also he did not serve us an advertisment ... unlike what this channel is starting and will become very shortly.. ( and props to munro sell out and make its money for old man and company ) but my eyes will stay in the future shill free..
did not need a battery bank.. dont need a shaver... come on.. if you want rev from advertiment.. make a separate sponsored video and tear it down and say whats good about it!!! we would find value in that and so woud they !!! KEEP TRUE TO YOUR VISION
Oh David! That was frakking awesome thank you for your kaupapa. Now repeat after me "I will start my journey to health with 6 squats a morning in the shower...😍😇😎 We need you dude.🤠
Hi Dave, quick comment on the memory modules - Those are GDDR6 DRAM modules, similar to what are found on graphics cards. They are significantly faster than traditional DDR RAM that you would put in your computer, and they are in what's called "clamshell mode" where there's a DRAM module on the top and bottom of the board. This is important for GDDR6, because it allows you to double the capacity for of a GDDR6 controller (which would be important for running larger inferencing models.) The part number on screen is for a 14Gb/s module, which is not the fastest model that Micron makes, but it's significantly cheaper and more reliable from a signaling and board layout standpoint. On each chip, there are 4 interfaces, each capable of 56 Gigabytes/second. I work on GDDR memory (not at Micron or Tesla) so when I saw a familiar part number, I figured I'd share what I can! Thanks for the presentation!
If I can tack on some more technical details, those are 16Gb (Gigabit), not 16GB (Gigabyte) modules. It seems like a LOT of people commonly conflate these two units, but the difference is a factor of 8 (a byte is 8 bits). The capacity of memory modules is usually measured in bits, not bytes.
Yes! Also, Cory pivoted the educational/entertainment point of the vid there at the end. He seamlessly morphed from teacher and lead engineer to salesman. There are a lot o companies around the world with endineers and thought leaders in their ranks. The Gordon Moores o this world weren't Titans o industry from the begining. There are genius young people watching, learning, and absorbing content like this like sponges. One or more of them is making up their mind on a degree program for lean design as we speak. The future of design and manufacture will be fascinating to watch!
I just received an an e mail from the regional sales manager categorically stating the the upcoming delivery of my new model s will have HW4 but no USS or RADAR !!!! Needless to say i cancelled the order forfeiting my deposit
If I have multiple notifications for UA-cam MUNRO LIVE always gets watched first. Always an articulate expert who can explain the unexplainable to us nerds out here.
Fantastic video! Personally I love hearing about the computer side of things, especially considering most people refer to Teslas as "computers on wheels". To that end, would have loved more technical detail on the CPU/GPU in the infotainment (if only for completeness), but I'm sure it's been covered elsewhere.
The reason for the extra space on the FSD computer is the form factor. They moved the GPU to the main board of the infotainment board, which was dictated the size of the infotainment board and thus the enclosure. Also, more connectors on the side diictated on one of the sides dictated that dimension. It's very unlikely that they'll do a new revision with same HW4 chip - Tesla had a LOT of time to design it and would likely placed everything they needed ahead of time. As for retrofit, we 'know' that initially HW4 will be used in cars with the same number of cameras, so it brings value even with existing sensor locations. If Tesla deemed necessary they can build a retrofit HW4 board that would have similar form factor and connector locations with HW3 computer. They can easily fit the HW4 chips and comonents around it as we know there's spare space on the current board. Finally, as for the radar, I've worked on car radars and can confirm that the original radar Tesla was using was useless. It had no or very little Y-axis resolution so could not differentiate a wall from a bridge, the resolution was low and the data rate was very limited. The new radar uses 2-generation ahead of the original one, has Y-axis resolution, more cameras, higher 'frame' rate and better z-axis (distance) and x/y-resolution. It makes a world of a difference for Teslas applications which are much more than not hitting the car in-front like what most vendors currently use radars for. Elon said several times that hi-res radars bring value and his comments for removing the old radar was due to false detects with low-res radars which caused more damage than good.
@@royh6526 I don't think it will be needed, but if at some point Tesla realizes that HW4 is required for RoboTaxi, it can build a HW4-based retrofit for older S3XY and only provide it to people signing up their cars to the RoboTaxi network. Building a retrofit is always possible (assuming it uses existing sensors) - the question is if it will be needed (I strongly believe HW3 can do robotaxi) and if the cost is prohibitive or not (including possible legal action to Tesla selling FSD and promising HW3 will be capable to support it)..
