I have a threadripper system and I have the following issue

When i am running a script pio uses only 65% of cpu https://prnt.sc/m0yeb1 I get the same issue with jesolver who should be at 100% cpu. I have two disks i moved pio folder to both of them. I dont run them at the same time ever but both disks have the same issue. One is ssd and the other hdd

Power performances are on high

https://prnt.sc/m12od4
https://prnt.sc/m12oj9

Hi. I have an issue with my PIO solver. I manually looked at 49 flops because I don’t have the pro version to run scripts. And now that I go back and look at the ones I found surprising and retype it in to see what happens on turns, it’s not giving me the same results as it did the first time. For instance. As Ks 3d was a 85% bet and now I’m only getting a 60% bet for OOP utg vs BU flat 100bb effective. Why? Also sorry if this is a reply to the above. First time posting

I'm on a Mac and I can't seem to open the downloaded FreePIO file.

It seems like your cores only work at 1.6Ghz instead of 3.4Ghz (I am using information from the email you sent us as well). The temperature is ok. Where Pio is located doesn't matter. All cores are in use. How fast a CPU core run is not something an application have control over. It looks like some configuration problems but it's very hard to guess what it is.

I am not sure what you mean by "retype it in to see what happens on on turns". Can you describe the process in some detail? You might be doing something there which is not correct theoretically but I need to understand it first to offer a solution.

Unfortunately PioSOLVER is Windows only. We have quite a few customers using it on Macs but they all have a way to run Windows there either parallel to MacOS (Parallels, other VMs) or booting into Windows (Bootcamp).

I've recently moved my solver from one server to another and something funky is going on with running scripts on the new server. After "set_board OK!" the script is seemingly taking a longer and longer pause at each iteration. After the first board in a script it moves on near instantly and then 50 solves in it will pause at this stage for minutes. If i kill the script and restart it where it left off it resumes running quickly again and slowing over each iteration. (solve times aren't slowing, just this increasing pause) Anyone else experienced this? Any ideas on rectifying it if possible?

Hi.
Few simple questions.
Is 16GB of RAM enough to solve most of the 6-max NLHE spots? Especially when we begin to gain intuitive sense into what the optimal bet sizing is, so we don't have to put say 3 bet sizes to the flop. I could go for 32GB as well.
Also as I move up the stakes, I can begin to make assumptions such as there would be no raising range on the flop.

Then I'm thinking about upgrading my processor. I'm thinking about:
Intel i5-9600k. 6 cores, 6 threads @ 3.7ghz (no hyperthreading)
Intel i7-9700k. 8 cores, 8 threads @ 3.6ghz (no hyperthreading)
Intel i7-9900k. 8 cores, 16 threads @ 3.6ghz.
Intel i7-8700k. 6 cores, 12 threads @ 3.7ghz

Is hyperthreading worth the performance upgrade?
I have just the Basic-version so I can only utilize 6 cores. Would 9600k be good then?

hey!

i use piosilver basic and made a preflop chart what was saved to .txt. pio cant read it and th pc froze after i click on it. what should i do?

Is there no way to construct a range out of a limited sub-set of hand combinations? For example: Constructing a 5b jam range that is limited by what's available to us after 3betting? I can only start from scratch. This means the solver will not have the proper proportion of hands no matter what I do.

This is unfortunately a problem with the way we allocate memory. Deallocation is very slow after a while (due to inefficiences of Windows meomry allocator). It makes sense to restart the solver sometimes to mitigate the problem. It's one of the things I will be working on in very near future.

Postflop: yes as long as those are reasonable trees. If you start inputting thigns like 3 bet/raise sizes everywhere then it will blow up to ~100GB very fast. You can use a free version to estimate configs. My view is that everything above 16GB is already too big. You should be able to solve things with 3 bet sizes/one rise size below it if you avoid very long chains of small bet/small raise/small raise/...

