I no longer think there will be a mid cycle refresh of Zen 5 to give the gaming/office series the same IOD as Strix. Its not necessary and it would canabalize Zen 6. When I look back at what was said shortly after the Zen 5 launch, I believe a ton of people did everything they could to say negative things about Zen 5. It was always clear to me, Zen 5 was a nice uplift from Zen 4. Linux saw a nice uplift. A bunch of server benchmarks saw a nice uplift. Games mostly didn't see much of an uplift, although a few specific games were boasted nicely by Zen 5. We're at or near the end of the line for design gains on CPUs. There isn't a lot more water to squeeze out of that rock for traditional computing. AI, maybe. We are at or near the end for process gains. There isn't much gain left to gain from process steps and the cost of 2nm is ridiculous. The bulk of gains moving forward will be brute force. If you need more memory bandwidth, that will be had by adding channels. If we need more compute, that will be had by adding cores. This is why chiplet production technique is so important. Zen 6 will have 12 cores per die. The big gun will be a 24 core part. All performance cores. That's a lot of cores.
I agree with most of this but I have to add that I believe that this logic applies to silicon as a material. https://www.industryemea.com/news/64892-beyond-silicon-–-what-will-replace-the-wonder-material After silicon, it might be a whole new ballgame. I still stand by my initial statement that Zen 5 was held back by a crappy I/O chip.
If dominating gaming, content creating, workstation, and literally everything except a couple of benchmarks that are memory constrained isn't enough for you, maybe you are looking for an excuse to talk crap? They were number 2. Now they're number 1 and the lead is stretching out as Intel's microcode and various fixes have come at the cost of performance while AMD has improved performance a bit since launch. Somehow you're trying to frame that as a failure? How much does AMD have to win by before you'll call it a tie?
I believe I have made my point very clear previously, so I am not going to litigate it again. Perhaps you view my wanting AMD to be better than they already are as me talking crap, but I assure you it's not.
Look, my good friend, I want more too. We both understand how identical our interests are, in this regard. But, this is a thread about $AMD. AMD has checkmated Intel and appear poised to cut into nVidia's long standing GPU dominance. AMD's GPU gains are extremely important for integrated graphics applications, which is almost everything, these days. That's great for AMD's business and important to AMD. AMD, even on the cusp of a big step forward, is of only passing interest to hard core gamers. This crowd is hugely pandered to but does not move the stock price in any significant way. AMD seems poised to move from 20% share to 40% of this market but time will tell. From what I can tell, AMD is irrelevant in the AI world. They may play in the same league as nVidia but they aren't as good (based on what little I currently know) and it's a winner take all proposition. If I was looking to spend a half trillion dollars to build the biggest AI network in the world, I wouldn't knock on the door of the second best supplier. This is a case where whomever has a minuscule advantage in compute per Watt will take 99.9% of the AI business. It very much appears that GPUs will become irrelevant to nVidia's income statement within 24 months.
Wow! That's pretty heavy. I didn't realize there had been a lawsuit involving you about this subject!
I haven't been able to go swimming since the last time he and I disagreed. This damn ankle bracelet is really inconvenient.
If the massive number of leaks can be believed, RDNA 4 is a 40% uplift over RDNA 3. They have come a long way with ray tracing, also. This is a big deal. As nVidia stumbles with the 5K series paper launch, AMD is holding back RDNA 4 so they can position their parts as well as possible. At this point, it seems near certain AMD will comfortably compete with nVidia 5070 and down. They might even be competitive with the 5080. If AMD can capture 50% of the market from 5070 on down, that's a huge chunk of the GPU market. I believe it's possible but I don't think AMD can get to 50% of the GPU market. They will be lucky to achieve 40%. This is a case where market position will make all the difference on how successful RDNA 4 is. Although, I think we're at the point where AMD has what they need to roll RDNA 4.
AMD has a deal with Samsung to manufacture some of their 4nm class product. Transferring some production to Samsung is the only way to gain GPU market share, as TSMC has zero spare capacity to ramp more AMD product. I would point out, however, the GPU market is shrinking. AMD owns the iGPU market so gaining GPU market share will be a larger percentage of a smaller pie. AMD must surely be feverishly working on ROCm (AMD's answer to CUDA). This is a software project of epic significance. I expect AMD to catch up or nearly catch up to nVidia in the GPU space but I don't see a path for AMD to catch up to nVidia in the AI space. I'm not saying AMD can't catch up. I just don't see how they can, at the moment. Meanwhile, AMD is set to have early Zen 6 engineering samples. People are talking about Zen 6 but they seem to have forgotten that Zen 5 is not fully released. Zen 5 products still to come: 9950X3D - the new flagship. rumored to be 18% faster than the 9950X. 9900X3D - an absolutely tremendous APU for the desktop and the flagship for most of us who don't have motherboards that will support over 200W of power consumption. 9600X3D - this will instantly take the affordable gaming crown and might even be the best CPU available for some single thread games. https://videocardz.com/newz/amd-ryz...cpu-spotted-in-factoriobox-benchmark-database
Early indications are that RDNA 4 is competitive with nVidia Blackwell from 5070 down, including the Ti, and not that far from the 5080 in some games. To be clear, this will have no appreciable impact to AMD's bottom line.