@@денисбаженов-щ1б Semi was planned to have 4680 cells from the beginning. 4680 cell development took years longer than expected. Finally Tesla decided to bring out the initial Semi with 2170 cells. Higher cost and increased weight penalty, but they couldn't wait any longer.
FINALLY someone that understands the problem with radars and how they are very much not created equal. It was fine for autopilot but an active detriment for FSD or even just getting more advanced features or AP unless the camera vision stack started disregarding inputs from the radar which at this point makes it sort of useless.
You have such good timing! So many times I take a break and want a bit of quality 'TV' and see a Munro post within the last 10 or 20 minutes. Good Job!
Cory, Dave, you guys make watching technical detail stuff like diving into HW3 and HW4 differences interesting, even for someone who has no knowledge of car internals. Thnx to the team at Munro making very interesting observations and showing design decisions I have been looking with a total different eye at my own car . Thanks for all your excellent and detailed critical reviews that you brought to us. I would never image siting through hours of video’s of guys showing me car bodies and how well or not so well they were put together. I especially liked what you showed how getting rid of a few small parts that cast a few dollars end up saving the company millions of dollars. And we’re talking just about the part cost. But leaving out the handling and manufacturing of these parts will make out an ever bigger cost saving and even time saving. This brings Elon’s mantra “ the best part is no part” even more home.
Great presentation. Could you do a show about the technical differences between the old Tesla radar and the new Tesla radar? For example: What the new radar can see vs what the old radar can see?
Great closing. Love it when engineers respect each other and the work they do for the client. Keep up the great work, we love seeing the value you guys provide to the community all over the world.
Is this is going to be a free upgrade to everyone that purchased FSD on the 3.0 hardware? Of course it is! I'm sure the Tesla cult members will be disgusted that I even asked that question.
Thanks for answering the question at the end about a retrofit for the current owners of Teslas. Great presentation. The future upgrades for FSD looks exciting with potential improvements.
Dam- as always another #masterclass. And.. more Corey / Dave duo please! First time I’ve watched Dave here and much like the whole Munro crew- just a true master at his craft. Not a single media channel even comes close! To steal a Elon quote- second place needs a telescope to catch a glimpse of you guys. Just awesome!
“Windfall for Tesla “. Is that the shareholder - or the owner of the vehicle ? What’s the second hand value of your FSD now that it’s not going to deliver as previously ‘suggested’ ?
Excellent review of changes, much appreciated! Wouldn't be surprised to see they moved to 7nm process, since the phone and gpu industry has left that node, it's got to be cheap from Samsung.
with more memory, it's safe to assume the memory bandwidth was also increased. Looking at the pastebin, HW4 has 2 ethernet ports and list 23 FSD sensors
You probably already have this thought. My thoughts on the more RAM is 1. Higher resolution camera= more needed memory 2. Often sometimes when you redesign something, the older version of chips is going to be end of life soon and so it makes sense to plan new items with the current version so it's supported for 5 years or so. Also, it may actually be cheaper.
Great video. Was wondering. Can someone take out the ai system and build their own mini tesla? I see a lot of ppl trying to build a mini (powerwheel size), programming everything themselves. Instead of figuring everything out on your own, just install the ecu? Maybe more complicated than it sounds lol. Great video
I feel bad for all the Tesla buyers who paid for FSD believing that they would get autonomous driving. My bet would be that even Hardware 4 is not enough to get there. As Cory and Dave pointed out, it’s the next step. When you buy a Tesla, don’t pay for any options that aren’t in the car when you receive it.
I think since I paid full price for my car and full price for my FSD, but my car does not and probably will not be able to get retrofitted with it. I think Elon should give me FSD FREE for my next Tesla I may get.
Thank you Dave and Cory, but especially Dave, for the insight and breakdown of the HW4 system. I don't come from an engineering background (but from a lower level tech background) but this channel has become my go-to for learning all kinds of engineering and is always a pleasure to watch. Should be required viewing for young and old aspiring engineers alike. Big thumbs up!
