GrinderPro FPGA

The new plot format will prevent grinding. We have many knobs we can turn. We expected this could happen eventually just not so soon.

Iā€™m sure CNI has lots of knobs; Iā€™ve observed some of them in action over the past few years

3 Likes

They claim it works on the new format as well, even requiring less VRAM, so potentially a bigger market than just M2 Ultras with 192G.

Of course itā€™s just a claim so far.

If you scroll up, itā€™s clear this isnā€™t a traditional grinding attempt where you do the full compute. I still have no idea how they can spoof so much netspace with so little compute, and so little pre-compute.

I wonder how they can make the claim that it will work with the new plot format when they donā€™t know yet what the new plot format even looks like.

2 Likes

Remember plot grinding has always been possible since day 1 but not economical. It was even done on a supercomputer shortly after mainnet just to prove it. GrinderPro developed a novel method that is economical. No one here has come close to guessing how it works.

While we arenā€™t sure if the new plot format will prevent plot grinding for sure we donā€™t think Chia has figured out what we are doing so there is a good chance they will make the same mistake.

1 Like

Awesome. Let us USA users try this, please.

One more question - why are you guys sharing this tech instead of just grinding the crap out of the blockchain yourself? Take all the XCH from farmersā€¦

As mentioned, ready to add credibility to the claimā€¦ā€¦ā€¦

Not sure why such radio silence or proof of this IS trueā€¦.

Iā€™m pretty sure this is a NESTED Hellmanā€™s attack.

Unfortunately the new plot format will have the same problem as they are limited by the fact verification needs to happen fast on consumer hardware.

My other guess is it is a weakness in their memory hard functions

The attack computes arbitrary malicious proofs using less than a megabyte of memory (about 1/3000 of the honest proverā€™s memory) at a relatively mild penalty of 170 in computation. This is more than 55,000 times faster than what is claimed by the designers.

The attack requires a one-time precomputation step of complexity , but its online cost is only increased by a factor which is less than 2 when spending precomputation time.

NOTE THE PRE COMPUTATION STEP. This is might be what GrinderPro is doing during initialization in the cloud

@brandon Do we know how big the files are yet?

1 Like

From what I can tell, they never do any plotting. Not in the pre-compute step and not while ā€œgrindingā€. 192G is not enough VRAM to do a full phase 1, and it would be way too inefficient anyways. Their pivot to GPUs was also executed quite fast, suggesting the algorithm is more simple.

That would mean they broke Chiaā€™s Proof of Space in a cryptographic sense. Not like a hellman attack, which is a tradeoff rather than an exploit.

I want in.

Am travelling again to China soon.

Note in beyond hellman they waive away the fact they canā€™t prove proof of space is secure by claiming if you nest or repeat the function it doesnā€™t matter

ā€œThis protocol has a significant
soundness and completeness error.ā€

When used as a replacement for PoW in cryptocurrencies, neither the sound-
ness nor the completeness error are an issue. If this PoS is to be used in a contex. Where one needs negligible soundness and/or completeness, one can use standard
repetition TRICKS to amplify the soundness and completeness, and make the cor-
responding errors negligible."

Let me tell you about this one simple trickā€¦

Yes we do.

Man great to see XCH pumping!!!

I wasnā€™t sure whether it was worth flying out to China to partner the GrinderPro team but now it is paying off.

Donā€™t worry normal farmers we aim to only sell a small percentage of the total net space to the public to keep price stable. Look for a YouTube review soon and then website launch in 2025. The software is still command line only and we are working on a GUI.

It is not economical to grind with the H200. Running 24 hours a day it would be around $72 a day and it cannot grind anymore space than the M2 Ultraā€¦

It makes much more sense to use the H200 for only the initialization step.

We donā€™t think it is worth it to buy an M2 Ultra just for Chia grinding but there is a huge user base of M2 Ultra users in the local LLM space we are targeting to run GrinderPro when their computer is idle. We expect this user base to expand with the upcoming plot format (the size reduction may allow Mac laptops with less memory to grind).

LOL so the new plot design is designing itself directly into a preferred grinding setup? lol

PM the info! Ready to test it and validate, more than happy to use command line

We are getting reports of random shutdowns. If you are getting an emergency shutdown error due to overheating please make sure your M2 Ultra has plenty of room around it (not on a shelf) and make sure the air inlet is clean.

Note it is only stable at 7 PiB if you have zero background tasks running. Make sure to disable adobe/dropbox/office/etc.

Happy to test it out. Machine would only be running thisā€¦ā€¦

Did anyone else on the waiting list get the discord invite? I got the invite but they want me to give them my phone number which Iā€™m not going to do.

1 Like