Bitcoin Forum

It's a great point I havne't heard discussed.

Litecoin was supposed to be designed so only CPU's could use it and not GPU's, then scripts were written to use GPU instead of CPU and get way better hashrates

what if same scripts can be utilized to mine litecoins on ASIC?! (or another script WRITTEN to do it)

no.

/thread

this.

And yes, it has been discussed many times.

whats the problem?

https://bitcointalk.org/index.php?topic=116987.0 (https://bitcointalk.org/index.php?topic=116987.0)

the word 'litecoin' does not exist on that page

so what. BTC algorithm is different from the LTC. So is there answer to your question.

NO !!! :)

Before there was LiteCoin there was TeneBrix (or maybe shortly after, I dunno they all sort of came at once these altcoins). They both use an algorithm called sCrypt rather than Bitcoin's use of SHA256. SHA256 is a series of outrageously simple integer operations that require practically no memory to perform. sCrypt on the other hand is explicitly designed to use quite a lot of memory. CPUs are a general purpose processing device with a pretty fair amount of memory on the chip. GPUs have quite a lot of memory available to them, but are designed to share that memory evenly between each of their several hundred processor cores so the idea was that sCrypt would be GPU resistant because GPU cores have a harder time getting their hands on a decent pool of working memory - I haven't looked over any code to see how exactly anyone has beaten that limitation, but it was sort of a soft limitation.

The TL;DR version, what you REALLY need to understand about LiteCoin to get this answer is this: The thing holding you back from mining faster isn't the bit of the processor that's doing actual math, it's the amount of RAM your processor has and how fast it can access that RAM. RAM that's on the CPU itself = crazy fast, the RAM clicked into slots on your motherboard = comparatively quite slow.

So now imagine you're building an sCrypt ASIC... You design a special purpose circuit that does the math and put it on a chip with a fair amount of RAM. Congratulations, you've just re-designed the CPU you already had, but at greater cost, worse power efficiency and it's only ever going to be good for one thing thus limiting its potential resale value. Woo.

In short, you already have LiteCoin ASICs - they're manufactured by Intel and AMD.

You obviously cannot mine LTC with a bitcoin ASIC (Application Specific Integrated Circuit), if that is what you are asking.


However, this just isn't true.  It would be possible to design a chip that only performed scrypt and pair it with exactly the right amount of ram, and it would crush generic hardware in terms of efficiency.  Specialized hardware is always going to be faster than general purpose CPUs and GPUs, especially for highly parallelizable  tasks such as mining.

Is it worth it at current LTC prices though? Of course not.

No, neither Intel nor AMD make devices with the sole function of mining Litecoins.

Maybe a better question would be:

Can I harvest FPGAs from existing BTC mining boards, and put them onto a new board alongside some fast memory and use that to mine LTC?

Harvesting SMT chips already soldered onto boards sounds somewhat difficult, but is probably practically doable.

Theoretically, you could.  I'm not sure if all FPGAs have the proper IO to interface with memory at a decent speed though.

Except that there's no "exactly right amount of ram" for sCrypt.

sCrypt, like bCrypt is defined over a largeish table which is constantly accessed and modified, both in size and content, during the algorithm's run. The amount of space this table will take up is effectively a pseudorandom value dependent on the message so each hash will take a different amount of RAM than the last.

That's the entire reason sCrypt based altcoins were supposed to be GPU resistant and should still remain ASIC resistant: their memory requirement is both large and random and the algorithm itself is designed to access that memory as much as possible, thus making performance of a brute force attack against it heavily reliant on memory latency.

Again, I haven't looked at code to say how the GPU miner actually works, but the theory was that since all but a very very tiny amount of GPU memory is "pooled" in such a way that only one core can access it at a time, any GPU miner would have most of its threads deadlocked for use of that one shared resource most of the time. I'm tempted to say that the miner probably uses a very clever workaround to solve the locking problem, but I sincerely doubt that they've found a way to sincerely perform brute force against sCrypt without maintaining and heavily utilizing that large table in shared RAM.

