I have a situation where i raise preflop and i want to fix the bb response. i lock from clipboard the 3 bet then i lock the fold from clipboard but the 3 bet changes. How can i lock both without changing each other while leaving call unlocked?

Yes, I would recommend Intel. I have no benchmarks whatsoever from AMD though, but I would expect performance to be worse with possibly the exception of ryzen.

I understand. This is on the todo-list but unfortunately not heavily prioritized. You can get charts (dials) of individual ranges by pressing the column headers in the table.

Yes, preflop solutions with constraints in the form of action reduction (specifying a game tree) and bucket abstraction (forcing similar hands to be played the same way.)

You select these constraints yourself to a large degree, so RAM requirements will vary heavily according to settings. It's not obvious how much RAM you need to produce accurate ranges in PLO, as there are no known full game solutions to compare with.

No, "Call committal threshold" means calls which commit x% of the players stack will not be allowed (when all in is available). He will instead have to go allin.

The same way with "Fold committal threshold". Folds will not be added if more than x% of the stack is invested.

This will add all actions selected in the checkboxes under the table graphic. So either you have no actions selected or they are all pruned out from the threshold settings.

This will be fixed in todays update.

I doubt anybody here is asking if they should use something before AM4 (so AM4/X399 vs X99/X299). I could be wrong though. Obviously X399 and X299 hasn't been released yet, but it will be soon.

So, are you saying you have some Intel benchmarks? I'd be interested in differences between some of the higher core Xeon processors and the X99 HEDT processors. To me it sounded like you were implying the 6950x (current X99 flag ship) would beat any Xeon configuration?

I need as much RAM to view/load a solution as to compute it right? There is no way around that?

Why is the iteration/second so slow on this calculation?



At this rate to get to 10x would take a month. This is on a AWS 16 core 128gb setup.

I have no benchmarks for those computers. Yes, but that is pure speculation on my part.

Not right now, except if you have saved with compression, then you can load a given number of streets (change the "Max streets" in the load dialog.)

Things to try:
1) Uncheck synchronize in the settings panel. You can do this while running.
2) Set avg streets to 0. You have to restart to do this.
3) The run seem to barely fit in memory. Make sure you have set the -Xmx value to your amount of free memory (122g minus whatever the OS uses, not 122g) or you may experience thrashing.

I have no idea if this is feasible, but I'm going to throw it out anyway...

It would be nice if when building a tree, you could restrict certain bet sizes to only be available on certain boards types.

For example, lets say I have 3 cbet sizes in single raised pots depending on the board. It seems like it would significantly cut down on tree size/ram requirements if I could say, "only use the 1/4 pot sizing on board type X (e.g. paired), the half pot sizing on board type Y, and pot sizing on board type Z. Seems like this could cut down a lot on RAM requirements/processing time as opposed to having all 3 sizes available for all board types unnecessarily.

How do I set maximum RAM in the MAC version? In the folder all I see is this:

This is a 20GB solution. It's been loading for about 10 minutes. How long are these supposed to take to load?

Update: Looks like it look 20 minutes to load approximately.

Any other ideas, made all three of these changes and now getting only 14k/s iterations

also does the iterations/s increase as the calculation goes on?

While this would be nice, it probably will not happen any time soon due to the complexity in implementing this.

Edit the "start2.sh" file, and also run it from that file. ("sh start2.sh" from a terminal).

With an SSD it should take less than a minute. 20 minutes sounds bad even for a HDD. If you have antivirus running, it's possible it's contending for the disk, hampering performance.

Yes, speed generally increases.

You could try to lower the settings to make the run fit comfortably in memory. If that increases the speed, then there's some sort of thrashing issue.

The speed sounds very low, but it's hard to say for sure without seeing the tree. Could you email the tree file to support@monkerware.com?

This time was from a Google Cloud Compute VM. I think it runs Windows Defender by default. I don't think it is using an SSD. I don't think I configured it properly to be used with an SSD. Do you know if the lack of an SSD might otherwise affect performance? Just load times and write times right?

