The Real Reason Elon Musk is Building The DOJO Supercomputer

The Real Reason Elon Musk is Building The DOJO Supercomputer The Tesla Space 458K subscribers Join Subscribe 3.2K Share Download Thanks Clip 90,356 views Sep 1, 2024 #TheTeslaSpace #Elon #Tesla Why Tesla needs their own Dojo AI training computer to rival Nvidia Last video: Neuralink Reveals Major NEW Human Trials Update! • Neuralink Reveals Major NEW Update! ► Support the channel by becoming a member: / @theteslaspace ►The Tesla Space Merch Store Is Live! Shop our first release while quantities last: https://shop.theteslaspace.com/ ► Patreon: / theteslaspace ► Join Our Discord Server: / discord ► Subscribe to our other channel, The Space Race: / @thespaceraceyt ► Subscribe to The Tesla Space newsletter: https://www.theteslaspace.com ► Use my referral link to purchase a Tesla product and get up to $1,300 off and other exclusive benefits. https://ts.la/trevor61038 Subscribe: / @theteslaspace Welcome to the Tesla Space, where we share the latest news, rumors, and insights into all things Tesla, Space X, Elon Musk, and the future! We'll be showing you all of the new details around the Tesla Model 3 2023, Tesla Model Y 2023, along with the Tesla Cybertruck when it finally arrives, it's already ordered! Instagram: / theteslaspace Twitter: / theteslaspace Business Email: sean@creatormill.com #Tesla #TheTeslaSpace #Elon Transcript Follow along using the transcript. Show transcript Transcript 0:03 one of the biggest gambles that Elon 0:05 Musk and Tesla have made in recent years 0:07 is the design and production of their 0:10 own bespoke supercomputer Dojo musk has 0:14 said repeatedly that Dojo is a risky 0:16 move but it's also a technology that 0:19 Tesla might not be able to live without 0:23 let's talk about why that 0:28 is what's really fascinating is that 0:31 when Elon Musk started talking about 0:33 Tesla's project Dojo back in 2019 the 0:36 demand for graphics processing units or 0:38 gpus was still limited primarily to the 0:41 video game and digital video rendering 0:43 industry supercomputers were pretty much 0:46 only thought of as machines that solved 0:48 medical Mysteries and made climate 0:50 predictions yet at Tesla's first 0:53 autonomy day we saw the company's AI 0:55 team standing on stage telling the world 0:58 that they were going to build their very 1:00 own supercomputer and use it to train 1:03 artificial intelligence very few people 1:06 at the time would have known how 1:08 significant this was flashforward of 5 1:12 years and now gpus and supercomputer 1:14 clusters have become the single most 1:16 important piece of infrastructure in the 1:19 AI Revolution that has swallowed the 1:22 world the rapid shift in the current of 1:24 the tech industry has catapulted Nvidia 1:27 to the top of the stock market making 1:29 the world's biggest GPU supplier into 1:31 the world's most valuable company 1:34 ironically Elon Musk and Tesla have had 1:37 a major role to play in the rise of 1:39 Nvidia Tesla has spent billions and 1:42 billions of dollars on Nvidia AI 1:45 Hardware this is most evident in the 1:47 company's latest construction project a 1:49 gigantic addition built onto the south 1:52 end of the Texas gigafactory that is now 1:54 home to a supercomputer cluster that 1:56 Elon Musk has named cortex about half of 2:00 the chips that power this data center 2:02 will be made by Nvidia amounting to 2:05 about 50,000 of their h100 gpus but a 2:09 significant amount of the computing 2:10 power in cortex will also be made by 2:13 Tesla and somehow at the same time as 2:16 cortex was being built Elon also found 2:18 the time to construct an even bigger AI 2:21 training computer for his xai startup 2:23 company at a location in Memphis 2:25 Tennessee that one is known to house up 2:28 to 100,000 th of nvidia's current h100 2:32 and h200 GPU chips with Elon already 2:35 threatening to buy an additional 300,000 2:38 units of nvidia's Next Generation chip 2:41 the 2:41 b200 so all that to say Elon Musk knows 2:45 the value of Nvidia Hardware better than 2:48 most which is why he has been pushing 2:50 his own staff at Tesla to develop 2:53 something that can compete with what 2:55 Nvidia is doing and this is about more 2:57 than just trying to cut out some of 2:59 nvidia's market share it's more about 3:01 preserving Tesla's own ability to remain 3:04 at the front of the AI race