Quote:
My thoughtprocess was, that in generell we know approx with how many Kombos we will "arrive" at the river. And while playing we know some approx percent numbers for for example betting the river or ch/folding. And then we execute our play on this information. Thats why I thought that unadjusted frequencies are much more usefull. But as it is not in that, I was curious what I miss out
Yes, exactly this. It's easier to reason about unadjusted number (you know that if you bet with a blocker you will be called less often) than about "real frequency" one which requires knowing your exact gto range and adjusting back for various combos.
Quote:
But imo it makes not really sense. Bc if we make for example a report for the river to see the influence of card removal.
It seemed to us at the time that it's easier to weight between flops/runouts using real frequency. It's true we could have implemented it to display unadjusted frequency as well (just take the average for runouts and weighted average across flops).
Quote:
So would it not just make sense if we can manuelly decide if we want adjusted or unadjusted freq in the aggregation report?
Yes, it's a nice feature idea
Quote:
I ran a preflop simulation after building a tree and input a 25flop subsets. I saved the full tree after getting the results.
Then later, to adjust the strategy to a certain player, i opened the file and went to 'set strategy and lock node'. After doing the changes, I just pressed 'GO' again to run the simulation. I then checked the preflop tab and no flops were selected, but they are in memory right ?
Yes. It's still better to:
-build the tree from scratch
-node-lock
-solve
than locking on already existing tree and re-solving. The reason is that the preflop solver remembers it's far away in calculations and may have trouble going back if you disturb the solution by node-locking.
Quote:
So in 1.9 i7 was 1,5x faster than i5 and now it is over 2x faster.
It's because Sandy Bridge generation is "old" CPU now. I will add optimized compiles for Sandy Bridge as well at some point, if you really need it now please email me. It won't help much in the preflop solver but will in postflop one.
Quote:
3. On Contabo server (2x Xeon E5-2630 v4 @ 2.2GHz) it is 4,8 bench in 1.10. Is it normal? I expected something closer to TR 1950x results.
This has 20 cores @ 2.2Ghz, 20 * 2.2 = 44.
Thread Ripper has 16 cores @ 3.4Ghz, 16 * 3.4 = 54.4 so Thread Ripper is faste.
Quote:
4. So if Threadripper is still about 2,0 bench is it still the best choose up to 128Gb?
I don't think people are getting 2.0s bench on not overclocked ThreadRippers
Your bench on the Contabo setup seems kinda low though. Maybe it's after the Spectre/Meltdown patch was applied? I still don't know how much those are going to influence performance in Pio case. So far one user reported a total performance disaster on an Online.net server after the patch was applied.
Quote:
Just thinking about new computer. If someone else or you punter has access to any PIO benchmarks results of other CPUs it would be nice to post it here or on the PIO website.
I tried collecting it at some point but sadly Pio engine changes too often. In general number of physical cores * frequency + 15% bonus for hyperthreading works pretty well assuming those are CPUs from similar generation.