I had the same initial impression, but over an hour of city driving at 30mph you'll use 6300 watt hours of energy (30mph * 210W/mi) which is comprised of both the 65watts of HW3 FSD, plus 6235 Watts for the drive motors plus everything else. If the new computer actually uses the full 160 watts, then the example of 1 hour of city driving increases to 6395 watts (160-65 = 95 extra watts for HW4) or about 213 watts/mile. At highway speed the slight increase in computer power is even a smaller fraction of total energy usage. Not too bad for better FSD performance
15:30 When they removed the radar, Elon said (June 2021) that "Vision has become so good that radar actually reduces signal/noise." He later (Feb 2022) said that "Only very high resolution radar is relevant." One year later, here we are with the relevant technology to improve the signal-to-noise ratio.
Could we see an in-depth analysis of the various charging stations, as this is turning out to be an important factor in choosing ICE Vs. EV, and one EV against another. It would be interesting to see how these units differ internally both individually and as part of a network of chargers, and how they interact with the cars they supply power to.
Thanks Corey and Dave, great info. I saw some chatter on social media about HW4 and I'm so glad to have a Munro Live video on the topic so I don't have to waste hours sifting through uninformed opinions!
Dave you are an amazing guy!! Notice how he looks down when Cory praises him. I don't know why but this was quite a weteyed moment for me. This shows the Munro work enviroment!!!!
Linux Device Trees! ❤ A software readable spec sheet enabling the kernel to use the board. But humans can also read it, making it an object of great value during reverse engineering. Device Trees is an enabler of a common Linux kernel version across multiple different hardware versions, making the software modular, easier, and in the end cheaper to develop and maintain. I don't know how one would be agile using Green Hills RTOS, but if anyone wants to compete with Tesla / iterate as quickly, they probably need to have a Hardware Abstraction Layer, and an update strategy using continuous delivery that decouples OS and electronics to a large degree. Legacy auto doesn't seem to get this. I'm highly interested to hear Daves perspective of Green Hill RTOS vs Linux DT with respect to the OODA loop of iterative development.
Imagine paying the price of good used car just to have full self driving that is always promised to be one year away but is never realized. Meanwhile Tesla keeps improving the hardware tech, but never actually achieves full self driving.
16GB of RAM total of HW 4.0 but what is the SSD capacity? Of course could disable redundancy in the future like they have with 3.0 to utilize the power of both nodes to effectively 32GB total RAM.
Great to meet and learn so much from Dave. He's super knowledgeable and a nice guy. Cory was great as always. All of you guys at Munro & Associates make a great team including the head honcho himself, the much admired and respected Sandy. Keep up the great work.
Great Video, was hoping to hear some of my Buzz words: ASIL level(s), Software Bill of Materials (SBOM), Neutrino or Muon Kernel, QNX, Blackberry, Samsung Fab, Hypervisor, Security, Privacy. TIA.
So the bottomline is once the FSD software outgrows the HW3 hardware (which it apparently has) to take advantage of the capabilities of HW4; just go out and buy a new car. Problem solved. 🙂
Thank you very much for the comprehensive summary on HW4. As to the memory: With internal bump of NN Cores from 2x to 3x per Chip plus also a bump of the clock on those cores I estimated an increase from 72 TOPS to ~132 TOPS raw performance. The internal SRAM is now (likely) from 32MB increased to 48MB. So desire of increasing external memory bandwidth I can understand very well. But from 68GBit/s to 4x56Gbit/s is not necessarily a good sign. Why: The biggest claim on efficiency of the FSD1 was the reduced need for external memory bandwidth. So, this seems to be at least "relaxed" as a paradigm ... but with the price of higher power consumption.
If they bring back Radar, does that mean the USS might come back? I’m actually annoyed they’re bringing back radar since they have a huge chunk of their fleet without it. Like wtf Musk…
I am just so glad I didn't waste my money buying FSD when I bought my model S, I went for the upgrade to FSD3 HW because the SW changes made the FSD2 HW useless, but clearly the statement "all model 3" will come with FSD HW that supports fully autonomous driving" is way off the mark. Don't get me wrong, they've done well to maintain upgradability, but HW4 suggests ALL of the earlier built cars are on a dead end evolutionary path.
Hi Dave, you mentioned BroadR-Reach used for connecting the HD Radar. Are they using Broadcom automotive Ethernet chip for the 100Mbps twisted pair BroadR-Reach connection?