during 3:07 Tesla's Q2 earnings called this year 3:09 Elon said that demand for NVIDIA 3:11 Hardware is so high that it's often 3:14 difficult to get the gpus he continued 3:17 by saying quote I am quite concerned 3:19 about actually being able to get steady 3:21 gpus when we want them and I think this 3:24 therefore requires that we put a lot 3:26 more effort on dojo in order to ensure 3:29 that we've about the training capability 3:31 that we need so one significant reason 3:34 that Tesla needs Dojo is simply because 3:36 no matter how big and viia grows one 3:39 company can't possibly quench the 3:40 world's thirst for AI Hardware as Elon 3:43 Musk would say the best supplier is no 3:46 supplier of course this is also about 3:49 money economics 101 says that when the 3:52 demand is high and the supply is limited 3:54 the price will go up and when a company 3:56 finds themselves in a position like 3:58 Nvidia does today where the wealthiest 4:00 companies in the world are demanding 4:02 more and more Hardware than you could 4:04 possibly produce then you can pretty 4:06 much charge any price you want now 4:10 that's not to say that designing and 4:12 building your own bespoke supercomputer 4:13 from the ground up is going to be any 4:16 cheaper than buying Hardware on the open 4:18 market not in the short term at least 4:20 but Dojo is a long game going back to 4:23 tesas q1 earnings call 2023 Elon told 4:26 investors that Dojo has the potential 4:29 for an order of magnitude Improvement in 4:31 the cost of training and also has the 4:34 potential to become a sellable service 4:36 that we would offer to other companies 4:38 in the same way that Amazon web services 4:41 offers web services he followed that up 4:44 by saying quote look at Dojo as kind of 4:46 a Longshot bet but a bet worth making in 4:50 the summer of 2023 Tesla's Dojo D1 chip 4:53 goes into mass production at tsmc in 4:56 Taiwan it's probably not a coincidence 4:59 that this timeline matches up with a 5:01 sudden rise of nvidia's stock price on 5:03 June 21st Elon Musk confirms that an 5:07 early version of the dojo computer rack 5:09 has already been online and running 5:11 useful tasks at the company for a few 5:13 months in January 2024 we get our first 5:17 indication of what a functional Dojo 5:19 training cluster will look like while 5:21 Elon also reiterates that Dojo is a 5:24 high-risk high-reward project he also 5:26 says that Tesla was pursuing the Dual 5:29 path of Nvidia and Dojo that Dojo is 5:32 working and is doing training jobs he 5:34 notes Tesla is scaling up and has plans 5:36 for Dojo 1.5 Dojo 2 Dojo 3 and whatnot 5:41 now Dojo V1 is a $500 million 5:44 installation at Tesla's production 5:46 facility in Buffalo New York we are not 5:48 exactly sure why Tesla chose an obscure 5:50 location but as someone who lives in the 5:52 same part of the world we know that it 5:54 gets cold during the winter which 5:56 probably helps with the cooling and 5:59 there's also Al an abundant stable 6:01 supply of green energy from 6:02 hydroelectric dams so in many ways 6:05 Buffalo is probably a better location 6:07 for a data center than Austin or Fremont 6:11 either way Elon continues to temper 6:14 expectations on dojo reminding us that 6:17 $500 million is actually not very much 6:20 money saying it's only equivalent to a 6:23 10K h100 system from Nvidia Tesla will 6:27 spend more than that on Nvidia Hardware 6:29 this year year the table Stakes for 6:31 being competitive in AI are at least 6:33 several billion dollars per year at this 6:36 point in June 2024 Elon puts these 6:39 numbers into an even more specific 6:41 context writing on X of the roughly 10 6:44 billion in AI related expenditures I 6:47 said Tesla would make this year about 6:49 half is internal primarily the Tesla 6:52 designed AI interface computer and 6:53 sensors present in all of our Cars Plus 6:56 dojo for building the AI training 6:58 superclusters and Nidia Hardware is 7:00 about 2/3 of the cost my current best 7:03 guess for NVIDIA purchases by Tesla are 7:05 3 to 4 billion this year he follows that 7:09 up by reiterating there is a path for 7:11 Dojo to exceed Nvidia it's a long shot 7:14 as I've said before but success is one 7:16 of the possible outcomes so even among 7:19 Tesla's own AI Hardware spending Dojo is 7:22 a lower level priority than inference 7:24 Computing Hardware this is the