- Joined
- Jun 19, 2003
- Messages
- 9,101
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Windows Bug with Zen 4/5 (HWUB)
- Thread starter Brackle
- Start date
it's ridiculous that this is a thing
it does appear to be a non trivial gain in a handful of cases.
... and the last generation seems to benefit as well and they basically wind up equal half the time. which is also ridiculous in a way.
edit: maybe I shouldn't say non trivial. more like beyond the margin of error. it's certainly a thing, at least.
it does appear to be a non trivial gain in a handful of cases.
... and the last generation seems to benefit as well and they basically wind up equal half the time. which is also ridiculous in a way.
edit: maybe I shouldn't say non trivial. more like beyond the margin of error. it's certainly a thing, at least.
- Joined
- Jun 19, 2003
- Messages
- 9,101
If this didnt affect Zen 4, i wouldn't of even posted about it really. But it does seem to affect Zen 4 and 5. Now not saying its an AMD or Microsoft issue....could be both who knows, but it does need to be fixed.
It’s possible it affects other CPUs as well but those were all they had time to test. AMD told Steve that it should be fixed in some upcoming Windows update but, considering this is MS we’re talking about, I’ll believe it when/if it actually happens.
- Joined
- Jun 19, 2003
- Messages
- 9,101
Exactly. And lets me honest, it could AMD blaming Microsoft when its an AMD issue? Who knows other than we now know there is a bug with Zen 4/5. Pretty stupid you have to use that admin account to get proper performance.
Run as administrator vs built-in administrator has been not exactly as advertised since they implemented it.
Armenius
Extremely [H]
- Joined
- Jan 28, 2014
- Messages
- 54,738
Is there an article about it so I don't have to watch a 13-minute video?
LittleBuddy
Gawd
- Joined
- Jan 3, 2023
- Messages
- 531
In this Hardware Unboxed video, the host addresses a potential issue affecting the gaming performance of AMD Ryzen processors, particularly the new Zen 5 series, on Windows 11. The host discovered that running the processors on a hidden Windows Administrator account, which has elevated privileges, results in notable gaming performance improvements. This issue was initially thought to affect only Zen 5 CPUs but was later found to impact both Zen 4 and Zen 5 processors, and potentially others as well.
Performance Gains:
- Cyberpunk 2077 Phantom Liberty: Zen 5 saw a 7% improvement in average frame rate, which closed the gap with Zen 4 when both were tested with the hidden administrator account.
- The Last of Us Part I: A 6% boost in average frame rate and a 10% uplift in 1% lows for Zen 5. Zen 4 also saw a 5% improvement.
- Hogwarts Legacy: Zen 5 showed a solid performance uplift, making it 9% faster than Zen 4, which did not see any improvement.
- A Plague Tale Requiem: Zen 5 became 9% faster, with no change for Zen 4.
- Horizon Forbidden West: Zen 4 extended its lead, becoming 5% faster than Zen 5 after retesting.
- Hitman 3: No significant change; performance remained identical.
- Homeworld 3: Zen 5’s performance increased slightly, making it 4% faster than Zen 4.
- Marvel’s Spider-Man Remastered: Zen 5’s margin increased, becoming 9% faster than Zen 4.
- Starfield: Both processors saw a modest improvement, with Zen 5 now matching Zen 4's performance.
- Watch Dogs Legion: Both CPUs saw a 5-6% uplift, with no change in the relative performance margin.
- Assetto Corsa Competizione: No change; Zen 5 remained 18% faster.
- Baldur’s Gate 3 and Counter-Strike 2: Little to no change in performance.
Overall Performance Gains:
On average, the Ryzen 7 9700x (Zen 5) saw a 4% performance improvement across the 13 games tested when using the hidden administrator account, while the Ryzen 7 7700x (Zen 4) saw a 3% improvement. This slightly increased Zen 5’s overall gaming performance lead over Zen 4 from 3% to 4%.Conclusion:
The issue appears to be a Windows bug that affects both Zen 4 and Zen 5 processors, rather than a specific problem with Zen 5. The bug can be mitigated by using the hidden administrator account, though it's unclear if it affects other processors like those from Intel. Microsoft may address this in a future Windows update.
