This topic is temporarily closed for at least 4 hours due to a large number of community flags.
This topic was automatically opened after 30 days.
any updates on this
firewalk studio concord?
âDr. Yangâ is now promoting his own, new scam coin. Strange that he injected himself into this conversation with no mention of that.
This thread and his new coin are somehow related. Beware of both.
Is there any registered business behind either? Do they have a business address? Can you even verify the existence of these people?
I have chased many scammers here. The easiest pointer to a scam is no real business or contacts. If they are real then you can easily find out their corporate information, who is writing their code, etc, etc.
The second largest pointer is when you are presented with a guaranteed profit. If they actually had such a system then they would use it, not share it. They are always having trouble securing financing and want to do it privately. Bullcrap.
2 Likes
bro im not here for that I just want to watch the world burn
I was able to visit the team last month. The technology seems to work but the total system efficiency is less than advertised. This is not a simple plug and play setup.
We are looking at producing another set of boards but it might not be worth it with the new format coming out. The boards are very expensive as they are 12-layer boards and you need 3 to make a stack.
2 Likes
What is their total system efficiency?
Yeah would be interesting to know how many orders of magnitude itâs off.
1 Like
Their advertised numbers are correct if you donât take into account the initialization. You have to take apart the FPGA stack to load the configuration. There is one âinitializationâ config and another âgrindâ config. Once you have it in the âgrindâ config the efficiency is as advertised. Itâs cool the same FPGAs can be used for initialization as grinding but it is a pain to setup and reconfigure. You need to be an expert.
How long does this âinitializationâ take? And does it require storing the result somewhere, like on HDDs 
?
Or it just fills the 2 TB of RAM? But in this case I wonder why you couldnât store it on SSD and just load it back into RAM after a reboot.
The technique is subtle yet simple. We are working on developing a software version where the initialization is done in the cloud (to prevent reverse engineering) and the user has a local GPU to âgrindâ the downloaded file. This will not be as efficient as the FPGA stack setup but it should still be worth it depending on XCH price. The software will be closed source so you should run it on a segregated machine.
It sounds like an extreme version of DrPlotter, I had this idea over a year ago. I also looked into doing it on FPGA, but concluded itâs too expensive and too much work.
The main issue with this approach is the time it takes for the âinitializationâ, as you call it. Because it requires plotting a lot more plots than you can spoof later, like 13x more.
So how big would this file be for letâs say a 4090?
We are not sure a 4090 with 24GB of VRAM will work. Our test system is apple studio m2 ultra with 192GB of unified memory.
Uff, but the bandwidth on that memory is low isnât it?
EDIT: Oh itâs 800 GB/s, not too bad then.
Yeah and itâs stable up to 184GB of VRAM if you manually allocate by setting iogpu.wired_limit_mb
So youâll code it in OpenCL or Metal?
Stupid question, is the prototype actually farming blocks right now, or just spitting out proofs?
If itâs not farming blocks, you might be getting fooled.
Brandon,
Please donât share anymore information untile release software. We making XCH now with software grind but still need FPGA for initial. Not ready to release. Right now plan mac software only. Team is working cloud infrastructure. We expect technique will work with new format based on discussions.
Thanks you,
GrinderPro
GrinderPro on unalive watch right now.
That be interesting lol. I still find it hard to believe.
Since you require lots of VRAM (~184G) it means youâre doing full grind or almost full. But this is very slow, at least 10x slower than DrPlotter Pro4x.
So somehow you know a plot will win or have a much higher chance (like 100x) of winning before doing the grind?