I guess even if we have HW4 in the 2023 Tesla vehicles coming out right now, there is still no radar and I will find out when I get my new Model Y long range in a week or so how many cameras does it have and their resolutions. Probably the 2024-2025 Tesla vehicles may adopt radar back into the mix for those that are doing FSD or commercial efforts, but at what cost. Looks like the HW4 system has more power, but all the external bells and whistles are not there yet. Time shall Tell where Tesla is going with their vehicles and software. They seem to be several steps ahead of the other manufacturers. Thanks for showing the HW4 architecture. Take care.
Should call it HW3.1. It connects a few more dots in the aspirational path to a safe and efficent autonomous driving. Why bother with dual simplex when modern ARM A76AE cores allow lockstep protection for a single safety processor. More performance, lower power, lower cost, smsller footprint, prospects for backwards compatibility. Nice to see T1 ethernet finally in the design for the radar.. But all their cameras need to go T1 for lower cost video over ethernet.
wow. too bad hardware 3 can’t be upgraded to 4---> wires would be wrong length and space issues. I told my son to wait for HW 4 before he buys a Model Y===> His patiences will be rewarded. Thanks Cory and Dave
Dave has all the skills of a master lecturer with all the knowledge necessary to explain things thoroughly. a true joy to watch and listen to..
also he did not serve us an advertisment ... unlike what this channel is starting and will become very shortly.. ( and props to munro sell out and make its money for old man and company ) but my eyes will stay in the future shill free..
did not need a battery bank.. dont need a shaver... come on.. if you want rev from advertiment.. make a separate sponsored video and tear it down and say whats good about it!!! we would find value in that and so woud they !!! KEEP TRUE TO YOUR VISION
Oh David! That was frakking awesome thank you for your kaupapa. Now repeat after me "I will start my journey to health with 6 squats a morning in the shower...😍😇😎 We need you dude.🤠
Exactly when he saw the General Public License he get a disorder!
Their entire team is so well media trained, it's impressive.
Hi Dave, quick comment on the memory modules - Those are GDDR6 DRAM modules, similar to what are found on graphics cards. They are significantly faster than traditional DDR RAM that you would put in your computer, and they are in what's called "clamshell mode" where there's a DRAM module on the top and bottom of the board. This is important for GDDR6, because it allows you to double the capacity for of a GDDR6 controller (which would be important for running larger inferencing models.) The part number on screen is for a 14Gb/s module, which is not the fastest model that Micron makes, but it's significantly cheaper and more reliable from a signaling and board layout standpoint. On each chip, there are 4 interfaces, each capable of 56 Gigabytes/second.
I work on GDDR memory (not at Micron or Tesla) so when I saw a familiar part number, I figured I'd share what I can! Thanks for the presentation!
If I can tack on some more technical details, those are 16Gb (Gigabit), not 16GB (Gigabyte) modules. It seems like a LOT of people commonly conflate these two units, but the difference is a factor of 8 (a byte is 8 bits).
The capacity of memory modules is usually measured in bits, not bytes.
One of the best episodes! Dave was great. I learned a lot.
Cory has a bad ass leader skills! And Dave did a great job here, what a treasure to have such talent available to us for free!
Yes! Also, Cory pivoted the educational/entertainment point of the vid there at the end. He seamlessly morphed from teacher and lead engineer to salesman. There are a lot o companies around the world with endineers and thought leaders in their ranks. The Gordon Moores o this world weren't Titans o industry from the begining. There are genius young people watching, learning, and absorbing content like this like sponges. One or more of them is making up their mind on a degree program for lean design as we speak. The future of design and manufacture will be fascinating to watch!
Dave seems awesome! More Dave tech spec content! !!!
Agreed
YES!
I just received an an e mail from the regional sales manager categorically stating the the upcoming delivery of my new model s will have HW4 but no USS or RADAR !!!! Needless to say i cancelled the order forfeiting my deposit
Great video and appreciate the work you guys and gals do...
Much appreciated!
Excellent teamwork. As someone who works in an adjacent sector to Dave's work, I give kudos for explaining it details very well.
