AMD Ryzen 9 9950X3D

[Draconis]

I've done it, but many of the games I play are a no-go due to the anti-cheat systems.

Ah, that is a pain point. Hopefully Valve's backing of Linux gaming in general will gather momentum and these issues will be resolved eventually.

Good luck on getting your GPU.

 

[JohH]

People keep clamoring for a dual X3D CCD version but I'd like AMD to take the binned X3D CCD from this SKU and sell it as a 9850X3D. It'd be the fastest gaming CPU, again.

 

[Onasi]

@Epaminombas
You’re in the wrong thread, sonny-boy. This is a discussion of the 9590X3D, not your personal anti-Linux soapbox or whatever you’re doing here. Just giving you a heads up before the mod team does in a less friendly fashion.

Anyway, me memeing on this CPU aside, I find it very curious how much (relatively) faster it is than its non-X3D brother in ML and AI tasks. That cache really helps that much for that? I might be out of the loop on that one, if someone more knowledgeable could elaborate I would be grateful.

 

[Chrispy_]

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.

Not sure why there's any controversy. The tests have been done by professionals across multiple channels and review sites, it's conclusive and unanimous with absolutely no grey area for doubt or misinterpretation.

The only people in denial are those who choose to ignore the hard evidence posted by numerous sites that clearly document their testing methods. What was true for the 7950X3D and 7800X3D still holds true for the 9950X3D and 9800X3D; Today's slew of reviews confirm exactly the same behaviour we've been seeing for years with dual-CCD processors in gaming loads.

I'm running a 9900X right now knowing full well that the gaming experience is probably less consistent and therefore inferior to the 5800X3D it replaced. If I was only gaming on that particular system I wouldn't have even bothered with the upgrade.

 

[Denver]

The AMD driver on Linux is so well-polished that it often outperforms Windows, even when MS OS benefits from running games natively. I never imagined this would happen

Windows vs Linux gaming using a 7900 XTX | CachyOS | Nobara 40 | linux performance | 9800x3D - YouTube

Linux Gaming Nvidia Vs AMD In 17 Games | 7900 XTX | 4080 Super | OpenSUSE Tumbleweed | GPU Benchmark

 

[JohH]

So your solution is to demand more and more powerful CPUs and completely forget about the lack of optimization from developers?

Yes, writing multi-thread interactive simulations in quick fashion is actually near impossible and will never be done on time schedules the game industry demands.

Binning a better a X3D CCD is already done for this SKU. So it is a trivial new SKU to add.

 

[Onasi]

I think I'm pretty normal but I don't do any of those things when playing a game.

Same, to be honest. Never understood this - if I am playing single player I want to be immersed and focused, so why have any background distractions? If I am playing MP with the lads it’s kinda the same, just with a VoIP software working somewhere in the background minimized. Like, why the fuck would I have my browser with 50 tabs open while I play a game? Should I also maybe F@H and render in Blender also while I’m at it?

 

[Feyd]

@dgianstefani :
Page 25: "Please note that the cores without 3D V-Cache are clocked a bit lower than the cores without cache."

 

[Makaveli]

The 9950X3D is 12.5% faster than the 7950X3D. Closets to the 15% IPC estimate that I've seen so far. Nice!

Edit: NVM. The 9950X3D has about a 7% average higher clock than the 7950X3D so the IPC gains are minimal. Good news is that the 9950X3D clock speeds are identical (slightly higher?) to the non-X3D 9950X. So that problem is solved.

If you look at the THG review they list clock speeds during the benchmark runs.

The gap between 9800X3D 5,257 Mhz vs 9950X3D 5,348 Mhz is smaller than I thought it was going to be.

9950X is hitting 5,590 Mhz

 

[leonavis]

These dual CCD chiplet CPUs are such powerhogs. Samish effiency as a 14900K in gaming so terrible in general while also drawing almost twice as much in idle & low power scenarios. For my use case I'd probably draw less power in a year on a 14900K than a 9950X3D.

Then you probably don't need a high end CPU

I was hoping to see 7950x results in the mix.

Is it worth going from 7950x to 9950X3D?

Depends what you need and already have. With a 5090 on 1080p/1440p for the last few FPS, useful. E-Sports etc.

Otherwise in gaming you're GPU-bound mostly anyway. So probably not.

Ps. GN has the 7950X in their mix on yt

 

[ThomasK]

Great so far.

