Modifying KNC Titans To Work with Scrypt N=20

[BitcoinPappi]

Hi Guys,

Is it possible to modify the KNC Titan to support Scrypt-N=20. They are said to support upto N=19.

Looking at the code. What if the number of threads are reduced by half to allow the chip more room?
https://github.com/KnCMiner/knc-asic/blob/d83f1e67785eff0af00175d85564ace840c28186/knc-asic.h#L43

With regards to N level scrypt the only difference is the memory requirements from what I have read.

N=10 128KB
N=11 256 KB
N=12  512 kB
N=13 1 MB  
N=14 2 MB  
N=15  4 MB  
N=16 8 MB
N=17 16 MB  
N=18 32 MB  
N=19 64 MB  --> We are here
N=20 128 MB  --> We only want to go here
N=21 256 MB

Since the only requirement is the memory requirement I believe we can tweak it just abit.  My idea is to try reducing the number  of threads to allow the chips more room?
https://github.com/KnCMiner/knc-asic/blob/d83f1e67785eff0af00175d85564ace840c28186/knc-asic.h#L43

What do you guys think? Would be willing to pay for functional solution.

--
B

[1715309438]

1715309438

[1715309438]

1715309438

[1715309438]

1715309438

[BitcoinPappi]

Bump. Anyone let me know

[BitcoinPappi]

N

[ekrififi]

I'm not sure if I completely understand what you were looking to do...

Are there different variations of Scrypt N... some that can be mined with Titans, but others that can't be.

And you're trying to modify Titan's firmware update to allow your Titans to mine those OTHER Scrypt N algos?

Am I WAY off here?

Either way, Did you have any luck?

[gaalx01]

What are you now mining on the Titans?

[lenart_o]

Hi, anyone find solution??

[lightfoot]

Hi!

I tried to do this. The short answer is no.

The longer answer I found after dealing with the GarlicCoin insanity: Titans can handle larger N factors because they have memory on each cell processor, that's why the chips are so damn big. An increase in N factor means that you either need memory or CPU to work. N=10 works super fast on Titans not just because the chips are fast (they really are only about the speed of Neptunes) but because they have so much memory per cell they can use that to speed up.

When you go to N=11, you need more memory so it slows down. By N=16 the speed of the Titan is about the speed of a 8gb graphics card. By N=17, a 4gb card. By N=19 you need too much memory to do the work, and thus N=20 can't be done on chip.

Which moots the whole thing. Even if you could, a 2gb graphics card would probably outrun you.

[Searing]

Hi!

I tried to do this. The short answer is no.

The longer answer I found after dealing with the GarlicCoin insanity: Titans can handle larger N factors because they have memory on each cell processor, that's why the chips are so damn big. An increase in N factor means that you either need memory or CPU to work. N=10 works super fast on Titans not just because the chips are fast (they really are only about the speed of Neptunes) but because they have so much memory per cell they can use that to speed up.

When you go to N=11, you need more memory so it slows down. By N=16 the speed of the Titan is about the speed of a 8gb graphics card. By N=17, a 4gb card. By N=19 you need too much memory to do the work, and thus N=20 can't be done on chip.

Which moots the whole thing. Even if you could, a 2gb graphics card would probably outrun you.

thanks for the clarification...I always wondered about that

[lightfoot]

Yours for better science. However at N=20, mining with anything less than a 8gb video card is kind of pointless.

[BitcoinPappi]

Hi!

I tried to do this. The short answer is no.

The longer answer I found after dealing with the GarlicCoin insanity: Titans can handle larger N factors because they have memory on each cell processor, that's why the chips are so damn big. An increase in N factor means that you either need memory or CPU to work. N=10 works super fast on Titans not just because the chips are fast (they really are only about the speed of Neptunes) but because they have so much memory per cell they can use that to speed up.

When you go to N=11, you need more memory so it slows down. By N=16 the speed of the Titan is about the speed of a 8gb graphics card. By N=17, a 4gb card. By N=19 you need too much memory to do the work, and thus N=20 can't be done on chip.

Which moots the whole thing. Even if you could, a 2gb graphics card would probably outrun you.

Thanks for finally providing a response with details. As I remember, I had asked you to look into this since 2017 but did not get the explanation above.

[lightfoot]

Right. It was overflowing the buffers, and no amount of coding fixed it. Finally figured out why while working on the GarlicCoin thing.

N=11 though.... works.

[jcoin]

Is N=9 possible ?

[Searing]

Is N=9 possible ?

many have tired...on this topic of something besides scrypt-N...but any other scrypt, or scrypt-N or whatever .....has not been profitable

the last was FJC (fujicoin) but has not had any buyers in 1 week or so on any of the 2 exchanges left (bleutrade and yobit)

thus not a lot of hope in these areas. When electric rates hit summer rates on May 15th in the USA, even FJC will no longer be profitable

(assuming you can find someone to buy some) and I assume all Titan Cubes will be off the network at that time (33% on top of FJC loss..bye bye)

Unless you know of some other scrypt-N coin or someone who is doing a 'black swan' event of a coin with N-9

we are at a loss.

On a good note all them Titan Cubes make 'nifty' bookends and doorstops..for that complete ASIC miner, I don't give a sh*t decor

brad

[lightfoot]

Is N=9 possible ?

Should be. The N factor is basically a trade off between memory space and processing power: One of the key elements of the Titan (and the reason it ran so fast on Litecoin back in the day) was that each core had a fair bit of memory space. As a result it could run Scrypt with a slower processor but hash much faster because of the extra memory. This also allowed it to run higher N factors: I seem to recall the interception with 8gb GPUs happened at around N=15 to N=16.

Why for so?

[lightfoot]

many have tired...on this topic of something besides scrypt-N...but any other scrypt, or scrypt-N or whatever .....has not been profitable

the last was FJC (fujicoin) but has not had any buyers in 1 week or so on any of the 2 exchanges left (bleutrade and yobit)

thus not a lot of hope in these areas. When electric rates hit summer rates on May 15th in the USA, even FJC will no longer be profitable

Well, is there a reason to own/trade FJC?

[Searing]

many have tired...on this topic of something besides scrypt-N...but any other scrypt, or scrypt-N or whatever .....has not been profitable

the last was FJC (fujicoin) but has not had any buyers in 1 week or so on any of the 2 exchanges left (bleutrade and yobit)

thus not a lot of hope in these areas. When electric rates hit summer rates on May 15th in the USA, even FJC will no longer be profitable

Well, is there a reason to own/trade FJC?

Not at this time. Been saying that for a month. W/O KNC Titans no FJC and with no heat now that winter is gone and the pump is over...and no buys for

the last week...and add to that fact a 33% increase summer rates in USA May 15th...that is all she wrote boys and girls...

should have sold when you could. FJC has not been modified in any way since 2014 beyond keeping up with BTC core. Thus w/o Titans it will be

deader than dead. No inovation = No use for FJC = death of coin ....hell ..there won't even be a network for this coin after MAY 15TH...


MAYBE if the developer of FJC would have added master nodes or SOME hook since 2014..but the chance the titan's briefly gave this FJC is gone..

hell, even garlic coin is 39x the price of this FJC

So back to topic, FJC is gone and Scrypt-N of any flavor is moot..the time has come and gone



NOPE....shut my knc titans off 2/14/19 with FJC ...have some left...sold the rest

glad I got out when I did

(however again, many doorstops and bookends can be made for friends and family with KNC Titan Cubes)