 So welcome everybody to the by Excel webinar. This webinar is number 61. And it's a pleasure with pleasure we we lost the Paul Bauer that is the promise developer manager, and he will tell us what is new in Gromax 2022. I'm hosting this webinar for by Excel and Alessandra Villa, and I work on the Royal Institute of Technology, and with me it's an approval from the University of Oedimburg. So Paul has done his PhD in Uppsala University. He was working on computational enzymology in the group of Lynn Kammerlin in try to understand more enzyme catalysis. Then, and there he will start to work in analysis simulation engine. In, he moved to the group of Eric Linda where was he was both a scientific programmer and a researcher. And in 2019, it took over the position of Gromax developed manager. And since they are, he has focused in on the smooth surrounding the project, as well as implementing new feature to enhance the visibility of Gromax. I will give the work to Paul. Thank you Alessandra so I'm just gonna start sharing the presentation now. Great. Okay, I hope everyone can see the slides and thanks everyone for joining today for this. I hopefully an informative overview about what has changed with the newest version of Gromax 2022 release. And yeah, gonna tell you what kind of new features we have implemented what other things we are working on. And I hope I can also give you a taste of what new interesting science you can do with this version. Quickly. I hope everyone here knows we use Gromax to do bio molecular simulations but not just by the molecular simulations but basically everything can simulate that you can describe with a pair was potential between entities particles whatever you want to call them. And I will give the most practical showcase here and this figure that I always like using the gestures and transmembrane receptor in the simulation books with different kinds of solutes and the lipid bilayer that it is embedded and all simulated atomic atomistic detail. And running very efficiently on all kinds of different hopeless. So if you're interested in talking about comments if you have questions and want to know an announcement about new things that we've been working on. I highly recommend that you join our user forum at Gormax bio Saudi view. There you get all the announcement about new things we have been working on the new new versions released and patches of released. We also have the community there that will help you hopefully be able to help you with answering questions that you have when it comes to usability. When you get an error or the program to something that you didn't think it should be doing. We also have a part there where you can find third party tools and extensions to grow max for example new force fields that we don't ship with our number of our main version. Or data analysis methods that we don't have included but the other people found useful. So, please go there. And maybe you will also see me answering some of your questions when I find time to do that. So if you really want to know everything what is new in the newest version, I just recommend that you go to online documentation. They find all the documentation for this version and all the previous versions that they have them. Everything I'm going to mention today is mentioned there as well and release notes. So if you just want to read about it and I recommend you switch off now and go to the view I have you. Hopefully give some more context to some of the things we have been looking on, however, then the rubber dry some of these we have in the case notes for it. You can also learn about the previous changes we had for 2021 version and 2020 version in the respective webinars for them. I've been talking about the development for the for the previous years. Let's go right to that. This version is has been released just about two weeks ago. And we are planning to have a few hatch releases on it for hopefully fixing bucks and getting them out to the fixes out to the users as soon as possible. I also want to remind everyone that the previous 2021 release is now just an extended maintenance. So that we are not going to fix minor bug stare that for example, affect data analysis tools. You're going to fix anything that affects the simulation correctness. So if you find physics breaking back, we are definitely going to fix it and included in the hopefully last patch release for this. In July or August. We do this so we can focus our effort on keeping the country from this as stable as possible and fix everything right there. And to also be able to focus on the development for next year's version that is supposed to be released also beginning of February 2023 or slightly after depending on how things are going in development. As a second note this means that for 2020 2020 version they're not going to be any more patch releases. Just a few weeks ago I made the final patch release there for 2020.7. So this is going to be the final version of that release bunch. The new things we have in 2022. There are several, but depending on what you're interested on maybe some people are going to be more interested in just listening to a collection of music we have from our cool quotes list. Someone made a playlist of all the songs we have in the grammar school quotes that you hopefully get after every time you want to grow much comment. And if you want to listen to something while doing data analysis of your simulation is running. Check it out. I thought it's supposed to be quite okay. But we also have been doing some actual work, and this includes a lot of work on under the hood that you hopefully won't have to interact with but will affect how your simulations are running and how fast they're running. And this includes that the speeder has sped up the calculation of the free energy kernels and the energy functions a lot by doing a lot of content maintenance and we can use the same kind of acceleration that we use for the normal one one interactions there. The same kind of templating on accelerated single corners. And if you also work on improving the control you have over free energy transformation pathways with a new implementation of soft core interactions between particles when you appear or disappear