autopilot 7:27 computer that is integrated into every 7:29 test TLA vehicle it is the hardware that 7:31 runs the incar hardware that takes input 7:34 from the eight cameras and uses that to 7:35 make real time driving decisions Tesla 7:38 is currently producing version four of 7:40 their inference AI computer with version 7:42 five expected to arrive Sometime Late 7:44 next year or maybe in 2026 given Elon 7:48 timelines but we know now that this 7:50 incar processing power is quickly 7:52 becoming more important than the data 7:54 center training hardware and that the 7:56 autopilot V5 chip will be pretty much 7:58 necessary to achieving full autonomous 8:01 driving Elon wrote on X our next gen AI 8:04 model after this has a lot of Promise 8:07 around 5x increase in parameter count 8:10 which is very difficult to achieve 8:11 without upgrading the vehicle inference 8:13 computer okay so what does Dojo actually 8:17 do that's the tricky bit to explain 8:19 because AI training is about the closest 8:21 thing to digital alchemy that we've seen 8:23 so far but Elon recently specified that 8:26 Dojo 1 which is the Buffalo cluster 8:28 that's being installed now will have the 8:30 equivalent to roughly 8,000 Nvidia h100s 8:34 of training power online by the end of 8:36 this year Elon shared a couple of 8:38 pictures that show the front and back of 8:40 the dojo cabinets but no specific 8:43 details so we can gather that Dojo V1 is 8:46 not massive but not trivial either it's 8:49 less than 150 power of Cortex at Giga 8:51 Texas and less than one10 the power of 8:54 the X a gigafactory of compute in 8:56 Memphis but in the case of Dojo raw 8:59 power power shouldn't matter much 9:01 keeping in mind that this is a custom 9:03 designed AI training system that was 9:05 developed entirely by Tesla so the 9:08 hardware is custom tailored to Tesla 9:11 specific applications which should make 9:14 it significantly more efficient at those 9:17 particular tasks than a generic 9:19 multi-purpose chip like those from 9:21 Nvidia musk has said that the first 9:24 version of Dojo will be tailored for 9:27 Tesla computer vision labeling and 9:29 training which is great for FSD and for 9:32 training Optimus Tesla's humanoid robot 9:35 but it wouldn't be useful for much else 9:37 and for right now that's totally fine we 9:40 know that the company's main push right 9:42 now is getting the robo taxi vehicle 9:44 ready to be unveiled in October of this 9:47 year and it would make sense that the 9:49 acceleration of AI training Hardware 9:51 with cortex and Dojo is being done 9:53 specifically to support the robotaxi 9:56 project this is one of the rare vehicles 9:58 that will be judged entirely on its 10:00 functionality its ability to drive 10:03 autonomously so the software absolutely 10:07 needs to deliver now that does leave an 10:09 open-ended question on what the future 10:11 of Dojo will actually be if Tesla has 10:14 already decided that they are going to 10:15 solve full self-driving within the next 10:18 year or at least in the relatively short 10:20 term then what do they really need dojo 10:23 for well many things really so first off 10:26 Tesla is pushing out the robotaxi in 10:28 America and potentially China but for 10:31 FSD to work in every part of the world 10:33 the AI will need to be trained on roow 10:35 data that Tesla hasn't even collected 10:37 yet if you think about the quantity of 10:40 video that Tesla has collected from the 10:41 US and China their two biggest markets 10:44 they need equivalent data from Brazil 10:47 India Thailand every language every 10:49 different variety of road sign or lack 10:52 of road sign there is no end in sight 10:55 for Tesla's driving Ai and beyond that 10:58 we know that Tesla will move into a more 11:00 generalized AI model approach for their 11:03 humanoid robot project right now they 11:05 are training the robots Vision 11:07 navigation and object detection networks 11:09 in pretty much the same way that they 11:11 train FSD but for the Tesla bot to break 11:14 out into the consumer market and fulfill 11:16 elon's bold expectations of bots 11:19 outnumbering people then they're going 11:21 to need a pretty staggering variety of 11:24 AI models that go Way Beyond Vision 11:27 which is where dojo is currently con 11:29 strained all we have to go on is a very 11:31 limited reply from Elon he wrote in June 11:34 2023 Dojo V2 will address these 11:38 limitations so stay tuned thank you to 11:41 drift for partnering with us on this 11:43 video check out the link Down Below in 11:45 the description The Tesla Space 458K subscribers Videos About Twitter Instagram Patreon Shop the The Tesla Space store Cybertruck 2077 T-Shirt $29.95 Fourthwall Excitement Guaranteed Mug $14.95 Fourthwall Excitement Guaranteed T-Shirt - White $29.95 Fourthwall Tesla Bot Mug $16.95 Fourthwall The Tesla Space Logo Dad Hat $24.95 Fourthwall The Space Race Logo Dad Hat $24.95 Fourthwall 159 Comments rongmaw lin Add a comment... @godassasin8097 4 days ago 2:35 bro said "threatening to buy" like it's a terrorist action 19 Reply 1 reply @CycleGirl-77 3 days ago (edited) NVidia is reporting about 74% gross margin on H100s. If Dojo offers similar performance / manufacturing cost, that's an incredible savings for Tesla and could give them a serious competitive advantage in offering AI services compared to companies relying on NVidia. 3 Reply 1 reply @franklinmargolis2051 4 days ago My basic contention for past 3 months… DOJO is going to be mass produced for colocation of a data center where FSD is accepted. Keeps sensitive driving data within a country/region. All while creating useful self driving program for the inference computers in the car reflecting local rules, signs, customs and habits of drivers. 6 Reply @MrSchweppes 4 days ago xAI's supercomputer in Memphis consists of 100,000 H100s only. It's not a mix of H100 and H200 GPUs. If it were, Elon would have emphasized this in his tweet on July 22, when the training of Grok 3 began. 4 Reply @danharold3087 4 days ago (edited) The NVIDA H100s are not generic multi purpose chips. They are not even general purpose GPU chips. They are more specialized than that. DoJo is said to be a better fit for what Tesla is doing. I have no reason to doubt that. 37 Reply 11 replies @ezbody 2 hours ago (edited) Musk is busy again, building another super duper something or another, sleeping on the office floor, no more time for tweeting every other second. What a hero, how lucky we are to live in his presence. Amen Reply @MrNaders 4 days ago Great informative video as always keep it up 13 Reply 1 reply @edfrancis7001 12 hours ago Be aware that Tesla has serious employee headcount requirements in Buffalo, so they are business there wherever practical. Reply @johannesdolch 1 day ago Elon said very clearly that Dojo is a hedge against not getting the chips they need. Reply 1 reply @pawelkapica5363 10 hours ago I want to see him actually finish a thing he promised instead of hyping up the next cash grab. 1 Reply @oidbio2565 4 days ago A generic multi purpose chip like those from Nvidia? Nvidia’s chip can be adapted to work efficiently with any training regime 3 Reply @jarodmoves 2 days ago Very good video. You are a great site to learn how to present information. Many site waste a lot of time. Keep it up! Reply @Xibot888 3 days ago Teslas biggest competitor is BYD and NVIDIA has already partnered with them to install AI driving systems in their cars. Reply @user-yi8uu1du3b 3 days ago Maybe there is a kW per Gigaflop issue - Elon did mention how much power a DOJO stack took, and he didn't say how it compared with NVidia. Reply @nealremy6635 3 days ago We Are in Unchartered Financial Waters! every day we encounter challenges that have become the new standard. Although we previously perceived it as a crisis, we now acknowledge it as the new normal and must adapt accordingly. Given the current economic difficulties that the country is experiencing in 2024, how can we enhance our earnings during this period of adjustment? I cannot let my $680,000 savings vanish after putting in so much effort to accumulate them. 92 Reply 6 replies @paulmcgreevy3011 3 days ago Not sure why you think Tesla isn’t collecting data in other territories? Vehicles collect data regardless of FSD subscription. 1 Reply @stillpaints 4 days ago Back in 2019 crypto mining created a large portion of the demand for GPUs. 3 Reply 1 reply @user-lm4nk1zk9y 4 days ago What would happen if TSMC suddenly for some reason stopped producing chips? 