Yeah, right... lots of bull from their marketing department - I wonder if it's the same for GPU and CPU divisions - that would explain a lot. The administrator account problem also hits Ryzen 4 so after testing both with admin account reviewers found out that it is basically the same (non) difference between the two gen and intel was also affected so this is basically a problem for everyone.

AMD adjusted its previous claims regarding the Ryzen 9000’s performance over the Ryzen 7000 series,

- "adjusted" --->so they lied but being the white knight of the industry and not a greedy corporation like the others of course they will get away with that by placing the blame on MS, reviewers and maybe aliens. Even if they did not lied on purpose and it was a honest mistake it is extremely embarrassing - amateurish - that you can't review correctly your own product and only after other third parties reviews it you find "a problem". Also they admit using in their marketing slides intel cpu's with default settings-baseline power profile - so 125W long and no reviewer used this since it is Intel own recommendation to use the performance or extreme profile and after all the "noise" with their test figures now they used intel extreme profile and they say that there is parity in gaming - oh really? - so the marketing slides were a joke? And why is AMD using for testing an administrator account since no user or gamer in the real world use this? - this is how they are replicating what people who buy their products will get in terms of performance? I ended up buying a 7950X (30% discount so I couldn't resist) with a MSI X670E ACE motherboard (still expensive, no discount here) - works beautifully. If anybody wants to have a good laugh below is one of the marketing slides used by AMD - the competition was with "baseline power profile". https://i.imgur.com/zuciPRw.png

barbacot:

And why is AMD using for testing an administrator account since no user or gamer in the real world use this? - this is how they are replicating what people who buy their products will get in terms of performance?

why should an admin account perform better than a local one? It shouldn't, thats the point. now a windows bug has been discovered and will be patched.

They even managed to make Steve at AMD Unboxed angry. When you enrage your staunchest fans it's time to fire your VP of marketing and hire some damage control asap.

[youtube=IaPo5io-Gfw] It seems some of AMD's clarifications are flawed. Their approach to benchmarking isn't great and they ran Intel's CPUs in a suboptimal configuration.

CPC_RedDawn:

why should an admin account perform better than a local one? It shouldn't, thats the point. now a windows bug has been discovered and will be patched.

What really bakes my noodle is that Guru3D users seem to have found this aways back (there's a thread on running games as Admin causing a CPU perf uplift) and it may even affect older Zen parts yet. Perhaps it relates to the implementation of a new security feature in Win11? Unsure, but yeah -- performance regressions like that are not what I want to see and the fact that nobody's internal Q/A caught it til now is concerning.

I dont understand how Intel CPUs should be benchmarked? Suicide mode for testing then baseline for gaming? I mean they are also missleading with their performance.

CPC_RedDawn:

why should an admin account perform better than a local one? It shouldn't, thats the point. now a windows bug has been discovered and will be patched.

It's a feature , not a bug :P

CPC_RedDawn:

why should an admin account perform better than a local one? It shouldn't, thats the point. now a windows bug has been discovered and will be patched.

I presume for the same reason why Spectre and Meltdown patches lowered performance.....security. Some of these side channel attacks exploit how CPUs perform, ie branch prediction, prefetching, shared caches etcetera. I'm assuming that the local account has security mitigations that reduced branch prediction efficacy. Windows 11 24H2 probably optimized it further to lower the penalty.

CPC_RedDawn:

why should an admin account perform better than a local one? It shouldn't, thats the point. now a windows bug has been discovered and will be patched.

This been around since win 7 days, if it has been fixed in all this time it either not bug or Microsoft, is in we dont care mode. I am sure there are thing less restrictive when on admin account and your more open to attack security issue when using an admin account, which why they dont want use using admin account. But at same time a lesser account should be being have there cpu and or gpu performance reduce cause of it. some game/programs are effect some are not. which to mean it some said developer are doing and other are not.

What an unecessary mess, AMD should have done all this before the launch. Now most people have already made their mind about the 9000 series and most won`t change even if AMD manages to fix thir performance.

Well im happy that Zen4 will also get faster with the new update, thats something 😀

Undying:

Well im happy that Zen4 will also get faster with the new update, thats something 😀

I wonder if this is just a Windows 11 problem or it goes back to 10 or lower. Also, I wonder if Zen 3 is affected or not...

