1 00:00:00,240 --> 00:00:02,360 Hi, and welcome to the Neil Ashton Podcast. 2 00:00:03,000 --> 00:00:05,880 In each episode, we explain some of the fascinating ways 3 00:00:05,880 --> 00:00:09,080 that science and engineering are changing the world around us. 4 00:00:09,760 --> 00:00:12,720 We talk to leading engineers from elite level sports like 5 00:00:12,800 --> 00:00:16,840 cycling and Formula One to some of the world's top academics to 6 00:00:16,840 --> 00:00:20,445 understand how fluid dynamics, machine learning and 7 00:00:20,445 --> 00:00:22,960 supercomputing are bringing in a new era of discovery. 8 00:00:23,920 --> 00:00:27,040 We also hear some of their life stories, their career advice, 9 00:00:27,640 --> 00:00:30,040 the lessons they've learned on the way that I hope will be 10 00:00:30,040 --> 00:00:33,800 helpful to you too. So sit back and enjoy this 11 00:00:33,800 --> 00:00:40,600 episode. Hi, welcome back to the Neil 12 00:00:40,600 --> 00:00:43,240 Ashton Podcast. I thought I'd do something 13 00:00:43,240 --> 00:00:44,605 interesting today, which I was. 14 00:00:44,613 --> 00:00:48,160 I was looking back at the previous episodes and I saw that 15 00:00:48,160 --> 00:00:53,195 almost a year ago today I did an episode called the Future of 16 00:00:53,195 --> 00:01:00,160 CFD Five key trends to watch. It was season 2 episode 2 and I 17 00:01:00,160 --> 00:01:04,879 thought I will try and do another version of that and 18 00:01:04,887 --> 00:01:09,112 basically see a year on, was I right or was I wrong? 19 00:01:09,120 --> 00:01:16,296 So what I had as the five were the rise of GPUs in CFD, AI and 20 00:01:16,296 --> 00:01:19,450 machine learning, shift to cloud computing, digital 21 00:01:19,458 --> 00:01:24,610 certification, and I think I had the last one which was mergers 22 00:01:24,618 --> 00:01:29,345 and acquisitions. OK, Now this was before I joined 23 00:01:29,345 --> 00:01:31,360 NVIDIA. Oh, I should just say. 24 00:01:31,368 --> 00:01:34,428 So obviously NVIDIA's you know, creates GPUs. 25 00:01:34,436 --> 00:01:41,630 So the rise of GPUs in CFD. Can I answer that while being 26 00:01:41,638 --> 00:01:44,960 completely unbiased? Probably not, but it's true. 27 00:01:44,968 --> 00:01:48,886 Basically, there has been an unbelievable rise in GPUs for 28 00:01:48,886 --> 00:01:49,560 CFD. I. 29 00:01:49,560 --> 00:01:52,040 Can say that even if I wasn't working for NVIDIA, I think 30 00:01:52,040 --> 00:01:54,240 people in the community know that to be the case. 31 00:01:54,960 --> 00:01:56,840 Has that accelerated in the past year? 32 00:01:56,880 --> 00:02:00,640 Yes, most definitely it is. Yeah. 33 00:02:00,720 --> 00:02:03,440 Just the number one question. And again, I know I work for 34 00:02:03,440 --> 00:02:06,040 NVIDIA, so I'm going to say that, but it's true. 35 00:02:06,600 --> 00:02:11,838 I go to conferences and I see it without me even talking to 36 00:02:11,838 --> 00:02:13,586 them. You know, they're showing slides 37 00:02:13,586 --> 00:02:15,902 saying, OK, it's 10 times faster on a GPU. 38 00:02:15,902 --> 00:02:18,000 I did not get involved in that study whatsoever. 39 00:02:18,000 --> 00:02:19,280 I'm just sat there listening to it. 40 00:02:19,760 --> 00:02:22,670 So it it is absolutely the case. 41 00:02:22,678 --> 00:02:26,125 And as anything there was, it was slower at the beginning. 42 00:02:26,133 --> 00:02:27,840 People were like, OK, is this really true? 43 00:02:27,848 --> 00:02:31,544 You know, is this marketing? I think now there's been so many 44 00:02:31,544 --> 00:02:34,094 independent studies showing it that people are fully getting 45 00:02:34,102 --> 00:02:37,190 behind. And I think what's also made the 46 00:02:37,190 --> 00:02:39,454 difference is 1, the hardware's obviously keeps 47 00:02:39,454 --> 00:02:43,076 getting better and better. The pace of GPUs, because 48 00:02:43,084 --> 00:02:46,060 obviously AI has helped. This is far out pacing. 