devtrackers.gg
7 months
ago -
EVE Online
-
Direct link
Transcript (by Youtube)
| 0s | good morning capsuleers and welcome to |
|---|---|
| 2s | the third day of FanFest |
| 5s | 2023 if you've ever wanted to peek under |
| 7s | the hood of Eve now is your chance as we |
| 10s | are about to explore the building blocks |
| 12s | of new Eden please give a very warm |
| 15s | welcome to CCP |
| 25s | vertex good morning everyone it is so |
| 28s | good to see you all my name is is Ben |
| 31s | also known as CCP vertex and I'm the |
| 34s | development director of our core |
| 36s | Technology Group known as |
| 38s | platform I've been around eve since the |
| 40s | early beta in 2002 so this is actually |
| 43s | my 13th FunFest overall it's really good |
| 46s | to see so many familiar faces and |
| 49s | friends having been around for the past |
| 52s | two decades means that I've had the |
| 54s | privilege of witnessing the evolution of |
| 56s | Eve over the 20 years and how we're |
| 59s | constantly tweaking and improving things |
| 61s | modernizing the game the look and the |
| 65s | feel crafting the best visual experience |
| 68s | is something we put a lot of effort into |
| 71s | we take a lot of pride in that because |
| 74s | we want you to have the most visceral |
| 76s | experience possible the most immersive |
| 78s | experience possible and the last two |
| 81s | decades we've come a very long way in |
| 83s | that Journey from Eve in the early Alpha |
| 86s | all the way back in 2001 where the look |
| 89s | and feel wasn't well def find at that |
| 91s | point to the open and closed beta |
| 93s | periods of 2002 we constantly evolving |
| 98s | that look and |
| 99s | feel modernizing the engine capabilities |
| 103s | the art pipelines we've released |
| 105s | countless 3D model redesigns texture and |
| 108s | Shader |
| 109s | updates completely overhauling the |
| 112s | lighting and visual |
| 114s | effects to the graphics API upgrades |
| 116s | moving from DX 99 to 11 and Nyon 12 |
| 121s | this transformation represents two |
| 123s | decades worth of modernization |
| 126s | effort all while needing to support your |
| 129s | record setting Fleet fights some of the |
| 132s | largest PVP battles in gaming history |
| 135s | something you should be so proud |
| 137s | of you don't make it easy for us but we |
| 141s | really wouldn't have it any other way |
| 143s | and we are definitely not done with |
| 146s | modernizing the look and feel of |
| 149s | Eve |
| 151s | CCP games presents |
| 155s | carbon a Next Generation game |
| 161s | engine pushing the boundaries of large |
| 164s | scale MMO |
| 167s | technology to create entire |
| 179s | universes forged over two decades carbon |
| 183s | Powers the eve |
| 186s | Universe creating Worlds Without |
| 197s | Limits all you need to |
| 200s | do is lose yourself in |
| 207s | them we are proud to present carbon Evon |
| 211s | Line's Next Generation game engine |
| 213s | building on our 20year Legacy of |
| 217s | breathtaking visuals carbon represents |
| 220s | our vision of a modern game engine |
| 223s | supporting the latest and rendering |
| 224s | techniques and features within an |
| 227s | efficient and flexible |
| 229s | framework this effort extends to our |
| 232s | tools and our pipelines so that we can |
| 234s | increase our developer velocity which |
| 237s | means we can get features to you quicker |
| 239s | than ever |
| 241s | carbon will release with upscaling the |
| 243s | best suits your choice of GPU |
| 247s | vendor so going off of our direct X12 |
| 250s | and metal work we will be supporting rri |
| 252s | shadows and |
| 254s | Reflections to complete that visual |
| 257s | experience we will be adding support for |
| 259s | 3D |
| 261s | audio and to ensure that the engine |
| 263s | performs better than ever we will have |
| 265s | an allnew GP driven pipeline which you |
| 268s | will hear more about later in this |
| 269s | presentation |
| 274s | releasing early next year carbon enables |
| 277s | new visual experiences for decades to |
| 280s | come ensuring Eve continues to be the |
| 283s | ultimate science fiction experience and |
| 286s | now I would like to hand over to our |
| 288s | technical director CCB tuxford to Deep |
| 290s | dive into the technology that drives all |
| 292s | of these changes thank |
| 299s | you |
| 306s | thank you Vortex uh I'm csb tuxford and |
| 309s | I'm technical director of platform like |
| 312s | Vex I've been here at csb for many years |
| 315s | uh in various role including game |
| 317s | designer software engineer and technical |
| 319s | director of |
| 323s | Eve and I'm here to talk about our |
| 325s | Basics engine |
| 328s | destiny |
| 334s | so that is all messed up sorry about |
| 337s | that |
| 350s | just right there we are little technical |
| 353s | difficulties so what is a physics engine |
| 356s | so a physics engine is a software |
| 358s | component that simulates physical |
| 359s | Interac action in a virtual environment |
| 362s | it calculates how objects move Collide |
| 364s | and react to |
| 366s | forces in Eve we have a proprietary |
| 368s | physics engine called Destiny it's |
| 370s | simple and efficient and it's a key |
| 373s | factor that enables us to have these |
| 374s | large Fleet |
| 378s | fights but how does it actually |
| 381s | work the server runs a simulation for |
| 384s | the entire solar system but the client |
| 386s | runs one as well for all items in local |
| 389s | proximity |
| 390s | basically anything you can see the |
| 392s | client is simulating as |
| 395s | well the client will send instructions |
| 397s | to the server instructions such as |
| 401s | orbit approach or a line basically |
| 404s | anything you can see in the UI all the |
| 406s | interactions you can do the client sends |
| 408s | that to the server the server will send |
| 410s | that straight back to the all the |
| 412s | clients that need to know about |
| 417s | it and then everybody will progress to |
| 419s | the the next frame now technically the |
| 421s | server doesn't really wait for the |
| 423s | clients to to progress to the next frame |
| 425s | but they all kind of do it in |
| 427s | sync this implies a few things and the |
| 431s | first thing is we only communicate |
| 434s | changes you will get the initial state |
| 436s | of the of the universe but after that |
| 439s | you only get some changes this means |
| 441s | that uh if there are no changes in the |
| 444s | simulation there's no actual network |
| 446s | activity going |
| 448s | on and this is because Destiny is |
| 451s | deterministic um you don't know what |
| 453s | determinism is it basically means given |
| 455s | the same input it will yield the exactly |
