I never really overclocked even back when it had more tangible benefits. Now everything comes pre-overclocked to within an inch of its life. The good old days are over, but all the automated clock speed control is good for people like me I suppose.
Zen 5 has progressed from engineering samples to production sampling. I had a dream that Zen 5 would be here in July but now it looks more like August. That is, unless reports of Zen 5 sampling are stale. The first Zen 5 are expected to be laptop Strix. It looks like the full Zen 5 line won't be online until the end of this year, at the earliest.
I see great things for AMD on the short term; perhaps until mid 2025. Longer term, Qualcomm's Snapdragon X Elite is an extremely impressive CPU for anything run off a battery. I don't know if Qualcomm will take over the entire industry but even a small bite of the mobile market will change the industry dynamic dramatically.
Tom's Hardware published an article citing Zen 5 trailing Raptor Lake 14900. Other sources cite Zen 5 being a 20+ percent boost over Zen 4. Since Tom's Hardware has been wildly dishonest in the past, I find the IPC boost leaks more credible. At some point in the near future, reality will be revealed and most web sites will continue publishing the same misleading claims.
I just saw a leaked CPUZ benchmark (WCCFTech) showing a Zen 5 engineering sample beating (but essentially identical) to Intel 13900/14900 while beating 7950X by . The CPU was operating at 5.8GHz. Let's unpack that. - Zen 5 is said to not do well on CPUZ while scoring very well on pretty much every other test - This benchmark could be fake - This is an engineering sample - This sample beat 7950X by roughly 20% - This is almost exactly what we have been expecting Other thoughts. - There is some evidence the fastest Zen 5 will be memory bandwidth constrained, perhaps making them ideal for gaming but provide little to no advantage over cheaper parts for productivity users. - DDR5 memory is starting to mature but there does appear to be room for performance improvement in this area (EPYC series CPUs have 8 memory channels) WCCFTech article: https://wccftech.com/amd-ryzen-granite-ridge-zen-5-desktop-cpu-leak-5-8-ghz-19-percent-faster-7950x/
First, everyone is struggling with 3nm. Everyone. TSMC does not have the node under control, although they are getting by with it. That's why I favor AMD in the short run, as Zen 5 will be released on 4nm with 3nm to follow. 4nm is predictable and scalable with high yield. Intel claims it will bring Lunar Lake to market in Q3, 2024. I hope they can do it but consider me extremely skeptical. Meanwhile, the dumpster fire that we call Samsung Foundry is also really struggling with 3nm. AMD and Samsung have agreed to partner on 3nm development with the first leading edge GAAFET architectures. This might be a minor coup, if they can pull it off. Longer term, I don't see much advantage to Samsung's future looking MBCFET architecture. Why shrink the production node if it doesn't shrink feature size? Feature position matters a bit but that seems like small potatoes compared to feature size, to me. I'm no process engineer so disregard these musings. If they can bring GAAFET online this year, they might have viable product shipping in 2025 when Intel and Nvidia look to once again threaten dominance. In these situations, I generally always bet on the encumbant. That would be TSMC. In this case, I would still bet on TSMC but I find this Samsung/AMD partnership and them taking on GAAFET could be a very real market advantage. This seems like a smart move for both companies. https://wccftech.com/amd-samsung-set-to-team-up-for-advanced-3-nanometer-chips-report/
AMD is hugely popular in the gaming community. I think this is because AMD is a substantially better value, particularly at the midrange. In terms of all out frame rates, 7950X3d vs 13900 (13900 beats 14900 for some games), I think Intel actually edges out AMD by an infinitesimal margin. For productivity applications, I think AMD is faster in far more tasks than Intel. Also, it appears that DDR4 is still faster than DDR5, when the best parts from each are selected. DDR5 will soon surpass DDR4, though. Some DDR5 speed levels were paper released and only exist for benchmarkers until recently. AMD needs DDR5 to evolve and become available in volume for Zen 5 to dominate. Currently, DDR5-7200 exists with CAS latency of 34. My gawd. Throw in a ton of L3 cache and you have a supersonic fighter jet that requires a continent to turn. Speaking of jumbo cache, it appears that 3d cache provides zero benefit to 90% of productivity apps or even in benchmarks. For gaming, 3d cache provides a noticeable boost. These massive latencies make X86 lousy for real time applications. ARM has become the go-to for RT but it is now clear RISC will replace it in the near future.
