x86 CPUs from AMD & Intel current/future releases.

[exoticspice1]

This can be used for all things x86. Current and upcoming CPU/SoC from AMD/Intel that are x86 based. Like Zen 5 and Arrow Lake/Lunar Lake.

Today Intel released the last LGA1700 CPU the i9 14900KS. Its not a good chip at all and fails in every metric that it targets. The screenshot below explains it all. The AMD X3D CPUs wreck Intel.

 

[Cmaier]

This can be used for all things x86. Current and upcoming CPU/SoC from AMD/Intel that are x86 based. Like Zen 5 and Arrow Lake/Lunar Lake.

Today Intel released the last LGA1700 CPU the i9 14900KS. Its not a good chip at all and fails in every metric that it targets. The screenshot below explains it all. The AMD X3D CPUs wreck Intel.

I’ll let them know you approve

 

[exoticspice1]

I’ll let them know you approve

. I know its a jest post but if you know anybody from AMD that still works there. Tell them AMD is keeping x86 relevant.

 

[Cmaier]

. I know its a jest post but if you know anybody from AMD that still works there. Tell them AMD is keeping x86 relevant.

I know lots of folks still there, and I’ll let them know.

 

[dada_dave]

So Raptor Lake is another Rocket Lake? How ... disappointing and yet expected.

Intel claims they'll be fabbing Arrow Lake compute dies on 20A at the end of the year with the rest of the dies being on TSMC N3. Let's see what kind of improvement they get.

 

[KingOfPain]

The Ars Technica headline says it all:

Intel’s 6.2 GHz Core i9-14900KS is a reminder of why the MHz wars ended

An impractical bragging-rights CPU tops Intel's 14th-gen desktop lineup.

arstechnica.com

 

[leman]

I am very curious about the next-gen Intel cores that are supposed to ditch hyperthreading and come with new u-arch that can dynamically balance execution resources between ST and MT scenarios.

 

[exoticspice1]

Well, its nearly computex and AMD will be unveiling performance details about Zen 5 soon. Excited to see how it improves.

 

[exoticspice1]

Apple still the IPC king.

 

[Jimmyjames]

Ahhh the sweet embrace of Geekbench 5.4 AES-XTS.

 

[Artemis]

View attachment 29762
Apple still the IPC king.

Oh Z5 was in no way going to pass M4, but even I thought they’d get to maybe M1 minus 5-7%.


They fell short of even that lol. And it’s not like power is good, and area ballooned for the cores apparently.

The IPC ranking is
Apple
Cortex X925
Qualcomm’s Oryon
Gap
Zen 5
(Likely also LNC)

But yeah the hype train looks worse than ever.

 

[Artemis]

That benchmark list is so dishonest it’s not even funny lol. The GB6 and GB5 subtest only….

 

[Cmaier]

Oh Z5 was in no way going to pass M4, but even I thought they’d get to maybe M1 minus 5-7%.


They fell short of even that lol. And it’s not like power is good, and area ballooned for the cores apparently.

The IPC ranking is
Apple
Cortex X925
Qualcomm’s Oryon
Gap
Zen 5
(Likely also LNC)

But yeah the hype train looks worse than ever.

I just don’t see how it‘s possible to get that close without burning a tremendous amount of power to do a lot of speculative decoding and scheduling. We made AMD64 a lot easier to decode than all the crap that came before, but there was only so much we could do while staying backwards compatible.

 

[casperes1996]

I just don’t see how it‘s possible to get that close without burning a tremendous amount of power to do a lot of speculative decoding and scheduling. We made AMD64 a lot easier to decode than all the crap that came before, but there was only so much we could do while staying backwards compatible.

I don't know too much about how big a disadvantage this really is compared to all the other parts of the design after decoding, but regardless of how they achieve it, I'm honestly impressed. Clock-for-clock it may not be Apple tier ISO IPC or anything but it sounds like they're doing really impressive work under the constraints they have to work with. Apple's chips have been incredible over the last many years, but the teams working on the various iterations of Zen have done very good work in the x86 space. To my mind, from a chip layout perspective, Zen 4c especially sounds like a really cool accomplishment at basically half the footprint of a full Zen 4 core. And the Zen 5 performance bump is, while not revolutionary or anything, a solid generational increase. Which we don't always see; Intel 11th gen springs to mind for example

 

[dada_dave]

Oh Z5 was in no way going to pass M4, but even I thought they’d get to maybe M1 minus 5-7%.


They fell short of even that lol. And it’s not like power is good, and area ballooned for the cores apparently.

While I don’t know this to be true for a fact, it could, at least partly, be because they went full AVX-512. Their previous solution was so good … and they didn’t burn area for it. When I saw that the “leaked” 40% gains was mainly for those AVX-512 workloads (and even that was generous it turns out), I suspected it’d be more like Zen 3 to Zen 4 again. I agree with @casperes1996 that these are solid, workmanlike gains that would’ve been more acceptable had the leaks, deliberately or not, not got (certain) people unreasonably hyped.