Wasn't it just a W11 issue? I seem to recall that W10 wasn't affected.

wavetrex:

I wonder if this is just a Windows 11 problem or it goes back to 10 or lower. Also, I wonder if Zen 3 is affected or not...

Doing my own tests on my 5900x disabling uac / lua I ended up with the snappiest system in the universe. It's weird. Basically how windows xp would've run back in the day just heaven for malware and viruses. Now in reality uac while sure security feature it won't protect people just clicking yes to everything. But clicking this off everything runs in admin so it's easier to get things infected.

AMD also discussed a Windows feature called the "Admin" profile, which enhances performance through branch prediction optimizations.

This makes no sense. Branch prediction is not under the control of the compiler or the code being run. It is a key part of a processor's design and there are no compiler options that can affect it. Upon further review, and reading this : https://www.devgem.io/posts/using-likely-and-unlikely-attributes-in-the-standard-library-algorithms, it seems there are things that can be done. The way it works is the code of most likely branch is placed as the destination of the first if test and the less likely ones are made secondary. My first statement is actually correct but there are a couple of directives that can be used to adjust the code that is generated and have the effect of optimizing branch prediction. The options don't affect how the processor operates. They just re-order the code to make prediction of branch targets somewhat more accurate, depending on how accurately the directives are used.

Ryu5uzaku:

Doing my own tests on my 5900x disabling uac / lua I ended up with the snappiest system in the universe.

My system must be full of malware, viruses, coin miners, spam senders, etc... etc... ... or maybe I'm not a moron who downloads every miley-cirrus-naked-photo.jpg.exe out there 🙄 -- p.s. For anyone crying in the wind "but but the firewall!!!!"... I'm on an internal 192.168.x.x network with a Linux router doing the heavy lifting.

Installed it on my 5800X3D system... & I got a 3% FPS increase in Elden Ring (Flawless Widescreen to unlock the FPS) in a very static spot, it went from super consistent 184 fps to a shaky 182-192 usually sitting closer to 190. That was CPU limited... in Ghost of Tsushima I got more like 16% in a scenario that was 100% GPU limited. ... How? The 5800X3D is homogenous, with 1 CCD, 1 CCX on that CCD, and all 8 cores on that 1 CCX. Also I don't recommend installing any preview anything unless you really need that extra performance, 24H2 comes with AIDS. MS managed to even butcher the Task Manager, it takes forever to load.

Neo Cyrus:

Installed it on my 5800X3D system... & I got a 3% FPS increase in Elden Ring (Flawless Widescreen to unlock the FPS) in a very static spot, it went from super consistent 184 fps to a shaky 182-192 usually sitting closer to 190. That was CPU limited... in Ghost of Tsushima I got more like 16% in a scenario that was 100% GPU limited. ... How? The 5800X3D is homogenous, with 1 CCD, 1 CCX on that CCD, and all 8 cores on that 1 CCX. Also I don't recommend installing any preview anything unless you really need that extra performance, 24H2 comes with AIDS. MS managed to even butcher the Task Manager, it takes forever to load.

Well the issues run deeper then just CCD communication in this regard.

wavetrex:



My system must be full of malware, viruses, coin miners, spam senders, etc... etc... ... or maybe I'm not a moron who downloads every miley-cirrus-naked-photo.jpg.exe out there 🙄 -- p.s. For anyone crying in the wind "but but the firewall!!!!"... I'm on an internal 192.168.x.x network with a Linux router doing the heavy lifting.

I removed my lua via registry 😀 and been living like so since. Got my precautions elsewhere for sure. Just way nicer to use the pc now.

Gomez Addams:

This makes no sense. Branch prediction is not under the control of the compiler or the code being run. It is a key part of a processor's design and there are no compiler options that can affect it.

Yes there are. There are instructions(fence) added by the compiler where if you have a branch prediction which passes security boundary it will add partioning/checkpoint instructions to disable or control the speculative excecution to prevent leaking data via a side channel. It is one of the main mitigations techniques they have specultive code execution vulnerabilities