RAM is simply required in order to perform certain calculations, it's not optional.

With Monkersolver you can specify simulations that run into the 100s of GBs of RAM, so whether 128 is enough depends on your intended use.

Keep in mind when selecting hardware that not all RAM is equal, and not all CPU cores are equal. Make sure to get at least DDR4 RAM. 24 cores on a platform that supports DDR4, and with a RAM limit of 128 GB should be fine.

Hi I'm considering buying the full version (Basically only interested in the preflop functionality), please kindly answer the following questions

1. My understanding is this solver uses heavy abstraction to build the trees (these multiway spots would be impossibly large otherwise) so just how close--if it's possible to quantify--are the Monker solutions to a real GTO solution (meaning one in which the ENTIRE tree is built, no abstraction)?

I am skeptical that something like a fullring preflop tree with any distribution of stacks can be solved to really any degree of closeness compared to an unabstracted tree given the RAM a lay person would have at their disposal.

I can only find servers that go up to 256gigs of RAM--just how much precision would we be giving up via the abstractions by solving say a fullring NLHE preflop tree with all stacks equal to 30bb? A full tree with no abstraction including all-in 3bets and not-all-in 3bets and flatcalls I'd imagine would be many hundreds of terabytes without the abstraction. To go from that to 256gigs is a massive reduction in complexity.

2. Can I specify an arbitrary range (like an open range) for arbitrary sizing for arbitrary position and solve the tree from there? In other words can I attempt to build literally any tree I want?

3. Can I customize all stacks and starting ranges in a preflop sim, or does it have to be equal stacks?

4. Can I customize the amount of abstraction the solver uses?

5. Any practical differences between ESCFR and CSCFR algorithms and which if any does the Monker team recommend to be used

6. If I want to be able to use up to 256gig of ram do I simply add the following to the MonkerSolverFree.l4j file?

Thanks!

At the login for MonkerViewer it says 'Server Offline' under where I'm supposed to input the email address and password. Any ideas on how to fix this?

I am looking at some monkerviewer files, and a set of solutions includes the EV for each decision. For example, folding KQo in the SB is obviously -.5 but it also shows calling is better at -.15. But I have another set of solutions for different stack sizes and it does not include the EV. Why would some solutions have EVs for each decision and others just have fold/call/raise?

thanks.

There is nothing to fix. You should still be able to login normally. Access to the store through Monkerviewer is no longer available.

@1 The GTO solution to full ring NLHE is not known, and therefore the distance in EV or the deviation in strategy to it is also not known.
The developer has made a comparison of the Monkersolver algorithm to the LHE CEPHEUS bot (GTO HU LHE), which you can find here.

@2 Yes.

@3 Asymmetrical stacks are possible.

@4 Yes.

@6 '-Xms3g', '-Xmx256g', you only need one definitions of what the maximum amount of assignable RAM is. Also '-Xmx256g' assumes you have more than 256 GB RAM in your system. For a system with 256 GB RAM installed, realistically you will be able to set Xmx to something like 240. Note the solver can use more than 240 RAM, even if you assign '-Xmx240g', so it's important that your system doesn't run out of memory to prevent out of memory errors, and crashes.

Yes the equilibrium not being known is a very good point.

So let me ask--is it really valid to compare LHE results to NLHE?

You made reference yourself to the CFR algorithm that was employed in the algorithm of Monkersolver. The objective of CFR is to iteratively remove dominated actions, and in doing so produce strong strategies.
The fact that Monkersolver's algorithm was tested against a GTO LHE bot shouldn't take away from the validity of the method when it comes to solving big bet games.

i am in the market for a solver. have a couple of questions.

1. i have lenovo legion y740 , intel intel i7-9740h 6 core-12 t, 16 gb ram. can i run it in this machine for solving individual hands.

2. Also i will be upgrading to an amd 16c-32t ryzen desktop in the future , will i be able to transfer license from current laptop to desktop ?

Thanks.

I'm interested in purchasing a solver but am not familiar with how to use it or anything to do with it.
1. Are there any step by step tutorials on how to get started?
2. How do I input the ranges for the players?
3. I downloaded the basic 6m trees but cannot seem to find how to load them in the free version.

I downloaded the free versions.

Hello,

When I tried to load saved runs I got the following error:

'ERROR: java.lang.NullPointerException: entry'

I already reinstalled the free monker software (and changed it to run as administrator) and installed the lastest java 64bit version but nothing changed.

The memory of my computer is 32GB RAM.

Can you please advise what the problem is?

Thank you!

Do I have to use the monkersolver or the monkerviewer?

Ok, my files have a .mkr extension so is monkersolver.

