Why everyone say Zen 5 is bad ?

[DAPUNISHER]

I am not seeing the performance boosts on either my AM4 or 5 3D CPUs. Biggest boost was 8% Gears 5 bench 1080low 5600X3D+7800XT. 7800X3D+7900XTX was 6%. I have no idea how Steve is getting those numbers. Maybe something to do with fixing Nvidia driver overhead? 4090 being more CPU bound?

 

[oxrufiioxo]

I am not seeing the performance boosts on either my AM4 or 5 3D CPUs. Biggest boost was 8% Gears 5 bench 1080low 5600X3D+7800XT. 7800X3D+7900XTX was 6%. I have no idea how Steve is getting those numbers. Maybe something to do with fixing Nvidia driver overhead? 4090 being more CPU bound?

On my X3D ccd the gains are minimal.... on the frequency ccd the gains are larger probably why you're seeing minimal gains.

Honestly this seems to make the 7700X perfom closer to the 7800X3D imho we Will have to wait for a large selection of games tested to know for sure.

I do have a 4090 as well though but at best I'm seeing 5% on the 3d cache ccd and 10% on the frequency CCD.

 

[Solaris17]

You don't have to run admin to get improved branch prediction in 24H2, right?

I don’t think so. That’s a core OS scheduling thing. That exists in builds themselves. This is why I take everything YouTubers say with grains of salt. Performance from running as admin (which doesn’t start certain apps and other “annoying” consumer stuff) sure I can see it.

But user accounts do not dictate core level OS functions. He likely just misspoke. Because the newer 24Hx builds DO INFACT come with the new scheduler but the one making the big impacts are insider slow ring right now; unless you own a co pilot P.C. which comes with the new builds standard.

 

[phanbuey]

i kind of want to pick up a 9700x for science

I don’t think so. That’s a core OS scheduling thing. That exists in builds themselves. This is why I take everything YouTubers say with grains of salt. Performance from running as admin (which doesn’t start certain apps and other “annoying” consumer stuff) sure I can see it.

But user accounts do not dictate core level OS functions. He likely just misspoke. Because the newer 24Hx builds DO INFACT come with the new scheduler but the one making the big impacts are insider slow ring right now; unless you own a co pilot P.C. which comes with the new builds standard.

I was thinking this too - is it possible the app doesn't get virtualized as admin? or windows defender just harasses it less in general?

 

[Solaris17]

i kind of want to pick up a 9700x for science




I was thinking this too - is it possible the app doesn't get virtualized as admin? or windows defender just harasses it less in general?

No virtualization inits before the OS does. You can disable things like VBS(which hooks before kernel loading) while still having virtualization enabled in bios. All it does is expose CPU bits and registers.

To be clear I don’t think he is lying. He just misunderstood or said it in a fashion that made it unclear to the masses.

AMD said

“Use the ADMIN account”
“Use the new insider build”

but they are not one in the same.

Better schedulers offer better performance.

Running as admin; which doesn’t start apps and services by default offers better performance.

you can retro actively see this yourself on anything with e-cores.

Get windows 10 1701 and Windows 10 1903 (or w/e the latest is) and a performance uplift can be seen even on 12th gen CPUs. With the same two methods.

It’s just easier to see now because OSs are heavy but performance could always be seen as far back as Vista service packs doing these two things. (As later service packs included scheduler upgrades for multicore CPUs).

You don’t need to take my word for it either. You can tangibly see this using any spare drive and a format. Go through OOBE create your user account do windows updates reboot and check your running processes and services.

Then format again and this time only enter the admin account. It’s not as in-depth as full scale testing (though you can!) but it is a good optical verification.

 

[mkppo]

Yep, right now there seem to be three things:

1) Admin account, which reduces the bloat affecting performance
2) 24H2 update, which offers better scheduling and BP optimizations
3) BIOS tweaks, like force enabling prefetchers etc like Wendell showed.

All of those increase game performance, but #1 has a bigger gain on Zen 5, #2 has only slightly higher gain on Zen 5 and #3 seems to only affect Zen 5.

 

[AusWolf]

I had moved on to another OS after my first experience with windows.