nVidia has announced next gen products, perhaps to take the wind out of AMD's sales. I'm even more bullish on AMD. While Rubin will probably be a step forward, AMD has Zen 6 and UDMA scheduled to come online before Rubin is released. They are looking at roughly doubling memory bandwidth (50% wider bus and higher clocks) for Zen 6. Even if Rubin is really good, I don't see nVidia pulling away. I think nVidia will have to push hard to keep the gap as it is and even that is becoming irrelevant. The reason I say the gap is irrelevant is because the 5090 XT is the only GPU that can do high frame rate at 4K. The 5090 barely exists so it almost doesn't matter. Once AMD achieves high frame rate at 4K, it won't matter all that much if nVidia can do 240 fps at 4K while AMD can only do 200; there will be choice for high frame rate 4K. Right now there isn't choice but we basically don't have high frame rate 4K since the 5090 XT barely exists. This is a weird situation where nVidia is in the lead but the GPU race is AMD's to lose. If AMD executes well, they will dominate GPUs in one generation.
Im seeing a lot more people buying the AMD GPUs this cycle, their sales are going to look good for the Q
AMD is absolutely rocking this Q. AMD books are going to be accounting porn. We will know more in 6 weeks. GPUs will help polish AMD's bottom line and they will strip significant market share away from nVidia. Oddly, this will help AMD a bit but won't hurt nVidia on the short term. I've heard people talk about why someone would buy AMD or nVidia and it's always a load of crap. You go to Micro Center and your option is to buy an AMD GPU or to leave without a GPU. AMD did a very good job with RDNA 4 and that's a big part of the equation but the real story is that AMD has product in the channel while nVidia does not. UDNA looks like it will be a significant step forward. I am not scared about Rubin, in any way. The really weird thing in the GPU race is that it only matters for AMD. Five years from now, if gamers are running AMD CPUs with nVidia GPUs, that will be just fine. If gamers want AMD CPU/GPU, that's somewhat better for AMD and maybe nVidia will be making the brain of our digital overlords.
Word on the street is, GPU prices will not be going down any time soon. This won't affect nVidia, since they don't sell many video cards. I doubt it will have much affect on AMD, either. Even if AMD boosted their GPU parts by 20%, it would make a difference but not all that much. More importantly, RDNA is doing extremely well. This should pave the way for UDNA to take things to the next level.
Zen 6. Wow. TSMC N2X node. 6+ GHz. 48MB L3 cache per chiplet. LPDDR5X. RDNA4. 24 core (12C per chiplet). 10% IPC uplift. There is also talk of a new 3D cache chip that will be given a heavy aluminum top layer and will act as a heat spreader. Plus, the next 3D cache will be able to stack 2 high. That means, they could add 240MB of L3 cache per CCX, if they want to. Can you imagine a halo part with 480MB of cache? 20% clock uplift, coupled with 10% IPC uplift, and 50% core count uplift. That's just on the desktop. Not many people are going to need that but it's wild they can do it. Hopefully, the halo parts will enable them to sell a mountain of office PC parts. What can Intel come up with in one year that can possibly compete with that? Even with the fancy new RAM that will no doubt be clocked very high and the big cache, the AM5 socket is going to be a major bottleneck. They really need 4 channel memory for HEDC but I suppose they need to leave something for Zen 7. Oh yeah... midrage GPUs will be dead in 12 months.
I will add that N2X is a high clock frequency node. It is extremely believable they could hit 6GHz with reasonably low power. My AMD 9700X can hit 5.6GHz easily with reasonable temps (83C) with a small undervolt of -0.35V. Without the undervolt, it throttles to 5.5GHz at 95C. That is the boost clock, of course. I would do well with a better cooler but I like low heat, low power, chips so I'm happy with the undervolt and eco PBO curves. Actual power draw rarely tops 65W.
Does anyone care about SoundWave? I think it is either shipping or will soon be available to OEMs for handheld development. It should be a pretty solid ARM product with a really decent AMD GPU. N3P. I have every expectation it will stomp on Qualcomm. Perhaps I will also mention there are more Zen 5 parts to be released. With Zen 6, they are talking about releasing those products over the course of a year. So, my Zen 5 desktop will be obsolete in 8 months but a Zen 5 laptop might be good for 18 months before Zen 6 laptop parts come out. Those laptop parts will be on N2P, by the way.
