AMD Ryzen 9 9950X3D

[Epaminombas]

For me its not video games rather my photo tools holding me onto windows as I find Mac too restrictive(even though every single of my software has Apple silicon versions available).

It must be very frustrating for Linux users to see image and video editors prefer to pay $3000 for MacOS rather than use Linux for free.
It must be unbearable pain.

Indeed, that is exactly what I mentioned. It scales much better under Linux, to the point I think using Windows is actually actively detrimental to this architecture. Their results are crazy good. Unfortunately, I cannot give up on Windows just yet, because I like my video games too much.



It was proven the moment AMD released the 5950X... 8+8 works, 8 3D + 8 3D would work the exact same. The problem is that resources are reported to Windows as available, but it is not aware of where they are physically present in the processor. Each CCD is effectively a different node, and there are penalties for accessing data on the adjacent node, which is what causes the loss of performance and efficiency seen with these chips under Windows, and arguably why Linux is so much faster.

I imagine and share your pain.
It's frustrating to see people prefer using Windows with lower performance than using Linux with 99999% more performance.

must be brainwashed by microsoft

 

[AVATARAT]

There are thousands of benchmark videos of Windows 11 winning in over 90% of games in terms of performance in all types of configurations.

Showing that there is no performance problem in Windows and that there is no superiority of Linux based on CPU task scheduling.

View attachment 389014



View attachment 389015

View attachment 389016

By default, there is a high latency between the chiplet cores. Linux handles this better than Windows by simply not transferring the data across the CCD, nothing more, nothing less.

Regarding gaming, you can jump all you want about how much "better" games run on Linux, but when I go to some screenshots in the Benchmarking section, somehow I can't see that you're right, but all the Linux results are far from reality And hey, that's pretty normal

 

[Onasi]

It's NEVER been proven that two would help much (because it would still be TWO CCD's), and that's why it must be the solution?

I don’t think you fathom my power level. I don’t actually particularly care if the impact on performance is there, if it’s minor, major, nonexistent, doesn’t matter. @Dr. Dro was kind enough to already elaborate on that if you are curious. I just don’t fuck with the XBox Game Bar or shit like Process Lasso and, worst of all by far, the non-symmetrical layout triggers my OCD something fierce. I couldn’t possibly run something that hideous in my system without suffering a stroke.

 

[mb194dc]

Ultra niche use case product. The 9950x cheaper and if you're gaming at 1440p+ and don't have a 090 Gpu you won't be able to tell the difference. Not to mention other options that are available.

If you've got 5k to burn on a system and want the regular desktop best, go for it.

 

[JustBenching]

The regressions are weird too. Everyone points to the fact the memory controller is on a separate die then the compute making for a lot of latency, but AMD does this too and isn't as impacted in gaming so there's obviously a difference in implementation. Suspect next Intel CPUs should fix that but we will see.

There's been some articles if you overclock the interconnect and E cores on the Intel CPUs the gains are pretty big. Kind of want one just to try and mess with for a bit. I'm in the tinkering mood these days.

How do you reckon amd isn't suffering from the same issue? The 9950x is as fast as the 285k (they are within 2% of each other in games) and both are slower than the 13900k. The x3d chips don't have that issue cause they are less reliant on memory due to the large cache.

 

[Epaminombas]

By default, there is a high latency between the chiplet cores. Linux handles this better than Windows by simply not transferring the data across the CCD, nothing more, nothing less.

Regarding gaming, you can jump all you want about how much "better" games run on Linux, but when I go to some screenshots in the Benchmarking section, somehow I can't see that you're right, but all the Linux results are far from reality And hey, that's pretty normal

Show me 1 game that exists natively on Ubuntu and that game exists natively on Windows. Show me the performance difference between the two.

Show me 1 image editing or engineering software that exists natively on Ubuntu Linux that also exists natively on Windows.

So far I haven't seen any benchmarks showing this and you won't see any.

 

[dgianstefani]

Process Lasso has per-app presets that you can set once and forget, and works almost flawlessly, so no need to manually reboot each time. Maybe someone will test a "lassoed" 9950X3D to see what the performance penalty is compared to a standard 9800X3D. But on Alder and Raptor Lake it did good on games that didn't like E-Cores.

Doesn't work with many games that use anti cheat, as they don't let you adjust affinities/priority.

 

[Draconis]

Unfortunately, I cannot give up on Windows just yet, because I like my video games too much.

If you have a spare SSD, time and the inclination I suggest test Linux and gaming. I'm not suggesting you switch, just give it a try to see how far it has come. If you install a Steam supported distro all you need to do is install the OS, install the Steam client and install a small game and give it a go.

