lookup tables difficult to predict.  you need to load it all into memory to hash it.  it requires a lot of on die memory to load the tables into.  there are no fpga boards on the market that have that much on die ram.  it would cost a fortune to develop.  with most scrypt coins having low value and low liquidity there is not incentive for an investor to deploy the several million it would take to build an fpga or asic just so that can make a few hundred grand.  they woould be competing with GPUs manufacturers which have already invented the wheel.

Pretty much spot on, except it wouldn't cost someone several million to make an FPGA for Scrypt. For an ASIC, yes it will cost a million+, but FPGA development can be done for much cheaper. Around/under $100k if you were going to outsource all development, and much cheaper if you were going to develop it yourself.

to build an fpga board with on die ram?  there is no market for it, so they have never been produced.  it would have to be done from scratch.  it would be as expensive as an asic investment.

update  the closest is the stratix v.  dev boards run at $10k each and even then they do not have enough memory to handle scrypt.

http://www.altera.com/products/devices/stratix-fpgas/about/trimatrix/stx-trimatrix.html

you'd have to go out to external SRAM, or that QDR++ RAM (which is like going to 100 miles for a trip to the grocery store).  Even that is small in size. So you'd have to build a chip from scratch.  Then when you talk about DDR3 or DDR5 RAM that is like traveling to the moon to buy a box of ho hos.
legendary
Activity: 1484
Merit: 1026
In Cryptocoins I Trust
Yes, Litecoin ASICs do not exist yet, but are not harder (or  costlier) to build than any Bitcoin ASIC. It's just a different hashing algorithm!

Scrypt has different parameters than SHA256, but those obstacles should be easily tackled.

really this old arguement again.  scrypt is sha256.  scrypt does a salsa20 8 times mix backwards and forwards, it makes the lookup tables difficult to predict.  you need to load it all into memory to hash it.  it requires a lot of on die memory to load the tables into.  there are no fpga boards on the market that have that much on die ram.  it would cost a fortune to develop.  with most scrypt coins having low value and low liquidity there is not incentive for an investor to deploy the several million it would take to build an fpga or asic just so that can make a few hundred grand.  they woould be competing with GPUs manufacturers which have already invented the wheel.

Pretty much spot on, except it wouldn't cost someone several million to make an FPGA for Scrypt. For an ASIC, yes it will cost a million+, but FPGA development can be done for much cheaper. Around/under $100k if you were going to outsource all development, and much cheaper if you were going to develop it yourself.
hero member
Activity: 714
Merit: 500
Yes, Litecoin ASICs do not exist yet, but are not harder (or  costlier) to build than any Bitcoin ASIC. It's just a different hashing algorithm!

Scrypt has different parameters than SHA256, but those obstacles should be easily tackled.

really this old arguement again.  scrypt is sha256.  scrypt does a salsa20 8 times mix backwards and forwards, it makes the lookup tables difficult to predict.  you need to load it all into memory to hash it.  it requires a lot of on die memory to load the tables into.  there are no fpga boards on the market that have that much on die ram.  it would cost a fortune to develop.  with most scrypt coins having low value and low liquidity there is not incentive for an investor to deploy the several million it would take to build an fpga or asic just so that can make a few hundred grand.  they woould be competing with GPUs manufacturers which have already invented the wheel.
Thank you for saving me from have to say this.


I have it saved in a text file for easy cut and paste  Grin Grin Grin
Pages:
Jump to: