I'm breaking this out of my blog because it is specific. Please feel free to counter or present alternative approaches. The vast majority of text on the Internet is total crap based on previous total crap. People like to pretend to know more than they do. Bad information and bad ideas propagate like syphilis in the mediasphere. How then to filter? I start with the idea of disregarding EVERYTHING. I ONLY onboard credible information. If there is no credible information, I move on with zero information. When I started following Tesla closely in 2014, there was a lot more good information than 2016. Hot topics have far lower good information ratios. In late 2017, there were many dozens of media streams reporting Tesla stopped production due to zero demand. It's a song they would sing countless times over the next several years. I found a guy on a sub Reddit posting pics of multiple semi trailers being loaded with cars, every day. There was a filter decision point. Did I trust 99% or the one guy who was just a guy like me with a camera, taking pics on his way to work? That's when I bought a few hundred more shares of Tesla. Conclusion: One guy who seems objective and has actual information trumps reports that are vague and do not cite sources, regardless of how many of them echo each other's made up garbage.
Fast forward to January of this year. I've been following some interesting activity in the CPU/GPU/NPU space. There is a ton of BS floating around. 100 Million words and all crap, IMO. There are a handful of people who have leaked information in the past and have excellent to perfect track records. These folks have a common posting style. Their comments are brief and specific. When I see those posts on Twitter or in a forum and I know they have been correct in the past, that makes them statistically far more likely to have valid information now. I have put some thought into how many of these people have direct involvement at AMD/Intel/Nvidia and how many have just been lucky. Direct involvement doesn't mean they work at those companies but it does mean they have either seen the designs and parts or they are somehow privy to communication at these companies. Some of these leakers are probably asked to provide the information they do by the companies themselves. Leaking is a way to manipulate the market. At the least, I expect tech companies are aware of the leaks and could stop them if they wanted. Specifically, I'm talking about the guys who have made 15 posts on a particular product line over 5 years and all 15 have been spot on. So, it's a statistics game. From there, much of the media BS becomes transparent. When I read Zen 5 will have a 40% performance boost, it seems near certain this comes from leaked reports that AMD's next gen AVX512 unit has about 40% IPC boost. Somebody saw the 40% boost number and was too stupid to figure out what it meant. Instead, they just blurted out that Zen 5 will be 40% faster. From there, other outlets cited the moronic number and now a lot of people are going to be sadly disappointment when Zen 5 comes out. AMD has publicly stated the 40% number is both wrong and dumb but bad ideas are hard to kill. So, filtering is really about statistics and cynicism. Trust no one. Try to verify with another actual source, not just an idiot parroting a number you are trying to verify. Basically, I try to be a journalist. Somebody has to. Certainly, few to no people in the media is doing the job properly.
I very much agree with pretty much all of this. Most stuff online is either drivel or clickbait nonsense. If someone is going to invest real money, they had better gather as much information as possible and separate the good from the crap. I was wondering about that 40% claim and whether it was pulled out of thin air or, as you said, a misrepresentation of a more precise claim. The AVX512 IPC boost will really help in the datacenter/enterprise market, but it will just about be useless for 99% of home uses. The 40% claim will disappoint those who take that at face value and not do their due diligence with online rumors. Having said all that, AMD is killing it with their CPU division.
My research method is not infallible. I've gotten it wrong the very odd time. Where I stumble is interpreting the research. Here is what I can gather about yield per company (I have no per fab data). Samsung claims 75% for their 4nm node. Rumors/leaks suggest TSMC N4P yield is roughly 70%. N4E is rumored to be roughly 80%. Intel 7 was well under 50%, a year ago. I can't find rumor/data on where it is now. The best 3nm process yield is usually cited around 60%. I can tell you every one of these numbers is a guess, at best. Actual yields tend to be a closely guarded secret.