Let's see if amd eventually comes up with the 9955X3D, 9950X3D+ or whatever it ends up being called, with another pile of cache below the 2nd ccd.

 

[BorisDG]

Can't wait for availability, so I can upgrade my rig.

 

[igormp]

The manufacturer creates the drivers; Windows does not have full control over how software wants to use CPUs for this purpose.

Baixar o Kit de Driver do Windows (WDK) - Windows drivers

Instruções para o download da versão mais recente do Kit de Driver do Windows (WDK)

learn.microsoft.com

This WDK Kit is used by manufacturers to create drivers and test and approve them in quality tests. People think that Microsoft is the one who creates the drivers.

You can create software and force it to use specific cores, bypassing what the AMD driver initially wants to do.

What Microsoft does is improve hardware compatibility in the latest Windows 11 BUILD released.

Windows 11 24H2 has improvements for new CPUs and GPUs.

Previous versions do not.

And there are a lot of people buying new hardware from 2024/2025 and installing outdated Windows 10 or an old Windows 11 version that is not compatible. Then these people come to the forum to complain about poor performance.

While it's true that AMD is responsible for the driver, the scheduler is on MS's end, and Windows' scheduler has been always bad.
As a comparison, the Linux scheduler deals way better with heterogeneous CPUs.

There is no such thing as a lack of CCD optimization in servers, because only server software knows how to deal with many cores.

Or servers have thousands of tasks at the same time that use several cores.

Servers also often run on Linux, and no server CPU has heterogeneous core designs.

Intel's thread director works around this issue seamlessly through hardware-based runtime feedback to the operating system by literally telling it where to issue and physically execute each instruction

And even that still has its issues, but it's also not really necessary on Linux, it's just a duct tape for window's schedulers to alleviate issues such as what AMD faces.

This is especially true with Nvidia GPUs, because the Nvidia drivers on Linux are decidedly worse than on Windows and also inferior to AMD's.

Totally disagree on that, but I believe you're talking just about games, so let's leave it at that

I actually speak English, but if you want to debate this in Portuguese, I'll be happy to oblige in DMs.

Include me, I want to be part of the fun lol

Anyway, me memeing on this CPU aside, I find it very curious how much (relatively) faster it is than its non-X3D brother in ML and AI tasks. That cache really helps that much for that? I might be out of the loop on that one, if someone more knowledgeable could elaborate I would be grateful.

There you go:

AMD Ryzen 9 9950X3D Delivers Excellent Performance For Linux Developers, Creators & Technical Computing Review - Phoronix

www.phoronix.com

It's a bit faster, but not by much. Be aware that the test in there for llama is only regarding prompt processing, not token generation.

 

[HD64G]

As expected, the greatest all-around CPU. I am less surprised by its performance and more surprised by AMD setting the same MSRP as they did for the less impressive back then 7950X3D. That's a nice marketing move and will help PC customers more if they keep making plenty of all their X3D CPUs that are in production today.

 

[Mr_Engineer]

Holy Crap! Over at Phoronix, Zen 5 Linux performance is getting crazy good!

AMD Ryzen 9 9950X3D Delivers Excellent Performance For Linux Developers, Creators & Technical Computing Review - Phoronix

The 9950X3D beats the 7950X3D by 25%!!!
The 9950X beats the 7950X by 20%!!!

I think it's time to admit that Windows is no longer able to recognize new CPU architecture changes like in the past. I suggest that this is due to frequent gen-to-gen IPC competition between AMD and Intel. Microsoft apparently cannot respond like Linux can.

Agreed. Especially Windows 11 performance is still broken, even after the Win11H2 updates. It's still not getting optimal performance out of the new architectures (Zen 5 and probably even Intel Arrow Lake).

Thank you Microsoft

 

[Wirko]

@W1zzard - An idea (involving more work of course): could you make another set of 3 graphs of clock frequency vs. thread count with the OC settings you described below?

 

[Dr. Dro]

Why did this topic take this Linux vs Windows war turn? The arguments also look like this;

View attachment 389070

I'll take the blame, since I brought up that the R9 X3D CPUs tend to perform better under Linux, as corroborated by the phoronix review. They are MUCH faster under Linux than they are under Windows. Same applies to the 7000X3D series, by the way. They're still great on Windows but you'll get the best out of it on an OS that can actually schedule and issue instructions to the correct physical resources on the processor. This problem is particularly egregious on the 7900X3D and 9900X3D, because they only have 6 cores on each side. Regardless of 6 or 8 core CCDs, without something equivalent to Intel's "thread director", it's about impossible for these to behave perfectly under Windows and as a result they will always trail benchmarks, with the higher end model just about getting a pass because most apps will run perfectly on 8 cores, but not all will on 6.