49 00:02:46,068 --> 00:02:48,675 CPUs in terms of like generational generation, you 50 00:02:48,683 --> 00:02:52,386 know, like the improvement that you get each generation is, is 51 00:02:52,394 --> 00:02:55,286 like this compared to CPUs, which is, you know, a bit 52 00:02:55,294 --> 00:02:58,075 slower. So you're, by moving to it, you 53 00:02:58,075 --> 00:03:01,320 keep getting better and better. So the reason to move has 54 00:03:01,328 --> 00:03:06,935 probably accelerated. I think the, there's been a 55 00:03:06,943 --> 00:03:10,304 healthy competition between different CFD companies each 56 00:03:10,312 --> 00:03:14,024 seeing GPUs to be part of that differentiating thing. 57 00:03:14,032 --> 00:03:18,784 And that has created a, a sort of an acceleration, you know, 58 00:03:18,792 --> 00:03:21,335 because there's one vendor releases a GPU version, everyone 59 00:03:21,335 --> 00:03:23,286 wants to make sure they also have one. 60 00:03:23,294 --> 00:03:26,515 So there's been a little bit of that, but for good reasons. 61 00:03:26,523 --> 00:03:29,410 It's not purely, you know, just for the sake of it. 62 00:03:29,418 --> 00:03:31,600 It's because people are genuinely seeing a speed up 63 00:03:31,608 --> 00:03:34,680 customers liking it and and therefore they're doing it. 64 00:03:34,688 --> 00:03:38,732 And I think most interestingly, which is I think something I 65 00:03:38,740 --> 00:03:42,504 commented on is there's a lot of start-ups now who are 66 00:03:42,512 --> 00:03:47,018 specifically focusing entirely on being GPU native. 67 00:03:47,026 --> 00:03:50,670 And that has also definitely created a shift because no 68 00:03:50,678 --> 00:03:55,205 surprise if you write code from scratch with a sort of 69 00:03:55,205 --> 00:03:58,014 software–hardware co-design, you're going to create something 70 00:03:58,014 --> 00:04:01,860 really good. And so because of those 71 00:04:01,868 --> 00:04:07,490 start-ups, they have shown in some ways the like most optimal 72 00:04:07,490 --> 00:04:11,190 solution and has really shown some very interesting 73 00:04:11,198 --> 00:04:14,040 performance. What I find more at a technical 74 00:04:14,040 --> 00:04:17,656 level interesting is where the whole like unstructured implicit 75 00:04:17,656 --> 00:04:22,968 finite volume, which I guess was the industry standard does 76 00:04:22,968 --> 00:04:26,400 that. Is that the most optimum 77 00:04:26,400 --> 00:04:34,840 numerical approach on GPUs? It could be, but because they're 78 00:04:34,840 --> 00:04:38,720 so much more powerful and because of that, you can start 79 00:04:38,720 --> 00:04:42,600 to do more transient high fidelity LES, then people are 80 00:04:42,600 --> 00:04:46,000 going, well, if I'm going to do an LES and I'm going to have to 81 00:04:46,000 --> 00:04:50,086 run lots of time steps and I've got this GPU, should I maybe 82 00:04:50,086 --> 00:04:54,760 now go back to explicit that previously wasn't maybe the 83 00:04:54,760 --> 00:04:58,000 optimum way of doing it, but now with the GPU it is. 84 00:04:58,840 --> 00:05:02,600 And then maybe could I do a, a wall-modelled approach if I'm 85 00:05:02,600 --> 00:05:04,734 doing a wall-modelled LES, and does that mean I should maybe 86 00:05:04,734 --> 00:05:06,716 use the immersed-boundary method because I don't need to 87 00:05:06,716 --> 00:05:09,975 resolve the wall? And so that it's this interplay 88 00:05:09,975 --> 00:05:14,206 between the hardware and the software is creating some really 89 00:05:14,206 --> 00:05:17,205 interesting changes. And and finally, one of the 90 00:05:17,213 --> 00:05:20,975 interesting was a high order side, You know, I think high 91 00:05:20,983 --> 00:05:24,055 order, without going into too much of the technical details, 92 00:05:24,063 --> 00:05:26,272 there's some interesting stuff around like tensor cores, 93 00:05:26,280 --> 00:05:29,370 etcetera. So the whole high order thing is 94 00:05:29,370 --> 00:05:34,082 coming back in to play. So yeah, was my prediction of 95 00:05:34,090 --> 00:05:39,026 key trends for CFD correct, GPUs and CFD, I'd say yes. 96 00:05:39,034 --> 00:05:43,147 And is it still for 2025/26? Definitely. 97 00:05:43,155 --> 00:05:49,104 It's only accelerating and I think you'll find that it 98 00:05:49,112 --> 00:05:53,750 continues to accelerate whilst there is so much more learnings 99 00:05:53,750 --> 00:05:57,127 to be had in terms of optimal architectures. 100 00:05:57,135 --> 00:06:00,198 And there's like a crossover point when your compute gets 101 00:06:00,206 --> 00:06:03,348 higher, you can go to a different modeling paradigm. 102 00:06:03,356 --> 00:06:09,015 So for example, and not wanted to speak about this for too 103 00:06:09,023 --> 00:06:13,680 long, you've gone from, let's say a wall-resolved RANS, like a 104 00:06:13,680 --> 00:06:17,438 low-y+ RANS, where you wanted to resolve the wall and where 105 00:06:17,446 --> 00:06:21,002 you need a certain gridding strategy to be able to resolve 106 00:06:21,010 --> 00:06:24,408 the wall to a wall-modelled LES where you say, oh, well, I don't 107 00:06:24,408 --> 00:06:25,953 need to resolve the wall because I'm modelling it. 108 00:06:25,961 --> 00:06:28,850 So maybe I can go to a different gridding paradigm and I can go 109 00:06:28,858 --> 00:06:32,088 to explicit. But if you go to the next stage, 110 00:06:32,088 --> 00:06:34,816 which is the wall-resolved LES, that maybe say, well, 111 00:06:34,816 --> 00:06:37,906 actually now my gridding paradigm that was good for 112 00:06:37,906 --> 00:06:41,475 wall-modelled LES may not be good for wall-resolved LES. 113 00:06:41,483 --> 00:06:43,246 Maybe I need to do something slightly different. 114 00:06:43,254 --> 00:06:46,468 So it's interesting how these waves happen and the, like I 115 00:06:46,476 --> 00:06:49,344 said, doesn't interplay. So what I find most interesting 116 00:06:49,344 --> 00:06:54,660 is a year ago or two years ago, the most bold predictions was 117 00:06:54,668 --> 00:06:57,680 more towards hybrid RANS/LES, wall-modelled and everyone said 118 00:06:57,680 --> 00:06:59,887 oh, wall-resolved is too far away. 119 00:06:59,895 --> 00:07:03,441 I'm actually seeing now people be a bit bolder and say, well, 120 00:07:03,449 --> 00:07:06,520 you know what with the latest GPUs, maybe wall-resolved LES, 121 00:07:06,528 --> 00:07:10,745 it started to become possible. And that's interesting. 122 00:07:10,753 --> 00:07:14,624 I think not for everyday applications perhaps, but now 123 00:07:14,632 --> 00:07:18,180 that that is coming up, some other stuff starts to get 124 00:07:18,188 --> 00:07:20,845 interesting like just the sheer number of time steps you need to 125 00:07:20,845 --> 00:07:24,824 do. And yeah, so still important. 126 00:07:24,832 --> 00:07:27,580 OK, AI, machine learning. Well, duh. 127 00:07:27,588 --> 00:07:30,255 Yeah, that's clearly still a key trend. 128 00:07:30,263 --> 00:07:33,120 Has things moved up in the last year? 129 00:07:33,128 --> 00:07:35,512 Absolutely. I'd say one of the big 130 00:07:35,520 --> 00:07:39,242 breakthroughs I personally seen is probably a year ago or a 131 00:07:39,242 --> 00:07:43,640 little before a year ago, one of the bottlenecks seemed to be 132 00:07:43,648 --> 00:07:46,012 volume prediction using AI surrogates. 