GSDragoon
Limp Gawd
- Joined
- Feb 24, 2004
- Messages
- 273
What does an admin account and extra privileges have anything to do with CPU performance?
Hmm these new Ryzen chips seem to do better in Linux vs windows https://www.phoronix.com/review/amd-ryzen-9950x-9900x/15
"In total I ran nearly 400 benchmarks across all the CPUs. When taking the geometric mean of all the raw performance results, the Ryzen 9 9950X came out to being 17.8% faster than the Ryzen 9 7950X"
"In total I ran nearly 400 benchmarks across all the CPUs. When taking the geometric mean of all the raw performance results, the Ryzen 9 9950X came out to being 17.8% faster than the Ryzen 9 7950X"
LittleBuddy
Gawd
- Joined
- Jan 3, 2023
- Messages
- 531
It's a bug with Windows 11. Using the Super Admin account granted additional performance but it's unclear why.
Super Admin has more permissions and unrestricted access to system files, and some security features disabled by default, so it differs than a normal admin account.
The bottom line here is Microsoft sucks. Using hidden admin accounts and seeing gains.... I mean how the hell is that an AMD screw up ?
AMD clearly sent all the Zen5 changes to the Linux kernel team months ago... it looks like 2 month old kernels know exactly what to do with Zen5. No stupid parking schemes, or terrible Xbox game bars required. Which means AMD sent all the changes to the Linux kernel team MONTHS ago probably 6-7 months ago. Which makes sense if I remember right the first mention of Zen5 in Linux kernel commits were almost a year ago. (not that it was ready then that is just when AMD started bringing up support) You can't convince me AMD wasn't sending info to Microsoft at the same time.
I don't know what is going on at Microsoft but clearly there hardware support teams are not up to the task of implementing modern CPUs. They have had smaller issues previously with Intel as well. We know Intels current arch doesn't rely as heavily on Microsoft scheduling... but this isn't something the CPU should have to deal with at all. Scheduling should really always be a software implementation. Makes you wonder if Intel putting more hardware scheduling on their chip is just them foreseeing the obvious issues that would come up with MS if they had left it to software.
AMD clearly sent all the Zen5 changes to the Linux kernel team months ago... it looks like 2 month old kernels know exactly what to do with Zen5. No stupid parking schemes, or terrible Xbox game bars required. Which means AMD sent all the changes to the Linux kernel team MONTHS ago probably 6-7 months ago. Which makes sense if I remember right the first mention of Zen5 in Linux kernel commits were almost a year ago. (not that it was ready then that is just when AMD started bringing up support) You can't convince me AMD wasn't sending info to Microsoft at the same time.
I don't know what is going on at Microsoft but clearly there hardware support teams are not up to the task of implementing modern CPUs. They have had smaller issues previously with Intel as well. We know Intels current arch doesn't rely as heavily on Microsoft scheduling... but this isn't something the CPU should have to deal with at all. Scheduling should really always be a software implementation. Makes you wonder if Intel putting more hardware scheduling on their chip is just them foreseeing the obvious issues that would come up with MS if they had left it to software.
I am starting to suspect the "quality" issue AMD was holding their launch off for was this.
Microsoft seems to have really screwed this up for AMD. AMD has to have been pushing MS just as hard as they were pushing kernel changes to the Linux team. Linux support is golden, bullet proof. Zen5 performs as expected in every workload under Linux including games. Microsoft is clearly dragging its feet updating what needs updated on their end.
GSDragoon
Limp Gawd
- Joined
- Feb 24, 2004
- Messages
- 273
I doubt it's extra permissions. More likely certain security features being disabled or perhaps less background bloat.
I'm curious if even older Ryzen CPUs benefit from this or even Intel CPUs, like mentioned. Should be easy enough to test. I might try this out on my 3900X.
For a good reason, this is the equivalent of sudo -n but for everything you do after logging in.
This account bypasses every ounce of internal protection and is basically telling Windows to have unprotected sex with each application and service it touches.