| 458s | the same results and because we're |
| 460s | working on diffs the calculations really |
| 463s | have to be |
| 464s | deterministic because any discrepancy |
| 466s | May for example cause collisions on some |
| 469s | simulations but not the others and those |
| 472s | discrepancies will only grow as time |
| 476s | passes you may have already experienced |
| 479s | this as d |
| 480s | sometimes it's because the client is |
| 482s | actually missing some messages but |
| 484s | sometimes it's discrepancy in the |
| 485s | calculations now we haven't had one in a |
| 488s | while now |
| 490s | but you you you may have already |
| 493s | experienced the ramifications of |
| 495s | it so the third thing is the client |
| 499s | needs to know pretty much everything |
| 501s | that it can see because in order to |
| 503s | simulate the state we need to know |
| 505s | everything so I have a small example of |
| 509s | this let's say we're simulating this |
| 511s | raven we just want to simulate where |
| 512s | this raven is going in order to do that |
| 514s | we need to know what its original |
| 516s | position is its velocity and what it's |
| 518s | actually |
| 519s | doing in this case it is approaching |
| 522s | that Tempest so we need to know its |
| 525s | state it's position velocity what it's |
| 528s | doing it's orbiting the Tempest and so |
| 530s | on and so forth we also need to know |
| 532s | about anything these ships can collide |
| 537s | with collisions are are an important |
| 540s | part of any physics simulations |
| 542s | collisions in Destiny are pretty simple |
| 544s | but more importantly they're really easy |
| 546s | to calculate and this is because the |
| 548s | geometry of the objects is pretty simple |
| 552s | all moving items are simulated as SES |
| 555s | and finding collisions between two |
| 556s | spheres is really really simple you just |
| 558s | compare the distance between the centers |
| 561s | distance between the centers to the |
| 563s | radius of the SP and this is kind of |
| 566s | fine for smaller |
| 568s | ships for state and larger structures |
| 571s | it's very noticeable that they aren't s |
| 573s | especially when you start maneuvering |
| 575s | your smaller ships around them we have a |
| 577s | pretty innovative solutions to that we |
| 579s | just add more SS to |
| 582s | it and since the structure doesn't move |
| 585s | the calculations are pretty |
| 587s | simple some shapes are also pretty big |
| 590s | and they don't have spherical |
| 592s | geometry they don't they have spherical |
| 595s | GE geometry but they don't look remotely |
| 598s | spherical regardless of what we try to |
| 600s | tell the |
| 602s | artists |
| 604s | so where are we now destiny has served |
| 606s | us for 20 years and we will continue to |
| 609s | develop it and work on it so in order to |
| 611s | prepare it for the third decade of EV on |
| 614s | line I'm going to tell you about some of |
| 616s | the modernizations efforts we've been |
| 620s | doing the tick is really where all the |
| 622s | magic happens in the |
| 624s | simulation we have an evolve step that's |
| 627s | where we move all the objects around and |
| 628s | we solve all the |
| 631s | collisions then we have everything that |
| 632s | happens before the tick and everything |
| 635s | that happens after the tick this is |
| 636s | pretty much what programming is about |
| 637s | it's like three lines of |
| 640s | code do the thing before the thing do |
| 643s | the thing do the thing after the thing |
| 644s | pretty |
| 646s | simple these are really poorly named |
| 649s | thing do preti do potic what really |
| 651s | happens there in the in the preck is We |
| 655s | Gather up all the messages we need to |
| 657s | send to the client all these approaches |
| 659s | or bits and a lines that the clients |
| 661s | told us they wanted to do we got them |
| 663s | sent them to the relevant clients so |
| 666s | let's take a look at what happens in the |
| 669s | preck |
| 671s | so it gets quite complex uh and it's a |
| 674s | lot more complex than what you would |
| 676s | reason reasonably expect for what it's |
| 679s | doing so what you're seeing here is |
| 682s | around 20 years of gameplay logic |
| 684s | getting intermixed with the physics |
| 686s | logic this goes on there was a limited |
| 688s | amount of time I wanted to spend in |
| 690s | PowerPoint making this animation but |
| 692s | it's like thousand lines |
| 700s | long |
| 704s | uh why is this bad so this is bad |
| 707s | because whenever you make changes to |
| 709s | this to this physics simulation code it |
| 711s | can cause unexpected failures in |
| 713s | different system and this will cause you |
| 715s | to lose confidence in making changes and |
| 719s | pretty soon you don't want to make any |
| 720s | changes at all and if you are making |
| 722s | changes you definitely do not want to |
| 724s | make any big |
| 725s | changes this drives poor cold because |
| 728s | you shy away from making the structural |
| 730s | changes to the Cod that are |
| 732s | required and then this cause unexpected |
| 735s | failures again so how did it really get |
| 738s | like that how did we make it like this U |
| 741s | reality is physics engine has been |
| 743s | staple there hasn't been a grown need to |
| 745s | make any changes to it so the changes |
| 747s | being made have been catered to other |
| 749s | systems and that allowed it to grow |
| 751s | around it this is really common in |
| 754s | software it's just part of the enthropy |
| 756s | of being a software |
| 757s | developer uh and it's not particularly |
| 760s | bad it just means that before we start |
| 762s | making any bigger changes to the |
| 764s | engine we need to do some maintenance |
| 768s | work as a starting point we've separated |
| 770s | the gameplay logic from the logic that |
| 773s | prepares and send the messages to kind |
| 775s | of encapsulate Destiny and make it uh |
| 778s | detangled from the rest of the |
| 781s | code and by simplifying and decoupling |
| 784s | our physics engine for gameplay we've |
| 786s | opened the door to creating targeted |
| 788s | benchmarks and this allows us to |
| 790s | precisely measure and enables us to |
| 793s | improve Destiny's |
| 796s | performance because if you can measure |
| 798s | it you can improve |
| 800s | it so all of these changes were made to |
| 803s | give us confidence the confidence to |
| 805s | change Destiny but what kind of change |
| 808s | are we really talking about |
| 812s | Destiny is efficient but there's still |
| 814s | plenty of room for improvement one of |
| 816s | the most resource intensive code in a |