 

[Makaveli]

I'll take the blame, since I brought up that the R9 X3D CPUs tend to perform better under Linux, as corroborated by the phoronix review. They are MUCH faster under Linux than they are under Windows. Same applies to the 7000X3D series, by the way. They're still great on Windows but you'll get the best out of it on an OS that can actually schedule and issue instructions to the correct physical resources on the processor. This problem is particularly egregious on the 7900X3D and 9900X3D, because they only have 6 cores on each side. Regardless of 6 or 8 core CCDs, without something equivalent to Intel's "thread director", it's about impossible for these to behave perfectly under Windows and as a result they will always trail benchmarks, with the higher end model just about getting a pass because most apps will run perfectly on 8 cores, but not all will on 6.

i've avoided these dual CCD's chips because of this.

I will move up in core count when we see Zen 6 and 12 cores on a single CCD.

 

[Niceumemu]

Amusing how AMD hasn't sent any 9900X3D chips out to any reviewers - goes to show that even AMD is aware you need to be stupid to buy that SKU.

Still think the 9950X3D isn't a good buy except in very specific cases - if you can actually benefit from the additional cores over the 9800X3D then you're better off having 2 PCs, one with a 9800X3D and another with a 9950X or a threadripper part. If two PCs doesn't make financial sense then a 9950X3D also doesn't IMO.

 

[Guwapo77]

I'm grabbing this mofo first thing in the morning! If that 5090 is still there...I might scoop that too (still can't wrap my mind around $3,330 GPU...so I'm going to let it sit in the store overnight)


Update: Thank goodness both of those Asus 5090s are gone, no temptation.

 

[freeagent]

i've avoided these dual CCD's chips because of this.

They run fine.. nothing special here just a stock windows install.

 

[Solaris17]

Man this thing is nice!

 

[A Computer Guy]

They run fine.. nothing special here just a stock windows install.

View attachment 389109

Similar to your results I've noticed when gaming on my 5950x Windows tends to use the first CCD for gaming and usually little to no activity on the 2nd CCD.

 

[MCJAxolotl7]

The manufacturer creates the drivers; Windows does not have full control over how software wants to use CPUs for this purpose.

Baixar o Kit de Driver do Windows (WDK) - Windows drivers

Instruções para o download da versão mais recente do Kit de Driver do Windows (WDK)

learn.microsoft.com

This WDK Kit is used by manufacturers to create drivers and test and approve them in quality tests. People think that Microsoft is the one who creates the drivers.

You can create software and force it to use specific cores, bypassing what the AMD driver initially wants to do.

What Microsoft does is improve hardware compatibility in the latest Windows 11 BUILD released.

Windows 11 24H2 has improvements for new CPUs and GPUs.

Previous versions do not.

And there are a lot of people buying new hardware from 2024/2025 and installing outdated Windows 10 or an old Windows 11 version that is not compatible. Then these people come to the forum to complain about poor performance.

Yes'nt, windows DOES have scheduler stuff, GNU/Linux has always imho had better cache scheduling, especially for 3D-V cache
Also, Dual/Multi CCD cache would rock on threadripper or even epyc
Edit: GNU/Linux, somehow i make the mistake as I type it on the OS itself

 

[Dr. Dro]

Amusing how AMD hasn't sent any 9900X3D chips out to any reviewers - goes to show that even AMD is aware you need to be stupid to buy that SKU.

Still think the 9950X3D isn't a good buy except in very specific cases - if you can actually benefit from the additional cores over the 9800X3D then you're better off having 2 PCs, one with a 9800X3D and another with a 9950X or a threadripper part. If two PCs doesn't make financial sense then a 9950X3D also doesn't IMO.

At the prices asked, it's a total upsell. That's the CPU that'll go to the bargain bin for below 9800X3D prices down the road, just as it happened to the 7900X3D... because it won't push the frame rates gamers want, and it won't pull its productivity weight for the creator camp. It's a CPU that no one wants, and AMD is releasing likely for binning reasons.