I'm not arguing that an ASIC couldn't be built, just that it's unlikely any ASIC will have significant performance gains over mass-produced and readily-available CPUs. CPUs are already very close to ideal for this application and any optimizations gained by building a custom one-use-only chip are unlikely to be economically meaningful, regardless of whether such an increase is technically meaningful. If you produce a miner that is twice as fast as a CPU but costs 4 times as much, I don't see people lining up to buy it.

There is an exactly right (optimal) amount of ram for a given set of computational units. CPUs and GPUs do not have their cache/memory specifically designed for scrypt, and therefore either have too much or too little of it.

I found this post on the subject: https://bitcointalk.org/index.php?topic=98535.msg1081219#msg1081219 (https://bitcointalk.org/index.php?topic=98535.msg1081219#msg1081219)

well its making more sense...

if litecoin becomes succesful, only then would we look for litecoin ASIC's

if one thinks it would, litecoin may be a good long term investment right now

Another thing to keep in mind is the price of bitcoin and litecoin.
Amount of bitcoins mined in a year (with current 50BTC/block) is 2,628,000 at current (BTC-E) price = $ 31,793,544
Amount of litecoins mined in a year is 4 times more, 10,512,000 at current price (BTC-E) = $ 777,888
Difference of a factor of 40, so a lot less money to spend to make hardware to mine it.

Thanks for that

Don't ask these questions about what an ASIC can do, you'll just get yelled at

 o.0

I have a few Xilinx ML605s coming my way soon, if I switch to lite coin, could I use those to mine?

Litecoin uses 128.5kB memory, I doubt the fpga itself has this memory, so you need some memory chip(s) besides it.
And even then, currently there is no litecoin bitfiles to load to the fpga, so those must be created also.

eventually it will be created then. minimum of 1year

I'm not sure about that.
A few posts back I calculated the total mined bitcoins vs litecoins converted to $, the value of bitcoins mined was 40 times higher than litecoin.
So if it will be created, the profit will be a lot less, unless ofcourse if the litecoin value rises a lot more compared to bitcoin.

The FPGAs I will be using have a few MBs of memory on them, how hard would it be to make the bitfiles and whatnot, that would have to be done in verilog right?

Wait, can anyone explain in greater detail why LTC mining only uses 128kb of memory, but yet any scrypt mining program use enormous (GBs) amounts of memory?

What kind of FPGA is that?
The Xilinx spartan-6 LX150 only has 18 Kb blocks if I'm correct.

The scrypt hash calculation only uses a little over 128kB, but if you run it on a quadcore cpu there are 4 calculations at the same time, so 512kB is used.
If you run it on a gpu, with for example has 1024 calculation units, you run 1024 calculations at the same time so you need 1024 * 128kB = 128MB.

So when I set --thread-concurrency to 8192 (4x 2048) for my 7970, it's using 128kb * 8192 = 1GB. So why does it use so much more than that?

how many gpu threads are you running? 1, 2 or 4?

I don't remember. I only tried it once, and went back to BTC. So in my case, will each thread use 1GB of memory?

I don't know in detail how the gpu litecoin miners work, but what I guess is indeed something with gpu threads.
Maybe when the gpu is calculating on the 1GB of work, the cpu already puts the next 1GB of work in place.
That way the gpu can start the new work immediately when the old is done.
Doing this with more threads would probably lower the chance all work is done before new work is ready.
But I'm no expert for litecoin gpu mining. Just a computer programmer thinking about what approach he would choose to do it.

And another thing, litecoin doesn't use 128kB but a little more, around 128.5kB.
It is possible it's faster to align on some boundary.
If the boundary for memory blocks is 64kB the total amount of 1GB becomes 1.5GB.
With 4 threads putting work ready it becomes 6GB.

Virtex-6 (ML605), not 100% sure exactly how much memory is on it, just going by what a friend told me who works there

I think I was wrong with the 18 Kb blocks.
According to http://www.xilinx.com/products/silicon-devices/fpga/index.htm all the 6 and 7 serie chips have enough memory.

