Today we are releasing PioSOLVER 1.4. This is particularly exciting release because it brings huge performance and memory efficiency improvements as well as new features to make understanding and analyzing GTO strategies easier. We are also adding a convenient updater.

Before moving on to usual feature list let's talk about performance for a while. The announcement we made 4 weeks ago is no longer relevant - the performance got way better since experimental version available at that time. We improved the algorithm itself, made it faster and added new optimization for suited/paired flops which brings on average 1.67x speed-up if you were to calculate all flops or 1.42x for a random flop. Mono suited flops are now 3x faster, suited or paired ones 1.5x faster and rainbow not paired one don't get the boost (but they are only 286 out of all 1755 or 6864 out of 22100 if you don't count isomorphisms). Overall expected speed-ups over previous version for typical cases look like this:

1. For heads-up ranges you can expect on average 2x performance improvements and 1.2x less memory usage
2. For 6max cases with wide ranges (about 50% vs 50%) it's on average about 3x faster and 2.2x smaller
3. For small ranges (like 25% vs 25%) it's about 4.5x faster and 3x smaller
4. For ranges smaller than that it's so fast it's hard to measure :)

This means many practical cases with one bet size are now solvable in 1-2 minutes to very good accuracy on run-of-the-mill CPUs. We will be posting exact benchmarks soon but for now we can say it's completely different experience and way more fun! This is especially true for computers with only 8GB of RAM - the limitations for those systems are now way less severe, especially for non-HU cases. Before the benchmarks arrive here are some typical cases: (all run on 3 years old quad, i7 3770 and to <0.25% of the pot accuracy)

1. Typical 3bet pot with heads-up wide ranges: 92 seconds to 0.23%
2. 6max BTN vs BB case with deep stacks: 166 seconds to 0.238%
3. 6max MP vs BB on a rainbow flop: 134 seconds to 0.24%

It's worth mentioning that 0.25% of the pot accuracy is a bit of an overkill for practical purposes and that all mentioned trees fit below 1.5GB of RAM. The improved algorithm is also way better at reaching near to 0 accuracy (0.1% or less) so the perfectionists among our users will have way more fun as well.

With performance like that the feedback loop for analysis is even faster and it makes analysis of tens of hands after the session very feasible.

Let's now look at other improvements we are rolling out:

1. Improved behavior for difficult trees. It used to be the case that the solver had problems with some trees and spent several iterations running in circles so to speak before finally finding the path to convergence. The situation improved a lot and it's not longer easy to find really difficult cases. This is especially important for trees with may bet sizes and overbet all-ins
2. There is now a convenient updater. All you need to do is to paste your download link there once and click a button to check for/download subsequent updates. No more manual copying and unpacking of the files. See GIF.
3. We introduced some new option to make tree-building even more customizable. Those are "force c-bet" option, "don't 3bet" option and "cap" option.
4. It's now possible to round strategies after solving. This feature was heavily requested. You can choose a "chunk" the solver uses (rounding to 0% 50% 100%, 0% 25% 50% 75% 100%, 0% 20% 40% 60% 80% 100% etc.), a player and a street (flop only, flop + turn, flop+turn+river). Those strategies are more exploitable than unlimited solution but it may be useful in practice. SEE HOW IT WORKS.
5. Vertical browsing: using ctrl+arrows (up/down to change rank, left/right to change suits) you can now easily change turn/river cards and see how strategies change. This is very useful feature to quickly compare how for example river calling ranges change depending on what card fall. We are very excited about this feature and there will be more comprehensive available later. For now QUICK DEMONSTRATION of me looking at calling range for in-position player against a 3 barrel. Notice how the range change depending on a river card. While this ends up at 6th position on this list this a feature we are very excited about as it makes visualizing GTO patterns accross various runouts not only possible but very easy!
6. It's now possible to look at the results for combos containing specific suits. Before it was sometimes hard to grasp what the strategy is for flush draws for example or a backdoor draws. NOT ANYMORE!
7. It's now possible to execute scripts from script creator window. It's not longer necessary to save them first.
8. Limit Holdem: not only the algorithm got way faster for limit but we are also adding Cepheus preflop ranges to our range collection. You can now easily recreate Cepheus solutions (PioSOLVER has no problems getting 1mb/hand accuracy) and see all the information not shown on Alberta University's website like expected values of plays, overall frequencies, range analysis and more. It's interesting to notice that even though we use completely different algorithm the solutions match almost exactly!

We are very proud of this release. PioSOLVER is finally at the stage we hoped it would be when we first released it. It's incredibly fast and functional. It solves huge trees in minutes (and sometimes seconds). It's not only the best training tool ever created for Holdem but also very fun (and addicting) toy. The difference comparing to previous version is staggering. If you are our customer you are in for a lot of fun, if you are not make sure to check free version update which is likely to be available in about a week!

The schedule for updates is usual. Edge version users already have it. Pro version users will get it tomorrow and Basic version users should get it this Wednesday. Make sure to check your email!