So, there’s a new (better?) plotter.
Uh, who has 416GB of RAM? I don’t even think my plotters can hold that much memory. I can’t even imagine the cost.
Nope, just checked. My two “budget” systems only have 4 slots that can hold 32GB sticks. So, 128GB total.
Definitely impressive, and highly efficient plotting on the right server. Made for filling PBs of storage and plotting as a service . Capacity and netspace difficulty will always be the deciding factor for chia profits, not plotting. Then again this goes back to chia really being a datacenter crypto. That is until datacenters can’t complete with the profit margins being below their operational costs.
In order to get that much RAM, you’d literally need a server that is dual Xeon E5-xxxx or dual AMD Epyc server that can do 16 DDR4 RAM sticks at 32GB each. The cost would be astronomical because you’d likely need ECC for whatever dual processor system you get with 16 slots. You can get DDR4-2666 ECC RAM for $150 per stick and at that cost, it is $2400 just in RAM. I could build 4 of my AMD R5 plotting systems for that price with money left over. What a waste of money in RAM alone.
I do agree that this is neat. If I had a system for this and the RAM to do it, I would probably experiment with it.
I have a server that can handle that, but it would cost me $1500 for the memory. And that would be just enough to be able to run that program. The server can handle 1.5TB of memory. So I could get 4 Samsung 980 Pro’s for that cost (almost). Of course, the memory would be dramatically faster and wouldn’t die after heavy use. So …
The system and ram are definitely an upfront cost but like you said it’s much faster and doesn’t die from using it so you still have the value of the system and ram when done. it’s like renting a Bentley for a road trip where all you pay is gas
Watching that is like … what can I say … chia ludicrous mode
I think this would be possible.
Wrap that NVMe as secondary RAM cache. And direct BladeBit to this directory.
Ugh the only wee bit of a problem is to find $$$ for a cool xeon server and shove 512 RAM in it
is there a Windows version? I have Threadripper 3970x with 512GB RAM I will test this…Amazing work
Single EPYC could do this, I have one that can do up to 2TB of RAM - but it’s still astronomically expensive, (only have 8 slots and 8*64GB ECC sticks = ~3000USD).
I just checked and the Threadripper 3000 series can do up to 512GB RAM depending upon motherboard support. Again, you’d need 64GB DDR4 sticks just like single Epyc or single Xeon. It would not be worth it to do it at the current DDR4 price.
also ,even with cheap DDR3 1333 memory stick of 32G found in 2nd hand market , it still cost around 1000 USD for 16X32G ,with current chia price, just not worth investing any more in this, my current 16X16G setup with full memdisk running madmax
At $3000 price point, it is worth looking at Intel Optane drive. You can stick it into any mainstream motherboard and don’t have to have a separate server build with all the expensive equipment that comes with it
I have a Dell Workstation (Windows 10) with 8 core Xeon, 512GB RAM and an additional 1TB SSD. Using MadMax plotter I get about a 43 minute plot time for both SSD and RAM. If a Windows version of BladeBit arrives I might try it to see if it actually produces valid plots in such a short time. If it does work, it might mean the early demise of k32 plots.
Thing is, you can resell RAM at a similar cost, it is not only about how much it costs to buy, but how easy it is to resell!
that’s kinda expensive but I think its possible to sell it (if something goes wrong) so that wouldn’t cost too much actually