If you are serious about pursuing this you might want to have a look a this: http://www2.engr.mun.ca/~howard/PAPERS/ccece07_yan.pdf (salsa20/8 is an underlying algo used in Scrypt)

They actually used an FPGA and 3 ASIC variants for Salsa20 in this paper, and the compact FPGA variant uses 194 slices and 4 Block RAM's on a Xilinx 2V250fg256

does that prove anything

that ASIC was made for litecoin mining?

No, it was doing salsa20, but not the full mining algo. This is the hard part of Litecoin though, the memory intensive part that makes stock FPGA bitcoin mining rigs unsuitable for Litecoin.

I was mostly linking to the resource for anyone who wants to incorporate parts of the design detailed in the paper into an FPGA or ASIC based LiteCoin miner. There is a ton of stuff out there already to springboard from, even an open source VHDL scrypt implementation: http://opencores.org/websvn,listing?repname=salsa20&path=%2Fsalsa20%2F&rev=1
This unit, a couple SHA2 units, some golden_nonce detection and you can have Litecoin a ASIC in 6 months with the right folks and a couple hundred thousand dollars. Most likely we will see an FPGA generation built before ASIC is profitable for Litecoin.

BTW, nothing says someone couldn't create an ASIC that could use SHA2, salsa20/8, and other future algorithms on the same chip, but each algo will need to justify it's own design time and die space.

So is this a yay or nay on using xilinx ML605s to mine?

If the is the full ML605 I think that has a DDR3 interface on it. That would in theory allow you to use some of your SP's to add load/store for a large number of inflight scrypt attempts.

I don't think this one will be as easy to build a bitstream for as BitCoin was, and any FPGA without a large high-speed RAM capability is going to be limited to an insanely low performance (think 1000 cycles per attempt without any pipe-lining.)

At a high level the challenge is going to be handling the random requests for a portion of the large random vector that is used by scrypt. This means that each pipline will need to manage enough parallel processes per engine to keep the pipeline full during the 30-50 clock cycle memory lookups, since we are not using Block RAM.

If someone wants to use this information to build the code, then at a high level I think it could work. I'll happily work with any experienced FPGA devs that have an appropriate board, but I don't have the programming skills to put this together solo, or a board staring me in the face demanding I put it to work.

I wonder if any ASIC manufacturers out there could be working on a LTC ASIC?  They could get into a dominant market position like BFL did if they got in early.

This would be somewhat harder to do than a Bitcoin ASIC miner.  For mining BTC, someone can just license an existing SHA256 IP core, or use an open-source one.  Scrypt, on the other hand, probably doesn't have ready-made cores "out there" like that.  A maker would have to design and produce their own.

I think the ASIC makers likely have their hands full and then some handling their current product efforts, but you're right, there is an opportunity here.

True, I don't think it's coming soon from the existing players, maybe a new party in 12 months? Existing in 18-24? if ever it is worth enough (to justify development and) to sell dedicated (and even more specialized) ASIC.

On the Scrypt ASIC thing, I think it's less hard than it sounds. Scrypt is just a Salsa20/8 core with a sha2 core before and after it, with enough DRAM to store the inflight variables or external RAM and a memory controller. It's all mainstream IP, there is even the VHDL I lined to over at opencores for Salsa20 (/8 means it is Salsa20 with limited iterations). So we really just need someone who thinks there in enough profit in plumbing them together and getting it fabbed, same as the Bitcoin ASIC's.

Right now there are ~$60k worth of LTC being generated each month (30 days at current difficulty, at current full network hashrate) compared to $2.6M each month worth of BitCoins, and we are just hitting the point where ASIC makes sense on Bitcoin. I think Litecoin mining is going to have to be worth a LOT more (30-40x?) before someone invests in the ASIC implementation.

How can you be a Hero member with 1900 posts and not know this?????

