avatar
最爱吃兽奶的兔🐰
408 Twitter followers
Follow
2018 年入场的小韭菜一枚｜ETH holder｜纳斯达贷还款者 | @theNextDAO 🦊｜DCA 重度爱好者｜ENTJ｜大厂后端资深研发工程师｜跆拳道黑带 & 古典舞十级选手 ❀
Posts
avatar
最爱吃兽奶的兔🐰
03-30
Thread
Can Google TPUs Shake Nvidia's Footsteps? I listened to an interview with a Google TPU engineer on Silicon Valley 101 while working out this weekend, and it was quite interesting. After listening, I gained some new insights and reflections. I've summarized the core points of the original interview and recorded my own thoughts. 1️⃣ Conclusions from the podcast (A more detailed AI text summary is also available; see Figure 2 for a longer version.) 1) Gemini has a counterintuitive characteristic: the more people using it, the faster it becomes. This is determined by the TPU architecture. Parallel computing + a reused caching mechanism achieves maximum efficiency when the computing power is fully utilized. Of course, this is a double-edged sword. Last year, with the release of Gemini 3, a large influx of GPT users caused frequent service crashes… The root cause was that TPU production capacity couldn't keep up; expansion couldn't keep pace with user growth. 2) Nvidia GPUs vs. Google TPUs: each has its own competitive advantages. Nvidia's advantage lies in software: the CUDA ecosystem is too mature, highly versatile, and difficult to modify. Google's TPU's advantage lies in its hardware-software integration: when running specific large-scale model algorithms, its performance directly surpasses Nvidia's. Apple has become the largest buyer of TPUs, and Anthropic is also purchasing them in large quantities. The reason is simple: they don't want to put all their eggs in one basket with Nvidia. 3) TSMC's moat is deeper than imagined. Whether it's Nvidia, Google, Apple… all chips are made by TSMC. No other company can replace TSMC's process technology and yield rate. Even more outrageous is that TSMC's production capacity is simply insufficient; major manufacturers are lining up to compete for it. Selling shovels is a sure thing, and I will firmly believe in and hold TSMC. 4) TPUs are similar to putting algorithms into hardware, somewhat like mining with ASICs. GPUs are general-purpose; TPUs are custom-designed. Google's chip team needs to place orders with TSMC one to two years in advance, meaning that the chips they are making now are designed for AI algorithms two years from now. Therefore, Google's AI and chip teams must be deeply intertwined. The algorithm direction they bet on today determines whether their chips will be usable two years from now. A correct bet is a game-changer; a wrong bet results in a complete waste of time and resources over two years. This constraint is not something Nvidia faces. 2️⃣ Some personal thoughts The so-called hardware-software integration means that if general hardware design capabilities are insufficient or too costly, then they simply design hardware specifically for a particular type of algorithm. In terms of hardware design capabilities, they cannot compare to general-purpose hardware. Why doesn't Nvidia develop NPUs/TPUs? Because they can't expand their advantages. Too many manufacturers can do it; if everyone does it, it'll be like the mobile phone manufacturers. For example 🌰: Almost all Android phones have infrared remote control functionality, i.e., universal remote control functionality. Why doesn't Apple do it? Is this function difficult? No; Do customers need it? Yes. Then why doesn't Apple do it? I think the form of a product may not necessarily be "entirely determined by demand." If two sets have a higher proportion of identical elements, they become less distinctive. Think about our impressions of mobile phones: Apple/non-Apple. If using Android, most people don't care about Xiaomi, Oppo, Vivo, or Huawei because they are too homogeneous, resulting in relatively low user stickiness. Analogize to the difference between a calculator and a computer: a computer performs general-purpose calculations, while a calculator performs specific calculations. A calculator naturally consumes less power and can only perform specific calculations. TPUs don't have hardware design barriers compared to Nvidia's GPUs. They're simply trying to reduce costs. At the same time, very few companies make huge profits by reducing costs; only those that propose new concepts and open up new directions can take off, and these are also easier to hype up. Current neural network calculations are mainly tensor calculations, and converting costs is a last resort for Nvidia. Therefore, we can't look at TPU's claims of advantages in isolation. From Google's perspective, developing TPUs is the best choice. However, from Nvidia's perspective, engaging in a cost war is a last resort. Because Nvidia has already secured a first-mover advantage, its CUDA ecosystem, and its top-tier GPU design capabilities. Can't Google make GPUs? Definitely not. If Moore's Law can do it, how could Google not? With a little more money, they'd be able to poach talent, right? The initial investment in chips is enormous. If Google starts making GPUs and nobody uses them, they'll face a situation where they can't even recoup their costs. At the same scale, Google's cost to make TPUs is lower than GPUs, and they can also bundle and sell their own Gemini, which is the better strategy.