133 00:07:46,020 --> 00:07:49,855 Stuff was coming out around like graph neural networks and being 134 00:07:49,855 --> 00:07:52,960 able to predict things, but all the examples were always on 135 00:07:52,960 --> 00:07:56,555 smaller cases. And one of the things that I did 136 00:07:56,555 --> 00:07:59,530 with some colleagues was to create that DrivAerML data set, 137 00:07:59,530 --> 00:08:05,296 which was really put out there as a data set to give 138 00:08:05,304 --> 00:08:09,534 realistically sized meshes and hopefully encourage people to 139 00:08:09,542 --> 00:08:13,300 see, right. Can my method take the entire 140 00:08:13,308 --> 00:08:17,762 volume and a year or certainly two years ago, it looked like 141 00:08:17,770 --> 00:08:20,805 not many people could do it. Everyone was just doing the 142 00:08:20,813 --> 00:08:24,080 surface or maybe a few slices and every time you saw 143 00:08:24,088 --> 00:08:26,424 something, you know in marketing online, it was always like the 144 00:08:26,432 --> 00:08:28,543 surface of the car or a slice of the car. 145 00:08:28,551 --> 00:08:32,522 What I found was a breakthrough in the past year is that various 146 00:08:32,522 --> 00:08:37,525 methods have come out and I guess the most, most most recent 147 00:08:37,525 --> 00:08:41,606 will be like Transformers that seem to be the most 148 00:08:41,614 --> 00:08:43,798 computationally efficient. Efficient for doing full volume 149 00:08:43,798 --> 00:08:46,982 predictions as in training on a full volume and then doing an 150 00:08:46,990 --> 00:08:50,175 inference and being able to get a full volume out. 151 00:08:50,183 --> 00:08:53,464 I feel like that really accelerated the realization that 152 00:08:53,464 --> 00:08:57,889 some of these AI methods could give you back a flow field that 153 00:08:57,889 --> 00:09:00,480 was more similar to a traditional one. 154 00:09:00,488 --> 00:09:05,060 And yeah, that transition from from graph networks to maybe 155 00:09:05,060 --> 00:09:08,200 neural operators and Transformers for me has been a 156 00:09:08,200 --> 00:09:10,300 really interesting 1. And I'm certainly excited to see 157 00:09:10,300 --> 00:09:14,424 what happens in the next year. I think we'll continue to see 158 00:09:14,432 --> 00:09:16,948 new architectures, new evolution of, of AI. 159 00:09:16,956 --> 00:09:20,840 So that is definitely a trend that's continuing. 160 00:09:20,848 --> 00:09:22,984 Yeah. And they're obviously linked, 161 00:09:22,992 --> 00:09:24,969 right. So if you have a faster solver, 162 00:09:24,969 --> 00:09:26,446 you can generate data more easily. 163 00:09:26,454 --> 00:09:29,460 You can then do the training. I think one of the big 164 00:09:29,468 --> 00:09:31,616 discussions now is on the right code architecture. 165 00:09:31,624 --> 00:09:34,100 Sorry, like programming environment, because online 166 00:09:34,108 --> 00:09:36,275 training is definitely a hot topic. 167 00:09:36,283 --> 00:09:39,100 You know, how can you take advantage of the transient 168 00:09:39,108 --> 00:09:42,884 nature of the flow? So at the moment, I'd say the 169 00:09:42,892 --> 00:09:45,325 state-of-the-art is normally that you would just generate 170 00:09:45,333 --> 00:09:49,064 those data and even if it was with a hybrid RANS or 171 00:09:49,064 --> 00:09:51,610 wall-modelled LES, you'll just take a time average and dump 172 00:09:51,610 --> 00:09:54,840 that and train on that. But I think what's emerging now 173 00:09:54,840 --> 00:09:58,560 is this