Where you get this information? What is partial grind?
Iâm replying to spacemesh user saying:
Iâm just assuming partial grind is where you drop entries on table 1 or 2.
If you tell me system spoofed total TiB sure.
How much external RAM per board?
The 3 stack system will spoof 297 PiB with water cooling. If you want more details please sir visit team. Please do not try copy.
Is this the part you are looking for - XCVU13P-1FHGB2104E AMD | Integrated Circuits (ICs) | DigiKey? And how many of those you need to support 1 server setup (297PiB)?
Itâs just not remotely believable. Iâm not far from China actually, but Iâm a bit scared to go there, and Visa is difficult to get.
2 Likes
$55,820 for one freaking FPGA! Good luck with that suckers! What a joke.
You will never get your money back before the new plot format comes out. No wonder they are sharing this now.
1 Like
$700 for used ones
Stick with mail order, you donât need to be picked up and put into a box!!
1 Like
How you find? Thank you sir very good price. Problem PCB house no allow used. Must buy from them. PCB supplier say no stock and no lead time estimator. Maybe we must try more PCB house. Try move fast with partner. Not much time.
Why not use a similar FPGA thatâs in stock? Nothing special about VU13P.
Bit longer timeframe than I was anticipating, but still not enough time to pay for any FPGA before it becomes 100% worthless.
Do a new layout is probably no small undertaking. If they are doing an all-to-all FPGA board stack each board could be well north of $100k for a small volume prototype run. With the new plot format and the crashing xch price investing any more money/time in a new design is probably not in the cards. I would love to get my hands on those gerber files. I wonder if they will ever open source the design.
1 Like
Theyâve got to be interconnected using the 8x 150G Interlaken IPs, otherwise custom boards doesnât make sense.
Thatâs exactly what I was looking into when I found DrPlotter method last year. I wanted to use HBM though, and the interconnect just wouldnât keep up with that.
GrinderPro seems to use regular external RAM on the other hand, maybe 128 GB per board. Which makes it even more strange, cause itâs slow.
So they need lots of RAM but somehow plotting really fast. All that RAM doesnât fill itself quickly. Weâre talking about phase 1 in 2 milli-seconds, that would normally be ~500 TB/s RAM bandwidth. Latest top of the line HBM can only do 1 TB/s.
1 Like
multiplexed fpgas with all-to-all complete graph functionality can be orders magnitude of faster than state-of-the-art GPUs when tackling p-compute tasks
These academics are probably Ising Machine researchers and have created a p-bit machine. Not sure why they are wasting time on Chia.
Are you saying I âthreatening them and their familiesâ ??? If so you need to point to exactly where that happened.
I never even heard of an Ising Machine until I googled it. I wonder if Shenzhen University has a probalistic computing department? We can probably find out who these guys are.
1 Like
Is it time to put rallbright into the river???
Not at all nor did I suggest that either. However someone making up absolute shit, then saying I said that made up garbage, is not okay on multiple levels. Thatâs gaslighting the fuckall outta people.
This thread is a larp until any proof is shown. No one has any proof to show on any of multiple fronts now. Max is right about the math, it just doesnât make sense. Sharing a picture of a machine or a screenshot of a terminal is a very low barrier that holds zero risks of someone reverse engineering anything.
actually Max calculated 2TB of ram would be pretty much what you need to store the plot ids to find enough ids that pass 3 signage points in a row at the required rate and gives it credibility. check the madmax discord
2 Likes