ok, maybe a different question then.

100BB stacks from CU,BU,SB,BB.

CU opens for 2,5BB. According to your solver what % should the BU call and what % should the BU 3bet to 8BB. Not asking specifc hands. Just overall gameplan.

Our solver (as well every other GTO solver available at the moment) only solves HU spots. For preflop it means:

-HU
-SB vs BB 6max
-steal vs BB with defined stealing range
-steal vs 3bet with defined stealing and 3betting range

While it's theoretically possible for GTO in multiplayer games it's very unlikely to be done with any precision in coming years. Equilibria in multiplayer games are also very shaky concept (unlike as in HU) and you may well lose if you follow them (there is a concept of collusion which might even be not intended).

Programs like Snowie have advantages here. They don't claim to solve for GTO, they just claim to get to reasonable strategy against many other reasonable strategies. The problem here is how good those are and what kind of strategies were used to optimize against. Basically it's a matter of trust. You either trust them or you don't. The only way to form an opinion here is to either follow those strategies in practice or ask good players who did.

I believe this post was overlooked. I have the same issue with respect to getting up to speed.

I think the answer to the last question is: Yes, you can stop/browse/start.

Yes, I must've missed it, responding to all the points now:

If you make a full save it will take as much on disk as it takes in memory. It's usually a better idea to make a small save (without rivers), they are 50x-150x smaller usually.
Still, it's hard to imagine how 5BB-deep simple tree could be 12GB unless we are talking about a preflop tree.

I am still assuming we are talking about preflop trees. It's usually a good idea to:
-not allow OOP's bet when they have a weaker range (so usually after calling preflop)
-set cap to something low, 3 is good for most situations to simplify the tree
-eliminating raises in position on flop/turn can someone be done without losing anything but you need some experience to know when it's safe (for example in 3bet pot in position it's safe to eliminate IP's raise on the flop)

There are some videos here:
http://piosolver.myshopify.com/pages/faq-videos

Specifically:
1)Quick start:
https://www.youtube.com/watch?v=JqGQoQKbCB8

2)How to use preflop solver:
https://www.youtube.com/watch?v=LX8n2E3OGok

3)FAQ (with links to specific questions in the description)
https://www.youtube.com/watch?v=VhG_YsEvfwU

4)New features in 1.8 (again, with links in the description):
https://www.youtube.com/watch?v=gJEPGSIpIBM

Yes you can do it. It requires some patience as for preflop trees stopping and calculating results takes some time. You can stop/resume as long as you have the whole tree in memory. In 1.8 version you can even make a small save after stopping and still resume solving (as the tree doesn't disappear after making a save).
The solver produces updates every 25 full iterations (in the tree building and calculation tab where you click "go") but it the intervals might be quite long (the reason it doesn't produce them more frequently is that it's a costly operation, especially for preflop trees).

This might be more of a GTO question than a question about the solver itself, I'm not sure. I've tried to simulate some situations where one player is using two river bet sizes, one that is very large and one that is smaller. Its surprising to me that in a lot of these spots, the bettors nut hands often have very similar EV betting say 2x pot as they do betting 1/2 pot. This seems to go against conventional wisdom that we generally want to bet larger with our strongest hands. Any idea why this is?

This conventional wisdom comes mainly from toy games where the aggressor has value bets/bluffs and the caller has only bluff catchers. In such games it's in fact true that you would always like to use the biggest possible size (well, in some rare cases any size is good enough). Once the ranges are more complicated and both players can have strong hands, especially if they can raise as well things become more complicated.

It's always useful to paste the config from the tree you are commenting on so other people can re-run it and see. You can do it by clicking "copy to clipboard" button and paste it, preferably use a service like pastebin or code tags here to avoid pasting a wall of text.

Hi,

As a husng player, i'm interested in the Edge product. Unfortunately, my current computer is very old, and i was about to buy a new one. I don't know much about hardware, could you please advice me ? What composants are important in order to use Piosolver effectively ?

In order to solve spots from preflop, i guess RAM is the main thing to focus on. How much RAM would you advice ? Is CPU important ? GPU ? Anything else ?

Thanks for your help

It's always a question of the budget. Assuming you are not buying a big work station (dual box Xeon setups) the best choice as of right now is probably 5820k i7, you can go for good cooling if you want to overclock (additional ~20%). If you want to spend more money then it's probably the best to wait for 10 core i7s to be released in Q2 2016.

I would advise against buying dual box Xeons as a setup like that is 10k+ at least and it's cheaper to rent one for a while.