TPU
0%
avatar
最爱吃兽奶的兔🐰
03-13
Thread
ByteDance laid off all its employees in Wuhan because of AI? Stop spreading rumors… AI is being unfairly blamed to this extent; even AI itself didn't expect this 🤣 The core reason: ByteDance has recently been addressing the "silo problem." What is the silo problem? ByteDance's bases are currently geographically dispersed, located all over the world. Sometimes a team of 10 people might be scattered across 10 countries, leading to extremely high communication costs. Although remote work is possible now, anyone who has truly collaborated across multiple departments knows how difficult cross-regional, multi-departmental communication is… It's practically Schrödinger's cat alignment 🤡 According to internal sources within ByteDance, this move in Wuhan has nothing to do with AI. In early December last year, there was an internal notification that by March of this year, all ByteDance employees in Wuhan would be relocated to first-tier cities, to follow the general direction of silo management. A friend of mine was interviewed for a position in Beijing, but he didn't want to go. He insisted on being based in Hangzhou, or he wouldn't accept the offer. ByteDance even created a special position for him in Hangzhou. This restructuring isn't about layoffs; it's either relocation or reassignment. Only if nobody wants them will they receive a major incentive package. Similar to the rumored Wuhan department, the team my colleague previously worked on in Hangzhou followed the same process. ByteDance insiders believe this is a good thing, because the fragmented organizational structure led to inefficient communication within ByteDance, unlike Pinduoduo. All of PDD's R&D teams are located in Shanghai, in one building, resulting in highly efficient communication. Therefore, ByteDance has recently been addressing the issue of isolated teams causing communication difficulties, this is part of an internal organizational optimization plan that was already in place… Don't always blame AI 🤡 twitter.com/0xMilkRabbit/statu...
avatar
最爱吃兽奶的兔🐰
03-13
Thread
Get a Gemini Pro account directly from Xianyu (a second-hand marketplace), cheap and plentiful! This is another must-buy deal that costs as little as two cups of milk tea! My previous post about getting a Gemini 3.1 Pro for a year for only 50 RMB was very popular 🥰, but some people were worried about security issues, such as account suspension or temporary expiration. Here's a clarification: The only risk with Google One is that the seller's account might be suspended by Google. They don't know how they got the account, maybe through student discounts or credit card fraud. However, there's no risk with the invited members. If they're suspended, only the main account will be suspended. After the main account is suspended, dissolving the family account is easy 😆, no need to worry about privacy. I often buy iCloud storage shared by family members from Apple, and it's been working fine for five or six years. The after-sales service from the seller I bought from is pretty good. With normal use, I'm not worried about not getting my money's worth. Several of my colleagues and I have been using it for two months now, and it's fantastic 🤤 And some users have commented that if the account expires, reliable sellers will re-invite you to continue using it. If you're still worried, group member @MMMusol recommended a direct top-up method. For those who prioritize long-term stability and are concerned about frequent outages, this is worth considering 🙌🏻. 36 RMB gets you a 3-month official subscription. It's a bit more expensive than the previous 50 RMB for a year, but the stability is a major advantage. However, it's still a great deal compared to the official website's 140 RMB per month! Payment Method: Direct top-up to your Google account, the whole process takes only 5-10 minutes. Operation Process: 1) Search for "Gemini 3.1 Pro direct top-up/gift card" on Xianyu (a Chinese online marketplace). 2) Choose a price between 6u and 7u, with a duration of 3 months (gift card direct top-up). (It's best to choose L3 or L4 for the seller) 3) After purchasing the product, the seller will send you a gift card link. 4) Link the bank card and address provided by the seller to your account. If there are any errors during the process, inform the seller. They will handle it promptly (reputable sellers with good after-sales service will provide excellent service). 5) Linking the gift card to your account completes the process. One option is a 50 RMB family invitation plan that lasts for one year, cheap and plentiful, but there's a possibility of being banned before the year is up. Another option is a 35 RMB plan that lasts for 3 months. The one-month account top-up option is more expensive but relatively stable. Purely ad-free sharing. Which option you choose is up to you. Both are great! twitter.com/0xMilkRabbit/statu...
L3
0%
avatar
最爱吃兽奶的兔🐰
03-12
Thread
Everyone says humanities students ask Gemini, and science students ask GPT. So grabbing a Gemini subscription on Xianyu/Taobao is a must! 🤤 For the price of two cups of milk tea, you can experience a year's worth of Gemini Pro AI. This includes access to Gemini Pro + NotebookLM + Nano Banana Pro. You read that right, a year! After all, the original price was 140 RMB for just one month. In terms of user experience, Gemini 3.1 Pro responds incredibly fast, much faster than ChatGPT 5.4. However, its performance is still slightly inferior to the ChatGPT 5.4 Thinking model. But it already easily outperforms various free basic models. And the Nano Banana Pro is incredibly useful for graphing! 🤤 Compared to the original price of 140 RMB per month, the current 50 RMB is a steal if you use it for more than half a month. This value is so tempting that even those in the next group who only use free AI are envious. Every little bit saved counts in a bear market, making this perfect for those who want to experience advanced AI but don't want to spend a lot of money. Operation process: First, you need a Google account. Search on Xianyu (a Chinese online marketplace) for keywords like "Gemini Pro annual membership" to find a reputable seller. You'll receive a family invitation verification email in your Gmail account; click to join. Family invitations are legitimate, and Google email accounts don't require real-name verification, so there are no privacy issues. 🥰 During a bear market, this low-cost, high-value tool is very convenient. I've been using it for two months now; it's a great experience for a small price (pure user experience, no ads 🫶).
XNO
0.88%
loading indicator
Loading..