interesting, well, can I use some of the transient data 174 00:09:58,560 --> 00:10:01,480 that's being created anyway, you know, dump them out at 175 00:10:01,488 --> 00:10:07,092 checkpoint and train on that. So I think the whole AI will 176 00:10:07,100 --> 00:10:10,120 continue and continue. The one thing that I would 177 00:10:10,128 --> 00:10:14,160 probably add to that AI bucket is the agentic AI side. 178 00:10:14,168 --> 00:10:18,140 So let's say you've got surrogates that's in one market 179 00:10:18,140 --> 00:10:21,400 and you've got LLMs for just general productivity gains and 180 00:10:21,408 --> 00:10:24,935 coding on the other. I put agentic AI somewhere in 181 00:10:24,943 --> 00:10:28,680 the middle, which is how can you have multiple agents working 182 00:10:28,688 --> 00:10:32,975 autonomously through some sort of master agent that can go off 183 00:10:32,975 --> 00:10:36,472 and do tasks for you. So instead of you having to 184 00:10:36,480 --> 00:10:39,240 manually click run this simulation or write a Python 185 00:10:39,248 --> 00:10:42,756 code that goes and runs, you'd said, give a text prompt saying 186 00:10:42,756 --> 00:10:45,936 I would like to go and run this simulation with these parameters 187 00:10:45,936 --> 00:10:49,598 and please create me a report. And you'll have a workflow where 188 00:10:49,598 --> 00:10:52,944 an agent will interpret that request will then go and let's 189 00:10:52,952 --> 00:10:55,320 say create a simulation setup for you. 190 00:10:55,328 --> 00:10:59,400 But crucially also then run the simulation, let's say an HPC 191 00:10:59,408 --> 00:11:03,504 cluster, use a separate, you know, agent could be a surrogate 192 00:11:03,504 --> 00:11:06,383 model essentially, you know, to go run it or to do the 193 00:11:06,383 --> 00:11:08,626 meshing, then another LLM, maybe do the analysis of the 194 00:11:08,626 --> 00:11:10,720 results. And the point being is that you 195 00:11:10,720 --> 00:11:12,480 have to kick that all off with a text prompt. 196 00:11:13,360 --> 00:11:16,040 That is the most cutting edge example. 197 00:11:16,040 --> 00:11:19,040 And that's what I feel is probably in the next 12 months 198 00:11:19,040 --> 00:11:22,000 will be the next big jump. And some start-ups are already 199 00:11:22,000 --> 00:11:25,200 doing it as some companies, but I feel that's probably one of 200 00:11:25,200 --> 00:11:27,987 the big next steps. I should add, by the way, for 201 00:11:27,987 --> 00:11:31,800 GPUs I didn't mention, I think precision is a very interesting 202 00:11:31,800 --> 00:11:35,400 topic now. I think there's huge gains if 203 00:11:35,400 --> 00:11:39,880 CFD can use lower precision for for obvious reasons, you know, 204 00:11:39,880 --> 00:11:44,865 that there's so many FLOPS in in the lower precision because of 205 00:11:44,865 --> 00:11:48,160 AI generally only needs, you know, FP8 or FP16. 206 00:11:48,560 --> 00:11:52,400 So it doesn't mean that you can't do 64 or 32, but if you 207 00:11:52,400 --> 00:11:56,200 can find a way to harness that power, then you can really 208 00:11:56,200 --> 00:11:57,848 unlock the next big accelerator. 209 00:11:57,856 --> 00:12:02,144 So the first one I had was the shift to cloud computing. 