Yeah, 64GB of RAM is minimum you should shoot for for a preflop solver (32GB is quite limited). You will still run into some limitations with 64GB if you want a lot of preflop exits. I think it is already very nice though and good for most practical purposes.

GPU doesn't matter. SSD doesn't matter for the solver although it's nice to get one for overall system performance and especially if you plan on using big database.

Another idea is to rent cloud instance/dedicated server for a while. You can get very good setups for 200E-300E/month (something which would cost you ~15k to assemble) or to to check the offer of PioCloud:

http://piocloud.******.com/

When it comes to postflop solving you probably don't need more than 16GB of RAM after recent optimizations (even full postflop limit trees which are humongous fit under that now).

I had 2k5 in mind. Is it too short ?

It should be more than enough for the hardware I mentioned but prices are very dependent on the area you live in so do your research.

Thanks a lot.

How can it be that when I give OOP the option to lead flop (and change nothing else) I see a decrease in his EV of almost 1bb/100? I don't see how having an extra option should ever be detrimental, since if checking is higher EV then the solver should figure that out on its own, no? Both sims were solved to an accuracy of 0.1% pot and the difference was .084 chips.

Text from "copy to clipboard"
.rar file containing very small flop save
http://www.filedropper.com/jc7c8dsmall

You can run it and then uncheck the "Include OOP bet" and run it again to see what I mean.

Unfortunately this is not that simple. This being complicated is the reason I was very reluctant to add rake and I don't really want to add ICM. Anyway, here are some facts:

1)In HU zero-sum game it's guaranteed that every equilibrium has the same EV. Playing an equilibrium strategy guarantees you won't lose. Even if there is more than one equilibrium every one of them is as good (guarantees not losing).

2)In HU non-zero sum game it's no longer true - there could be many equilibria with different payoffs. This means that an equilibrium in non-zero sum game is not that valuable strategy. It's still true that no player can improve in an equilibrium but still there might be more states like that, some terrible for IP player, some terrible for IP player, some somewhere in the middle

3)Introducing rake (or ICM) makes the game non-zero sum.

4)The hope is that it doesn't matter that much and all the equilibria are kinda close to each other, unfortunately especially with big rake (or very steep ICM adjustments) the hope is just that: hope.

5)Your care has a very steep rake (probably trying to simulate 0.5/1$ game)

6)It's likely that the solver finds a different equilibrium if the OOP bet is disabled. There is very little we can do about it, methods for finding all equilibria are unknown.

Notice that if you disable rake, everything is back to normal:

EV with OOP lead on the flop enabled:

running time: 131.860
EV OOP: 30.824
EV IP: 29.176
OOP's MES: 30.887
IP's MES: 29.233
Exploitable for: 0.060
SOLVER: stopped (requested)

EV with OOP bet on the flop disabled:

SOLVER: stopped (requested)
Results:
EV OOP: 30.635
EV IP: 29.365
OOP's MES: 30.689
IP's MES: 29.417
Exploitable for: 0.053

While the concept of having multiple equilibria (and thus lack of something you could call "GTO") in non-zero sum games is a difficult one I would like to give one toy example:

-pot is 10 chips
-stacks are 100 chips
-both players have AA only, it's preflop and we play fold or shove
-it's tournament, 200 chips are worth 180 in utility (typical ICM considerations, similar situation to big rake), 100 chips are worth 100 in utility
-for simplicty assume 110 chips are worth 110 in utility
-OOP is first to act

Imagine two set of strategies the players have:

a)OOP always shoves and IP always folds

This is an equilibrium because calling a shove to flip is -EV in ICM settings so IP can't improve. OOP ends up with a stack worth 110 utility every time, IP ends up with 100 utility.

b)OOP always folds but IP would spit call every shove

This is an equilibrium as OOP can't improve (shoving would trigger a spite call and both players would lose). OOP ends up with 100 utility and IP ends up with 110 utility

Both sets of strategies are equilibrium but EVs vary a lot.

A game with rake is full of those situations on the river. You may make just profitable bluff which can't be called because of rake, on the other hand if your opponent is already calling (their strategy is to call too much) you can't even make marginal value bets because you would lose w/e you gain to rake. This may happen in many places in the tree resulting in large number of possible equilibria.
It is quite likely that removing an option just triggers the solver to find a different one as the OOP player is quicker to "take the territory" faster (that is start making plays, the opponent can't defend against because of rake).

This is what math is, there is very little we can do about it unfortunately as we can't find every single possible equilibrium. I hope that makes sense and explains my skepticism about introducing rake/ICM to the solver.

