Hello !
When it was possible I put my 7950x in a cluster, 1 master and 4 slaves.
At that time, the quil quantity of the 4 slaves disappeared from my dashboard, keeping only the (300 at the time) quil of the master which started to increment faster, logically with the power of the 4 others. I told myself that you had foreseen this situation, since we could no longer do --balance.
Today, talking with friends, I realize that there is an increase in their daily Q as this 2.0 approaches, with 7950x, going from 30 Q/day to about 120-130. And my cluster no: if I divide by the number of nodes, I am at 40Q per day.
We could make a cluster with the principle of gaining a little more Q, but I think that is not really the case.
So I would like to break my cluster and go back to my 5 independent cpu. What do I risk? I get back my 300 x4 cpu plus the number of Quil that I did on the master (i.e. 300 + the Q done during the cluster)? Please enlighten me :).
Kind regards,
Tinou
Hi, do you have an idea pls ?
Tinou
Hi, going from mono to a cluster or viceversa, if you do it properly there is no risk.
Bear in mind that once your nodes enter into a cluster, the increment of your slaves doesn’t increase anymore (only the master).
If you consider that and the latest update Cassie provided saying that 2.0 should reach us in a matter of days, unless you have very high increment stores that you can use in your slaves, you may be better off as you are in the cluster.
The key during this period was to grow as fast as possible the increment, and then move those stores to high computation machines.
I dont see much of a downside. He isnt losing any rewards and he is gaining more rewards per machine.
The controller machine will retain its increment and the linked machines will go back to their own.
Agreed that he won’t lose any rewards if he does the switch from cluster to mono properly.
I don’t agree that he will earn more rewards per machine, that would only be true once the current slave’s increment will surpass the one of the master, and probably at this case this will never happen if 2.0 is around the corner, as he effectively paused the increment of the slaves when they entered the cluster. If the delta with the master is too big by now, won’t catch-up.
I did very specific tests with clusters, both with machines with the same specs and different specs, with very high increments, and my findinds were that the cluster was yielding more rewards than running the nodes separately, not something dramatic, but about 20-30% more. The biggest reward increment (in line with the 2-3x factor he mentioned, was due to the high increment). Same specs machines cluster was easier, because different specs ones required workers tuning though.
Good to know, but he has his misconfigured or something if he is still only making the reward of one machine.
That is why im saying at this point he would be better off getting the rewards from each machine.
But you’re right, if he had it set up right, technically it should be performing much better than each individually
Maybe, but i do all that here : Kingcaster | Quil Parallel Nodes Guide
My workers are 7950x x2 and 7950x3d x3. Cluster rewards are about 170Q/day (against 130 for each friend’s machines 7950x)