We have an annotated die shot of STX, but GNR's FPU is somewhat bigger still.People keep complaining about the AVX512 because that logic and register file is taking up all the area. But is it? Do we have annotated die shots?
The hugely increased front end buffers/queue (which seems to still operate like 4 wide most the time) might take up even more area. And what good are all those duplicated resources except for servers with SMT?
The real defect of the AVX512 is the latency increase for 128/256 bit instructions. And that doesn't sound like it was intentional...
STX FPU is small, but we'll have to wait for Granite Ridge pic.We have an annotated die shot of STX, but GNR's FPU is somewhat bigger still.
View attachment 104869
Nehalem to Sandy Bridge might have been the last one.
NetBurst to Core 2 as well.
You can always buy Intel.It's Nvidia's tensor cores all over again, except there's no DLSS to try and convince consumers that charging them more for hardware they can't use is a good thing.
This is why I have a gigantic beef with the pricing and I hope (Mindfactory gives promise) that market reception is awful and these processors don't sell well at all. AMD fudged this, no way around it, and they need to learn a lesson here.
I keep seeing people compare this to Intel's 4 core reign. It's not really comparable because market pricing on Zen 4 is so low, there's nowhere near a performance gain for consumer workloads to even come close to justifying the increase to MSRP over Zen 4 parts.
Intel kept pretty good control over pricing, without giving massive discounts and without keeping old gen parts around on the market. The new gen would come and slot in at practically the same price the old part was at, fully replacing it and the old gen part would just disappear from the market.
AMD has historically kept the last gen parts around to help fill in lower pricing tiers and gaps between new gen products, since they release a pretty sparse stack initially and only fill in over time. If AMD keeps Zen 4 around, there really is no reason for any consumer to choose Zen 5 at MSRP. If AMD intends to eliminate Zen 4 from the market, it would be the first time. If they're trying to pull an Intel, I don't think they really have the market control, market share and mind share to pull it off.
This saved Intel. Also where is my promised 10 Ghz CPU.Netburst to Core 2 was on that came to mind.
I don’t think you’re being charitable with the opposite side of this argument.I'm getting more and more annoyed at this attitude of "should have given it more power" after we spent years lamenting that CPU power usage was getting out of control. AMD will learn to appease reviewers instead of the average consumers and Intel is definitely taking notes for Arrow Lake from this. The same people who complain that overclocking is dead are the ones complaining the 9700X did not come with the pedal already nailed to the floor.
High power computing should be opt-in, not opt-out. Put conservative settings as stock for the average user, let the enthusiast make the monumental effort of enabling PBO or using ECO for custom builds.
As far as review guidance goes, it was probably bad. They had bad marketing, bad BIOS health before launch, why would review guidance be any better...
I don’t think you’re being charitable with the opposite side of this argument.
I don’t see many people advocating for running unlimited power limits or *higher* power limits than last generation. The argument is that they should have kept the TDP the same as the SKU that it is replacing.
They had 2 options:
1) Keep the same TDP as last generation, show improved performance in multithreaded benchmarks.
2) Reduce TDP, show zero multithreaded performance improvements (or sometimes a regression) but market it as efficient.
Considering it’s a desktop SKU (not mobile or server) and last gen only had a TDP of 105W, many people believed option 1 was a better choice.
15% improvement on what is a 400 benchmark, productivity oriented list??? Up to 25% if you aren't power starved?The Ryzen 7 9700X delivered 1.195x the performance of the Core i5 14600K competition or 1.15x the performance of the prior generation Ryzen 7 7700X. The Ryzen 5 9600X came in at 1.35x the performance of the Core i5 14500 and 1.25x the performance of the Ryzen 5 7600X.
Stop simping Michael. This is 2 years after Zen 4, on a node that's supposed to be ~20% less power draw. I'm far from impressed. If I was tasked to build a new server farm for a cloud provider sure, but for client?The raw performance of these Ryzen 9000 series processors was extremely impressive. These new Zen 5 desktop processors showed significant uplift in areas such as gaming and single-threaded workloads commonly led by Intel like Python, NumPy, Cryptsetup, audio encoding, and web browser performance. The Zen 5 generational uplift also showed great strides in even better AVX-512 performance for helping more AI workloads to a lot of other strong finishes in technical computing and HPC workloads. Whether you are just a heavy web browser user and running lots of Python scripts to doing a lot of creator workloads and software development, Zen 5 is an exceptionally well rounded design.
