I can probably use Google but this completely caught me off guard. Did Chia not say they are asic resistant? Asic proof as it were? This completely caught me off guard. Am I out to lunch on this topic? Asics have been GPU killers. My point being, asics are never a good thing to hear about unless you are a corporate miner (farmer) or industrial size operation. Thoughts? Maybe this is a moot point because storage cost is the biggest barrier, not how fast or easily you churn out plots.
I didn’t see the latest AMA, but quite sure they would be referring to the Asic Timelords they are developing. Fastest Timelord wins, so for this part you can create Asics.
Interesting.That is on a level I don’t quite comprehend. I wouldn’t know the difference and how or what impact that could potentially have when it comes to the average farmer. I swear when asics were mentioned previously, a while back yes, Gene or Bram was very blunt in their claims regarding asiscs. They didn’t imply there was some suitable asic use for Chia. I’m definitely baffled.
ASIC for timelord is not equal to plotting nor space. If im not wrong, its to improve the VDF, and VDF is used to calculate that every farmer read the plot in the same time or something like that.
Run an ASICdont make you earn any XCH.
Well yes. For Chia farming (having plots and finding blocks) no asic is really possible. Storage space is storage space so if someone invents a better way of making storage space this can be adapted for any general storage purpose and therefor will never be an asic.
Timelord run the VDF function on the Chia blockchain (above my level of understanding really but I do know that that the fastest one wins, so there is always effectively only one Timelord at any single moment.)
These are computational, so you can create asics for it
As it doesnt really matter too much who has the fastest Timelord, it’s not a super big issue and there is no point in having millions of them anyway because you only need one.
I appreciate the extra explanation. Definitely a technical subject. I see the word asic, sorry, it means threat. This happens when you get teeth knocked out. You get scared…
1 Like
I read about an attack long ago where plots with correct proofs could be made on demand, now if an asic was created to do this were all fooked, but I very much doubt they were the asics being mentioned in the ama.
Such an asic would kill the coin and the profit I’d imagine as it would become a pow coin and not be at all green.
1 Like
Same here. I have ignored that issue, assuming it was not feasible. However, maybe this problem has to be revisited.
The original claim was based on Chia’s plotter that was taking 10 hours to generate a plot, where the final proof needs to be produced in 30 seconds. As that plotter was rather primitive, potentially that statement was worth as much.
Then came MM and instantly shrunk plotting times to about an hour. The first reaction was how close that was from generating proofs on demand, and as such the first response from Chia was that MM is bad, don’t use. That one hour plotting time was still far away from what was needed to break the system.
Then people started optimizing their systems bringing plots to 10-20 mins with MM, and then BB came along shrinking plotting times to ~5 minutes per k32 plots. Now we are only about 10x away from being able to generate plot on demand on off the shelf H/W.
Plotting effort can be broken down into three components: 1. CPU crunch time, 2. RAM speed / memory access, 3. Final delivery to NVMe/SSD/HD. Assuming that someone wants to build a custom H/W (ASIC based proof finder), #3 is gone. Standard CPUs are rather bad at handling floats (long divisions / multiplications), and this is where ASICs shine. So, we can assume that #1 (crunch time) potentially can be reduced by 10x on a custom ASIC (Max mentioned that the next step is using GPUs for plotting - would be interesting to follow him, how he will address that with his MMX). That leaves us with RAM speeds, and those can also be improved (not sure if 10x, but today usually RAM is not the choking point for plotters). So, maybe such H/W is not that far away. Although, as you mentioned, once introduced, there would be no room for us farmers, but rather a competition between such ASIC based harvesters.
What I don’t understand is whether it may be a way to further speed up such on demand “plotting” by producing sub-k32 plots. Whether such “plotter” would need to generate a whole k32 plot, or rather knowing what it needs to find, generate a sub plot. If such sub plot could be predicted (based on a given challenge), that would further shorten the proof producing time, making such efforts potentially feasible.
I am not saying that I do understand the proof generation process, so all what I wrote could be just garbage, though.
4 Likes
I dont think what you wrote is garbage, but i don’t fully understand it either…
Some asics now have ram with them, I can’t see why one couldn’t be created with nand added.
