Is there a way to make scripts to calculate from turns (for every turn card) for a given set of precalculated flops (so that you already have turn starting ranges for after each flop)?

Speed is more or less proportional to base frequency x number of cores. There is around 15% bonus for hyperthreading (i7 over i5) and a slight bonus for newer generations of CPUs.

That means that your CPU has a score of 4 x 2.9 = 11.6
Something like i7 5820k would be 6 x 3.3 = 19.8 and additional 15% so 22.7 so that's around 2x faster than your current one.
You can get something faster than that if you go for big machines but those are quite expensive to rent. As one example we are renting 16 core Xeon @ 2.6Ghz which is around 2x faster than 5820k so about 4x faster than your machine.

The most reliable way to get the solution faster though is to make the tree smaller. Reducing number of possible bet sizes, all-ins etc. are good ways to start.

There are too many providers out there with their own ordering process and options. We can't really keep up to date with that and they have their own support to help.
That being said it's a good idea to ask about it on our Skype group as a lot of people went through the process of setting up Pio on either cloud instance or a dedicated server.

In general dedicated servers are more convenient (you don't need to deactivate the license before closing down the VM) and cheaper if you run it a lot. The downside is that you pay per month so if you only need it for few hours here and there cloud instances are a better option.

There is no way to generate it automatically right now. You would need to somehow generate it yourself. This means either quite a lot of typing/copy pasting or making a tool that's going to generate it for you.

was trying to load a friends sim and got this error


server error: ERROR: load_tree couldn't open...
the file doesn't exist or isnt a valid save

Probably one of the following happened:

1)You are using an older version of the solver than your friend (the saves are backward compatible but not forward compatible) - in this case you need to update to 1.9.2 version and try again

2)The save is a preflop save and you are trying to load it with the postflop solver - to solve that you need to connect to the preflop browser (or solver in the edge version) by going to solver->connect in the top menu; when you do it once successfully it should automatically switch between them in the future

A curious Q;

Are people seeing drastically improved solve times when renting cloud servers (ie Amazon E2C)?

If I for instance get PIO Edge and rent a 40 cpu core server, will PIO be able to run on all 40 cores and have the performance scale accordingly?

It depends on the instance. You can for sure get one much faster than home computers. That is especially true for the dedicated servers.

Pio scales almost linearly with number of cores. Just make sure you are getting fast Xeon cores and not some old genration AMD ones (everything AMD before current Ryzen generation was really slow for the solver purposes).

Thank you. Just wanted to make sure there is no theoretical cap where the core count stops scaling with performance. E2C seems to top out at 64 cores, so you're saying that the same old rule of core count x frequency still holds true with 64 cores?

My understanding is that by "64 virtual cores" they mean really 32 cores with hyperthreading. The solver doesn't scale perfectly linearly but it's pretty close.
That being said I never tested those cloud instances myself so you would need to ask around. If you can get a dedicated server with 20, 24 or 32 cores then yes - it will scale close to linear with number of cores.

Thanks! Much appreciated. One last question;

Is the advent of the (relatively) cheap 14-16 core AMD Thread Ripper CPU's likely to have any effect on the 12 core cap for the Pio Pro version?

Older AMD CPUs are slow, don't buy then nor rent them, you will regret it.
It's a different story with Ryzen but everything older than that is just painfully slow. You can expect that those CPUs clocked at the same frequency and having the same number of cores are going to be around 3x slower than Intel ones.

I should have been more clear. The Thread Rippers are upcoming continuations of Ryzen. Bigger Ryzens with 14-16 cores @ up to 3.6 GHz, due this summer.

I was wondering if the introduction of these more mainstream >12 core CPU's could potentially lead to an increase in the 12 core limit for Pio Pro? Let's say a Thread Ripper ends up being $1k. That I could justify. But if another $1k would be needed for a Pio Edge upgrade to use all cores it's getting a bit more dicy

Hi, how can I tell if my PIOsolver pro is using my 8 cores? Is there a way to tell? A tree that is for example 27gb is taking approx 1.5 hours. My processor is Ryzen 3.0 GHz and 8 cores. I have 32gb RAM for now. Does this time it's taking seem right to you? I just wanted to make sure all my cores are being used, presume they are but needed to double check.

Also if I ever upgrade my ram to 64gb for example, this will let me build bigger trees but will it decrease the time it takes to run an exact same size tree?

Thanks for your time.

That's with 0.35% of the pot accuracy fwiw.

I am sorry I am not up to date with names. I've just seen a lot of benchmarks for older AMDs and they were bad. Ryzen's benchmarks are very good so yeah that will likely be quite a good option. Some people reported systems with Ryzens being unstable though (it looked like it's mainly memory issue though). For some others it works perfectly even if they run Pio 24h/day.