Another articulate, personable and well organised expert. You guys are loaded with talent. 🤩😍🤩
Thank you so much 🤗
If I have multiple notifications for UA-cam MUNRO LIVE always gets watched first. Always an articulate expert who can explain the unexplainable to us nerds out here.
Fantastic video! Personally I love hearing about the computer side of things, especially considering most people refer to Teslas as "computers on wheels". To that end, would have loved more technical detail on the CPU/GPU in the infotainment (if only for completeness), but I'm sure it's been covered elsewhere.
The reason for the extra space on the FSD computer is the form factor. They moved the GPU to the main board of the infotainment board, which was dictated the size of the infotainment board and thus the enclosure. Also, more connectors on the side diictated on one of the sides dictated that dimension. It's very unlikely that they'll do a new revision with same HW4 chip - Tesla had a LOT of time to design it and would likely placed everything they needed ahead of time.
As for retrofit, we 'know' that initially HW4 will be used in cars with the same number of cameras, so it brings value even with existing sensor locations. If Tesla deemed necessary they can build a retrofit HW4 board that would have similar form factor and connector locations with HW3 computer. They can easily fit the HW4 chips and comonents around it as we know there's spare space on the current board.
Finally, as for the radar, I've worked on car radars and can confirm that the original radar Tesla was using was useless. It had no or very little Y-axis resolution so could not differentiate a wall from a bridge, the resolution was low and the data rate was very limited. The new radar uses 2-generation ahead of the original one, has Y-axis resolution, more cameras, higher 'frame' rate and better z-axis (distance) and x/y-resolution. It makes a world of a difference for Teslas applications which are much more than not hitting the car in-front like what most vendors currently use radars for.
Elon said several times that hi-res radars bring value and his comments for removing the old radar was due to false detects with low-res radars which caused more damage than good.
Don't agree with your retro-fit statements. Agree with all the rest.
@@royh6526 I don't think it will be needed, but if at some point Tesla realizes that HW4 is required for RoboTaxi, it can build a HW4-based retrofit for older S3XY and only provide it to people signing up their cars to the RoboTaxi network. Building a retrofit is always possible (assuming it uses existing sensors) - the question is if it will be needed (I strongly believe HW3 can do robotaxi) and if the cost is prohibitive or not (including possible legal action to Tesla selling FSD and promising HW3 will be capable to support it)..
@@денисбаженов-щ1б Semi was planned to have 4680 cells from the beginning. 4680 cell development took years longer than expected. Finally Tesla decided to bring out the initial Semi with 2170 cells. Higher cost and increased weight penalty, but they couldn't wait any longer.
@@денисбаженов-щ1б You are wrong.
FINALLY someone that understands the problem with radars and how they are very much not created equal. It was fine for autopilot but an active detriment for FSD or even just getting more advanced features or AP unless the camera vision stack started disregarding inputs from the radar which at this point makes it sort of useless.
You have such good timing! So many times I take a break and want a bit of quality 'TV' and see a Munro post within the last 10 or 20 minutes. Good Job!
Cory, Dave, you guys make watching technical detail stuff like diving into HW3 and HW4 differences interesting, even for someone who has no knowledge of car internals. Thnx to the team at Munro making very interesting observations and showing design decisions I have been looking with a total different eye at my own car . Thanks for all your excellent and detailed critical reviews that you brought to us. I would never image siting through hours of video’s of guys showing me car bodies and how well or not so well they were put together. I especially liked what you showed how getting rid of a few small parts that cast a few dollars end up saving the company millions of dollars. And we’re talking just about the part cost. But leaving out the handling and manufacturing of these parts will make out an ever bigger cost saving and even time saving. This brings Elon’s mantra “ the best part is no part” even more home.
More Dave videos, please! He's a great teacher. Thanks.
Thanks, Dave! That was super informative, with a lot of great insight.
Glad it was helpful!
44 years engineering in big auto electrics world wide and at last, my kinda channel.
Great job Dave. Phenomenal information.
Excellent work as always from Munro team. Much appreciated.
Our pleasure!
Clear. Concise. Technical as needed. Thank you Cory and Dave. Wonderful brain trust. The World is fortunate - because you share.
Great presentation. Could you do a show about the technical differences between the old Tesla radar and the new Tesla radar? For example: What the new radar can see vs what the old radar can see?