particles or change their one of us and one of us parameters. This will hopefully make it much more stable to run this kind of transformation transformation reactions or simulations. Because there could have been issues before with the previous implementation again this is something you shouldn't really notice it should just look if it doesn't then it's something. And it comes to new features we have implemented the ability to use one summation coordinates something I'm going to go into more detail on the next slide. There's a new interface you for running quantum mechanic molecular mechanic habit simulations with DCPK package. And we have worked a lot on the accelerator support that means you can use different acceleration devices aka graphics cards with a different frameworks to actually upload the and calculations to those devices. And we'll also go and get more detail in this later part of the presentation. Now, what is the main main meat of this release. The first thing is really the implementation of transformation reaction or core coordinates. So the full for pulling particles or find forces to different part of this new simulation was relatively limited to a few defined mathematical interactions they could do distance based pulling you could do angle based pulling not that much else. And that was very limiting so you were not able to do all the kinds of simulations you're probably interested in and had to use similar packages like hatching grounds with plumes or using foot party extensions for example with covers. So we have been working on this to improve this and one of the improvements is the transformation for coordinate implementation. They can use any mathematical transformation that you can think of. Can you can apply to a set of particles. So for anything that you can think of. Any kind of mathematical formula, any kind of interaction, you can pass this to remarks, and we will use this to calculate forces between particles for all kinds of linear nonlinear combinations of functions, different functional types. Any kind of weird angle distance combination interaction. And you can think of or not think of. And I hope this will improve improve the usability of comments dramatically for calculating different reaction profiles or investigating behavior of systems that you previously limited off by the implementation. I think it will completely remove the need for some of the advanced features you have for example include, but I'm hopeful that we will at least make it possible for some users of the external patches to use our main because with this they will be able to use all of the features that comes to offer the simulation to accelerators, because this is fully fully implemented in our code and fully integrated. So it should just work, you could should just be able to use your mathematical expression for your for your for the force you want to apply. One comments with it and everything but everything should be done automatically under the hood. There may be still some issues because this was a major effort to get this in so please try it out and let us know if something doesn't work or something works on the way to expect. But I hope this will enable a lot of new, a lot of new and interesting science that can be done in this form of similar simulations. This is a major addition to the code that has been in the book for several years and has now finally come to fruition is a new interface as a set for quantum mechanics mechanics hyper simulation with DCP2K package. So this, we have been working on our collaborators in Finland at University of Eurasia. Make it possible to just do a standard and the simulation setup where you use your protein structure that you want to simulate here as example of your fluorescence protein. You apply use your normal molecular dynamics topology parameters whatever to solve with the system and generate an input. And then you add to this a description of your QMNM setup. Again, Gromax automates a lot of this and gives you a range of different settings that are known to work in interfacing with CP2K. And then you just want it in the version of Gromax that is coupled to the DCP2K package. And it will do its job and you will get a quantum mechanically hybrid machine hybrid hybrid description of your system. And we'll hopefully be able to study a lot of very interesting signs in the realm of catalysis or fluorescence behavior. Anything again you can think of that is lived that is something that you can only study with QMNM and their molecular mechanics is limiting. This is actually a feature that I myself am very happy that we have finally integrated because I'm coming from this firm. And being able to do this now with Gromax is very exciting because I hope that people will be able to use the computational power of Gromax to apply this to more and more interesting and larger systems. There are still some limitations of this because we were more interested in getting it actually stable and safe that people can be sure that the inputs Gromax generates are accepted by CP2K and work level Gromax. The plan is to extend it to be fully flexible and generating those inputs so that any kind of QMNM simulation that is possible with the similar to the QM part of CP2K can be done with Gromax in this interface. This is a very interesting and exciting example that has already been done by again our Finnish collaborators is actually using basically all the new features together with one of the main features in Gromax the accelerated rate histogram method of your age to simulate an enzyme reaction and source COVID-2 I think it's a phosphatesterize but I would need to check again because I don't want to say anything wrong. And this combines both the QMNM interface running the QMNM simulation using transformation coordinates to describe the reaction coordinates that is the working informing bonds in the reaction in the system. And AWH to sample the surface of the reaction coordinate exhaustive and very efficiently. So I didn't put more information on the slide because I think this is about as beautiful as it can get but the tests that our collaborators have shown that it's not just more, it's not just easy to use this in Gromax it's also way more