AMD's keynote at Computer Taipai was underwhelming. This is the problem with good research. When announcements happen, its rarely surprising. The biggest surprise to me is the absense of the 880 chipset. The 870(X) chipset is the 670(X) chipset with a new license that makes USB4 mandatory where it is optional on the 670(X). The silicon is the same. YouTubers have been playing with Turin 192 core for two months so that's nothing new, either.
BTW, Zen 5 is back to looking like July. It never wasn't July. I was coerced by ignorant FUD. Speaking of 40% gains, that was also proven to be complete bollocks as an average IPC lift but SIMD instructions will execute roughly twice as fast as Zen 4. Cinebench users will be lifting their bench without using their hands. The most germaine point regarding the Computex Taipai AMD keynote is that every system makers were tripping over each other to get a picture of themselves standing beside Lisa Su. I don't blame them, I would too. I'm in love with Lisa Su. Lisa... call me. Dell has specifically tied themselves to Intel. That decision is going to cost them for the second half of this year.
That is one thing I am not a fan of with regards to such long socket longevity. I am not sure there is much that can be done to spice up a new chipset if the socket has to be the same. Couple that with such a blatant money sucking market segmentation for those who want more PCIE lanes and RAM channels being forced to upgrade to newer, workstation priced TR systems, and this new chipset is "meh." Great CPU, but limited lanes and RAM channels. Frustrating. What I would really like is less m.2 slots and more PCIE slots. Such a waste of PCB real estate and lane allocation. If one is serious about having multiple m.2 SSDs then get a PCIE to m.2 adapter and be done, and give me back my 7 slots! I keep pushing back an upgrade since I am not in dire need of a speed upgrade. I am still rocking an 18 core Xeon on an x99 chipset, and for my money, the product I would want is an 7950X3D. I could get my gaming fix (whenever I have the time) and still fly through editing apps. I miss the days of the OG TR and Intel x99/x299 boards. Very powerful and does not completely break the bank.
Also, I want to point out the great analysis you have been doing lately regarding the semiconductor industry as a whole. I find it to be pretty much spot on. So thank you, TomB16. Waaay better than Tom's
There are rumblings of a major v-cache upgrade at AMD. It sounds like they are targeting later this year to release. This rumor has some credibility but 95% of things said about AMD are crap so hard to say... Some of the numbers mentioned on Reddit are wild. If true, it will be a whole lot of really, really fast cache. I wouldn't comment on it, except two of the sources have been accurate in the past so there is some chance this vague rumor is true. AMD v-cache already does extremely well in the gaming market so it would make sense AMD would try to improve and secure this position.
There are quite a few declarations of Zen 5 being a failure. Most amusingly, it has been called "Zen 5 %". Perhaps these are the reason for $AMD dropping several percent in the last few days. Before subverting to this outright religious crap, there are a few things to keep in mind. First, Zen 5 is now the fastest x86 CPU on any benchmark. For gaming, it's the fastest. For server, EPYC is the fastest. AMD now holds all the records. What the liars are trying to obfuscate is the fact that AMD just rolled out productivity CPUs at Computex. Those weren't gaming parts. Productivity parts just beat the 13900/14900. AMD is going to release v-cache variants of Zen 5 shortly. I'm not aware of any leaks, regarding L3 improvements. Even if the only improvement is a die shrink, it will be a considerable gaming uplift. Zen 5 L1 cache has vastly improved latency: over halved from Zen 4. When they bring in a massive L3 cache, the numbers will be compelling. AMD is a value, right now.
I'm choking on the idea that AMD is still down, after the scandalous revelation that 9950X is about the same as 7950X3D. This year's Camaro is the same or slower than last year's top model Corvette. Scandal? Traders are morons.