 

[AVATARAT]

Show me 1 game that exists natively on Ubuntu and that game exists natively on Windows. Show me the performance difference between the two.

Show me 1 image editing or engineering software that exists natively on Ubuntu Linux that also exists natively on Windows.

So far I haven't seen any benchmarks showing this and you won't see any.

Just tell me which libraries emulate Proton and Wine, and then tell me when the emulation might be faster than your native.

I can agree here.

Yeah, you won't see any because you can't get a better result under Linux, and that's why benchers don't waste their time with nonsense

 

[Chrispy_]

...and it's no surprise to anyone that the 9800X3D is still better for gaming, because even in 2025 there's no way to guarantee that a game won't get split across two CCDs with massive latency penalties that hurt performance. I mean, I guess you can disable half your CPU, but then you're just paying more to buy what is effectively a 9800X3D anyway.

Roll on Zen6, it's about time we had a 12-core, single-CCD, X3D part.

 

[Dristun]

Doesn't work with many games that use anti cheat, as they don't let you adjust affinities/priority.

Oh. Didn't think about it. Well, I guess CS2 pros with 540Hz screens better stick with the 8-core then

 

[Epaminombas]

Just tell me which libraries emulate Proton and Wine, and then tell me when the emulation might be faster than your native.

I can agree here.

Yeah, you won't see any because you can't get a better result under Linux, and that's why benchers don't waste their time with nonsense

You don't understand.

Tell me a famous software or game that has developed its native software on Linux and also developed its native software on Windows without ANY emulation.

Which of the two will have better performance? Is it Linux? If it is Linux, how much better is it?

 

[Daven]

...and it's no surprise to anyone that the 9800X3D is still better for gaming, because even in 2025 there's no way to guarantee that a game won't get split across two CCDs with massive latency penalties that hurt performance. I mean, I guess you can disable half your CPU, but then you're just paying more to buy what is effectively a 9800X3D anyway.

Roll on Zen6, it's about time we had a 12-core, single-CCD, X3D part.

Disabling the non-X3D CCD when gaming is one way but that seems controversial here with some for and against it.

Really all enthusiast CPUs just need large pools of cache accessible by all cores which is only possible by stacking.

 

[Epaminombas]

...and it's no surprise to anyone that the 9800X3D is still better for gaming, because even in 2025 there's no way to guarantee that a game won't get split across two CCDs with massive latency penalties that hurt performance. I mean, I guess you can disable half your CPU, but then you're just paying more to buy what is effectively a 9800X3D anyway.

Roll on Zen6, it's about time we had a 12-core, single-CCD, X3D part.

There is no game that uses more than 6 cores.

You can launch a Zen6 with a 20-core CCD, the game will not use more than 6 cores.

 

[JustBenching]

There is no game that uses more than 6 cores.

There are games that use 16. What are you talking about man?

 

[Epaminombas]

Disabling the non-X3D CCD when gaming is one way but that seems controversial here with some for and against it.

Disabling CCD can cause the system to overload.

Because CCD without cache is handling other tasks than your game.

 

[Daven]

Disabling CCD can cause the system to overload.

Because CCD without cache is handling other tasks than your game.

View attachment 389021

View attachment 389022

Are you serious? Disabling the non-X3D CCD will overload the system? Then how the hell does a regular 9800X3D work?

Anyway, back to reality, this test has already been done by W1zzard. No overloads reported.

Ryzen 7950X3D with One CCD Disabled - The 7800X3D Preview

Ryzen 7 7800X3D will be the most affordable Zen 4 X3D processor, but it doesn't release until April. We reconfigured the Ryzen 9 7950X3D to match the specs of the 7800X3D and ran it through our test suite, which includes 14 games at 720p, 1080p, 1440p and 4K. Results are impressive.

www.techpowerup.com

 

[AVATARAT]

You don't understand.

Tell me a famous software or game that has developed its native software on Linux and also developed its native software on Windows without ANY emulation.

Which of the two will have better performance? Is it Linux? If it is Linux, how much better is it?

Of course

Break the libraries into DX or VC++ and wait for the game to start.

I'm waiting for the day when I jump to Linux to get better game/bench results, and I'd be happy if you could prove me wrong with real game data. But sadly, you can't.

 

[Epaminombas]

Are you serious? Disabling the non-X3D CCD will overload the system? Then how the hell does a regular 9800X3D work?

Anyway, back to reality, this test has already been done by W1zzard. No overloads reported.