efficient. Using this combination of QMNM transformation coordinate and AWH than using standard and below something to exhaustively sample the full reaction coordinate that is going on here. So one of the advantages of AWH as I hope at least some of you know because we have this had this for quite a while now is that you can distribute the sampling of the reaction coordinate to several independent simulations that all independently sample the surface that you're interested in. And then are able to share this information in between to make sure that they explore the reaction the surface, the reaction surface to parts that haven't been completely sampled yet. Which makes it way more efficient than using standard and below something simulations where you are limited to a set window of for example distance coordinates here that describe the forming and breaking bonds, and are not always able to fully sample the regions that you're interested in and the plot shows this quite well because with AWH you get a very smooth reaction profile with very well defined transition states. Everything while umbrella sampling gets the same overall plot but shows that there's still some regions that probably would need more sampling and that's actually accurately described. And yeah as I said, this doesn't include the performance data but with AWH you're able to achieve this kind of accuracy that you're interested in in this relation to understand this way faster than in other ways. And then I hope people will find this useful and will try this out and if they find bugs they let us know. But you hope this will allow many many new interesting synthetic studies done with Gromax and help the groups that want to do this doing it easily and efficiently. So with more feedback from methods we're going to go to actually hardware support and where you can run Gromax efficiently on what kind of product. As was mentioned already a year ago when we had the webinar for 2021. He said that he have been working on adding the SQL, adding offloading to inter devices using the SQL standard to Gromax. We had this already implemented back then, but mostly just in the way to see that we can do it and not fully supported in the main release. In the last year we have been very busy, the GPU people know what you have been extremely busy trying to extend the support for SQL and making sure it works well on different devices different implementations and being able to accelerate the calculations of more interaction types. And has been very successful we have now extended the support that we had with open the deprecated open CL implementation on and into devices to more interactions that hadn't been able to be calculated before on accelerated devices. But we're still not there that the support the support all interaction types that we have support for with India devices using food. This is because that's not an easy task not just in adding new implementations for the actual calculation corners but also making sure that it looks well with all kinds of calculations that are done in Gromax and that's always the correct calculation as and the correct way to actually communicate between devices and our calculation should be done on those devices. So this is something we are currently working on to get better hardware support for example into devices, but it will likely be already available for 223 and hopefully maybe before as part of making making possible to use comments on the new big supercomputer resources that are currently being implemented in Sweden, especially in this audience, especially interesting for people in Sweden that will have access to the to those working computers. Also to others that want to just use to use their graphics cards and their machines in the workstations. Another thing we have been working on and got extended is the ability to directly communicate between good advices and running simulations on multiple GPUs. This has previously been possible just when you use the internal grammar extract MPI writing library. It's also no possible to do this with regular MPI. So the decision there to not have this available by default just before the release because I wasn't completely convinced that we got it to divide in our cases. It is possible to use this by using the environmental flags that are already there in 2021 and documented in our guides for getting good performance from and using cook for the device and video devices and it was combined with Buddha Buddha support. Again, people that are interested in using this I highly recommend you try it out but you have to be very careful and please let us know if you want into anything that is unexpected there unexpected simulation behavior or box that you encounter because we need to fix them as soon as possible. So that we can enable this for next release or maybe as part of the patch release for the next one. One more thing when it comes to sickle this that we now not just support the interval MPI implementation of the secret standard. But also the obstacle implementation and using both of those we are now able to target devices from all of the big GPU vendors and media in the end Intel for floating. We used and we decided on going this way to support new and devices because it is more sustainable for us to continue working on one major interface for accelerators. For example, instead of trying to support individual interfaces for example, the CUDA implementation secret implementation for Intel and the implementation that could also be possible for me. So we're going with that. And we hope it will work fine for everyone that uses it. Again, this is partially experimental store. It will work, you will be able to get the working comics version compiled either one API cycle or the cycle. I would ask you to again be careful when using those accelerate using those for accelerating the simulations. Or if anything doesn't work or doesn't behave as you think it should. In other news, there is a new site for people that you want to learn how to use Gormax or back to usability. And that's tutorials comics to talk they can find some introduction direction to it but also tutorials for more advanced methods in Gormax. Not just hopefully not just give you an