It was Windows XP on a Pentium 4 computer. Anytime I tried to change the look of the system (via 3th party tools) I noticed that every few days I had a complete system crash. I think windows XP just wasn't stable on the hardware because we had a second PC with the same hardware that I hadn't changed anything on and it also crashed very frequently.

Despite having already said goodbye to windows for good after windows XP, I do have an idea of its successors since people around me use it.

Windows Vista: this was far too heavy for the hardware that was popular at the time. Like windows XP, it was also not completely stable on much of the hardware at the time.
Windows 7: I found this to be the only windows that felt professional. It was reasonably light for the hardware at the time. It was also reasonably stable and had few flaws. And I still use it now in VirtualBox, it (fully) boots in 7 seconds (login included) on FreeBSD via VirtualBox. It kind of feels then that Window 7 starts up as fast on FreeBSD as e.g. opening Photoshop on windows, which is impressive.
Windows 8: The worst UI, worse than all the UIs I have used on Linux. Many loyal windows fans dumped it as soon as possible for windows 7 or windows 10.
Windows 10: Feels a bit like they mixed the start menu of Windows 8 with the classic start menu of Windows 7/XP. I personally found this an ugly windows and didn't think it worked very well either. The UI of Windows 7 worked smoother.
Windows 11: A windows that uses a lot of RAM. I find the UI pretty OK for windows but a bit ugly and I also notice that Linux UIs perform many daily operations with fewer intermediate stops. Windows11's UI is less efficient than LXQt, KDE Plasma, Cinnamon, etc.

I used Linux for a very long time. Ubuntu 10.10 I really liked and was one of the first Linux systems I used.
The emerald themes that were available then are still not (visually) matched by other UIs I have used since then.

At one point, I had switched to FreeBSD, which I found to be similar or better than Linux in many areas. Something that bothered me was the default security. Eg Intel microcode from the CPU was not used automatically, I had to install and activate it manually.
Now I use OpenBSD which places more importance on the out-of-the-box security aspect.

What is often said about Zen 5 is that it is more stable than Intel gen 13&14. But it must be said that in experienced hands, FreeBSD and OpenBSD are more stable than Debian stable and 'any' windows. By which I mean that the Intel 12700KF on OpenBSD is more stable and reliable than Zen 5 + windows 10/11.

The ranking on stability of operating systems based on my extensive experiences.
1. FreeBSD (bspwm and Polybar)
2. OpenBSD -current (bspwm and Polybar)
3. Debian stable, GNU Guix
4. Windows and macOS (many open-source apps are very buggy on macOS)

I'm a big fan of bspwm (window manager) which feels a bit like a genius determined what the all-important features are that I will need. I think I only use the default configuration file with a few minor tweaks.
But as I use it, it makes multiple screen setups fairly redundant (or no great value).

The keybinding to switch workspaces works smoothly and very fast. It is a manual tiler so I can easily scale the ratio of windows to the size I want. I can also easily switch to floating or fullscreen modes.

What I also find very useful is that I can determine in advance whether a window will be scaled horizontally or vertically via Ctrl+Mod+h/j/k/l

I also find Ctrl+G incredibly powerful and I don't know if there is an equivalent in Windows and macOS. It swaps your current 'small tiled window' to the largest window that is open in another workspace.

There are also shortcuts to swap tiled windows directly and many other useful keybindings.

Furthermore, I have tweaked Polybar so that I can close/minimize/maximize any window with one mouse click. I have a configuration that I can no longer improve in terms of productivity and that I could reuse forever on any Unix-like system that supports bspwm. (=almost all Unix-like systems)

When Windows 10 stops supporting my games or my hardware, I'll be Linux-bound myself. There's no way I'm gonna get Win 11, or any other bloated, spyware-ridden crap.

 

[chrcoluk]

On my X3D ccd the gains are minimal.... on the frequency ccd the gains are larger probably why you're seeing minimal gains.

Honestly this seems to make the 7700X perfom closer to the 7800X3D imho we Will have to wait for a large selection of games tested to know for sure.

I do have a 4090 as well though but at best I'm seeing 5% on the 3d cache ccd and 10% on the frequency CCD.

I did have a suspicion X3D wouldnt fare as well, as the extra cache in my own brain logic is already band aiding poor predictions.