I want that video too!
great videos as usual, nice to see another member of the Monroe team.
Much appreciated!
Great closing. Love it when engineers respect each other and the work they do for the client. Keep up the great work, we love seeing the value you guys provide to the community all over the world.
Wow!!! Wonderful presentation. More Dave and Cory thanks Sandy
Is this is going to be a free upgrade to everyone that purchased FSD on the 3.0 hardware? Of course it is! I'm sure the Tesla cult members will be disgusted that I even asked that question.
Whos that guy on the video with Dave?
Thanks for the great video Dave!!
Thanks for answering the question at the end about a retrofit for the current owners of Teslas. Great presentation. The future upgrades for FSD looks exciting with potential improvements.
Dam- as always another #masterclass. And.. more Corey / Dave duo please! First time I’ve watched Dave here and much like the whole Munro crew- just a true master at his craft. Not a single media channel even comes close! To steal a Elon quote- second place needs a telescope to catch a glimpse of you guys. Just awesome!
“Windfall for Tesla “. Is that the shareholder - or the owner of the vehicle ?
What’s the second hand value of your FSD now that it’s not going to deliver as previously ‘suggested’ ?
❤I’m a sucker for a clear concise slide presentation.
Thanks! Amazing stuff, we can always count on Munro to be informative and interesting to watch, thanks!
We try our best, thank you for watching!
Interesting
all of these non-sandy videos have been getting even better!
Excellent review of changes, much appreciated! Wouldn't be surprised to see they moved to 7nm process, since the phone and gpu industry has left that node, it's got to be cheap from Samsung.
So I guess Munro has a new Y/3 to pick up in the future lol. I wonder how long it'll take to trickle down from the S/X though
They word is "All cars produced in the summer", so Q3?
@@rogerstarkey5390 whose word said "summer" ?
with more memory, it's safe to assume the memory bandwidth was also increased. Looking at the pastebin, HW4 has 2 ethernet ports and list 23 FSD sensors
You probably already have this thought.
My thoughts on the more RAM is
1. Higher resolution camera= more needed memory
2. Often sometimes when you redesign something, the older version of chips is going to be end of life soon and so it makes sense to plan new items with the current version so it's supported for 5 years or so. Also, it may actually be cheaper.
Thanks guys. Dave you did a great job. Thanks for breaking that down for us! Super valuable
Thank you for watching!
11:40 Its not 16Gbyte but 16Gbit. Thats 2Gbyte configured as 512Mbyte/32bit and its slow DRAM and obsolete.
Any plans for EV charger tear downs?
We will be doing an Home EV charger teardown series soon.
Great video.
Was wondering.
Can someone take out the ai system and build their own mini tesla?
I see a lot of ppl trying to build a mini (powerwheel size), programming everything themselves.
Instead of figuring everything out on your own, just install the ecu?
Maybe more complicated than it sounds lol.
Great video
Great to see another Munro expert on the channel. 👍🏽
I feel bad for all the Tesla buyers who paid for FSD believing that they would get autonomous driving. My bet would be that even Hardware 4 is not enough to get there. As Cory and Dave pointed out, it’s the next step. When you buy a Tesla, don’t pay for any options that aren’t in the car when you receive it.
Dave is awesome!
Some manufacturers should make kids car with ADAS, if full self driving becomes real will we still need DL( driving licence) to drive them?.
Thanks for this interesting video.
By the way: What's the story behind the olympic rings?
I think since I paid full price for my car and full price for my FSD, but my car does not and probably will not be able to get retrofitted with it. I think Elon should give me FSD FREE for my next Tesla I may get.
Thank you Dave and Cory, but especially Dave, for the insight and breakdown of the HW4 system. I don't come from an engineering background (but from a lower level tech background) but this channel has become my go-to for learning all kinds of engineering and is always a pleasure to watch. Should be required viewing for young and old aspiring engineers alike. Big thumbs up!
160 watts idle consumption is notable. For context, when I am being mindful I am able to get one mile for 210 watts in my Model 3.
160 watts maximum, not average. However your point is well taken. Really don't know what average will be.