introduction to what you're doing in this relation but also gives you an idea how it looks under the hood and what you can do to more precisely target the problem you're interested in. So I would recommend that you have a look there. You can run most of those tutorials in on your own in the browser using the mind mind my binder page that is linked here. Or you can just download them to your workstation or your laptop if you want to just one from the stand type of look. Those tools have been developed in cooperation with the NCCS in Sweden. So thanks for their efforts in getting this done. I just want to note that those tutorials are not meant as a replacement for the awesome tutorials done by Justin Lemko that has have taught me how to use comments and I think most of the people have had at least some knowledge of them. So, if you want to learn how to use comments, use either of them, maybe both. I think that's the best way. We have all tools because then we are able to always make sure that they work with our latest version that we keep them consuming. And that new features gets tested and explained to users there, where we have control over them. Another interesting thing that is coming hopefully soon within the next few weeks is that we're going to have a replacement for our web page that I'm not sure how many people actually go to Bromax.org and see the very old page that we have. We have been working on getting this updated so you actually have recent and not completely outdated information there and are able to access all of the materials we have on our other side separately from the web page. So, I'm very excited to actually get this done as well so that users that are interested in comments will have a nice entry point into where to start their simulation career or where they can start learning how to contribute to comments that they're interested in this and contributions of their, of course, always very welcome so we can continue working on new things. Now, there's also some Miss Angelos information that doesn't really belong in any of the other categories. One of them is again something you're probably not going to be affected with directly but it's very exciting for us that we have extended the re implementation of a lot of our first corners and put forth corners for non bonded and listed forces or bonded angle, the heat work calculations with the envelope team, and we are hoping to continue this so we can actually move the calculations done in Gromax to this format because it will help us both confirming accuracy of our calculations and also making it easier to extend it to new formats in the future. So, hope you can stay tuned there for more exciting use, but right now that's something you're probably going to get directly affected if you're not interested in writing your own simulation engine for example then have a look at and we live and maybe it's useful for you as well. So, in Gromax, Python API GMXF has also been extended. Again, that's not some so much new things you will probably see directly. Then try and run when using it or I'm trying to use it with something that you're interested in doing. I hope everyone has heard about it by now because I think being able to control Gromax simulation directly from Python using this API, able to set up simulation workflows, controlling data input or modifying some of our one one control panel directly from Python and very efficiently is something that is very exciting for me as a usability point of view and hopefully make running simulations much easier for the average user and scripting also making it possible to script a complete simulation setup efficiently. Now just a few gotchas to people that's probably maybe not that well worth with reading or release notes. If you if you use AWH and have been using cover diameters for angles and wondered why it's everything else in Gromax is always in degrees and this was in radiance. Yeah, we moved this now to also use degrees as everything else to be consistent. So that means if you're trying to set up new simulations with new version. Don't be confused if this now also has to be in degrees and make sure it is. This doesn't affect any old simulations via check in this automatically so that everything under the hood gets actually converted to the right unit. And that is, I think very interesting I hope for many people is that our simulation setup engine grump is now a lot faster for not just for some force fields when setting up and run inputs but for basically all of them. I hope that you won't be much interested in that other than oh yeah it's faster now because it's has been a bottleneck and we have have been trying to improve this time we are trying to improve this further so that grump becomes less and less of a bottleneck when actually setting up simulations. Also when it comes to usability there we have been trying to make grump more expressive when it comes to invalid simulation setups or simulation setups that could lead to invalid physics so they're more warnings now. This not just includes the inclusion process here but also some combinations of coupling algorithms so if you were a dedicated user of the parents and Thomas that embarrassed that. Grump will not tell you that this is a very bad idea and but not let you set up a simulation. If you're not very if you're not telling it explicitly that you want to do this. And we're also moving to the point now that grump does not modify what you're fine in your one input. And we're also moving to get to the point that everything you said as a one input information is something comics will always take as it is even if it's not efficient or can be maybe two issues but is this often in a valid simulation. This is ongoing work, but we hope that with this it will be more understandable also for people that use use comments that's for example a setting that they have is will be accepted and will be used, but grump will tell me that it's not the most efficient. You can't of course just add features to grammar sometimes we also need to be able to remove things or make sure that features that don't work well are advocated. Noted in last this webinar with the application list we have now removed the ability to build just