210 00:12:02,152 --> 00:12:06,140 Now, you know, people who know me or listened to this before 211 00:12:06,148 --> 00:12:09,180 know that you know, I currently work from NVIDIA, but I used to 212 00:12:09,180 --> 00:12:11,730 work for Amazon Web Services, which is obviously a massive 213 00:12:11,738 --> 00:12:15,420 cloud computing company. And there is no doubt whatsoever 214 00:12:15,420 --> 00:12:18,580 that cloud computing has totally transformed the IT 215 00:12:18,580 --> 00:12:23,908 sector and the HPC sector. When I left AWS. 216 00:12:23,908 --> 00:12:28,782 I definitely saw an increasing momentum in people adopting the 217 00:12:28,782 --> 00:12:35,430 cloud cause of the fact that they could essentially focus on 218 00:12:35,430 --> 00:12:38,480 the stuff that mattered to them and not on the sort of 219 00:12:38,480 --> 00:12:44,960 manual data centre stuff. I still see that increasing, but 220 00:12:44,960 --> 00:12:49,080 probably the bit that maybe I was not brainwashed because I'm 221 00:12:49,080 --> 00:12:53,560 joking aside a little bit. But you I definitely, I'm seeing 222 00:12:54,520 --> 00:12:59,480 I'm still a more hybrid sense where on one hand people are 223 00:12:59,480 --> 00:13:03,857 absolutely seeing the value of the cloud, you know, through AWS 224 00:13:03,857 --> 00:13:06,680 or GCP or Microsoft or you know, all the cloud players. 225 00:13:07,120 --> 00:13:08,760 And by the way, there's an interesting angle now with all 226 00:13:08,760 --> 00:13:13,785 these neoclouds like CoreWeave, etcetera, which is mainly I 227 00:13:13,785 --> 00:13:19,495 guess to the AI and GPU angle, but I'm still seeing a hybrid 228 00:13:19,495 --> 00:13:21,000 use. You know, there's still quite a 229 00:13:21,000 --> 00:13:24,480 lot of on Prem, there's cloud and and I still think there's a 230 00:13:24,480 --> 00:13:30,760 lot of room for improvements in the way to access cloud 231 00:13:30,760 --> 00:13:34,707 resources. You know, I think SaaS from a 232 00:13:34,707 --> 00:13:38,150 CFD point of view now SaaS products are still not fully 233 00:13:38,150 --> 00:13:41,230 there. You know, if you ask most people 234 00:13:41,230 --> 00:13:43,230 are they doing it software service? 235 00:13:43,238 --> 00:13:47,270 No, they may use the cloud to have a cluster, but are they 236 00:13:47,278 --> 00:13:49,590 doing it in a true SaaS like manner? 237 00:13:49,598 --> 00:13:55,300 I think that it's slower than I thought people still seem to 238 00:13:55,308 --> 00:14:01,710 prefer and there's good reason. So we'd have to get into of 239 00:14:01,718 --> 00:14:05,716 having a local experience in that you're not just doing 240 00:14:05,724 --> 00:14:08,495 through a web browser. And I think that's probably just 241 00:14:08,495 --> 00:14:11,950 decades of engineers being used to just SSH-ing and stuff 242 00:14:11,950 --> 00:14:14,674 and and having like, yeah, running scripts. 243 00:14:14,682 --> 00:14:19,264 And it's something that you think would change, but it, it 244 00:14:19,272 --> 00:14:25,090 does seem remarkably stubborn. And yeah, so I, I definitely see 245 00:14:25,090 --> 00:14:29,040 the cloud continuing, but I'm still waiting for some company 246 00:14:29,048 --> 00:14:32,124 to come out there and really make it easy from a workflow 247 00:14:32,132 --> 00:14:36,488 point of view in like a true hybrid way or to just simplify 248 00:14:36,496 --> 00:14:39,394 maybe the cloud experience. You still have to be quite an 249 00:14:39,402 --> 00:14:43,370 expert to do it. So I am, I'm not changing that 250 00:14:43,378 --> 00:14:47,062 prediction. I'm just seeing that it seems to 251 00:14:47,062 --> 00:14:50,107 be a bit slower than I fully appreciated. 