By the way, I moved the files to /MonkerSolver/savedRuns/ and when I click Solve and the Open Folder icon I see the solved trees but when I try to load them, even only one street, I get the error I mentioned above.

Is Monkerviewer currently working for anybody?
It does not connect even with hours of loading time, 24GB assigned, few ranges to load for around a month now.

@1 Your laptop is suitable for viewing preflop and flop solutions, as well as solving for the turn and river. Some small preflop and flop trees may run as well.

@2 Yes, you can transfer the licence.

@1 The monkerware website has a basic guide on how to get started (https://monkerware.com/solver.html).

@2 You can use ProPokerTools syntax (http://www.propokertools.com/oracle_help/range_of_hands).

@3 The free version only supports the turn and river.

The error you are receiving is that of a corrupted file.

Monkerviewer works, but can struggle if you are trying to load a large library. You could try loading with less files in your 'ranges' directory.

1. What's the difference between the "-Xms" and "-Xmx" definitions?

2. Does the solver have a function that lets you estimate how big a tree will be before building it?

3. Do we get to select a subset of flops to feed into a preflop tree? If not, how does the solver solve preflop without considering flop subsets?

4. I've looked for an explanation of what the "Avg strategy" and "Synchronize" functions do...can someone explain for me?

5. What do I set the "Threads" value to in order to use all threads available?

6. Is it correct that the fewer "Strength buckets" you use the less abstraction there is?

7. Is "Texture" a further refinement of how much abstraction is used? What's the difference between "None" and "Perfect"?

8. Lastly, any general comments on things we can do (like different settings) to maximize performance?

@IsaacAsimov IDK if you're affiliated with Monker or not but in any case you've been quite helpful in this thread and you are owed a big thanks

what does the "To act" value do in the custom blinds setting? The default value is set to -1.

Trying to build a preflop tree with 50/100 blinds but the solver keeps giving me 5/10 blinds. That has the effect of making stacks 10x as large as I want them to be.

Why isn't my custom blind specification being saved?







When I try using 5/10 blinds, the solver instead gives me blinds of .5/1

I've built a large preflop tree (202 gigs). Is it normal for the "initializing payoff data" step to take quite a while for really large trees?

Not sure if this has been asked before but what are the drawbacks of re-saving an already saved run to save disk space?
For example a I have a 100bb preflop range, which is around 17GB but if i open it and chose to save it for storage it reduces the size of to about 250MB.
I realise I only open one street and this is what I am saving but does this really affect PFR ranges?
I had a quick look and cant see any difference.

Same applies for already ran post flop sims. Opening (as one street only) and re-saving them saves a ton of space and since I'll be running new sims for future streets I only need that street ranges.

Hope this makes sense.

Any input would be much appreciated.

I am not affiliated with Monkerware.

1. -Xms specifies the starting memory pool for the Java virtual machine, -Xmx specifies the maximum amount of memory that can be assigned to an application by the Java virtual machine.

2. Yes.

3. The solver buckets strategically similar boards according to your texture settings.

4. In order to calculate mixed strategies, you need to collect an average strategy. Synchronize refers to the synchronization of CPU threads. (You can find definitions of a lot of features by hovering over them with your mouse.)

5. This depends on your CPU, but typically it's either the amount of CPU cores, or, if you have hyperthreading, CPU cores x2 that will peg your CPU.

6. Fewer strength buckets means more abstraction.

7. Yes, texture is an abstraction tool. 'None' texture does not take any board card into consideration, whereas perfect takes every board card into consideration.

It changes which player acts first.

For 50/100 blinds input '500,1000'.

Depends on how long you are talking about, but it can take some time to initiate larger game trees.

Resaving to 'save for storage' should not affect any of the data you are saving.

Thank you @IssacAsimov.

It is almost unbelievable the amount of space this simple trick saves, and I may not even need to buy a new drive now.

Them sims were eating up disk in a very rapid pace on the first save.

Do you know if the tree size estimator function is particularly accurate?

I have 256gig of RAM and, set to use up to 240gigs, and I routinely get out of memory errors on trees estimated to be well below that.

What factors affect the k/s solving time?

Seems like some trees I can get upwards of 1,500k/s others the solver struggles to break 200k/s. All cases I have the entire CPU dedicated to solving the tree with max threads set.

Does it vary based on tree structure?

How do I create preflop trees using the tree building filter with the following characteristics?

-I want only one flatter of a preflop raise EXCEPT button and/or BB; these can be allowed to coldcall even if another player has already coldcalled

-I want no coldcalling of 3bets or any higher order bets

Is this something I just have to do manually?

Tried messing around with the Cold Calls setting in the filter with no success.


What do the ">2 players" and "2 players" buttons do? These do not have definitions when you hover over them.