6 Reply 7 replies @amondhawes-khalifa1949 7 hours ago All the biggest billionaires and trillion dollar companies racing toward ultrapowerful supercomputers and ASI, and everyone else is just watching like, "I guess we're headed for the season finale now, huh?" Reply @TheStatechamp2010 3 days ago There won’t be a single Dojo chip sold to a 3rd party unless Elon has a supercluster twice as big as any other hyperscaler 1 Reply @alkimball8920 1 day ago DOJO will be needed on Mars. I'm serious. Will need AI robots on Mars to build infrastructure. Reply @realjoecast 4 days ago My understading, Qualcom is making devices that are hitting personal devices that seem to be doing just as good as actual GPUs with higher efficiency pound for pound. someday they may catch up in this space. I could be wrong though. Reply @Swizzenator 4 days ago Who else has created their own Ethernet protocol? They tear apart everything to improve it. 1 Reply @WilliamBusuttil 2 days ago 3-09-2024. Hi yesterday I saw a 3 wheel car, at Nassenhausen , Germany, today l saw your picture of the robot taxi, it's here. Reply @soundslight7754 4 days ago Another application of GPU's in 2019 was crypto mining 1 Reply @EngIlya 10 hours ago why then he buys 100k H100 GPUs for xAI? 1 Reply @user-sf7lv4jm4c 3 days ago Elon needs to start a company that builds it’s own gpus. And chips for everyone else. Gigafactories worth of chips. That would solve the chip problem. Reply @chrismacdonald9472 4 days ago Investing in quantum computers make more sense. I know basic computing is available. But. Seems like alotta money. Reply @Starship007 1 day ago Stocks, investing, is a high risk possible high reward later. Invest in companies with a man that has his heart in the right place. He is a true humanitarian. He even stated it will take 25 years before most have EV’s. We still need nuclear and Fossil fuels. We can not bankrupt companies by forcing change tomorrow. Hawaii over 30% renewable NOT due to government but private residential rooftop solar. USA has cleaner grid. Other countries just starting their Industrial Revolution they need help Reply @salty_berserker_channel 4 days ago Make sure you always capitalize the X in CorteX. Reply @lowgun410 3 days ago Why not attach newer more accurate and advanced gps sensors or satellite cameras to work with the new Dojo network? Reply @lourdessilva6442 4 days ago Sem palavras esse documentário Reply 3 days ago As opposed to the bushlit reason? Reply @AndrewGwilt 4 days ago From $37K to $45K that's the minimum range of profit return every week I thinks it's not a bad one for me, now I have enough to pay bills and take care of my family. 17 Reply 22 replies @simonscofield8825 3 days ago Does anyone believe what is said about the delay in the robotaxi - that its AI is getting online and posting random stuff on X for no reason??!! I guess we may never know as the truth is often wilder than the actual fact.... Reply @DanFrederiksen 3 days ago Here's the thing, nvidia h100 costs around 35k$ for a card which is a chip comparable in size to a 4090 and 80GB very fast ram. That chip might cost 200$ from TSMC and 80GB of GDDR6 ram costs 170$, maybe you can spot a price disparity. Meaning the potential for cheaper compute than nvidia's is extreme. Compared to current prices it will soon be comparably cost free. Like you get a data warehouse for free compared to what they pay now. And since AI is not about a lot of compute but the right algorithm, all these FOMO billions will soon be entirely wasted. 2 Reply @OdinReactor 4 days ago It's the Mojo Dojo Casa House Super Computer! Reply @andririan6342 1 day ago So why Elon musk prefer use nvidia compute instead of DOJO? Reply @jabarigraham 4 days ago LFG 2 Reply @steveblomefield9513 4 days ago With nvidia chips, tesla can redesign the dojo ai chip Reply @JoepSwagemakers 3 days ago I don't think the FSD model will need to be trained on the same amount of data from every country. The model will at some point probably recognize and know how roads in general work and will be able to apply that to most of all roads. As a human you also don't have to re-learn how to drive when you're driving in another country. Reply 2 replies @monkeyperson9392 4 days ago AI day was 2021, not 2019 1 Reply 1 reply @freebird7284 1 day ago in his spare time! Reply @woolfel 4 days ago lets be blunt and honest. Writing the software stack to optimize AI training is very hard. Anyone silly enough to think "I just need my own hardware" clearly doesn't understand how the entire stack works. The hard part about building a software stack for training machine learning model isn't the hardware. It's optimizing and dividing up the work to make sure the hardware is running efficiently. The parallel processing problem is a long standing problem in computer science. Go watch talks by Jim Keller on parallel processing. To anyone that believes Elon's BS lies, try to implement a compiler to optimize math heavy computation. 