 

[eidairaman1]

Windows is so bloated that i'm betting every CPU is seeing a bump with the update -- they probably found a scheduler incompatibiliity on ryzen systems ot they patched in a detect for AMD and shut off one of the 12400591759 nanny threads running in the background that was throwing off the prefetcher etc. Cyberpunk did something similar with their game where SMT on AMD was busted.

Unfortunately, for AMD this does not change the Zen5 value proposition vs the 7700. 9800X3D will be a beast though.

Everyone makes fun of the Fine Wine argument - but it's kind of classic AMD at this point.

Windows still has so much fat on it.

Also wIntel bias from ms.

 

[kapone32]

This might explain why my 9950x 24h2 based IoT build has seen fine performance as far as I can tell...

I mean the TPU review pretty much show it is one of the fastest CPUs for computing.

 

[DAPUNISHER]

The update is now available for 23H2 https://www.tomshardware.com/softwa...users-will-see-ryzen-performance-improvements.

When Windows 10 stops supporting my games or my hardware, I'll be Linux-bound myself. There's no way I'm gonna get Win 11, or any other bloated, spyware-ridden crap.

I believe you. 95% (pulled that number out of my bum) of the bellyachers will upgrade. It happens every windows version. Most diehards hold out as long as they can. I clung to 2000pro as long as I could.

 

[SL2]

AMD said

“Use the ADMIN account”

Where did they do that, do you have a source?

I mean, where do they explicitly tell users to do that (not how they did their own tests).

It would be highly irresponsible to tell people to use admin.

When Windows 10 stops supporting my games or my hardware, I'll be Linux-bound myself. There's no way I'm gonna get Win 11, or any other bloated, spyware-ridden crap.

If Windows 1.0 have a spyware rating of 0, and Windows 11 have a spyware rating of 10, what rate would you give Windows 10? 7? 9?

I mean, it's bad in W11, but it's not like it's a new thing.

 

[AusWolf]

If Windows 1.0 have a spyware rating of 0, and Windows 11 have a spyware rating of 10, what rate would you give Windows 10? 7? 9?

I mean, it's bad in W11, but it's not like it's a new thing.

To be fair, W10 sits right at my tolerance level. I know it takes anonymous data from me, but W11 goes too far by needing an online login, and I can't be asked to bother with custom installers and modified versions. I'd rather just swap over to Linux.

 

[SL2]

To be fair, W10 sits right at my tolerance level. I know it takes anonymous data from me, but W11 goes too far by needing an online login, and I can't be asked to bother with custom installers and modified versions. I'd rather just swap over to Linux.

AFAIK, autounattend.xml works. It's not enough of a hassle for me personally to switch OS.

The day I can't find any workarounds I might consider something else.

(Yes, BypassNRO seems to be working for some.)

 

[R-T-B]

Yes, BypassNRO seems to be working for some.

It's an officially supported solution on Pro SKUs IIRC.

 

[AusWolf]

AFAIK, autounattend.xml works. It's not enough of a hassle for me personally to switch OS.

The day I can't find any workarounds I might consider something else.

(Yes, BypassNRO seems to be working for some.)

Nah, this still seems like too much hassle to me. Downoad iso -> create bootable media -> boot from said media -> install. That's how an OS should work. If Windows doesn't offer this anymore, then Linux will.

 

[A Computer Guy]

Nah, this still seems like too much hassle to me. Downoad iso -> create bootable media -> boot from said media -> install. That's how an OS should work. If Windows doesn't offer this anymore, then Linux will.

I forgot what tool it was (maybe rufus) but basically during the step "create bootable media" you literally just uncheck some boxes to bypass the typical Win11 install requirements: TPM, online account, data collection, etc...

 

[R-T-B]

I forgot what tool it was (maybe rufus) but basically during the step "create bootable media" you literally just uncheck some boxes to bypass the typical Win11 install requirements: TPM, online account, data collection, etc...

Yeah rufus really makes it a cinch if thats really what is holding you up.

 

[AusWolf]

I forgot what tool it was (maybe rufus) but basically during the step "create bootable media" you literally just uncheck some boxes to bypass the typical Win11 install requirements: TPM, online account, data collection, etc...