I agree, they should have just left the 105 W TDP. I havent used any recent AMD products, but it seems to be pretty easy to enable PBO. It is still confusing though. Tom's Hardware showed a big boost with PBO enabled, while other reviews showed lesser boosts. It kind of reminds me of Meteor Lake. Overall looks disappointing, but there is the occasional review that makes it look comparatively much better.I don’t think you’re being charitable with the opposite side of this argument.
I don’t see many people advocating for running unlimited power limits or *higher* power limits than last generation. The argument is that they should have kept the TDP the same as the SKU that it is replacing.
They had 2 options:
1) Keep the same TDP as last generation, show improved performance in multithreaded benchmarks.
2) Reduce TDP, show zero multithreaded performance improvements (or sometimes a regression) but market it as efficient.
Considering it’s a desktop SKU (not mobile or server) and last gen only had a TDP of 105W, many people believed option 1 was a better choice.
If Intel hadn't wasted money on fab related costs trying to make their maligned version of 10nm work, they sure could've released a Pentium 4 10 GHz Anniversary Edition on Intel 7 or Intel 4 that downclocks to 2 GHz with AVX-512This saved Intel. Also where is my promised 10 Ghz CPU.
Meh. The power efficiency gain is overblown because of the the 7700X comparison. It's efficiency is not nearly as impressive compared to the 7700 and 7800X3D in most workloads.
The 7800X3D absolutely kills the 9700x in both gaming performance and efficiency.
Wonder why he didnt test any PBO settings? Gaming tests are problematical since he used a 4080 instead of a 4090. It looked like a lot of the games were gpu limited. The 9700x is an OK processor, possible suffering from the usual overhype expected of Zen 5. It looks really good in productivity, but kind of meh for gaming.Leo has a balanced review of it. Includes a fun little look going back to the 2700X.
Most reviewers don't bother with 0.1% lows and for good reason. They're highly variable run to run. If TYC isn't using an average of multiple runs then the data is likely anomalous..1% frame times seem to have been missed by a few bigger outlets...
Page 4 - Discussion - Zen 5 Builders thread
Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.forums.anandtech.com
I was just going to say this in a reply. Nehalem was a very good CPU. I had it and did not upgrade until Haswell. Because I didn't need to. Core 2 Duo came out in July 2006. Silly me upgraded my Athlon 64 3200+ Winchester with an Athlon 64 X2 4400+ Toledo. Maybe it was a drop in. All I know is that I should have gone with the Core 2 Duo instead of the Athlon X2 for my first dual core processor.Netburst to Core 2 was on that came to mind. And Piledriver to Zen. I don't think Nehalem to SNB was big enough to justify a new system as they were on different sockets and Nehalem was very capable.
That was one of the reasons I liked his review. Everyone tests for science. But few own a 4090, and when a $1000 GPU is the bottleneck most of the time at 1080 and 1440 max settings it provides some perspective.Wonder why he didnt test any PBO settings? Gaming tests are problematical since he used a 4080 instead of a 4090. It looked like a lot of the games were gpu limited. The 9700x is an OK processor, possible suffering from the usual overhype expected of Zen 5. It looks really good in productivity, but kind of meh for gaming.
Exactly. 0.1% get nerfed for all kinds of dumb reasons; like when you die in CS2. They rarely say anything useful, because the longer you play, the higher they tend to be. Some games start out rough and get smoother over time. Take away: games are a mess and all of the benchmarketers info is of limited utility. For those of us that play the games for many hours the experience is often different than their results would lead you to believe.Most reviewers don't bother with 0.1% lows and for good reason. They're highly variable run to run. If TYC isn't using an average of multiple runs then the data is likely anomalous.
BG3 is the only significant outlier here. I have over 250 hrs in the game split between 13700K/6900XT and 7800X3D/7900XT systems. Framerate can vary wildly depending on the area of the game you are in, but I don't recall ever experiencing any annoying stuttering that would indicate poor frame pacing.
I personally don't hold TYC in very high regard as a reviewer. If GN does some frame time graphs with Zen 5 I'll pay attention.
Simply looking at the last generation to establish naming and power rules isn't charitable either.I don’t see many people advocating for running unlimited power limits or *higher* power limits than last generation. The argument is that they should have kept the TDP the same as the SKU that it is replacing.