I had the same initial impression, but over an hour of city driving at 30mph you'll use 6300 watt hours of energy (30mph * 210W/mi) which is comprised of both the 65watts of HW3 FSD, plus 6235 Watts for the drive motors plus everything else. If the new computer actually uses the full 160 watts, then the example of 1 hour of city driving increases to 6395 watts (160-65 = 95 extra watts for HW4) or about 213 watts/mile. At highway speed the slight increase in computer power is even a smaller fraction of total energy usage. Not too bad for better FSD performance
Love the video! "Unique New York" is an old acting exercise btw. Anchorman was using that as it's a standard pre-show warm up :)
I think they should remove door handles and then I'll sponsor a 3rd party door handle alternative...
The Joy of watching intelligent people speaking on interesting things. Glad to be here.
love this dude.... more of him.. he brings no BS and facts... also he did not serve me an add.. (cory)...
15:30 When they removed the radar, Elon said (June 2021) that "Vision has become so good that radar actually reduces signal/noise." He later (Feb 2022) said that "Only very high resolution radar is relevant." One year later, here we are with the relevant technology to improve the signal-to-noise ratio.
Great video. Thanks Cory and Dave for sharing details of the new Hardware HW$.
Cory looken fit
Great job David, you were very well spoken and informative, I hope we see you again.
5th haha. Hope everybody that reads this attain your goals, and also get a Tesla too.
Dave Great video and insights. Munro is truly the gold standard and offers excellent insight and information! Thanks very much
Wow great 👍 More of Dave please 😊😊
Could we see an in-depth analysis of the various charging stations, as this is turning out to be an important factor in choosing ICE Vs. EV, and one EV against another. It would be interesting to see how these units differ internally both individually and as part of a network of chargers, and how they interact with the cars they supply power to.
Thanks Corey and Dave, great info. I saw some chatter on social media about HW4 and I'm so glad to have a Munro Live video on the topic so I don't have to waste hours sifting through uninformed opinions!
Our pleasure!
Dave you are an amazing guy!! Notice how he looks down when Cory praises him.
I don't know why but this was quite a weteyed moment for me. This shows the Munro work enviroment!!!!
Dave rocks!!! Thanks for the video.
You bet!
This was extremely insightful. Thank you Dave!
Glad you enjoyed it!
Thanks Munro!
Fascinating piece of tech and very nice explanation from Dave!
My pleasure!
I normally dont comment. But he was cool. I liked that presentation. Keep up the good work :)
Thanks, will do!
Nice to hear from another HW/SW nerd. Always bigger, better and faster. Also nice to hear about using Linux as the Basic OS.
Linux Device Trees! ❤ A software readable spec sheet enabling the kernel to use the board. But humans can also read it, making it an object of great value during reverse engineering.
Device Trees is an enabler of a common Linux kernel version across multiple different hardware versions, making the software modular, easier, and in the end cheaper to develop and maintain.
I don't know how one would be agile using Green Hills RTOS, but if anyone wants to compete with Tesla / iterate as quickly, they probably need to have a Hardware Abstraction Layer, and an update strategy using continuous delivery that decouples OS and electronics to a large degree. Legacy auto doesn't seem to get this.
I'm highly interested to hear Daves perspective of Green Hill RTOS vs Linux DT with respect to the OODA loop of iterative development.
Spectacular content, thanks! Just knowing it's Linux is fascinating, better understanding automotive electronics packaging equally cool. All the best
Imagine paying the price of good used car just to have full self driving that is always promised to be one year away but is never realized. Meanwhile Tesla keeps improving the hardware tech, but never actually achieves full self driving.
Could you do a episode on the differnt communication protocols and how they reduce wire size and amount? I likes davids analysis on this.
Thank you Munro team great informational video! 😁🙏🏼🙋🏻♂️
Our pleasure!
After this BS Tesla MUST offer those whom purchased FSD to be able to TRANSFER it to a future Tesla purchase should they choose to do so.
Really enjoyed this. Thanks to you both.
16GB of RAM total of HW 4.0 but what is the SSD capacity? Of course could disable redundancy in the future like they have with 3.0 to utilize the power of both nodes to effectively 32GB total RAM.
More Dave, please.
"60% of the Time, It Works Every Time."
@ 1:18 _"Talk to our team"_ You might have been right the first time ;)
Inspiring work! Not part of the industry, but still entertaining and educational.
A wonderful watch from the UK. Would love to watch a live show of you guys in the UK. Good stuff Dave.