Update: I stopped the time properly today, and it took 29 minutes to load (actually the software tells you how long it took to load, because it's added to the run time). Would you recommended disabling Windows Defender?

I get the same speeds when trying to run huge preflop trees which makes them not really feasible. While it is annoying, if you think about it, it is both a blessing and a curse, really.

My suggestion is to be creative and find other ways to test all the bet sizes you want to run using multiple, smaller sims.

How about running a postflop tree for a subset of flops? Say I can specify 10 flops that I want the same game tree for. To my understanding, as it is now I can only run flops for a single specific flop right?

I noticed that parts of the solution can be exported to PPT. I tried inputting these ranges into PPTOO, but the software can't handle such large ranges. I tried increasing -Xms and -Xmx directly in the JRE configuration software, but it had no effect.
I'm guessing that people are using these ranges in some other way to feed into some other software other than PPTOO or is there a way I can get PPTOO to run simulations with such large ranges?
Or is there perhaps some other way to see how much equity a specific hand has at any given point in the solution vs villain's range?

Those instances seem to use a network based filesystem by default. That will be quite slow compared to a normal local disk. Windows defender should not affect anything.
EDIT:
Right, that will only affect start/load/save times.

Running 10 flops at the same time would probably not offer any speed improvement compared to running it sequentially, while increasing RAM requirements. Running 10 flops sequentially (scripting) is on the todo-list.

The equity graph takes the filter into account. So, you could filter on a specific combo and see the equity there.

Would it be possible to manually set the location of the saved runs folder? It would make it a little easier to share the results between machines by setting it to a synchronized folder.

When saving runs, and selecting compression, there is a warning that says that this is not entirely lossless. I'm not sure how to interpret that. What sort of errors would this lead to potentially?

Do the pre solved solutions available for purchase in the cloud store take rake into account?

Do you have an nGTO solution for Omaha preflop ranges?

"MonkerViewer is a simple program used to navigate preflop ranges. Hold'em and Omaha nGTO ranges can be purchased from the store."



In the viewer itself there are no Omaha ranges for purchase.

1) I noticed that when I load a tree with saved ranges, the percentages never quite match up with what they were when I initially input them. Today for example the ranges showed as follows:



However if I reinput the ranges, they are now as follows:



2) Do you think that a hardware configuration with dual-socket CPUs would underperform vs a single CPU with the same spec?

You can change that in: settings->default saved run folder

No major problems, but frequencies of mixed hands may change slightly on streets where you don't have average. Unaveraged frequencies are not exact in any case, but tend to oscillate around the correct frequency, which is why averaging is useful.

No rake.

There used to be, but that information is no longer correct.

When starting the run, the ranges will change due to combos blocked by the board cards being removed.

If they change without starting the run, there's some problem which I would be interested in how to reproduce.

2)
Yes, but I'm not sure how much. If you go back a few pages someone with a dual-socket CPU seemed not very pleased with the performance. I'm not sure how much of this could be attributed to the dual socket though.

The screenshot shown was from before starting any runs and before inputting any board cards.
I am relatively sure that the combos from the saved tree are never the same as when I input them initially. I initially dismissed this, because I thought it might somehow be related to blockers. But in this case, as shown in the screenshots, the change is so dramatic, that that hypothesis didn't seem plausible any more.

I think you are referring to the 64-cores vs 10-cores post. I thought about that too, but I'm not sure if OP was using vCPUs or a local machine. I don't know enough about CPU architecture, but intuitively, I wouldn't assume that those would perform equivalently necessarily.

I'm thinking that RAM architecture might also have a relatively large impact on Monker performance right? As in DDR4 over DDR3.

Is it not possible to specify ranges in trees that start before the flop?

How do I add a fold to a tree? I know that folds are added by default, but I removed a fold, and later decided I wanted that part of the tree back, but don't see how to do that.

Right click on "POT" in your example, and edit.

As far as I can tell, no. But you can lock them afterwards.