This account is for doing deep-level diagnostics and driver development testing stuff like that, this is not something you want to be attempting unless you 100% know exactly what you are doing and what you are there for.
The hidden account bypasses every ounce of UAC and API protection, it allows everything running unchecked access to the OS, with every protection method Windows has completely disabled, It's an account designed for driver development and working with deep-level OS integration.
It's the "OH Shit" account and it is disabled by default for a lot of good reasons.
It is not a safe account to activate let alone use.
So you get better performance because there are no safety checks on memory access, driver access, CPU access, everything you do is just raw dogging the system.
The fact it makes a difference means there is a permissions problem, be it on the Chipset Driver or Windows who's to say, but that's weird for sure. But Windows 10 also has a hidden Admin account, I wonder if Zen 4 does it there too?
Last edited:
That is a lot of good questions.
I'm sure AMDs engineers could figure it out in a few days and fix it for good.... BUT ya know closed source and all.
GSDragoon
Limp Gawd
- Joined
- Feb 24, 2004
- Messages
- 273
The default behavior is the built-in local administrator account still leverages and prompts for UAC. Go check the Local Security Policy. Most rules are set on the Administrators group, not the local admin account. It still has restrictions on what it can do. Local System and Trusted Installer have more permissions than the built-in admin account too. It's not equivalent to root.
Last edited:
GDI Lord
Limp Gawd
- Joined
- Jan 14, 2017
- Messages
- 338
This, IMHO.
- Joined
- Oct 7, 2000
- Messages
- 79,330
thats the video in the op...
Yes at did not understood the even older as zen3 and before....
Maybe but that doesn't explain the odd behaviour. I mean if its just security features... why do some gains show no gain. Some games show big gains, and them all showing differing levels of gain on two different generations. For me it gets confusing when at least one game shows a big gain for Zen4 and nothing for Zen5 at all? It can't just be one low level security check being disabled... it has to be deeper.
My takeaway is that it has to be some scheduler issue still... possibly the scheduler gives precinct to some windows security features, or perhaps it thinks its scheduling that stuff to less performance impacting cores? I don't know the results are odd... the lack of uniformity is the confusing part.
The "easy" way to narrow it down would be to test it with Zen 3, 4, 5, and an older Intel CPU that only has P cores so a 10'th or 11'th Gen for a baseline. Everything after that has Thread Director so it's bypassing/augmenting the Windows scheduler.
If Intel 10/11 sees performance differences then that's a clear indication it's a Windows issue, if they don't then it points to something specific to AMD, if the AMD chips used are single CCD units then it removes that as a variable, so it would then be something specific about the Zen architecture or the drivers themselves.
Which is a crapload of work for any reviewer who chooses to undertake such a test.
Sounds like a job for Microsoft... but we all know they don't actually test all that much.
d3athf1sh
[H]ard|Gawd
- Joined
- Dec 16, 2015
- Messages
- 1,477
yeah, but that's some BS, another good reason to stay on win10 for gaming.
that and hopefully someone figures out exactly what needs to be disabled to have full access to the hardware performance you paid for. talk about windows bloat
1. Copy video link -> https://youtubetotranscript.com -> Paste link -> copy transcript
2. https://chatgpt.com/ -> ? -> Improve Model for everyone -> Off
3. Prompt: "Give me a 50 word summary of this text:" -> paste transcript
"The text explains that a Windows 11 bug affects gaming performance for Ryzen CPUs, notably Zen 5, due to issues with user account privileges. Testing revealed that using a hidden administrator account improves performance, suggesting that the problem isn't exclusive to Zen 5. AMD is working on a fix, and results may vary."
LittleBuddy
Gawd
- Joined
- Jan 3, 2023
- Messages
- 531
Pretty much what I did, but I copied the transcript directly provided by youtube, then told ChatGPT to focus on the performance information.
D
Deleted member 386090
Guest
There is ZERO performance to be gained by using the disabled administrator account as long as you clean up the amount of background task you are running in your base install. I spent half a day testing it against my typical Windows configuration.
Any performance gains seen in HWUB video is due to the significant increase in background tasks on a out of the box Windows install.