| 818s | cluster in a fleet fight is sending the |
| 822s | destiny updates to the |
| 824s | clients let's take a look at what |
| 825s | happens there so Destiny runs on both |
| 828s | server and |
| 830s | client there's a server server site |
| 832s | called |
| 833s | beyon it gathers the |
| 835s | instruction uh all these approaches |
| 837s | orbits Alliance and everything think |
| 839s | explosions that happen and sends them to |
| 842s | a client side service called |
| 844s | Michelle these are definitely Destiny |
| 847s | Child's references by the |
| 849s | way and it gets those to the client side |
| 854s | destiny the destiny code is all written |
| 856s | in C++ but the services Beyond CN Mell |
| 859s | are written in |
| 861s | Python and if you know nothing about uh |
| 864s | programming then C process is really |
| 866s | really fast python is kind of slow there |
| 868s | are top people working on that but |
| 870s | that's the reality of the situation uh |
| 873s | the logic for determining which uh this |
| 876s | means that the logic for determining |
| 878s | what messages to send to which client |
| 880s | along with the marshaling unmarshaling |
| 882s | and the data transmission is all handled |
| 884s | in the slower python layer and if you're |
| 887s | unsure about what marshalling and |
| 888s | unmarshalling is you can imagine |
| 890s | marshalling as boarding a plane dead |
| 892s | transmission as the flight and un Maring |
| 894s | as the |
| 898s | onboarding |
| 899s | by separating the gameplay logic from |
| 901s | the physics logic we're in a pretty good |
| 903s | position to send all the destiny packets |
| 905s | within the C++ layer this because all |
| 908s | the information we have is really inside |
| 913s | Destiny it's difficult to say what the |
| 915s | net effect on performance is because |
| 917s | there's a lot of other things happening |
| 918s | other than physics simulation but for |
| 920s | all the code we've seen the efficiency |
| 922s | of the data transmission go up at two |
| 923s | orders of |
| 926s | magnitude the other side of performance |
| 928s | is |
| 930s | and by that I mean how realistic the |
| 931s | simulation |
| 933s | feels it's clear to us that representing |
| 935s | ships of sere does not fit all |
| 939s | scenarios we use different shapes and |
| 941s | even multiple shapes to represent static |
| 943s | objects could we not use those for ships |
| 945s | as |
| 945s | well we could but at what cost to |
| 949s | Performance with investment to |
| 951s | modernization such as Benchmark and more |
| 954s | test automation we can know exactly what |
| 957s | that cost is |
| 962s | so for 20 years Destiny has been a |
| 964s | reliable Foundation when it comes to |
| 965s | physics a recent code modernization set |
| 968s | us up to boost performance and Fidelity |
| 971s | but what about other kind of more |
| 973s | gamechanging improvements now quick |
| 975s | caveat before I say that the upcoming |
| 977s | feature is purely |
| 978s | hypothetical it would need design |
| 980s | validation which has not happened at all |
| 983s | uh think a bit more as what could be not |
| 985s | what will be imagine being trapped in as |
| 989s | field surrounded by hostile ships out |
| 991s | Gone But Not |
| 993s | outsmarted you spot a nearby cluster of |
| 995s | large |
| 1004s | asteroids you T towards toward it |
| 1008s | evading incoming fire As you move behind |
| 1010s | the asteroid the hostes no longer have a |
| 1012s | clear line of side which caused them to |
| 1014s | lose their target lock on |
| 1017s | you seizing the chance you line to a |
| 1020s | nearby Planet initiate warp just as the |
| 1024s | enemies regain the luck you hear the |
| 1025s | sweet words warp drive |
| 1030s | active so now that you're up to speed a |
| 1033s | little bit on the server side I would |
| 1035s | like to introduce U Mario who will talk |
| 1038s | about some of the updates to the |
| 1047s | client |
| 1052s | all |
| 1053s | right um hello everyone uh my name is |
| 1056s | Daniel also not truly known as CCP maku |
| 1058s | just yet uh this is actually my first |
| 1061s | FanFest |
| 1062s | um thank |
| 1065s | you um I am a rendering engineer on the |
| 1068s | engine team uh and I'm here to you today |
| 1070s | to present a little bit about some of |
| 1072s | the rendering work that we have been |
| 1073s | doing and uh to start it off I want to |
| 1075s | introduce you guys to Trinity uh e on |
| 1078s | line is a very unique game not just for |
| 1080s | you players but also for for us |
| 1082s | Engineers working on the rendering uh we |
| 1084s | have a very large world uh we have a |
| 1086s | very high view distance in this world |
| 1087s | and there's a large number of things to |
| 1089s | render uh it's also massively |
| 1091s | multiplayer online which means that we P |
| 1093s | control about what you the players do in |
| 1095s | the world so we need to be prepared for |
| 1097s | anything uh Trinity is the the graphics |
| 1099s | engine inside of carbon and it's our |
| 1101s | solution to |
| 1103s | Graphics um so some of the visual |
| 1106s | improvements that we've been working on |
| 1107s | lately um |
| 1109s | I'm sure you're familiar with the |
| 1110s | improvements to Shadows uh we now have |
| 1113s | an unlimited number of uh Shadow casters |
| 1115s | and Shadow receivers and and the Shadows |
| 1118s | overall are just much much sharper than |
| 1119s | they used to be uh these Shadows Shadow |
| 1122s | changes go hand inand together with the |
| 1124s | the new volumetric Cloud effects that we |
| 1125s | have uh which are obviously very pre |
| 1128s | prevalent in the in the new area that |
| 1129s | just unlocked |
| 1131s | yesterday um it's a it's a really really |
| 1133s | beautiful effect and I would love to |
| 1135s | talk a little bit more about it uh |
| 1137s | unfortunately don't have time for |
| 1138s | everything |
| 1139s | um so I will move on to probably why |
| 1142s | you're here which is the new |
| 1143s | stuff um rate tracing um so what is r |
| 1148s | tracing well it's the ability it's a |
| 1149s | hardware feature of your graphics card |
| 1151s | that allows you to shoot uh race |
| 1153s | arbitrarily into scene and see see what |
| 1155s | they hit uh we're working on support for |
| 1158s | both uh Windows and |
| 1159s | Mac um and the cool thing about rate |
| 1162s | racing is that it enables us to do uh a |
| 1164s | bunch of new techniques that we simply |
| 1165s | weren't able to do before uh this |
| 1168s | includes stuff like improving the |
| 1169s | reflections uh and improving the Shadows |
| 1171s | even more uh as well as hopefully |
| 1173s | improving the lighting overall in the |