Reviewers can still test the performance option first, just like most of them test with XMP enabled. Intel seems to be heading towards this direction as well with their Baseline/Performance/Extreme profiles, though as always they manage to make it complicated for both them and the user. They'll probably get there eventually.In an ideal world, high power computing would be opt-in, not opt-out. Unfortunately CPU releases don't happen in a vacuum so we're stuck with comparing new CPUs to old CPUs with higher power limits.
However I'll concede that the 9700X, within the current apparent rules for TDP / core for Zen 5 could use a bit more power so we can have ~linear progression from one SKU to the next. Something like 90W TDP and 120W PPT would work and bring consistency to their entire lineup. I like the idea of increasing stock TDP when moving throught the lineup 6->8>12->16 cores, this is both desirable and easy to understand for the user, but we still need a conservative baseline to start from and progress ~linearly.
Going back to fairness though, I'll remind everyone we have witnessed a very unhealthy trend on the desktop in the last 7+ years, with power ballooning from 65/90W for the obvious reason of looking better in reviews. Intel started the trend and AMD followed suit. Like I said before, my stance is to have the product's default config in a very conservative place in terms of power/heat, and let the user learn how to unlock performance or power saving modes. It's a matter of user experience and a more predictable behavior out of the box. The best example I can think of in terms of user choice is memory XMP profiles: default setting for memory is the conservative JEDEC profile and the user needs to manually enable XMP. Reviewers can still test the performance option first, just like most of them test with XMP enabled. Intel seems to be heading towards this direction as well with their Baseline/Performance/Extreme profiles, though as always they manage to make it complicated for both them and the user. They'll probably get there eventually.
Most reviewers don't bother with 0.1% lows and for good reason. They're highly variable run to run. If TYC isn't using an average of multiple runs then the data is likely anomalous.
BG3 is the only significant outlier here. I have over 250 hrs in the game split between 13700K/6900XT and 7800X3D/7900XT systems. Framerate can vary wildly depending on the area of the game you are in, but I don't recall ever experiencing any annoying stuttering that would indicate poor frame pacing.
I personally don't hold TYC in very high regard as a reviewer. If GN does some frame time graphs with Zen 5 I'll pay attention.
That was one of the reasons I liked his review. Everyone tests for science. But few own a 4090, and when a $1000 GPU is the bottleneck most of the time at 1080 and 1440 max settings it provides some perspective.
That said, Leo is testing, not playing the games. I know for a fact you can better expose some of the CPUs in that list in games like Cyberpunk'd, Spiderman games, MS flight sim, Hogwarts, etc. Particularly with RT on.
Exactly. 0.1% get nerfed for all kinds of dumb reasons; like when you die in CS2. They rarely say anything useful, because the longer you play, the higher they tend to be. Some games start out rough and get smoother over time. Take away: games are a mess and all of the benchmarketers info is of limited utility. For those of us that play the games for many hours the experience is often different than their results would lead you to believe.
My overclocked 4770K looked good in Assassin's Creed Odyssey in the benchmarks. But I could make it crap the bed in big fights with mercs, soldiers, and citizens all going ham. Even PBO boosted Ryzen 2600X did not manage to be buttery smooth all of the time. It took a Ryzen 3600 to completely eliminate frame pacing issues; thank Denuvo!
Bryan is a good guy, but his degree is in biz/financing iirc. Props for being one of the first to callout raptor ringbus issues and swap back to a 10th gen. But like the rest usually benchmarks games, instead of playing them.
I liked the comment in the Zen 5 thread that posited AMD is separating work and play more. Vanilla for budget and work. 3D for serious gamers.
Unless you write reviews and pretend to provide accurate information or ignore people that correct your mistakes, no, that's not an issue. The problem with a lot reviewers is that some people trust them blindly while they are sometimes clueless about what they write or the data they get.I minored in computer science, so, is all my contribution to this forum moot?
100% this. Could not have said it better. There is some obvious shilling for dollars going on too.Unless you write reviews and pretend to provide accurate information or ignore people that correct your mistakes, no, that's not an issue. The problem with a lot reviewers is that some people trust them blindly while they are sometimes clueless about what they write or the data they get.
Strawman fallacy. I said nothing of the sort. We all engage in one or a combination of - provide user to user help, talk shop, discuss and share experiences of a common hobby. Once money gets involved you are held to a higher standard. It goes with the territory. Massive difference between Dr. Cutress and Toasty Bros, you dig?I minored in computer science, so, is all my contribution to this forum moot?
On topic: Does anyone know what happened to tech testers? She hasn't uploaded in a month it looks like. I was expecting a 9000 series review.