I don't think there will be a "patch" to fix this at all, because Microsoft would have to disable all the privacy settings, background apps, and default security features.
Any performance gains seen in HWUB video is due to the significant increase in background tasks on a out of the box Windows install.
I don't think there will be a "patch" to fix this at all, because Microsoft would have to disable all the privacy settings, background apps, and default security features.
I can say that all the same background tasks run on the hidden admin vs any other local account.
If your using a win 11 with a local account and not tying it back to an online service there is no difference between the background tasks by default.
So unless HWUB went the extra mile to disable services and applications in the hidden account that they didn’t do for the regular account (which would skew results and be intentionally deceptive) the only difference is the account permissions and access rights.
The effect of better cpu priority (if it is part what this is), would only matter if it compete with background task.
If it is lesser security checks for those running in admin, in that case it could have nothing to do with background app and privacy, just be the extra security....
Would be fun to check if installing trusted game on a devdrive could show part of the improvement (devdrive are faster pretty much because security scan is off for the files you put on it)
If it is lesser security checks for those running in admin, in that case it could have nothing to do with background app and privacy, just be the extra security....
Would be fun to check if installing trusted game on a devdrive could show part of the improvement (devdrive are faster pretty much because security scan is off for the files you put on it)
D
Deleted member 386090
Guest
I disagree and definitely do not think the bypassing UAC has any performance benefit at all, and I believed in it after reading this thread. But wanted to test it myself. I was impressed with HWUB's video seeing a 7%-10% performance gain on both the 9700X and 7700X.
My testing was limited to an X3D, maybe there is a difference on non-X3D chips, but I definitely saw a ZERO performance change using the disabled administrator account. Could be because most of my Windows changes from my main account carried over into the administrator account, the administrator account did re-install some of the bloatware, but it was not running in the background. Task manager in the administrator account had a similar number of apps running by default as my main account.
I tested Cyberpunk 2077 at 1080p High, The Last of Us: Part 1 at 1080P Ultra, Baldur's Gate 3, 3DMark Time Spy. Cinebench R23. No performance gain.
Update:
I don't disagree with Lakados.
Last edited by a moderator:
Windows 11 vs. Ubuntu 24.04 Linux Performance For The AMD Ryzen 9 9950X
"But at least in some areas we did indeed see Linux offering more uplift when going from the Ryzen 9 7950X to 9950X than when testing the same applications under Windows 11."
I feel this does not follow with the rest of the message, you say: Task manager in the administrator account had a similar number of apps running by default as my main account.
Which sound a lot like: I can say that all the same background tasks run on the hidden admin vs any other local account.
D
Deleted member 386090
Guest
My disagreement was in error I actually misread Lakado's post lol.
But my reference point is bad since I didn't do a clean install prior. So my initial assumption was the administrator account had reduce background processes. That's why I mentioned that the background processes were similar between the account in my second post. But the administrator account did not automatically run my startup tasks.
Not sure why I saw the same exact performance though, Steve from HWUB said he will test more CPUs soon, but didn't give a specific date, it will be interesting if he tests the X3D and it sees no change.
My 2 cents, AMD probably blamed HUB for the bad review and now they are trying to shift the blame to MS.
It's quite normal to get more FPS once you stop background tasks. The next step may be to stop the local network (works), or core isolation, or pfff so much "bench" tricks...
So just more crap from the HUB.
It's quite normal to get more FPS once you stop background tasks. The next step may be to stop the local network (works), or core isolation, or pfff so much "bench" tricks...
So just more crap from the HUB.
Sounds like you made random assumptions without watching the video.
GoodBoy
2[H]4U
- Joined
- Nov 29, 2004
- Messages
- 3,144
HU said productivity apps are unaffected, only Games are known to be affected so far.
So it's too early to say for certain what the cause is. It's just as likely to be a windows bug as it is a chipset driver bug or video driver bug. Linux uses different drivers so if the games run at the higher observed benchmark, it is not a smoking gun to blame Windows with. That is if the game can even run on Linux in the first place... so making this into a WIndows vs Linux is a moot point.