DX

1.7K posts

DX banner
DX

DX

@DaevidX

There is only one question of all questions. # Meaning of all life and matter #

🇸🇯 Norway เข้าร่วม Şubat 2018
197 กำลังติดตาม80 ผู้ติดตาม
DX
DX@DaevidX·
@Tslachan Only information you got from Elon is AI5 single chip is 5x compute equivalent to 2SoC AI4. Easy math means: 1xAI5 is 2.5x powerful than 1xAI4
English
0
0
2
57
Tsla Chan
Tsla Chan@Tslachan·
$TSLA Tesla's AI4 (HW4) features a dual SoC design with two identical chips for full redundancy and safety. In contrast, the upcoming AI5 (HW5) adopts a more powerful single-chip architecture with radical improvements: approximately 8-10x higher compute, 9x more memory, 5x higher memory bandwidth, and up to 40-50x better inference performance compared to AI4. Power consumption is expected to rise significantly, requiring enhanced cooling systems. The transition from AI4 to AI5 is anticipated to enable much smoother, more capable FSD with fewer hesitations, accelerate Robotaxi deployment, and boost Optimus robot intelligence, while strengthening Tesla’s advantage in custom AI silicon for vehicles and robotics.
Tsla Chan tweet media
English
22
66
378
17.9K
Alex
Alex@SoonyXboneUhh·
@RobGrieves Why just 150w in a car where you have space, massive amounts of power / ernergy and water cooling. Should run at 300 to 450 watt.
English
2
0
0
316
Rob Grieves 🇦🇺
Rob Grieves 🇦🇺@RobGrieves·
Let's talk about the AI5 picture. Memory From what I can gather from the grainy low res photo, those RAM chips appear to be SK Hynix H58G66DK9QX170N 8GB LPDDR5X with 9600Mbps bandwidth. 12 modules = 96GB @ 1.15TB/s The Die Size appears to be half-reticle (~430mm2). This gives it a yield and cost advantage over full reticle dies such as Nvidia's H100 (<800mm2). If we assume Tesla is using TSMC 3nm process, this gives the chip 108-125B transistors. Performance With that many transistors and memory performance, when constrained to ~150W such as a car or Optimus, we're looking at 2000-2500 TOPS which matches up nicely with a H100. Unconstrained, such as in a datacentre it could be much more. The Package This packaging is pretty cool. It includes memory on package which gives huge memory latency advantages than the typical memory on board config. To me, this amount of RAM packaged in this way is way overkill for a car. I believe we're looking at the datacentre version in this image. I believe for the cars or Optimus, we will see a traditional memory on board config (with less of it - e.g 32GB).
Rob Grieves 🇦🇺 tweet media
English
65
105
1.2K
109.9K
DX
DX@DaevidX·
@KRoelandschap @teslaeurope @Tesla_AI @niccruzpatane @wholemars @JoostJansen6 It's not about override. You overriding, your car will not break. Why FSD is unable follow defined street/road/highway speeds, by default? If is able, then it should. If for someone is better to drive slower, FSD should provide button to select driver speed profile.
English
0
0
0
8
Kees Roelandschap
Kees Roelandschap@KRoelandschap·
Everyone keeps saying FSD drives too slow, like a grandpa 😂 Challenge accepted. I just tested it. And yes — if you actually tell the car you want to go faster, it will. I take full responsibility for my actions. This is purely educational, to show what FSD is truly capable of and how you can use it. In a 100 km/h zone I pressed the accelerator to signal I wanted more speed. Even when there was a gap, I gave it an extra nudge so it wouldn’t merge right immediately. Basically driving like a dick on purpose, staying left until the road forced the merge. And the car handled it beautifully. Then came the best part: a flawless 4-lane change in under 1 km with medium traffic. Smooth AF! 🥵 Tesla’s FSD isn’t fully autonomous yet, but it’s an incredibly mature ADAS, already bordering on real ADS territory. Safety first: FSD always follow the rules. But if you decide to bend them yourself… that’s on you. Even then, FSD stayed calm, competent, and above all, very safe. This is how you actually push the system and see what it can do.
English
49
75
723
52.9K
DX
DX@DaevidX·
@teslaeurope Will FSD react on turn signals?
English
1
4
2
860
Elon Musk
Elon Musk@elonmusk·
Congrats to the @Tesla_AI chip design team on taping out AI5! AI6, Dojo3 & other exciting chips in work.