Because I'm 420 man

Maybe it's time for a bets of bicoin

"A company will be working on a litecoin mining prototype ASIC by the end of 2013"

do it

i'm too busy to take the time to word it right and specific to not be rejected by betsofbitco.in

I'd do it myself, but I don't have enough BTC xD

It seems that we will see FPGAs before ASICs... aren't they easier and cheaper to develop?

Therefore, ASICs by the end of 2013 for LTC seems doubtful.  ???

Also, there's a lot less money in LTC compared to BTC due to current values of each.

Yep, FPGA is generally used to prototype until ASIC designs are locked down and can be produced. Sometimes the transition to ASIC comes before shipping the first product, sometimes it never happens. It all depends on the comparative economics and performance you need. In the case of LiteCoin we want to do as much work as possibl, so ASIC is likely to have a speed/cost/density performance in the long run, but a stop at FPGA (or a year or more of just waiting with just GPU's) market more mature before the economics make sense. Heres some numbers I did on the last page to show the scale of difference:

"Litecoin ASIC miners will be available to pre-order in 2013"

http://betsofbitco.in/item?id=824

That's a harder bet to call, nice.

pre-orders could be up to 4-6 months ahead based on BFL, which means we are more like 18 months out. I'm not betting this early!

kick booty. i had to vote agree.

i think if bitecoin asic's come out within 3 months, there'll be a rush for maybe up to 5 months of releasing new versions...could be a lot longer

but somewhere people are going to think...its hard to keep up with bitcoin asic hardware to remain profitable, but theres this big market demand over here for new litecoin mining hardware...

and apparently all they need is the idea and some feasibility to start the pre-order process where they would get the money to then RESEARCH AND DEVELOP it LOL

I think you are seriously overestimating the profit potential right now., as I said earlier:


This means that you can make the same amount of money as a vendor capturing 2.5% of the BitCoin mining hardware market as you could capturing 100% of the LiteCoin market.

The price of LTC is going to have to shoot up like a rocket ship to get the kind of potential goods and services market built around it that BitCoin does. Right now it is a tiny fraction.

420 is 420 thinking we see ASIC in 3 months ;) maybe FPGA in 4-6, and ASIC in 18 would be more my guess.

There's already been a lot of talk about this in the alt chain forum. Basically any ASIC made now, unless it has a massive memory bus and high speed ram e.g. gddr 5, will be really slow. I would guess it'll be a little more efficient, maybe 2x-10x. But anyone making an ltc asic really has an uphill battle against gpus because they're so damned cheap and fast.

I can hardly wait :)

I so wish I would've stayed in LTC.

Yep, once you are out of LTC you are out for good.

http://alpha-t.net/product/scrypt-asic-miner/

They are verry near.

ASIC for scrypt is real now. Most of people who said no in this topic look stupid  ;D

Were pre-orders already taken?

They look stupid for correctly explaining the intentional difficulties scrypt introduces for scaling on relatively cheap core density like basic sha2?

Note, basically everyone in this thread made it perfectly clear that the discussion was based on the question of whether SHA2 asics (which do not contain sufficient cache by design, it's the most expensive component of any consumer compute device)  would ever being capable of mining scrypt effectively.

Perhaps you don't know the definition of 'stupid'.


Also, until one of those devices is in the hands of a regular customer, then we have nothing more than the same rubbish BFL fed you guys for months and there's most likely not going to be nearly as huge of a price/hashrate advantage as we saw with sha2 asics from most of these first gen designs.

Reading this thread was like jumping in a time machine.  :)

gridseeds are already outdated http://techcrunch.com/2014/03/20/kncminer-sells-2-million-worth-of-scrypt-mining-machines-in-four-hours/?ncid=tcdaily
100mhash for 10k

pre-orders are for noobs  ;)

Lol... just joking. But not really though...

how to invest with the adjustments to these altcoins

How are litecoin ASIC's these days?

You can make more BTC by pointing ASIC miners at other alt coins if you are smart :-)