252 00:14:50,115 --> 00:14:54,525 And maybe in hindsight, that's because I was seeing all of the 253 00:14:54,525 --> 00:14:58,379 positive and success stories at AWS and maybe I was not hearing 254 00:14:58,379 --> 00:15:02,868 the bits where people weren't, you know, doing it. 255 00:15:02,876 --> 00:15:09,012 The 4th 1 I had was on digital certification through 256 00:15:09,012 --> 00:15:14,568 higher-fidelity methods that, as I mentioned on the 3rd of it is 257 00:15:14,568 --> 00:15:18,335 definitely the case. You know, you're having a huge 258 00:15:18,343 --> 00:15:21,932 increase of people going to higher fidelity methods because 259 00:15:21,932 --> 00:15:25,869 of #1 you know, the fact that you can access to compute more 260 00:15:25,877 --> 00:15:28,642 easily. And that is definitely a topic 261 00:15:28,650 --> 00:15:32,420 that I think will continue to increase sector by sector. 262 00:15:32,428 --> 00:15:36,285 So the automotive has definitely pioneered that and all have 263 00:15:36,293 --> 00:15:37,680 moved to higher fidelity methods. 264 00:15:37,688 --> 00:15:40,420 I think now, like I said, they're, they're looking how can 265 00:15:40,420 --> 00:15:42,870 I go from hybrid RANS to wall- resolved LES. 266 00:15:42,878 --> 00:15:46,485 I think the aerospace sector is probably still trying to get 267 00:15:46,493 --> 00:15:49,990 from RANS, you know, to hybrid RANS or wall-modelled LES and 268 00:15:49,990 --> 00:15:53,680 the work that you know, we did in the last High Lift Prediction 269 00:15:53,680 --> 00:15:56,694 Workshop has definitely helped to push that forward. 270 00:15:56,702 --> 00:15:59,758 And there's some interesting work going on now to try and 271 00:15:59,766 --> 00:16:02,000 push it even further again to like wall-resolved LES. 272 00:16:02,008 --> 00:16:07,000 But I see that continually, you know, increasing and now it's 273 00:16:07,008 --> 00:16:11,580 more around building up best practices and understanding of 274 00:16:11,588 --> 00:16:14,440 those methods, you know, robustness that everybody knows 275 00:16:14,440 --> 00:16:16,660 where they go wrong, where they go right. 276 00:16:16,668 --> 00:16:21,209 So it probably will take time for that and for people to see 277 00:16:21,217 --> 00:16:24,308 the value because it is undeniably more expensive to do 278 00:16:24,308 --> 00:16:26,765 with the low fidelity. So I think there needs to be 279 00:16:26,773 --> 00:16:27,970 success stories that have been created. 280 00:16:27,978 --> 00:16:31,230 But I think probably one of the biggest challenges is 281 00:16:31,238 --> 00:16:33,060 transition. Certainly I'm speaking more from 282 00:16:33,060 --> 00:16:36,645 a CFD and external aero point of view, but one of the 283 00:16:36,645 --> 00:16:39,772 challenges still with even going to wall-modelled LES to 284 00:16:39,772 --> 00:16:41,120 hybrid is transition modelling. 285 00:16:41,760 --> 00:16:45,160 And I, I feel that that's why the wall-resolved LES is still 286 00:16:45,160 --> 00:16:48,200 very tempting because it doesn't fully solve it, but it, it 287 00:16:48,200 --> 00:16:50,880 definitely helps to overcome some of those challenges. 288 00:16:51,920 --> 00:16:54,200 So I think that will continue, but I would look more at how do 289 00:16:54,200 --> 00:16:58,160 we make wall-resolved LES approaches more, more 290 00:16:58,160 --> 00:17:01,480 affordable. I am going to be doing an 291 00:17:01,480 --> 00:17:05,118 episode where we're going to talk more on the combustion side 292 00:17:05,118 --> 00:17:07,358 of things. I do realise that this podcast 293 00:17:07,358 --> 00:17:10,000 just tend to be a little bit focused on external aero. 294 00:17:10,200 --> 00:17:11,768 That's basically my background. 295 00:17:11,776 --> 00:17:16,194 So I am purposely trying to, yeah, diversify a little bit and 296 00:17:16,194 --> 00:17:19,440 make sure that we're covering other areas. 