Great job of showing the inside OUT💥
Dang what a awesome surprise was wanting to take a peak at hw4 for some time now love the work y’all do. Also yea tesla for L they are amazing.
Great job Dave!
Great to meet and learn so much from Dave. He's super knowledgeable and a nice guy. Cory was great as always. All of you guys at Munro & Associates make a great team including the head honcho himself, the much admired and respected Sandy. Keep up the great work.
Are you guys going live on march 1st for the Tesla’s announcements
Working in IT I very much appreciate the technical breakdown. Dave is great! Not all geeks can explain things well
11:30 16Gb so 2GB per chip *8 so a total of 16GB of 4.0 per redundency
That was a really informative video, I hope Dave improves his health a little to be more comfortable and to keep providing his great knowledge.
Great Video, was hoping to hear some of my Buzz words: ASIL level(s), Software Bill of Materials (SBOM), Neutrino or Muon Kernel, QNX, Blackberry, Samsung Fab, Hypervisor, Security, Privacy. TIA.
Marie Osmond's 7 Favorite Nutrisystem Foods for Weight Loss · 1. Chocolaty Fudge Graham Bar > · 2. White Cheddar Popcorn > · 3. Rotini & Meatballs >
So the bottomline is once the FSD software outgrows the HW3 hardware (which it apparently has) to take advantage of the capabilities of HW4; just go out and buy a new car. Problem solved. 🙂
Thank you very much for the comprehensive summary on HW4. As to the memory: With internal bump of NN Cores from 2x to 3x per Chip plus also a bump of the clock on those cores I estimated an increase from 72 TOPS to ~132 TOPS raw performance. The internal SRAM is now (likely) from 32MB increased to 48MB. So desire of increasing external memory bandwidth I can understand very well. But from 68GBit/s to 4x56Gbit/s is not necessarily a good sign. Why: The biggest claim on efficiency of the FSD1 was the reduced need for external memory bandwidth. So, this seems to be at least "relaxed" as a paradigm ... but with the price of higher power consumption.
If they bring back Radar, does that mean the USS might come back?
I’m actually annoyed they’re bringing back radar since they have a huge chunk of their fleet without it. Like wtf Musk…
I am just so glad I didn't waste my money buying FSD when I bought my model S, I went for the upgrade to FSD3 HW because the SW changes made the FSD2 HW useless, but clearly the statement "all model 3" will come with FSD HW that supports fully autonomous driving" is way off the mark.
Don't get me wrong, they've done well to maintain upgradability, but HW4 suggests ALL of the earlier built cars are on a dead end evolutionary path.
Expecting new model y LR with HW4 . Superexcited. I hope model Y will also have RADAR
Hi Dave, you mentioned BroadR-Reach used for connecting the HD Radar. Are they using Broadcom automotive Ethernet chip for the 100Mbps twisted pair BroadR-Reach connection?
I guess even if we have HW4 in the 2023 Tesla vehicles coming out right now, there is still no radar and I will find out when I get my new Model Y long range in a week or so how many cameras does it have and their resolutions.
Probably the 2024-2025 Tesla vehicles may adopt radar back into the mix for those that are doing FSD or commercial efforts, but at what cost. Looks like the HW4 system has more power, but all the external bells and whistles are not there yet.
Time shall Tell where Tesla is going with their vehicles and software. They seem to be several steps ahead of the other manufacturers.
Thanks for showing the HW4 architecture. Take care.
2040 teenagers pull up at light. Roll down the window. What kind of processor you runnin?
Should call it HW3.1. It connects a few more dots in the aspirational path to a safe and efficent autonomous driving. Why bother with dual simplex when modern ARM A76AE cores allow lockstep protection for a single safety processor. More performance, lower power, lower cost, smsller footprint, prospects for backwards compatibility. Nice to see T1 ethernet finally in the design for the radar.. But all their cameras need to go T1 for lower cost video over ethernet.
wow. too bad hardware 3 can’t be upgraded to 4---> wires would be wrong length and space issues. I told my son to wait for HW 4 before he buys a Model Y===> His patiences will be rewarded. Thanks Cory and Dave
Disagree for the board for the infotainment. I will be playing games all the time that flu is a game changer with it for all!