in the one for comments because the need for this has been long time removed with the addition of the gym x weaponary. And there wasn't really any not logic behind just having an empty one executable or a GMX proper plan of the executable, because you can always run and be one from the GMX proper. And this made our life actually much easier than it comes to development. So, we have to get rid of this. So, if you have been trying to run planets on 32 bit platforms, I have to be very sorry and say that the official support for them has been removed together with a lot of the Cindy acceleration that we have for those kinds of platforms. So, to you to build planets on this platform, it may still work perfectly fine, but we can't guarantee this because we don't have the ability to test it also. Those platforms are not really useful in high performance computing anymore, and being able to work exclusively with 64 bit platforms makes our life much easier. Again, if you affected by this, then you can, of course, that fire if things are working, if not, yeah, I'm sorry, we have to focus all the relevant efforts on this. Another maybe minor thing for people that do what it is to come analysis method if you use the outdated analysis input files. We even move this again to make our life easier. Those files have from a version I think 15 years ago. So please use something more modern if you need to be able to still analyze those files you always you can always use an older version of comments that is the perfect time for the next development cycle there. Probably gone. There's going to be a few things we're going to move one of them is the internal molecular molecule review. GMX fuel. This tool has not seen any development and they have been so many very well developed external tools to actually visualize molecular structures available for many, many years. So I think that we will not try to compete with them and instead of please use one of the properly maintained molecular viewers so we can remove GMX real and again focus our efforts and other things. The same goes for the analysis to a GMX try. I don't really know myself what this tool has been used for and again has been not developed and deprecated and not used for extensive human years so we don't ask people if they wanted to data analysis that was possible just to use an external tool it would probably do things better than our own implementation. So I think that is probably more important for people that use GMX exclusively for data analysis, you will now be able to supply one input file that has the correct masses and atomic radii for particles for some analysis tools. This will remove the ability to reduce those parameters from the atomic names. This will make the tools much more robust because we will always be, they can always assert that they have the correct inputs. And it will make all that easier because we don't need to maintain a probably always always outdated list of induction rules that deciding which atom corresponds to which mass and atomic radii. I hope that won't affect too many people but it's something again that is important for usability of all the tools that I want. And of course we're not just maintaining things we also have a lot of interesting plans for the next development cycle and people are already actively working on this both on Stockholm and all of our external collaborators. One of the main things that we want to be able to enable is that you can use GPU dial communication between multiple accelerated devices for all accelerators and all acceleration frameworks that they're actively developing. This will mean CUDA and SQL by default. And then this will hopefully need to impressive improvement in performance as we have seen for CUDA devices. We have postponed this as I said before because we need to make sure that this works reliable on all simulation setups and with all one inputs that's possible to support this. And this is really cool because it's a lot of code plus we can we have to validate but this is one of the goals we have. So that with the next version performance performance will be even better. And this will target for a new thing that we want to improve with the next release and on to extend usability is the multiple time stepping implementation that we have in governments. It looks quite well but not again when you want to use accelerators to offer some of the calculations to them. Because it's again a difficult problem trying to do this correctly making sure that correct calculations are done at the correct time and the data from this calculations are available for calculations at the correct time. So this is not something that should affect how people running simulation. It's just it should be able to run more kinds of simulations with multiple time stepping and efficient acceleration. And the other thing that has been under the same active development for a long time but we are hoping to get done for the next release is that the change input by reading to using modern performance. So here we can validate in inputs already on the platform at level and don't have to run a complete file through one before making sure realizing oh yeah, a section was missing here. That is important so you need to start again and maybe after you have already waited for going to process your five million system that's not a good thing. So again, stay tuned. I hope this is something I can talk in more details when we come to the next release webinar in about a year. One more major feature that has been under development and that we are working on integrating is a new constant page lambda dynamics implementation that has been worked on by again out the library just in Finland. And it has been already published as preference or if you're interested in this, I recommend to look at this. And we hope to get this integrated properly in remarks and available for all kinds of simulations and all kinds of your problems you want to start with the next version. This is a major effort so I can't promise it will go into the next release but we are just hoping and working very hard to make making this a reality because the amount of different essential problems you can then study just increases. And one more thing