Ryzen 7950X3D with One CCD Disabled - The 7800X3D Preview

Ryzen 7 7800X3D will be the most affordable Zen 4 X3D processor, but it doesn't release until April. We reconfigured the Ryzen 9 7950X3D to match the specs of the 7800X3D and ran it through our test suite, which includes 14 games at 720p, 1080p, 1440p and 4K. Results are impressive.

www.techpowerup.com

The testing is in a controlled environment.

A real person would have multiple browser tabs open.

A Twich person would be streaming to other people while playing games and having other programs running in the background.

A normal person is doing a lot of stuff on their Windows PC.

 

[Wirko]

That applies for any operating system with processors with mixed cores and structures. So many weeks have passed and there is still no proper scheduler and kernel support for this. A processor should never need so many extra lines and kernel and scheduler code to barely work properly.

AMD needs a thread director too. Ruby's coming back to apply for the job.

 

[Epaminombas]

Of course

Break the libraries into DX or VC++ and wait for the game to start.

I'm waiting for the day when I jump to Linux to get better game/bench results, and I'd be happy if you could prove me wrong with real game data. But sadly, you can't.

You caught an important point that I was going to make.

DirectX is far superior to OpenGL and Vulkan.

This is one of the reasons why Linux has lower performance when using Vulkan or OpenGL natively.

Are you going to tell me that you want native DirectX on Linux? Make me laugh.

 

[mkppo]

Here's one.

In fact after seeing that video I was half convinced to try linux again but life sort of got in the way..

9950X3D looks fast no doubt but some benchmarks don't make sense. 9950X3D is slower than 9800X3D in CP2077 but faster with RT on.

Also strange that Space Marine 2 and Spiderman 2 still cross CCD's, those two games mainly cause the performance difference between 9800X3D/9950X3D. How are those two still not fixed..

How do you reckon amd isn't suffering from the same issue? The 9950x is as fast as the 285k (they are within 2% of each other in games) and both are slower than the 13900k. The x3d chips don't have that issue cause they are less reliant on memory due to the large cache.

Now that all CPU's are on 24H2, I can see around a 4.5% difference at 1080p between 285K and 9950X. In 'TPU chart speak', 4.5% is significant at 1080p. You were basing your arguments (earlier as well) based on older TPU benchmarks without 24H2 and I told you the same. Now that 24H2 is standard, the difference is larger.

 

[AVATARAT]

You caught an important point that I was going to make.

DirectX is far superior to OpenGL and Vulkan.

This is one of the reasons why Linux has lower performance when using Vulkan or OpenGL natively.

Are you going to tell me that you want native DirectX on Linux? Make me laugh.

Obviously.

So then Windows is better? And will be in the future, because somehow I don't see companies changing their minds and forgetting about DX.

Are you serious? Disabling the non-X3D CCD will overload the system? Then how the hell does a regular 9800X3D work?

Regular users lose FPS because of background programs. Of course, it depends on the game.

 

[Dr. Dro]

If you have a spare SSD, time and the inclination I suggest test Linux and gaming. I'm not suggesting you switch, just give it a try to see how far it has come. If you install a Steam supported distro all you need to do is install the OS, install the Steam client and install a small game and give it a go.

I've done it, but many of the games I play are a no-go due to the anti-cheat systems. Not to mention the Nvidia drivers are still completely ass, and I expect them to be double ass on new hardware like the 50 series... and I'm one of the nutters who actually bought a 5090. When will I get it... that's another story altogether

 

[Epaminombas]

I've done it, but many of the games I play are a no-go due to the anti-cheat systems. Not to mention the Nvidia drivers are still completely ass, and I expect them to be double ass on new hardware like the 50 series... and I'm one of the nutters who actually bought a 5090. When will I get it... that's another story altogether

Everything in Linux is the fault of something, except Linux
It's Nivdia's fault that Linux performs worse.
It's Intel's fault that Linux performs worse.
It's AMD's fault that Linux performs worse.
It's some god's fault that Linux performs worse.
It's the game developer's fault that Linux performs worse.
It's Vulkan and OpenGL's fault that Linux performs worse.
But never, ever blame Linux, because it is perfect.

Obviously.

So then Windows is better? And will be in the future, because somehow I don't see companies changing their minds and forgetting about DX.


Regular users lose FPS because of background programs. Of course, it depends on the game.

All that's left is for the all-powerful Linux community to create an API that's superior to DirectX.

It's so simple and easy.

They just don't do it because they don't want to see Linux on the PC rise.

But back to reality. Even MacOS's Metal API is better than OpenGL and Vulkan.

So the all-powerful and all-knowing Linux community wants Microsoft to make DirectX work natively on Linux for free.