| 1175s | future um the cool thing about even |
| 1179s | online is that it well space is fairly |
| 1181s | empty so uh you can count on rate |
| 1184s | tracing not having the same kind of |
| 1186s | performance hit that you're used to |
| 1187s | seeing when enabling it in other |
| 1189s | games so looking a little bit at the |
| 1192s | rate Trace Shadows well we already have |
| 1195s | shadows of course uh but rate tracing |
| 1197s | allows us to simil imate the size of the |
| 1199s | light source in a in a much more |
| 1201s | accurate way um if we add a shadowcaster |
| 1205s | to this little scene here you can see |
| 1206s | that only some of the Rays will be |
| 1207s | blocked and the actual shadow that you |
| 1210s | see from this very large light source |
| 1211s | will be very uh soft and and |
| 1214s | diffuse uh and and um yeah this is this |
| 1218s | uh this is fully realistic and fully |
| 1220s | Dynamic uh which means that if you have |
| 1222s | a for example an antenna sticking out of |
| 1224s | your ship uh it will cast a perfectly |
| 1226s | sharp Shadow onto your own ship while |
| 1228s | the uh distance space station uh will |
| 1231s | cast this very very soft Shadow on your |
| 1233s | ship and these two kinds of shadows will |
| 1235s | seamlessly |
| 1236s | mix um we have a prototype working as |
| 1239s | you can see uh but it's still very much |
| 1241s | work in |
| 1242s | progress uh we also really want to want |
| 1244s | to improve the reflections we of course |
| 1246s | already have support for limited |
| 1248s | Reflections uh but rate tracing will |
| 1250s | truly help us take them to the to the |
| 1252s | next level uh you can imagine the the |
| 1254s | metal parts of your ships uh reflecting |
| 1256s | the dynamic environment around you in a |
| 1258s | in a way you haven't been able to see |
| 1260s | before um as well as or or for example |
| 1263s | uh seeing a perfect mirror-like |
| 1265s | reflection of your ship as you pass by a |
| 1267s | frozen asteroid no prototype here just |
| 1270s | yet but it's something we're very |
| 1271s | excited |
| 1272s | about um but of course graphical |
| 1274s | improvements aren't everything uh we |
| 1277s | also need to make sure that it all |
| 1278s | performs well uh we do always try to |
| 1280s | optimize our code and our shaders that |
| 1282s | like that uh but there is an upper limit |
| 1284s | to how far you can go with just minor |
| 1285s | code changes uh sometimes we need to |
| 1288s | take your completely different approach |
| 1289s | to uh to uh to rendering something to |
| 1292s | make sure that we maintain the |
| 1294s | performance and one of the biggest |
| 1295s | challenges there is just pixels it's |
| 1298s | just the sheer number of pixels that we |
| 1299s | need to render uh the number of pixels |
| 1302s | is of course determined by the |
| 1303s | resolution of your monitor uh 1080p |
| 1305s | monitor is about 2 million pixels 4K |
| 1307s | bumps that up to 8 million and that's |
| 1309s | the number of pixels we need to render |
| 1311s | every single |
| 1312s | frame um of course the cost of each |
| 1315s | pixel will also keep going up we are |
| 1317s | always adding new shading improvements |
| 1318s | such as the Shadows uh we're adding new |
| 1320s | special effects and post-processing |
| 1322s | effects to make sure that the game looks |
| 1323s | as good as it ever can um and on top of |
| 1326s | that we want to add rate tracing and all |
| 1327s | of these increase the amount of time we |
| 1329s | need to spend on each pixel uh and we |
| 1332s | can't forget about anti-aliasing uh |
| 1335s | anti-aliasing usually comes in in in the |
| 1337s | form of some kind of resolution increase |
| 1340s | uh again increasing the number the |
| 1341s | effective number of pixels we need to |
| 1342s | render uh so it kind of multiplies the |
| 1345s | cost yet again and obviously solution |
| 1348s | isn't get everyone to get a really |
| 1349s | expensive GPU would be nice but that's |
| 1352s | not the reality we live in um but then |
| 1355s | looking into anti-aliasing specifically |
| 1357s | well what is aliasing and why are we |
| 1359s | trying to get rid of it well uh aliasing |
| 1362s | is essentially an artifact of the way |
| 1364s | that uh gpus render uh triangles and |
| 1367s | things in general uh you basically have |
| 1370s | this binary test for each pixel that |
| 1372s | checks if the if the center of the pixel |
| 1373s | is covered by the triangle and if it is |
| 1376s | it's completely filled in and the end |
| 1377s | result is the image to the left here |
| 1379s | where uh you can see these ugly Jagged |
| 1382s | edges this kind of staircase pattern |
| 1384s | it's very typical of aling and uh |
| 1387s | anti-aliasing is essentially any kind of |
| 1389s | method that attempts to generate the |
| 1391s | image to the right which is something |
| 1393s | that just represents the actual physical |
| 1395s | shape of this triangle much more |
| 1397s | accurately but aing doesn't just happen |
| 1400s | at the edge of triangles uh it also |
| 1401s | happens it can also happen in the middle |
| 1403s | of triangles uh such as in the case of |
| 1405s | these specular highlights uh so what's |
| 1408s | happening here is that the the |
| 1409s | highlights on this hole here is uh they |
| 1412s | are very very thin and therefore it's |
| 1414s | essentially completely random if just a |
| 1416s | row of pixel centers happens to kind of |
| 1419s | randomly end up on this reflection and |
| 1421s | when they do they pop into existence and |
| 1423s | then briefly disappear again uh whenever |
| 1425s | the camera moves uh so it's essentially |
| 1428s | a problem of details that are smaller |
| 1429s | than a pixel that we can't capture with |
| 1431s | this uh one of the most naive and |
| 1434s | simplest uh approaches to anti-aliasing |
| 1437s | is super sampling it's a very simple |
| 1439s | idea uh instead of doing a single sample |
| 1442s | at the center of each pixel we simply do |
| 1444s | multiple samples uh it's effectively the |
| 1446s | same as just rendering at a higher |
| 1448s | resolution then at the end uh we kind of |
| 1450s | just average together all of the samples |
| 1452s | inside of a pixel which again |
| 1454s | effectively the same as of scaling the |
| 1455s | image down again um this is really |
| 1458s | really effective uh but it also is |
| 1460s | really expensive uh if we're doing four |
| 1462s | samples for each pixel for example |
| 1464s | that's four times the amount of shading |
| 1465s | work and lighting all that good all that |
| 1467s | good stuff um top of that we need four |