The IPC ranking is
Apple
Cortex X925
Qualcomm’s Oryon
Gap
Zen 5
(Likely also LNC)

But yeah the hype train looks worse than ever.

I’m not a fan of singular IPC values but yeah given what I’ve seen that’s probably true almost regardless of particular test. I mean I can imagine there are instances where the ARM and Qualcomm chip trade spots and maybe even with Apple who knows some benchmark somewhere where the IPC get close, but yes.

That benchmark list is so dishonest it’s not even funny lol. The GB6 and GB5 subtest only….

Ahhh the sweet embrace of Geekbench 5.4 AES-XTS.

I mean yeah … I’ve argued several times that any collection of benchmark subtests is a little arbitrary but it’s never good when a vendor does it. Were I to play devil’s advocate, I could argue that the other GB5 and 6 subtest workloads left out are replicated in other benchmarks they chose: eg GB6 object detection vs GB5 XTS and GB6 ray tracing vs CB R24, etc … plus they wanted to throw in benchmarks that matter to their end users like gaming. If a truly independent third party reviewer did that, I would be okay after all any collection of tests is up to them! But when it’s the vendor itself … well … obviously that opens up one up to not totally unwarranted criticism of cherry picking even if the picks are individually reasonable.

Again it’s why I really dislike the average score. In the end, the “average” is just marketing and so easy to manipulate to tell whatever story marketing (or as we’ve seen a la object detection removal some random person on the internet) wants.

 

[diamond.g]

Am I reading the (marketing) slides correctly, there is no clock speed gain going from Zen4 to Zen5? So the improvements are literally straight IPC?

 

[Jimmyjames]

I think the jokes about ipc are more aimed at fans who claimed ridiculous gains for Zen5 while dismissing everything else. It’s a solid chip no doubt.

 

[Artemis]

Post in thread 'Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)'
https://forums.anandtech.com/thread...ranite-ridge-ryzen-9000.2607350/post-41223667

Lmao, Andy I think from Twitter is pulling his M4 sucks and Strix is fine. Read down where I address his BS, I was even kind to him and used GB5 (while really AMD has benefitted from AVX AI subtest gains in GB, but ofc when Apple does it everyone goes nuts).

 

[Jimmyjames]

Post in thread 'Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)'
https://forums.anandtech.com/thread...ranite-ridge-ryzen-9000.2607350/post-41223667

Lmao, Andy I think from Twitter is pulling his M4 sucks and Strix is fine. Read down where I address his BS, I was even kind to him and used GB5 (while really AMD has benefitted from AVX AI subtest gains in GB, but ofc when Apple does it everyone goes nuts).

Oof this guy. Agree with what you said to him and would add Geekbench 5.1 added AVX-512 specifically for AES XTS, which adds significantly to the IPC average. Something Anandtech points out.

Geekbench 5 Release Notes

He can’t have it both ways, either AVX512/SME is misleading in terms of IPC (with the caveats about an overall score discussed previously) or it isn’t. He also is pontificating on the “M4 runs at 4.5 during multi-core tests and therefore will consume +100W”, even after @leman clarified that this is incorrect. So frustrating. I don’t mind people being wrong, I don’t mind people making mistakes, but this kind of dishonesty is infuriating.

 

[dada_dave]

Speaking of that guy (actually a bit unfair to deconstruct60 at MR), I feel like I was being clear in the below why I chose to compare the Surface Laptop and the MacBook Pro rather than to the Air and why I was making the comparison at all (i.e. to highlight how the Snapdragon Elite/Pro doesn't really align neatly as an SOC to the M3/M3 Pro making direct comparisons difficult - though it is perhaps closest to the binned M3 Pro):

Qualcomm revealed X Elite's benchmark scores

It has been pointed out to me that the iPads needed a new chip to run the tandem OLED display and were already really long in the tooth with an earlier rumored launch of March. So it's possible this is all just coincidence but it was impossible (for me) not to feel their NPU statements: "we were...

forums.macrumors.com

And you know what ... when I posted the above I knew deconstruct60 was going to reply and I even knew pretty much what he was going to say and he did indeed do so, but I just want to double check that his confusion is not because I was being unclear. Because I recognize sometimes I'm not coherent, I'm sometimes quite tired and rushed or overly verbose, but I reread the above and it seems clear to me that I'm not arguing which device is the best value or that there aren't other better comparable Snapdragon devices. So I'm thinking it's not really worth responding to but I just want to make sure that this is just deconstruct60 being deconstruct60 - i.e. being technically correct (though they aren't always) but badly missing a fairly straightforward point. That maybe I spent too many words explaining.