1 Reply 1 reply @slwiser1 4 days ago Is it Dojo 1.0 or Dojo 1.5? I read somewhere they were doing 1.5 right now. Reply @elbow005 1 day ago Watermelon Chip Reply @cappybenton 4 days ago Wunderbar Reply @wayofflow955 4 days ago (edited) Bots are scary if you see them in real life, my brain keep thinking the bot could become like the terminator! Software tool is why Nvidia is dominating, all the researchers are used to using cuda and it only runs on Nvidia hardware! If the researcher said I need Nvidia hardware to make it work the vp will go get it! Does the tesla fsd broken up into many nets and can be train in parallel? Dojo might train a percentage of all the nets for fsd, any one know that? Reply @specialkonacid6574 4 days ago so glad i now know the real reason Reply @anonymousperson7536 2 days ago Elon "only high T males should govern" Musk Reply @SlipKnotRicky 4 days ago Anybody know if DOJO is a 64 bit system? Reply @KangoV 4 days ago (edited) AI has cost companies around $5 trillion so far, but only made a return of around $50 billion. A bubble? 2 Reply 7 replies @oscarwindham6016 4 days ago Still, junk in, equals junk out..., unless these supercomputers are 'allowed' to learn on their own, and so far, that has not really been permitted. Reply @azraelonyx 4 days ago Green energy ! Reply @zbigniewteterycz1571 4 days ago POZDRAWIAM WSZYSTKICH BARDZO SERDECZNIE . 1 Reply @fathermerrin1145 3 days ago What is it that you want to say? Reply @user-lg4wn3vw1i 3 days ago listen carefuly fsd that tesla car have can be used for tesla robots Reply @AmuXta 34 minutes ago All he is doing is taking on Invidia. Reply @AnIdiotAboard_ 3 days ago H100's are general purpose? LMFAO Quick Someone tell NVIDIA Share owners. Reply @lisaronquillo4214 3 days ago AI for 🇧🇷 Reply @EDM-Jigger 4 days ago Yeah, I honestly do not think they're capable of building their own supercomputer. using in-house CPUS and GPUS. literally their team assembled. to build the supercomputer aren't Semiconductor designers program. their programmers. and with the way Elon Musk has been treating Taiwan I really do not think TSMC would be willing to work with them. as easily and you can say, well, he can just start his own semiconductor fabs. Yeah, he could but that project would take him between 10 to 20 years just to get the Fab up and running. And that does not. sum to An equivalent level of? semiconductor manufacturing as Taiwan. since 90% of the world's most advanced. semiconductors come from those factorys. It's not like building a rock. engine, because if it was, then the Russians would have figured out how to manufacture their own advanced semiconductor. fabs, And when it rely on buying the technology. from us to do it and also. other European nations since those fabs can only exist In conjunction with those European nations that. build the tools. that are necessary for TSMC. to produce the world's most advanced semiconductor products. This is going to be another one of those many promises. that Elon Musk says. but never comes to fruition. 1 Reply @abdukalikov 4 days ago Anyone waiting on those sex bots?!?! Cmon dont be shy. ;) 1 Reply @treasuredev 4 days ago Second comment 2 Reply @thevillain8151 4 days ago Good luck, Nvidia is so far ahead of the game that catching up to them could take a decade if they all of a sudden stopped releasing products and just did development. 2 Reply 4 replies @ws7001 4 days ago Maybe he will stay with the carnivore diet. It resolves many brain issues. So far it resolves autism ( younger the better), bipolar, anxiety, schizophrenia, depression, and others. Reply @Johnsmith-p1v 3 days ago A Robo taxi will not happen in the next twenty years. Reply @jlvandat69 4 days ago Watching a genius disintegrate in real time is just brutal.......many great minds have followed the same script, ironically Nicholas Tesla, e.g. The lesson is obvious.......no matter who you are, if you don't take care of your mental health, your issues expand with time and it never ends well. Just ask DJT, as another example. Reply @mikes-wv3em 4 days ago TSLA STONK GOIN DOWNNNNNNNNNNNNNNNNNNNN Reply