that is important for people that want to develop their own tools that work together with Gromax or want to use Gromax as a library we are working on providing better interoperability with external tools for new to the API Gromax API. And you're hoping that with beginning with the next version we'll be able to provide a stable API that you can use to develop your tools again. Your tools against Gromax so that you don't need to adapt to our glimpse of changes to the current API that we're having. Yes. So, of course, this is not something that is done by a few people but we have a very large team that work. On our on Gromax both here in Stockholm, our parts of Sweden and all of the world. I'm not going to read the names here because it would take a lot of useful time I actually want to leave for answering the questions and maybe having some discussions. But when you use the new Gromax version then think and please thank the people that have been working on it. Of course, this has been again major effort and a lot of work for everyone of us. And I want to thank you for your attention. I hope this has been interesting for you and I'm waiting for your questions. Paul, thank you very much. Very interesting. Very good to have this overview. I have a number of questions already for anybody who's not yet asked your question you can use the Q&A functionality to ask your question and I will read it aloud. And Paul will answer. So with the first question is from Anita asking for any particular years Gromax version is it best to switch to the last version of that year for example 2021.6 as it becomes available to avoid bugs related to the physics itself. And you should not switch major versions for a particular study there I would recommend you stay with the 2021 branch. But you should try to use the latest patch release because we don't do those for fun, we actually every one of them fixes some kind of bug that is hopefully happened people have been running into this before. You should check the release notes to see if it's anything affecting you. But in general, I think it's best practice to keep on the latest version of a particular release bunch. Okay, now we have a few questions that are all about the QMM interface. I think I could have a start but answering some of them and I might be able to share something useful if I share a screen as well. So, shall we do that and then you can, you can also jump in. If there's anything that you want to add that of course and then when you want to go back to the presentation. Yes. Okay, so the first question that we have, I will share the screen. The first question we have is, does Chromax add the link atoms automatically for the CP2K QM setup. Now I think, and you might correct me, I think that with a new version, with this release version, the interface does. Yes, it does, it automatically generates your valid inputs. So you don't have to do those things manually. Okay, great. And then the next question about QMM is, do you have a tutorial where we can try this QMM with CP2K out? I'm not sure if we have an official Chromax tutorial for this yet, Alessandro. I can share, yeah, so I'll share, I'll indeed share as I said I would, which is that if people go to GitHub, bioexcelgithub, so github.com slash bioexcel. And if they search for in the repositories, a search for CP2K, you will find a number of tutorials that were developed for the interface. Wherever it says Chromax and CP2K, the most recent ones will be for more recent versions of the interface. And I suspect that's the most recent one that's shown here, which was used in a recent workshop in February, was probably using Chromax 2022. But maybe, no, maybe not, okay. No, that tutorial is a special tutorial that was developed for a workshop in Finland and where Dimitri is a combined CP2K with AWH. Right, yes. So that is a special tutorial and that, I mean, people are, but those tutorials, as Paolo said, they are not, yeah, they are in biocell repository, and they are not up to now, I don't know how much they are on the use of everybody. I mean, they can use but they have to have some resource to be able to run that. That's what I was meaning, while that is the main difference, while the tutorial that Paolo was presenting are just to avoid confusion, you can just run directly using my binder that I just mentioned to avoid confusion. So, we don't have an official Chromax tutorial for the human interface yet, but I think this is probably going to be added at some point in the future. I think the point about the resource is a very important one to make because, I mean, some of these runs with CP2K can use a lot of resources, especially if you have a lot of QM atoms. So actually, that brings me to another possibility that people can use is that there is, so I'm based in University of Edinburgh and we host an HPC machine called Archer 2, and we actually run a self-service course based on the course that biocell run using Chromax together with CP2K. Anybody can sign up for this and it gives you access to the machine to use sufficient resources to be able to run the associated tutorial. So if you were to register, if you were to find this, you can register for this and you will get an account on the machine and you'll be able to run this. And this uses perhaps a slightly older version of the interface, but it will get you familiar with it. There are associated recordings that have advice from Dmitri and other colleagues about some tips. So that's also an option, but yet indeed the difference is you cannot, you need access to some amount of resources to be able to run this. Okay. So then the next question is, what is the performance of the new Chromax CP2K QMM compared to the QMM implementation in CP2K itself. So the performance itself within Chromax you can, everything when it comes to the molecular mechanics part, this is done by our highly optimized calculation that you have for this. So I'm not sure how performance the QMM implementation in CP2K is itself. So I would hazard a guess that Chromax is faster for the MMPart. If you have a large MMM system, it's very likely more efficient to use Chromax to run this. Yes. So I think, having looked at the systematic benchmarking of this, that's