| 1469s | times the amount of memory to store all |
| 1470s | these samples and we also need four |
| 1472s | times the amount of bandwidth to read |
| 1473s | and write these samples so 1080p |
| 1476s | effectively becomes 4K 4K effectively |
| 1478s | becomes 8K everything just becomes so |
| 1480s | much worse but it is really really |
| 1484s | effective uh you can see the comparison |
| 1486s | here uh to the left we have no anti- |
| 1488s | lasing and to the right we have four |
| 1489s | time super sampling as you can see all |
| 1491s | of this ugly shimmering and like all |
| 1493s | this distracting shimmering is |
| 1494s | completely gone it looks absolutely |
| 1497s | amazing unfortunately because it's so |
| 1499s | expensive uh Eve has so far been forced |
| 1501s | to rely on something called multi |
| 1503s | sampling so multi sampling is uh an |
| 1506s | optimization of super sampling that |
| 1507s | essentially only tries to Super sample |
| 1509s | the triangle |
| 1510s | edges um so we're still only shading |
| 1514s | once per pixel uh but we actually choose |
| 1516s | which samples to actually write to uh we |
| 1519s | write the same color to all of the |
| 1520s | samples that are covered by the triangle |
| 1522s | you still need to pay for the memory and |
| 1524s | the bandwidth cost uh but the shading |
| 1526s | shading cost increase is ENT essentially |
| 1528s | negligible and that's where the the bulk |
| 1530s | of the cost comes the big drawback of |
| 1533s | this is of course there's no anti- eling |
| 1535s | for the shading in that case so if we |
| 1537s | again look at this shimmering image here |
| 1540s | the only thing that multi sampling |
| 1541s | really fixes is that top Jagged Edge |
| 1544s | like that staircase pattern on the edge |
| 1545s | of that that uh the H there you're not |
| 1548s | going to see any any kind of impact at |
| 1550s | all on the shimmering inside of |
| 1553s | triangles it would be really really nice |
| 1555s | if we could afford super sampling but we |
| 1557s | can't |
| 1559s | but what if we just we do we still do |
| 1560s | one sample per pixel but we kind of |
| 1563s | start moving it around inside of the |
| 1565s | inside of the pixel so we do a different |
| 1567s | Center every frame and then we kind of |
| 1569s | just add them all up at the end well |
| 1572s | that's the the foundational idea behind |
| 1574s | behind temporal |
| 1576s | anti-aliasing the idea is basically that |
| 1577s | you spread these samples out over time |
| 1580s | uh you do one sample per pixel but a |
| 1582s | different sample every frame uh the idea |
| 1585s | then is to blend it all together really |
| 1586s | carefully to get the effect the result |
| 1588s | of four time super sampling uh it's |
| 1591s | really effective and it's actually even |
| 1593s | faster than four times super sampling uh |
| 1596s | sorry four times multi sampling simply |
| 1598s | because we only need one sample per |
| 1599s | pixel just like when we have no |
| 1601s | anti-aliasing at all uh but you do need |
| 1603s | to be really really careful when doing |
| 1605s | this uh if you end up blending uh when |
| 1607s | you're not supposed to you can get |
| 1609s | blurriness you can get ghosting or you |
| 1611s | can see parts of the previous frame |
| 1613s | looks very terrible uh and if you and in |
| 1616s | fact if you blend too little you also |
| 1617s | get get problems you can't just disable |
| 1619s | the blending completely because then you |
| 1621s | will suddenly start to be able to see |
| 1623s | this kind of jittering motion around as |
| 1625s | you you switch you move where the pixel |
| 1626s | Center |
| 1627s | occurs and uh to take an actual visual |
| 1630s | example of this here's a very sumed |
| 1632s | image of an anger hanger you can see |
| 1634s | that it's very pixelated and you can |
| 1636s | also see that it has this these ugly |
| 1637s | staircase patterns everywhere uh this |
| 1640s | looks even worse in motion where you |
| 1641s | will have a lot of shimmering effects uh |
| 1644s | as this moves so when we apply temporal |
| 1647s | anti-al we will first start to uh uh |
| 1650s | sharpen the image as much as we can and |
| 1653s | then start kind of moving it around a |
| 1654s | little |
| 1655s | bit um and all of these images are |
| 1659s | slightly different so when we add them |
| 1660s | all together we end up with this really |
| 1663s | really nice and smooth result we still |
| 1665s | have all of the sharpness of the texture |
| 1667s | detail all of that good stuff but we |
| 1669s | have managed to eliminate the uh the uh |
| 1671s | staircase patterns and and on top of |
| 1674s | that it's much much more stable under |
| 1675s | motion looks very very good |
| 1678s | um but can we actually take this |
| 1680s | approach even further and it turns out |
| 1682s | that we can it's called temporal |
| 1684s | upscaling um the idea is to start at an |
| 1687s | even lower resolution we don't actually |
| 1689s | start at the full resolution uh you you |
| 1691s | can imagine moving this single sample |
| 1693s | that we have around in a in a 2X two |
| 1695s | area instead so over four pixels instead |
| 1698s | uh we then temporarily upscale this |
| 1700s | beyond the full resolution to again get |
| 1702s | this kind of super sampling effect uh |
| 1705s | and then scale it back down to full |
| 1706s | resolution while what your monor has and |
| 1709s | this works it works really well uh but |
| 1711s | it is even more challenging uh any kind |
| 1713s | of issues that you would normally have |
| 1715s | with temporal anti-aliasing uh such as |
| 1717s | uh flickering or uh blurriness or |
| 1720s | ghosting just becomes so much more |
| 1722s | obvious uh when you're doing when you're |
| 1724s | when you're kind of blowing it up to |
| 1725s | twice the size basically fortunately |
| 1728s | this isn't a problem that we really have |
| 1729s | to solve ourselves uh there are numerous |
| 1731s | temporal upscaling or at least only at |
| 1734s | least temporal upscaling functions uh uh |
| 1737s | algorithm that already exist and we are |
| 1740s | planning on supporting pretty much all |
| 1741s | of them so that you can pick the best |
| 1743s | solution for your specific graphics card |
| 1745s | uh we will we will be supporting AMD FSR |
| 1747s | 3 we will be supporting Nvidia dlss 3.5 |
| 1751s | Intel Intel's xcss as well as metal FX |
| 1754s | upscaling on Mac so that kind of uh |
| 1758s | makes sure that even as we're adding all |
| 1760s | of these new features even on weer |
| 1762s | graphics cards you will you will be able |
| 1763s | to enable these new goodies with a |