297 00:17:19,448 --> 00:17:25,690 And the final one was mergers and acquisitions and 298 00:17:25,698 --> 00:17:29,280 innovations. And this has definitely kept 299 00:17:29,288 --> 00:17:34,590 going. We have seen Ansys and Synopsys, 300 00:17:34,590 --> 00:17:42,640 so two big companies merge. We have seen, you know Siemens, 301 00:17:42,640 --> 00:17:48,538 Altair, Cadence, BETA CAE. So this has kept going and was a 302 00:17:48,538 --> 00:17:52,820 trend that I predicted and is still true today that there is 303 00:17:52,828 --> 00:17:57,320 so much interest in the community and engineering and 304 00:17:57,328 --> 00:18:01,160 semiconductor businesses that, you know, it's a growing 305 00:18:01,168 --> 00:18:04,620 business, really important. And big companies are realizing 306 00:18:04,620 --> 00:18:07,520 the value of acquiring or merging with others to, you 307 00:18:07,528 --> 00:18:10,500 know, bring some of their technology in and speed up. 308 00:18:10,508 --> 00:18:14,320 You know, that that time. I think that will continue. 309 00:18:14,328 --> 00:18:17,420 I would say the prediction for next year, it's not there's not 310 00:18:17,420 --> 00:18:19,748 many other big companies that can merge now. 311 00:18:19,756 --> 00:18:24,270 You know, it has consolidated a little bit to Cadence, Siemens 312 00:18:24,278 --> 00:18:28,772 and, and Synopsys, but where I do see that being definitely 313 00:18:28,780 --> 00:18:33,542 move is there's a lot of start-ups and I'm sure that the 314 00:18:33,542 --> 00:18:38,166 many of those start-ups will be acquired by some of these bigger 315 00:18:38,166 --> 00:18:40,482 companies. So they can start to really 316 00:18:40,490 --> 00:18:43,204 bring those to a more enterprise level and integrate it. 317 00:18:43,212 --> 00:18:46,315 So I, my prediction would be in the next 12 months that some of 318 00:18:46,315 --> 00:18:50,049 the start-ups that you, you know, you may know of, I won't 319 00:18:50,057 --> 00:18:52,918 mention all names, but the start-ups, you know, I think 320 00:18:52,926 --> 00:18:55,850 you'll see some of them being acquired by high profile 321 00:18:55,858 --> 00:18:59,728 companies just because that is, that's a tried and tested route. 322 00:18:59,728 --> 00:19:00,760 You know, when I spoke to the 323 00:19:00,760 --> 00:19:03,560 CTO of Ansys and he openly admitted it, he said, that's 324 00:19:03,560 --> 00:19:05,520 what we do. We look for great start-ups. 325 00:19:05,520 --> 00:19:07,200 We may even, you know, encourage them. 326 00:19:07,200 --> 00:19:09,960 And once they reach a certain level, we might acquire it. 327 00:19:09,960 --> 00:19:12,560 And it, that's a great way of developing new technology. 328 00:19:12,560 --> 00:19:14,920 Sometimes it's better to let a start up do it. 329 00:19:15,480 --> 00:19:17,640 You have a, you know, faster way of doing than a, let's say, a 330 00:19:17,640 --> 00:19:19,640 big enterprise. But then they acquire them, they 331 00:19:19,640 --> 00:19:21,600 bring them, and that's how customers get to use them. 332 00:19:22,440 --> 00:19:23,560 So I think that will also still continue. 333 00:19:24,600 --> 00:19:28,000 So where do I think things are going? 334 00:19:28,000 --> 00:19:31,040 Well, I think actually those five topics will still continue 335 00:19:31,360 --> 00:19:37,880 to be a thing for 25 and 26, and I'm excited to see how they 336 00:19:37,880 --> 00:19:40,360 accelerate. So I hope that was interesting. 337 00:19:40,360 --> 00:19:42,680 I thought I'd keep it short. I know these podcasts tend to go 338 00:19:42,680 --> 00:19:45,360 to an hour or two hours, so I thought I'll keep this one short 339 00:19:45,760 --> 00:19:49,200 and got a couple of really good guests coming up for the next 340 00:19:49,200 --> 00:19:52,560 two episodes. So yeah, look out for those. 341 00:19:52,600 --> 00:19:54,440 And for now, I hope you enjoyed this episode.