The big question is, would asic manufacturers bother? If they can’t sell the completed product en masse, I dont think it likely.
Nice to see your still about, been a while, be well.
1 Like
I agree that at this point that route is potentially not worth to pursue. Assuming that there is one such H/W produced, and it takes 50% of all wins, that will put most of us farmers out of business. If we are gone, then the number of nodes will collapse basically making the whole thing worthless. To me, this makes ASICs not feasible at the moment for chia, not really feasibility of getting such H/W produced.
The reason that it makes sense for POW coins is that the only thing in place is raw CPU speed (GPU vs. ASIC), so having pools of GPU miners maybe cannot compete with those with ASICs, but at least be able to participate in the game. Where with chia that would be the end (at least based on what we have today - small ecosystem).
“ASIC” is a shorthand for a custom based computer that instead of an off the shelf CPU has a specialized silicon (for simplicity, let’s say it can divide really fast, but additions are on par), so whatever memory makes sense can be included. We (farmers) are dealing right now with DDR4, where the latest GPUs have already DDR6. If those engineers can produce custom processors (ASICs), I would assume that they can also produce custom memory controllers and use the fastest RAM out there. My take is that the cost of such effort is not really in producing one box, but rather the very first silicon. So, the first such H/W price tag could be few hundred thousand dollars (what is actually zilch comparing to how much we farmers spent on HDs - $500-$1,000 millions).
1 Like
Lol, the cryoto noob trying to school the old work horse!
Made me giggle.
I understand exactly what an asic is.
Really? I’ve only heard of ddr5, and thats crap so far price for performance.
To be fair I’ve not bought a new gpu since 7970s, and they still work fine.
1 Like
Wrote that on purpose Actually, not to target you, but eventually others (as I saw some answers in another thread about C++ that didn’t make any sense). Still, the point I wanted both of us to agree (what we do) is that using 'the best" RAM for such box is not a problem at all. Basically, just restating that if someone wants to build it, I don’t think the H/W is the problem, rather the ecosystem.
Maybe GDDR6 is what is used. Early last year, I read that Samsung is working on DDR6. It should be 2x faster than DDR5 (thus 4x than DDR4). As such ASICs are not mass produced, and potentially cost is not the issue, there is no need to have DIMM modules, as one can create a custom board.
1 Like
Bitmain mass produce asics dependant on your definition of mass.
But they won’t invest in a new one, or order wafers
unless there’s a market for them.
1 Like
That was with respect to demand for chia ASIC - basically no demand, as even a single unit would be a chia killer.
1 Like
The replotting attack was discussed at length with the advent of MM. You only have to create phase 1, the rest is mostly compression.
1st thing to note is that a replotting attack only make it so that a plot always passes the plot filter, it still doesn’t guarantee you find a proof.
With that in mind, you need expensive hardware just to get one plot to pass the filter every time. When you compare this to the economics of having enough plots to always pass the filter, it 's just much cheaper to have hard disks with plots than it is to make an expensive asic or something similar that can plot that fast.
4 Likes
Just to clarify a point, not sure relevant, but in the AMA, Gene spoke of Chia manufactured asics. That was how I understood the comment. I hope there isn’t funny stuff going on or some other “surprise” that puts the average person on the outside. Like, thanks for building up the backbone, but now we have some asics we’re going to sell to the corporate audience. This is just my negative side speaking and those with far greater understanding have already weighed into this. From the get go, there has always been the undercurrent that Chia farming is not going to be for the regular folks. I’m just getting mixed messages is another way of saying this. I listen pretty hard to what the leadership of Chia says and I’m going with words spoken from the horses mouth. For the most part.
2 Likes
These are the Timelord ASICs we’re creating - Chia Partners With Supranational to Create Industry Leading Proof-of-Space-Time Security | Business Wire
These are all of the attacks and their mitigations: 3.14 Relevant Attacks and Countermeasures | Chia Documentation
ASIC Timelords make the network and the space much safer.
-Gene
9 Likes
I appreciate the clarification on this. The official clarification.
Thank you so much for chiming in. We all appreciate it a lot.
What incentive is there to be a time lord then? What incentive is there to run a faster time lord?