| 1765s | minimal loss in in image quality from |
| 1767s | the upsale |
| 1769s | but uh your GPU isn't the only thing |
| 1771s | that's kind of struggling um what about |
| 1774s | the CPU well um the the the GPU and the |
| 1778s | CPU have a very interesting |
| 1780s | relationship uh they're essentially two |
| 1783s | completely separate processors that |
| 1785s | coexist inside of your computer and it |
| 1787s | kind of works the way that the the CPU |
| 1789s | sends commands to the GPU and tells us |
| 1791s | what to do and in some cases we can |
| 1793s | actually end up with a case where the |
| 1795s | CPU simply doesn't have the time to |
| 1796s | generate enough commands for for the GPU |
| 1798s | and the GPU will just sit idle and not |
| 1801s | have anything to do and that's where our |
| 1803s | GPU driven pipeline project comes into |
| 1806s | play uh it's quite a quite a bit of a |
| 1809s | complex topic so we will need to take a |
| 1811s | deep dive into into how this |
| 1812s | relationship between the CPU and the GPU |
| 1814s | work so to understand this let's have a |
| 1817s | look at the Timeline where we can see |
| 1818s | what the CPU and the GPU are up to at |
| 1820s | any given time when rendering a frame so |
| 1823s | the CPU is of course responsible for a |
| 1824s | lot of different things it's not just |
| 1826s | responsible for graphics |
| 1828s | uh things like physics uh networking uh |
| 1831s | Network code uh as well as the user |
| 1834s | interface all this kind of good stuff in |
| 1835s | a lot of other things this is a bit |
| 1838s | outside of my area I work with the |
| 1839s | graphics so I'm just going to group all |
| 1841s | of these up into this big block and call |
| 1842s | it |
| 1843s | gameplay uh but at some point the carbon |
| 1846s | engine uh will have processed all of the |
| 1848s | other subsystems and it's now time to uh |
| 1851s | start rendering something and at that |
| 1853s | point it will gather up a bunch of |
| 1854s | commands and it will kind of send them |
| 1856s | off to the GPU |
| 1858s | and the GPU will then start processing |
| 1860s | this and rendering that frame but the |
| 1862s | CPU doesn't just sit idle as it waits |
| 1864s | for this to to complete no it will |
| 1866s | immediately move on to the next frame uh |
| 1868s | and the frame after that once once |
| 1870s | that's done and in this case uh this |
| 1872s | works really really well uh the CPU even |
| 1875s | has a little bit of breathing time and |
| 1877s | the GPU is is taxed to its full capacity |
| 1879s | it's working 100% of the time but there |
| 1882s | are cases in Eve uh quite a few cases |
| 1885s | where this is not the case where this |
| 1887s | submitting of the Gathering up all of |
| 1889s | these commands and submitting them to |
| 1891s | the GPU actually takes longer than it |
| 1893s | takes for the GPU to render them uh and |
| 1896s | this can happen simply because of |
| 1898s | massive Fleet battles or a huge amount |
| 1900s | of detail and it also depends on your |
| 1902s | your your specific Hardware like the |
| 1904s | relative power of your CPU and the |
| 1907s | GPU um and in this case well we're |
| 1909s | wasting a lot of time on the GPU uh the |
| 1911s | GPU is essentially sitting idle for a |
| 1913s | very large amount of time uh and we |
| 1915s | could in theory push the GPU to render |
| 1918s | maybe maybe for maybe twice as many |
| 1920s | frames per second in this |
| 1921s | case um but to understand why this |
| 1924s | happens we need to look into what's what |
| 1926s | actually happens in this submit block to |
| 1928s | make it make it so expensive uh and to |
| 1930s | do that we need to figure out how we're |
| 1932s | rendering a single specific like a |
| 1934s | single Eve ship like a single ship in |
| 1936s | the in the universe of Eve so I present |
| 1938s | to you an eve ship I promise this is |
| 1942s | definitely a ship uh like a like a real |
| 1945s | Eve ship it has a mesh which is |
| 1946s | determined by the ship shape here it |
| 1948s | also has a Shader which is determined by |
| 1950s | the color here and it also has a |
| 1951s | material which is represented by the the |
| 1953s | texture applied to this uh so to render |
| 1956s | the ship uh we actually need to submit |
| 1958s | four separate commands to the GPU we |
| 1961s | need to set which color we want to |
| 1962s | render it with we need to set set the |
| 1964s | shape that we want to render and then we |
| 1965s | need to set the texture and once those |
| 1967s | are done we can submit what's called a |
| 1969s | draw call which is essentially a command |
| 1972s | that says yeah Draw Something using the |
| 1974s | previously said |
| 1975s | stuff uh and if we only we were only |
| 1978s | rendering one ship uh this would not be |
| 1980s | a problem whatsoever this is just four |
| 1982s | commands problem is Eve has a lot of |
| 1984s | ships some might even say dozens of |
| 1986s | times um so we need to we need to really |
| 1989s | be able to render this many many things |
| 1992s | and well the first thing we're going to |
| 1993s | do we're program I'm a program at least |
| 1995s | I love lists I'm going to move them all |
| 1997s | into a list uh this will make it much |
| 1999s | easier to work with and once they're all |
| 2001s | in the list well it isn't that difficult |
| 2003s | to just kind of do a naive Loop over |
| 2005s | this and render them we'll just go |
| 2006s | through all of the ships that we to |
| 2007s | render we'll set the color we'll set the |
| 2009s | shape we'll set the texture and issue a |
| 2011s | draw call uh there's actually 100 uh |
| 2014s | ships in this in this image or in the |
| 2016s | scene so uh we will end up with a total |
| 2018s | total of four four four times 100 |
| 2020s | commands which is 400 very simple um |
| 2024s | however this is not the most efficient |
| 2026s | way of doing this we are actually |
| 2027s | switching between the same colors |
| 2029s | multiple times for example so we can |
| 2031s | actually eliminate a bunch of commands |
| 2033s | uh by kind of just reorganizing this |
| 2034s | data if we move this down here and then |
| 2037s | we make new lists we make one list for |
| 2039s | each color and then we kind of just take |
| 2041s | all of these ships we go through them |
| 2043s | and we sort them into their Rel into |
| 2045s | into the color we sort them based on the |
| 2047s | color that they have and put them in the |
| 2048s | corresponding list well it turns out |
| 2050s | that we can render these lists much much |
| 2052s | more efficiently now we no longer need |
| 2054s | to switch the color for every ship we |
| 2055s | can just set the color once and then |
