The recent batch of AMD CPUs are really really good! I know a lot of friends that are currently on a PC they built with the latest ones and they have amazing performance!
I think the hay days of x86_64 are soon over. I don't predict x86_64 will go away but we are about to enter the era of ARM. ARM is mature, cheap, and scales well. Ampere Computing just launched a line of data center CPUs with 64~192 ARM cores that I suspect will carve out a piece of the server market to the detriment of Intel/AMD. Intel and AMD are tooling up their ARM systems, also. I suspect they will have both data center and single chip ARM solutions on the market within 12 months, or so. AMD has been knocking on the ARM door for years but they've been scared to release the dragon that will kill their cash cow. On the other hand, I don't think ARM eliminates the need for x86_64. The problem is, x86_64 is a goose that can fly at mach 3 because it has had infinite resources poured on it for 45 years. If ARM starts soaking up all the R&D money, the gap could close quickly and we would only need one instruction set. I'm not sure how this will effect AMD. These massively paralleled compute platforms are made possible by CXX which AMD pioneered (they didn't invent it but they pioneered it's use). We can stack a lot of cores using CXX modules glued together.
I've been following process nodes for decades. I remember the brick wall that tripped us on the path to 64um. It doesn't seem all that long ago. As we got super small, traditional transistor structures stopped working. We had to use multiple transistors or new designs such as GAA (gate all around). I thought we were going to hit a brick wall at 3um. We are literally up against atomic limits. I just read an article in which Pat Gelsinger said there is no end to process shrink. He cited work being done on 1um and he could see a path to shrink indefinitely beyond this node. I assume he is referring to quantum computing. This is where I make the jump from having an idea to being an old man who begins all comments with, "In my day..." I'm just not up to speed on quantum computing. Because of my lack of knowledge, it is time for me to exit this space.
It's a shame TTT isn't still alive to see how big they've grown since he brought them to everyone's attention here at $2/share
I started following AMD when Jerry Sanders was CEO. Things were going extremely well Intel out competed AMD. If Intel hadn't been on top of its game, AMD would have done very well. I bought it below $2 and sold at a pretty significant profit, several years later. Suffice to say, if I had held AMD, I could be retired by now.
For what it's worth, AMD is set up to dominate a desktop refresh cycle and absolutely explode in the server space. There is noise they will lose in the console space but I think that is negotiation theater. 4Q24 should be good and all of 2025 should be amazing.
At this point, it seems certain that Intel will paper launch Arrow Lake at the end of this year. They will be lucky if CPUs start trickling out the door in January with the meat of the ramp in February. Meanwhile, at Nvidia, they have committed to gddr7 with their next GPU. I'm sure it will be formidable but the large size GDDR7 modules won't be widely available until mid 2025. On the short term, AMD is going to have 6~12 months of unanswered new products. Arrow Lake is planned for TSMC N3B node. If there are any bumps in the 3um ramp, this could end up being a huge problem for Intel. AMD Zen 5 will be produced on N4P. 4um is cheaper and has already scaled. With AMD's chiplet architecture, they don't need big die sizes. To me, this looks like a strategic win by AMD. I don't think there is much performance gain from N4P to N3B, either. The real bump will occur at N3P. Yeah, I see AMD running at full boil by the end of this year. 2024 should be very good with 2025 even better. Even if Intel and Nvidia respond with strong, great selling, platforms, it is going to take a while to turn the tide away from AMD.
Regarding CPUs, AMD is kicking Intel's ass. They earned this, and I hope they make a ton of money from Zen. Intel should be ashamed for their decade+ of complacence and their current failure to make a better, more power efficient product. Regarding GPUs, AMD has a long road ahead. Nvidia is outselling them at least 3:1 in the consumer market and more so in the enterprise market. I love competition for the consumer, and I want AMD to well here because I really like Lisa Su, but it's going to be very hard for AMD to make inroads here unless they can pull a rabbit out of their hat. Nvidia is not messing around like Intel did all those years.