Notice that it's quite likely that all the sensible equilibria are close to each other. In your case the difference is only 0.84bb/100. It would probably be way less in higher stakes games where rake is a smaller factor.

Makes sense. Thank you for the detailed explanation.

Question regarding computer power to run PIO Solver:
My laptop has 2.1 GHZ and 8 GB RAM, will it be enough to run the pro version? How long to calculate a flop solution and another turn/river solution?
Cheers

It depends what exact CPU is it. You can find the CPU name if you go to control panel -> system -> system. PM or post here and I will give you more information.
8GB of RAM is plenty after recent optimizations, you will be even able to calculate a lot of trees with 2 bet sizes (or even 3 with smaller ranges). The CPU will be a bottleneck with those bigger ones but if it's an i5/i7 quad it should be decent enough.

PioSOLVER calculates the whole tree at once so there is no delay when browsing from flop to river.
You can go back and forth or even skip branches vertically (changing turn/river cards but staying in the same line) without recalculating anything.

If I pm you my specs tomorrow can you tell me what upgrades I might need for pio to run very smooth, thanks in advance

i ve been toying with pf solver a bit and seems like 64GB ram is enough for any 100bb tree without limping using 49flopset, is this correct?

.

Yeah, without a limp you should probably be able to get more flops if you cut a bit postflop.
Trees with limps are a problem because limp-limp branch is usually huge. If you simplify it heavily you can probably get trees with 5-6 exits but again it depends how much you are comfortable cutting from postflop play.

I'm buying RAM right now and deciding between 128gb (for $720) and 64gb (for $350). I play MTT's and will be looking at a lot of SB vs BB spots with antes, where I think that there will probably be a ton of limping in a GTO solution (as opposed to HU where the SB is in position). So given what you said above about the limp-limp branch, I'm guessing I should go with 128GB (even though 128gb kit are annoyingly more than double the price of 64GB kits)?

hey,

this is my notebook cpu is it enough for your basic version and 3 different betsizes (notebook has 16gb ram,so that shouldn't be a prob)?

i5 4330m 2.8ghz cpu (it's only a dual but with hyperthreading so not sure)

In general this CPU is too slow for the solver. You may be able to calculate some simple trees (probably 2.5x-3x slower than benchmarks on the website done done on a desktop quad i7) but it won't be fun once you start adding bet sizes.

Trees with 3 bet sizes everywhere are rather big and they take a lot of time to solve even on fast CPUs. To be honest I didn't think 3 bet sizes would be even possible when we released the solver a year ago. It is now due to many memory optimizations but the speed is still a problem for them.

Hi,

I studied these 2 spots on many (many many) boards :
-BTN calls 17% vs SB polar 15% 3B
-BTN calls 13% vs SB linear 15% 3B

I don't think I need to detail precisely the ranges in each scenario as anyone can imagine it more or less, and I don't think this changes much the answers to my questions.

So now I try to understand the results.

My 1st surprise was to realize that the average equity realization for the group of hands BTN defends in each scenario is higher vs SB linear 15% 3B than vs SB polar 15% 3B.
Here, we face a stronger range (in terms of equity : SB linear 15% 3B > SB polar 15% 3B), but our equity realization is higher (1.05 vs 0.97).
I don't think the fact that our defending range is tighter impacts our equity realization, does it?
So can this result be explained by the fact that because the SB polar 15% 3B offers more barrelling spots for SB Villain?

My 2nd and bigger surprise was to observe that ATs equity realization was higher vs SB linear 15% 3B (1.1) than vs SB polar 15% 3B (0.95). Same for KQs.

I don't get it : both ATs and KQs are dominating many weaker draws in SB polar 15% 3B, so I expected the equity realization of these hands to be higher in this scenario than vs SB linear 15% 3B.

Any idea to explain these observations?

Thanks

Hi punter11235,

Idea of new feature for gaining time in PF solving : I would like to be able to select the basic strategy (first OOP action and eventually first IP action) for the range of cards and set it as a starting point for solving.

Currently, the starting point for solving (assuming that four actions can be made for OOP) is 25% fold, 25% limp, 25% raise, 25% shove for any hand in the range. Instead, setting 100% fold for 82o or setting 100% shove for pocket 2's (yeah I am computing shallow stacks in HU play) would start the solver in a state that is closer to the solution than starting 25/25/25/25.

I am playing a lot with the preflop limp/check tree, which plays a very important part in solving the SBvs.BB game when others folded, or the HU game. Saving time while solving would improve drastically the efficiency of PioSolver.

Thank you in advance for your help.