| 2057s | render all of the ships that use that |
| 2059s | color and the end result is that we |
| 2061s | eliminate almost 25% of our commands and |
| 2064s | we can actually take this one step |
| 2065s | further uh we can start sorting Things |
| 2066s | based on shape they're they're using as |
| 2068s | well um and that that will allow us to |
| 2071s | to eliminate even more draw calls uh in |
| 2073s | in fact we can we we no longer need to |
| 2075s | change the shape as often now so we've |
| 2077s | even though we're rendering the exact |
| 2078s | same pixels on the screen the exact same |
| 2080s | triangles we are now doing it with |
| 2082s | almost uh like by we we sorry we are now |
| 2086s | doing it with with almost half the the |
| 2088s | the draw calls of what we were doing |
| 2090s | before uh and this process of kind of |
| 2092s | just reorganizing the data to to render |
| 2095s | it more efficiently is called batching |
| 2097s | and it's it's a very important thing so |
| 2099s | yeah uh problem solved right let's see |
| 2102s | what what happens with our timeline well |
| 2104s | our submit block is not really really |
| 2106s | tiny the problem is that now we have a |
| 2108s | baching block instead and that it turns |
| 2111s | out that just reshuffling all of this |
| 2112s | data and taking care of it is really |
| 2114s | expensive as |
| 2115s | well what if we just kind of take this |
| 2118s | original list of ships I we just ye it |
| 2121s | directly at the GPU with no regards for |
| 2122s | its safety we we just don't care and we |
| 2125s | kind of just let it figure it all out |
| 2126s | for itself |
| 2128s | that is what the GPU driven pipeline is |
| 2130s | all about uh so it turns out that gpus |
| 2133s | nowadays are really really flexible they |
| 2135s | can actually generate commands for |
| 2137s | themselves uh they don't need a CPU to |
| 2139s | tell them what to do anymore so we can |
| 2142s | actually have the GPU do dispatching for |
| 2144s | us and the best part of this well one of |
| 2146s | the good parts at least is that it's |
| 2148s | even faster than the CPU at doing it um |
| 2151s | the actual best part is that there's |
| 2153s | minimal CPU involvement uh the CPU is |
| 2155s | now free to do other things while the |
| 2156s | GPU keeps itself busy so if we look at |
| 2160s | our rendering Loop that we had before we |
| 2161s | can kind of eliminate everything inside |
| 2163s | of that Loop there and basically is set |
| 2166s | the colar and then tell the GPU to do |
| 2168s | its thing and we eliminate more than 90% |
| 2170s | of the the CPU side draw calls and |
| 2172s | commands that need to be sent there |
| 2174s | while the GPU still gets to execute all |
| 2176s | of those 200 plus draw calls that we had |
| 2179s | before but it's fine because the GPU was |
| 2181s | already not being taxed at full capacity |
| 2183s | so that that's all good and what happens |
| 2186s | to our timeline well we used to have |
| 2187s | this batching block that was on the CPU |
| 2189s | timeline that just magically moves down |
| 2191s | to the to the gpus timeline and in fact |
| 2194s | the batching block is actually |
| 2195s | significantly smaller than this this is |
| 2197s | just the smallest block I could get in |
| 2198s | that could still fit the word |
| 2201s | batching um so what is this ultimately |
| 2203s | mean for Eve well it's a complete |
| 2205s | restructuring of how things are rendered |
| 2207s | inside of Trinity uh it's a lot of work |
| 2210s | there's essentially no Shader that's |
| 2211s | going to come out UNSC UNSC by these |
| 2214s | changes uh it's a lot of work but but |
| 2217s | once it's done we are confident that we |
| 2219s | will be able to have a lot more objects |
| 2221s | on screen and we can really get let our |
| 2223s | artists go crazy with this and also more |
| 2225s | variation inside of those objects we no |
| 2227s | longer need to have copies of the same |
| 2229s | objects to make sure that it works well |
| 2230s | with with batching or something like |
| 2231s | that we can just have more unique |
| 2233s | objects and let the artists the artists |
| 2235s | just get more free hands but the best |
| 2237s | part about this is just that it frees up |
| 2239s | the CPU for game logic it means that the |
| 2242s | entire Graphics Engineers kind of moves |
| 2244s | out of the way of all the other systems |
| 2246s | that also required the CPU time which |
| 2248s | will for example uh enable for the new |
| 2251s | physics improvements uh that uh CCP |
| 2253s | tuxford was talking about |
| 2255s | before but then the big question is |
| 2258s | when and well CCP verdicts already |
| 2261s | spoiled this early |
| 2263s | 2024 uh we are we are uh we want to get |
| 2266s | all of these improvements uh including |
| 2268s | um the upscaling uh the rate tracing as |
| 2272s | well as the GPU driven pipeline done in |
| 2274s | early |
| 2275s | 2024 that leads me to my final question |
| 2278s | how do artists actually work with all of |
| 2281s | this new stuff that we're giving them |
| 2283s | and that's where I hand over to CCP |
| 2296s | Amino thank |
| 2302s | you |
| 2306s | hello |
| 2308s | so my name is CCP Amino I'm a tools |
| 2311s | developer with CCP and so our job is to |
| 2315s | try and take the power that carbon |
| 2317s | brings to our game and put it in a |
| 2319s | package that will allow artists and |
| 2321s | designers to leverage that to build |
| 2323s | great experiences for |
| 2324s | you so with the big investment in carbon |
| 2328s | we've also been making a big investment |
| 2329s | in these tools and I would like to |
| 2331s | introduce you to our new very internal |
| 2333s | only tool |
| 2336s | graphite |
| 2338s | so |
| 2339s | graphite is a sort of editor for |
| 2342s | everything you might want to do in the |
| 2343s | engine whether you're making a new ship |
| 2345s | scanning an old one building a whole new |
| 2348s | scene for you guys to explore in new |
| 2351s | Eden or even badly trying to make a |
| 2354s | trailer like I am |
| 2355s | here you can basically use it to |
| 2358s | Leverage The Power of carbon in order to |
| 2361s | build all of the experiences that we |
| 2363s | want to deliver to you in the game and |
| 2365s | we've been working on this for quite a |
| 2366s | while we've actually started from the |
| 2368s | the ground up with the UI here we've |
| 2370s | been trying to modernize our tools |
| 2372s | internally overall try and bring all of |
| 2374s | the capabilities of carbon together |
| 2376s | under one tool set and Empower our |
| 2379s | artists and designers to deliver bigger |