Elon Musk tweet media
English
6.7K
12.1K
117.2K
18.1M
DX
DX@DaevidX·
Fuck Unsupervised yet (if it's still not ready for any kind of HW) HW3 owners are eligible to have same rights for Supervised iterations as HW4 owners. So run and compile the lite version every 3 months. Problem solved. In Europe it looks like there's v14.x mandatory to run (v12.x is not designed to EU market, so new version is needed)
English
0
0
1
59
DX
DX@DaevidX·
@arangoool @DavidLi54813379 @NotATeslaApp Based on latest post from Elon, No 😄 He's restricting AI4 for cars and AI5 for supercomputers and Optimus. Unsure what we will get, since this post is few minutes fresh
English
0
0
0
12
Not a Tesla App
Not a Tesla App@NotATeslaApp·
Tesla has announced its Spring 2026 Software Update, and it's a big one. We're getting a new Self-Driving app, "Hey Grok" wake word, Pet Mode, and more — but still no CarPlay. Learn more: notateslaapp.com/news/3966/tesl…
Not a Tesla App tweet media
English
7
6
70
6.8K
DX
DX@DaevidX·
@SawyerMerritt We will save this post for future. Just, for, sure 🫶😄
English
0
0
1
44
David Lindsey
David Lindsey@DavidLi54813379·
@arangoool @NotATeslaApp Highly doubtful but would be crazy awesome if they stopped updates for HW3 and was just like the reason why is because they're just going to retrofit all to HW4 A man can dream 😂
English
1
0
3
65
DX
DX@DaevidX·
Some wishes from customers, maybe next time? 🫶 You should give more love to legacy customers where's possible (HW3/Ryzen). Propagating new sales by limiting old vehicles is bad, always. - FSD v14lite (really hope it's Q2) - Grok for rest of Europe (even only English version is OK) - Let Grok adjust basic car functions - improve in-car UX! - maybe some UI redesign - B pillar sentry (HW3/Ryzen) - add own phone screen mirroring function into car (when in P it's enough, for fullscreen media playback - photos, videos) - add ability to play next radio station from favourite list in Tune-in in-car App, by a wheel button (now is blocked) - add ability to change all car sounds, including volume where it's allowed (beeps, reverse, slow speed forward,..). - Improve Track mode (add temperatures, add 0-100 acceleration timer, 400m or custom drag timer, lap timer by selecting GPS location as counter point) - add actual altitude, add GPS coordinates (we really no need see Google logo) - new maps&data worth 2026 year! - fix map route planner - fix broken speed limit signs logic - add button to stop battery preheat function when destination set to charger (don't want loose 5% of battery if i have no rush) - start with Tesla ECHO in-mobile App - add in-car screen screenshot function (for troubleshooting) ... - and finally fix dry windshield wiping nonsense. @Tesla @Tesla_AI @teslaeurope @elonmusk
English
0
0
2
123
DX
DX@DaevidX·
@KRoelandschap Maybe it's 05/05 or maybe 30/06 or 06/10 or maybe 01/12 Don't say it's scheduled at 5/5 with no evidence 😉
English
0
0
1
170
DX
DX@DaevidX·
@Tslachan Maybe it's 05/05 or maybe 30/06 or 06/10 or maybe 01/12
English
0
0
1
30
Tsla Chan
Tsla Chan@Tslachan·
$TSLA 테슬라 FSD의 유럽 전체에 대한 확장을 위한 EU 투표(TCMV)는 2026년 5월 5일 해당 회의를 통해 진행 예정?
Tsla Chan tweet media
한국어
38
79
503
135.8K
DX
DX@DaevidX·
@vad3rt3sla No other company selling FSD unsupervised with HW3 and then realizing they can't run unsupervised model. Get the fact straight. No excuses. That's not about good will. It's obligatory to deliver!
English
0
0
3
111
Vad3r
Vad3r@vad3rt3sla·
Imagine your iPhone can’t be updated to iOS 26 due to hardware limitations, and Apple says they will do a iOS 26 lite version. Yeah, Apple would never. Tesla said they are working on FSD v14 lite. No other company does this.
English
55
6
119
14.1K
DX
DX@DaevidX·
@eu_bike @iamtobi They patented but just for repairability improvement. Still no chip change like a PC CPU 🙂 It will be needed when FSD model is in feature complete stage and model change requirement is once a year or more. Then they can make the chip cost very low with HW neural model into it.
English
0
0
0
22
Tobi Mülhauser 🍕
We need a retrofit to upgrade from HW3 to HW4 🙏🏻
English
143
56
889
186.1K
DX
DX@DaevidX·
@yunta_tsai Where we can write our tesla wishes?🙂
English
0
0
0
21