Yeah rufus really makes it a cinch if thats really what is holding you up.

Maybe I'll have a go then.

How does it do with advertisements and auto-installing apps? Those are my other no-go areas for an OS that Windows has been infamous for lately.

 

[LittleBro]

That 5800X3D and 9600X uplifts are insane in some games, puts even 7800X3D to shame:

Optimierte Sprungvorhersage: Zen 5 mit Windows-Update KB5041587 getestet - Hardwareluxx

Optimierte Sprungvorhersage: Zen 5 mit Windows 11 24H2 getestet.

www.hardwareluxx.de

To sum it up, to really get the performance you paid for, these steps are inevitable:
1. Install Windows 10 (wait for Zen 3/4/5 branch predictor fix) or Windows 11 24H2 with branch predictor fix; (+11% perf.)
2. Disable VBS on Windows 10/11 or use hidden Administrator account which bypasses VBS; (+5% perf.)
3. Adjust few settings related to CPU in BIOS. (+?% perf.)

Could someone post a link to the article/video regarding Zen CPU BIOS settings (prefetchers, ...)? Thanks.


BTW, 9700X at 105W TDP shows 13.4% improvement over 9700X at 65W and 16.9% improvement over 7700X at 105W in Cinebench R23:
7700X @ 105W: 19 800 (baseline: 100% TDP, 100% perf.)
9700X @ 65W: 20 409 (-38% TDP, +3% perf.)
9700X @ 105W: 23 153 (+0% TDP over 7700X, +16.9% perf. over 7700X) (+61.5% TDP over 9700X 65W, +13.4% perf. over 9700X 65W)
This, indeed, shows that 9700X at 65W should have been released as 9700 without "X" as those models are always much more efficient.

MSI adds "TDP to 105W" mode for Ryzen 9700X and 9600X: up to 13% higher Cinebench performance - VideoCardz.com

MSI first to offer 105W TDP mode for 65W Ryzen 9000 SKUs The updated firmware adds an option to increase TDP over AMD specs. This feature is not officially provided by AMD but is available through the BIOS settings on MSI 600-series motherboards with the latest AGESA 1.2.0.1 firmware. The “TDP...

videocardz.com

This performance gain has nothing to do with branch branch prediction fix and bypassing VBS security. With those two issues fixed the results may be even better.

 

[Assimilator]

I'm just sitting here with a Zen 2 Threadripper and popcorn, knowing that I can effectively stay on Win10 forever. And by the time I do decide to upgrade my system (note: not necessarily GPU), it will be to something that is at least a generation old and thus actually works OOTB, so that I don't have to be a beta tester for a CPU company.

 

[SL2]

Yeah rufus really makes it a cinch if thats really what is holding you up.

Rufus does the same thing as the one in my link, both uses autounattend.

That 5800X3D and 9600X uplifts are insane in some games, puts even 7800X3D to shame:

The update works on the 7800X3D as well according to your own link.

 

[LittleBro]

The update works on the 7800X3D as well according to your own link.

Sure but at some games 5800X3D's performance gain is better than 7800X3Ds. For a 2 generation old chip, not bad.

 

[SL2]

I'm just sitting here with a Zen 2 Threadripper and popcorn, knowing that I can effectively stay on Win10 forever. And by the time I do decide to upgrade my system (note: not necessarily GPU), it will be to something that is at least a generation old and thus actually works OOTB, so that I don't have to be a beta tester for a CPU company.

I can't really say anything about having to be a beta tester for Windows in general, but this update showing up too late doesn't seem to be much about testing.

If it's really true that the update was included in Zen5 laptops (July 28), it means that someone messed up with the distribution of the update in WU, but nothing suggests that it was in need for any testing after it was released.

 

[Assimilator]

I can't really say anything about having to be a beta tester for Windows in general, but this update showing up too late doesn't seem to be much about testing.

If it's really true that the update was included in Zen5 laptops (July 28), it means that someone messed up with the distribution of the update in WU, but nothing suggests that it was in need for any testing after it was released.

I'm talking about being a beta tester for AMD failing to have the required Windows update already shipped before they released Zen 5.