| 2381s | and Bolder |
| 2382s | experiences leveraging all of the new |
| 2385s | benefits with the new |
| 2386s | engine some of the improvements we've |
| 2388s | making are actually just really |
| 2390s | simple in some cases it's just about |
| 2392s | taking a system we already have fitting |
| 2394s | a turret to a ship and shooting another |
| 2396s | one making it explode making that feel a |
| 2398s | little bit more in our tool set like it |
| 2400s | does in the game the important part here |
| 2402s | is having artist being able to use this |
| 2405s | like they're going to like you're going |
| 2407s | to experience it will allow them to |
| 2409s | hopefully build a higher quality version |
| 2411s | of that turret something feels better to |
| 2413s | use is more fun to play with and they'll |
| 2416s | be able to get that experience a lot |
| 2417s | earlier in their development |
| 2420s | cycle in some cases it's about bringing |
| 2422s | tools |
| 2424s | home when it comes to doing things like |
| 2426s | defining Collision geometry we've |
| 2428s | historically leveraged third party tools |
| 2431s | have artists work in there build the |
| 2433s | kind of data that we need for carbon to |
| 2434s | consume in order to do uh the Collision |
| 2437s | work and that's a quite a a long- windy |
| 2441s | pipeline the artists will spend a lot of |
| 2443s | time and in this third party tool be |
| 2444s | Maya or blender try and author this data |
| 2447s | and then they've got to go through the |
| 2448s | process of getting it into the engine to |
| 2450s | see if it works see if it plays nicely |
| 2453s | to see if they've introduced some really |
| 2454s | weird bug where you get stuck in the |
| 2456s | middle of a station |
| 2457s | so by bringing these things all Under |
| 2459s | One Roof and bringing them home we plan |
| 2462s | to uh leverage this to deliver higher |
| 2465s | quality to you in the game |
| 2469s | experience in some cases we're actually |
| 2472s | looking to improve our developers |
| 2473s | experience as |
| 2475s | well many of the places they'll spend a |
| 2478s | lot of time doing things like crafting a |
| 2480s | beautiful scene for you to |
| 2483s | explore they want to visually tell a |
| 2485s | story use their art assets and show them |
| 2487s | off to build big cool things for you to |
| 2489s | discover and enjoy and we want them to |
| 2492s | spend more of their time expressing |
| 2495s | themselves and designing what they want |
| 2497s | to deliver at the end rather than |
| 2499s | spending time fighting against the |
| 2500s | engine and the tools that they have at |
| 2501s | their fingertips so we put a lot of time |
| 2504s | and effort into leveraging some of the |
| 2506s | performance benefits that come with |
| 2508s | carbon to give them a much nicer overall |
| 2510s | experience smoother prettier UI make it |
| 2513s | a little bit more functional make it a |
| 2514s | little bit more adaptive make it feel |
| 2516s | good |
| 2517s | if it's a pleasure for them to work |
| 2520s | with then hopefully they will deliver |
| 2522s | something very pleasurable for you in |
| 2524s | the |
| 2525s | end in some cases we are actually trying |
| 2528s | to make use of and Leverage The Power of |
| 2531s | some third party tools you have seen a |
| 2533s | lot of uh the new Cloud Tech being |
| 2535s | used this technology is uh authored by |
| 2538s | artist in something called Ember |
| 2541s | genen and that tool will spit out data |
| 2543s | that carbon can consume and we can put |
| 2545s | in the game |
| 2547s | before we can actually get it into the |
| 2549s | game there's a lot of other little |
| 2550s | things that go into a little bit of |
| 2551s | overhead and ultimately they end up |
| 2554s | working heavily in isolation inside of a |
| 2556s | tools set outside of our ecosystem |
| 2558s | crafting those |
| 2560s | clouds so what we're trying to do here |
| 2562s | with uh graphite is make sure that they |
| 2564s | have as smooth and as fast a pipeline so |
| 2567s | when they're using these third party |
| 2568s | tools to build big new clouds we are |
| 2571s | very quickly getting them into the |
| 2572s | engine in this case preferably almost |
| 2575s | instantly they make make a change in |
| 2577s | Amber gen and we want them to instantly |
| 2578s | see that in graphite but along with that |
| 2581s | by having all of our tools under one |
| 2583s | place having the tools that they use to |
| 2585s | author scenes this allows them to see |
| 2587s | not only their Cloud very quickly in |
| 2589s | engine and see what that's going to look |
| 2590s | like but see it in the kind of place |
| 2591s | that they're going to use it allowing |
| 2593s | them to craft more bespoke assets that |
| 2596s | fit the scene and better tell the story |
| 2599s | or better describe the experience that |
| 2600s | they want to give |
| 2604s | you in some cases we're leveraging whole |
| 2606s | new |
| 2607s | capabilities so carbon brings a lot of |
| 2610s | benefits in terms of new |
| 2613s | performance um better options and this |
| 2615s | provides new opportunities for designers |
| 2617s | and artists to take advantage of every |
| 2620s | time we save a couple of frames in the |
| 2622s | engine maybe that means explosions can |
| 2624s | be bigger maybe that means scenes can be |
| 2626s | more complex there's less restrictions |
| 2628s | on the work that they |
| 2630s | do and so this is quite an experimental |
| 2633s | feature for us but we're trying to |
| 2635s | Leverage The Power of third party Tools |
| 2637s | in this case blender again with a really |
| 2639s | direct pipeline so the artist can work |
| 2642s | in blender and craft very large assets |
| 2646s | hit one button and see it very quickly |
| 2647s | ported straight into the game try it out |
| 2650s | how does it feel to fly around how does |
| 2651s | it feel to bump into it how does it |
| 2654s | look and by giving them this pipeline |
| 2657s | plan is to allow them again very early |
| 2659s | and very experimental allow them the |
| 2661s | power to build much bigger and grander |
| 2664s | experiences overall and delivered to you |
| 2667s | the best possible experience that they |
| 2669s | can manage to make Eve awesome |
| 2673s | forever thank you very |
| 2684s | much so if you have any questions you |
| 2687s | find many of us around we're all very |
| 2688s | friendly come and say hi have a chat |
| 2690s | we'll happily talk your ear off for |
| 2691s | hours about all of this stuff uh but we |
| 2694s | also have a panel on the second floor so |
| 2696s | if you're in the mood we'd love to see |
| 2698s | you there at 1:30 thank |
| 2705s | you |