The AMD chipset strategy is starting to become clear. I was wondering why they only released the 870 with Zen 5, since 870 is 670 with a different license. There was plenty of talk of an 880 chipset, all last year. When it was released with a repackaged 670, that took me by surprise. Stix Halo has four memory channels. We just found the new IOD. With quad channel memory, the 9950X3D will be the monster that eats even more of Intel's lunch. The real question mark for AMD right now is RDNA 4. They did the right thing to hold back the release during CES. There is plenty of drama to come out of nVidia and some of it will not be positive. Once that blows over, they will slide in with RDNA 4. Also, this could easily mean Zen 6 will be pushed to mid 2026. AMD can release Zen 5.5 in late summer with the new IOD and milk that hard until the spring of 2026. That will provide time for either TSMC to sort out N2P and get the ridiculous cost under control or for Samsung to bring their 2nm process fully online. I really like AMD's strategy. Lisa Su is an absolute rock star.
I see this as a phenomenal laptop and micro PC product. It will sell really well there. In it's current form; however, I am not seeing how this product, with it's emphasis on extreme bandwidth sharing between the CPU and iGPU, will translate into a high end desktop I/O chip. Lots of silicon used for an iGPU that will mostly go unused. Perhaps an iteration of this will be used for Zen 6?
Investors do not appear to be as enthused as we are, TomB16. We're pretty much at 52 week lows. I don't get it. I do not own any AMD for this very reason. They are a red-headed stepchild, and no matter what great things they do, they do not get rewarded.
AMD uses a chiplet design. CPU/GPU/IOD/NPU are all independent chiplets. The key will be to mount a pair of 8 core CPU chiplets on top of a quad channel IOD and possibly add an iGPU, if it suits them. Given the limitations of the current IOD, this should be easily good for 10+% performance gain to their marquee CPU complex. That's a generational uplift and one they can easily productize and sell as a new product. To be clear, I don't think a new IOD is going to help a 6 core 9600(X). I doubt it will help the 8 core to any significant degree. At the 12 and 16 core level, I expect there will be value. Because of this, I expect a late summer (approximately) product introduction of a faster 16 core and possibly 12 core CPU.
Imagine if you were Intel, struggling to catch up, and AMD has a 10% uplift in their back pocket, waiting to deploy.
For what it's worth, Zen 6 is intended for the N2P process and 16 core CCX. At that point, we will see 16 cores with 3D cache and 32 core CPUs on the top end. That will be little more than one year away. What's more AMD is not pressed for time so they have time to develop Zen 6 into a world beater. I would go as far as to say, if Zen 6 isn't a strong uplift with few to no bugs and a smooth launch, AMD will be dead to me.
I was going off the AMD supplied graphic shown here: Looks like you have the two CPUs and the iGPU/NPU integrated in with the I/O chip. I pulled that from here: LINK If that is indeed the case, then I am not sure that this particular chip is what I would want AMD going with for a high end desktop. Too many wasted transistors on unimportant features.
I am convinced that TB16 and RTN should start their own chip company. I enjoy reading your posts and thoughts about it.
The more I read, the more I am convinced of two things: 1. Strix Halo, as it is right now, is strictly for mobile and micro PC purposes. It will absolutely thrive there. 2. The advanced interposer used in Strix Halo will be used going forward on the desktop, as it will allow AMD to maximize performance on the (in my opinion gimped) AM5... and hopefully starting with Zen 6. The idea is to drastically cut down on latency and bandwidth between chiplets, and given the constraints of the AM5 socket, I'd say that's just about the best AMD can do without moving to AM6 before they said they would.
Nah, if I knew as much as actual engineers do, I'd overdesign everything and we'd go bankrupt in no time
I have a fantastic marketing idea for AMD to include a slice of pizza with every GPU. Gamers and coders are always starving. TSMC has a gymnasium sized room full of wafer ovens. All they need is some flour, a little yeast, some tomato sauce.... I'm pretty confident I could get AMD to 100% market share.
With Arrow Lake refresh, Intel promised performance would improve as they sorted out their microcode and drivers. Arrow Lake R performance has regressed, slightly. Meanwhile, AMD has taken a couple of noticeable steps forward. AMD is now winning benchmarks they've never won. AMD has hit a performance level at which it is impossible for nearly any benchmark to be written in a way that let's Intel win, without going to jail for straight up fraud. Intel still legitimately wins in a couple of niches but AMD holds all of the broad categories (Gaming, content creation, productivity, etc.) The purpose of this post is to speculate this AMD benchmark dominance will swing hard and become tremendously more pronounced. This will send AMD stock to the moon. Intel has dominated the industry with excellent hardware but they have also paid board makers to ONLY offer their high end boards with Intel sockets. They even paid motherboard makers to reserve the cool colors (ie: black, red) for Intel only. I remember when AMD boards were green, yellow, and weird citrus colors. Also, anyone who understands the benchmark industry can see it is fundamentally dishonest. There are some objective, open, benchmarks but many are not. It's not just the benchmarks, it's also reviewers. When AMD starts winning, those benchmarks are dropped from the testing schedule. It's getting really difficult to do that, now. There are only so many creative names to call a single threaded memory bandwidth test before it's obviously the same test. Now that Intel has run out of money, there will be less motivation to skew results in favor of Intel. What's more, now that AMD is making huge profit gains, I expect them to start behaving like Intel used to. In other words, AMD will pull out a 10~25% performance lead and benchmarks will show their CPUs as twice as fast. Anyone who thinks Intel just needs a process node step forward to get back in the game does not understand the industry. Intel can be saved, and it might happen, but it's a long shot at this point. AMD is poised to absolutely dominate.