In general hyperthreading gives around 12-15% but it may change with recent Intel mitigations. I would look into AMD CPUs as they offer much better price/performance ratio these days. You can use a formula of:

(number of physical cores * base frequency)

and add around 12%-15% for hyperthreading.

Yeah, this won't use hyperthreading.

We need a bit more detailed description of the problem, sending us the .txt file would be a good start. Please send it to support@piosolver.com and we will take a look.

I am sorry but I don't understand what the problem is. Can you describe in a bit more detail? You can always choose hands out of 3betting range so I am very likely not getting the point of the problem.

9700k: 8 * 3.6 = 28.8
8700k: 6 * 3.7 * 1.135 = 25.197
9900k: 8 * 3.6 * 1.135 = 32.688
9600k: 6 * 3.7 = 22.2

So if my math is correct, 9900k vs 9600k should be a 32% performance increase?
9900k seems like a better choice over both 9700k and 8700k also in terms of pricing and new technology.
But is 9600k "just enough"?

I would be interested to study such spots as donk betting turn, which increases the game tree a lot. I would also be interested to solve heads-up 2-max in the future just for the fun of it.

Hi,
First time using PIO Preflop. I'm getting error message:
"Server Error:Error Command add_all_flops not recognized."
For my flops I have selected PioSolver_2016_20__flops.txt
Any help would be greatly appreciated

I calculated a tree. OOP Bet% was ~77%.

I used Tree>>>Round Strategies (1/1, OOP, flop) so that each card pair was bet or checked at 100% frequency for OOP on the flop.

I then chose Tree>>>Set Strategy and Lock Node, selected "lock all hands", and recalculated the tree.

1. Is this the correct way to lock OOP strategy so that the remainder of the tree is recalculated?

*********************

I am confused by the terms "MES" and and "Exploitable for".

While calculating the above tree, one of the iterations is:

running time: 553.649
EV OOP: 127.245
EV IP: 97.755
OOP's MES: 155.550
IP's MES: 97.942
Exploitable for: 14.246

My thinking (which is obviously incorrect) is that if IP does not deviate from his strategy, OOP can exploit him for an additional (155.550-127.245) = 28.305 chips (!!!). IP can only exploit OOP in the same manner for (97.942-97.755) = 0.187 chips.

2. How am I wrong in the above paragraph?

3. But if exploitability is already accounted for by MES, then what is "Exploitable for"? Is this the maximum possible error. In other words, in the above iteration, then OOP's actual EV = 127.245 +/- 14.246 ? [I know this is wrong!]

4. When I solve the node-locked tree in the above example, "Exploitable for" starts to plateau at a large number (>6% of the pot). What does this mean?

***********************
I then recalculated the original tree by node locking OOP flop at 100% Bet.

"Exploitable for" quickly reached a low value.

Here is one iteration:

running time: 144.469
EV OOP: 131.167
EV IP: 93.833
OOP's MES: 131.591
IP's MES: 94.250
Exploitable for: 0.421

EV OOP was practically identical to the original mixed strategy (also 131.16).

6. MES for both players is very slightly above EV. What does this mean and why the huge difference between this tree and the last?

7. Is the previous tree better (as OOP) for opportunities to exploit an opponent?

Yes, that's correct although currently with the basic version you won't see the difference between 6 and 8 cores but it's likely basic version will use 8 cores in the future.
If it's worth it or not is a judgement call. Usually with Intel CPUs the most high end ones are overpriced and it makes sense to go for middle of the range one but you need to check prices and decide for yourself.

If you mean from preflop then 16GB of RAM won't be enough.

Your are trying to build a full preflop tree with either basic or pro version. The preflop solver is only available in the edge one and requires really serious hardware to run.

Unfortunately once you round to 0-1 the solver won't be able to recover from that.

MES is EV of a strategy which exploits the opponent in the best possible way.
Exploitable for means how much you would lose to MES per hand if you play once OOP and once IP according to current solution.