certainly true. But essentially the main bottleneck in the amount of compute time spent on the QMPart is much more than MMPart. So indeed it's essentially what Chromax is doing is just calling out to CP2K to do the QMM computation. So they're solved within the QMM region and also the QMM coupling. So that's literally directly using CP2K's own subroutines for that. So I guess the answer to the question is how similar is the performance? It's very similar to CP2K's own QMM performance simply because the QMM part dominates. But if you have a very large system and a small QMM region, then Chromax would be faster or less important. I think that's probably right. So we have a few other questions, but perhaps it's useful that I stopped. They're not about QMM. So maybe I should stop sharing screen and you could share your presentation. So the next question is curious about the implementation of constant pH molecular dynamics in Chromax. Has that been implemented and how can users try this out? So right now I just said there's a pre-print part for this, but the implementation itself has not been made public. But I've been involved in it and we are hoping to have this available to people who want to try out relatively soon. But just a note of caution that of course it's not up to the standards we have in Chromax. When we have this released as part of main Chromax when it's done, it's like maybe going to be a bit different to what is going to be released there. So I actually have to ask it to be patient until this is available. I'm sorry. Okay, and then there's a question. Is there any plan to bring back support for simulation without periodic boundary conditions? This is something we are looking at. The problem is that those relied on the group screen that we removed a few years ago. And the work on this has not been that's intensive to try to bring some of the deprecated removal function. But we are working on it. So again, please be patient. Next one is which flaws could arise from usage of the flags gmxgpu dd coms and gmxgpu pme pp coms. Does it only affect the stability of performance of the simulation run or could also affect the correctness of the simulation? That is what we, that is actually the reason why I decided not to have this enabled by default, but you have to use the environment flag. Because we cannot fully guarantee that you have that there are no correctness issues with this. And we had some just before the release we had, I found one bug that actually influenced the stability of estimation as well. That was related to the way this communication actually happens on between the GPUs. That was a very minor edge case and that should not affect people that use normal kind of simulation setups, but that's something we need to be able to account for. It may be that you never run into any issues because we know that our collaborators from NVIDIA that have been contributing this code have been using it extensively for production simulations and they haven't run into issues there. But we cannot guarantee it. And that's for me, I want to ensure that the gmx version get something that is enabled by default is not allowed to fail. Thank you make sense. Then what are the plans for non default non bonded interactions tables are currently not supported. Yes, and I expected this question so the table support is something we have been trying to get back into the comments now, since the day that we actually moved them. But this is, it's not as easy as people may think to actually get this back into the code because again, the old code relied on the group. Group scheme kernels that have been removed. And getting it done reliably for the role that they're using to crack a moment actually is a difficult undertaking. But I hope that he can get this done this way. Because it has been a long time and we need this kind of ability so people can use more flexible interaction types for the system and it's a major feature that the company is. Okay, thank you. We have only one final question which is something that I can probably tackle about GPU version of Chrome with CP2k so asking somebody's asking, how good I presume that means how fast will be the GPU version of compared to the CPU only version of Chrome, CP2k. Obviously, Gromach is great with GPUs, but since these QM calculations are dominated by the QM part QM part in CP2k the question. The question is really what is the weight element and part of the simulation step there. If you're not having a giant system that is millions of atoms for compared to a very small QM region. I don't think you're going to see much improvement between the CPU and the GPU version of Gromach. I'm not sure how well the CP2k runs with GPU if it can and to offer its regulation for the GPU. Yeah, it can but it's often not necessarily beneficial and especially the relatively small number of QM atoms that you might tackle in a QM memory for biomolecules. We've not seen evidence for strong benefit from use of GPU offloading in combined Gromach CP2k calculations. Not simply because it's just not as efficient as Gromach is for the particular part of the calculation that it's trying to do. So I think that's I think that's all the questions are a lot of people saying thanks as well so thanks is really great. So really appreciate that overview and talk and answering all these questions. So, Alessandra do you want to round off. Okay, and the next webinar will be given by Szilard Paul, and he will speak about improvement in the Gromach's heterogeneous parallelization, and it will be the fifth of April. So I hope to see you in almost one month. So we have a little break in the webinar and then after this webinar we will have a series of webinar that will address all the use case that has been run in by Excel but more information will become. And I highly recommend people actually joining us if they want to you know more about the ability to run Gromach's efficiently on accelerator with different ones that helps us. Szilard will be able to tell you exactly how to do the most efficient simulation that. Okay, I thank you pal for the participation in particular also thanks everybody or all the participant for the active and asking questions. I see you in a while. Bye bye.