Yes, the pro version will support 16 cores in the next release.

Open task manager and see how much CPU usage solver process reports. In general everything above 50% means it's using all the cores (assuming those are cores with hyperthreading). Usually it should be in 75%-90% range.

That sounds like a great time tbh. 27GB is a really humongous tree. Full limit holdem tree (with cap 4 at every street) is around 15GB these days and those were trees I didn't think would ever be solvable on a home computer.

Not right now. Maybe in the (distant) future.

Why is it reraise shoving here on the flop? I thought I have set it up so when I get check raised I reraise 3x, but it is shoving?

https://gyazo.com/66a07ab1124a341c5d2fdf35b71333bf

https://gyazo.com/a17bdc59352697e359e3ca08f5e1f64a

Thanks

There's a more general "add all in" type feature at the bottom of the input interface allowing for all ins once the pot reaches a certain size. Perhaps related to that? (just guessing here as I don't have PIO on this computer, just from memory.)

I have an i7. Getting the pro version vs basic version would run 15% faster? Is that correct?

I need the config to be able to tell. The screenshot is not enough as not every setting is visible on it and it's difficult to read anyway.
To share the config do the following:

1)fill the config (or load the tree)
2)click "copy to clipboard" button
3)paste the text to pastebin.com
4)link here

alternatively you can just copy it here in code tags so it doesn't take the whole screen.
I can the load it locally using Tools->paste treebuilding config and take a look at the full configuraiton.

Without seeing the config the best guess is that either you have cap set to 3 (so 3rd bet is rounded to all-in) or all-in threshold is set in a way that 3rd raise is already rounded to all-in.

I have silently increased the number of threads the basic version works with to 6 so there wouldn't be much improvement for i7 quads (that was the idea to not force i7 quad users to upgrade). The performance difference is only visible starting from hexa core CPUs.

i am trying to find turn bet optimal size for IP. I ran 4 different simulation with 40-45-50-55%
In order to get the conclusion i am looking at IP ev in aggregation reports. Do i have to run the reports in the root to compare evs or on the turn after oop is checking?

The EV differences are likely to be very small so you need very good accuracy to get reliable comparison. It's worth considering if that's the most fruitful area of research (in terms of practical benefits).

It's the best to just look at overall EV at the beginning of the tree. You can see those results after clicking "calculate results" in treebuilding and calculation tab.

This will show the ev of only one tree not the entire x trees i have in the folder correct?

just got an offer to buy 2 z800 workstations.

he said each one had 12 cores and 48gb ram and i could combine the two for 24cores

any idea what benchmark this would give me for pio?

im currently using an i5-4670k w/ a benchmark around 12

fwiw he says its the xeon x5690 processors 3.47ghz . need help in deciding if this is a good buy.

Hi,

It's quite hard to read the whole 101 pages of this topic, so I'm sorry if my question was already asked.

Since my computer have low RAM/CPU, I would like to perform a whole postflop computation in multiple steps, as follow:
- make a first run with multiple bet/raises sizes on the flop, but only 1 or 2 sizes on the turn and the river
- then take ranges calculated on the flop for any spots (such as check/bet/call) and rerun with a larger variety of sizings for the turn and the river.

I know this approach won't give me accurate GTO strategy on the flop, however it seems that's something we already do for preflop solving (running with few bet sizes for postflop sub-trees).

Of course I could do it by hand, but is there a way to do it automatically, in a script for example ?

Regards,

- Jurik

Yes, I am sorry. If you want to do it across many trees then it's the best to run the report in root. You for sure need to run it before any decision by either player is made.
I am a bit skeptical about being able to tell which bet size is better if they are so close in size though. There is a lot of noise from subset choice, accuracy as well choice of the sizing on subsequent decisions/streets.

It depends on the price. This is quite an old CPU which doesn't support modern instructions. I would avoid it unless the price is a real bargain.

As you can see here:
http://ark.intel.com/products/52576/...-GTs-Intel-QPI

Those are hexa core (6 cores) CPUs, not 12 core ones. Their performance will be worse than a modern i7 quad. My guess is that they would be slower than your i5 as well although that requires testing (running Pio benchmark on it).

This is in general a good idea however I would aim to reduce turn/river number of sizing as well. You will need a lot of experiments to know what is and what isn't safe to remove though. Those things are very difficult to test in general case.

Yes, there is a feature designed for that exact case. You can find it under Tree->Create subtree configuration in the top menu. I show how to use it here:

https://www.youtube.com/watch?v=1PEmsGaGLvk (3:00 mark, see the video description for a link to that part).

Not yet but it's an idea I like