Yes, it looks like you got that after rounding or node locking because usually the solution is much more symmetric when it comes to how much IP or OOP can win in comparison to the solution.


Again "exploitable for" means how much you would lose against a prefectly adjusting maximally exploiting opponent per hand. In your case you lose ~28 chips on one side and ~0 chips on another side which loses on average 14 chips per hand.

The exploitability with locked nodes is calculated with the assumption that one can't exploit if one is locked so it in theory should go to 0%. The main reason it doesn't is that the solver is unable to recover ones the hand is already 100%-0% on some actions even if that's no longer optimal after locking. That's why it's in general more reliable to lock nodes on unsolved or barely solved tree. You can build a fresh one, lock and solve and it should be mitigated if you are running into this problem.

The reason is imperfection of our engine. We use very little RAM in comparison to more mainstream algorithms but this is the price we pay.

I was under the (false) impression that solver starts from scratch each time "Go" is pressed. This is what I wanted. Thank you.

Is there any way to copy and paste the entire grid of the original flop solution to the new tree in one action, or do I need to copy Bet range and Check range separately into "Set strategy and lock node (this is what I'm currently doing)?

You can
- open "Node Locking" window in the old tree
- Menu -> Copy/Paste -> Copy
- open "Node Locking" in the new tree
- Menu -> "Copy/Paste" -> Paste

ThanX Juggler!

At what stakes does it makes sense to pursue GTO solvers?

Hi,

My AMD Threadripper only uses ~30% (40% (20%*2) when running two sims simultaneously) of its total CPU using PIOSolver. I use the PIO pro and have the thread count to 0 (default).

Any help would be appreciated.

Thank you!

is there anyway to change flop weight after a script was solved ???
I solved a very simple tree with all 1755 flops but all were weighted at :1 and would like to weight it properly to get accurate aggregation reports

It's a tough call. Some of our customers don't play professionally at all, some play low stakes, most play mid/high stakes.
For questions like this it's the best to seek opinions of other players instead of software developers.

Pio pro is limited to 16 cores as of right now. If it's 16 core Threadripper then all the cores should be in use though, you are just not using hyperthreading which doesn't have that big impact on the performance (even though task manager is going to show it as max 50% CPU usage).

Yes, you can do it like this:

1)Find the folder where the saves and the script are saved
2)Make a back-up copy of the script file (.txt one)
3)Edit the script file with a text editor, you will see list of all flops with weights there
4)Edit those weights to correct ones
5)Run the report

If you are using the newest version it will show you the weights when running the report so you can confirm those are correct. Don't forget to follow point 2) carefully in case you mess-up editing the script.

Do you know of any ways to ensure the PC uses its full CPU capacity instead of just the ~40% mine currently uses?

The problem is that Windows calculates CPU usage incorrectly. If you use all the cores without hyperthreading it will never go above 50% even though hyperthreading only contributes to around 0-15% of the performance depending on the application (in Pio case it's around 12%-15%). The best way is too look at the task manager and see if all the cores work during solving and that they work at high enough frequency.

https://i.gyazo.com/e9612f1146cbffcf...0b1b686813.png

Does that picture help?

Also, to be more specific on the issue, if I run a PIO benchmark test, I'll get a result of approx 2.5s (task manager shows that my PC uses ~90% of its CPU for this). However, if I run a series of them back to back without stopping, the results will slowly increase in time until it plateau's out to ~5s. The windows task manager will correspond with CPU usages (steady decline from 90% CPU usage to ~45%). The same will happen if I run a normal SIM in the PioViewer: lots of CPU usage (~70%+) and a steady decline to 40%.

A bit but I need per core one.

Is it really a pro version and not edge one? That's a fantastic benchmark.

Interesting, I really have no idea why is that. Does it also happen if you restart the solver before every benchmark?

I would like to know the bets in console, not the accumulated ones. it's possible
thanks