Review AMD Ryzen 9 9950X Review: Zen 5 at Full Power

[jeremyj_83]

It'll be the same 6400 as Zen 4 (they changed this with the AGESA that introduced higher speed memory support), but it'll be silicon lottery. Quite frankly I'm a bit concerned about the memory support because GN had a 9600X that couldn't run EXPO and TPU had a 9900X that wasn't stable with higher speed DRAM without a SoC voltage increase. If that crops up with reviewers it makes me a bit leery about how the overall CPUs are.

I know that some places had instability but I figure that can be fixed with BIOS updates. This is why I'm not running out and buying a new CPU right now and waiting a little while. Hopefully it will be rectified around the X3D launch, however, I am slowly getting the components together right now to replace my venerable 11 year old 4770k. RAM was a bit question as prices looking like they will go up a bit shortly and 6400MHz RAM is only $10 more than 6000MHz RAM. Thanks for the info.

 

[Mattzun]

Interesting review.
Great to have a review with a base config similar to a Dell or HP

The threading and core parking issues seem like bigger issues than gaming or pricing.

Seems like AMD should have waited a month or two to release the 9900/9950

 

[King_V]

Anyone know if Zen 5 will allow for a mclk/uclk 1:1 with DDR5-6400 RAM or will that max out at 6000 again?

This announcement regarding new G.Skill DDR5 CL30 RAM seems to suggest that it will. I wouldn't call it definitive, though.

G.Skill's new memory options will be optimal for AMD's AM5 platform, particularly for users running AMD's Ryzen 7000 and Ryzen 9000 series processors. Due to fabric limitations in AMD's Zen 4 and Zen 5 architectures, DDR5-6000 to DDR5-6400 is considered the limit of what AMD's AM5 Ryzen processors can achieve while running the Infinity Fabric at a 1:1 ratio.

 

[ezst036]

The differences in performance between Windows 11 and Ubuntu 24.04 LTS is quite striking. Does make you wonder what is Ubuntu doing differently that Windows and can Windows 11 get parity with Ubuntu on this front with just an update or will we have to wait for Win 12.

Well, for one, Windows is now an advertisement delivery system so that Microsoft can make an $extra $ buck $.

Ubuntu has no such overhead. APIs, graphics delivery, etc etc. Whatever is needed to make the advertisements work. Ubuntu isn't trying to adware you.

 

[stuff and nonesense]

Ok, but why the change is irreversible ? Never hear of something like this before with any peripheral. Usually there is some dll to replace or some registry setting to change and so on, but a reinstall of the OS is absurd. Think of a professionist that work with the PC and have plenty of softwares and peripherals to reinstall and reconfigure.

You could ask the TH cpu reviewers that question.

 

[thestryker]

Ok, but why the change is irreversible ? Never hear of something like this before with any peripheral. Usually there is some dll to replace or some registry setting to change and so on, but a reinstall of the OS is absurd. Think of a professionist that work with the PC and have plenty of softwares and peripherals to reinstall and reconfigure.

From the TPU review it's guidance direct from AMD (and from my recollection this guidance has been the same since they debuted it with Zen 4 X3D):

Just two days ago, AMD notified us that "Ryzen 9 9900X and 9950X have Windows Game Mode core parking optimizations installed by the AMD PPM Provisioning File Driver," and that "Windows game mode must be enabled," "Xbox Game Bar must be enabled and up to date in the Microsoft Store," "the legacy Control Panel Power Options must be set to the default 'Balanced' scheme," and that "sometimes Windows does not apply the correct provisioning after the CPU installed has changed. You can try uninstalling then re-installing the AMD Chipset Driver as a workaround, but a fresh install of Windows is ideal."

 

[domih]

Gamers on Windows are disappointed 🙁

Professional users on Linux love it 😀 As reflected on Phoronix, Level1Techs and Chips and Cheese, vs. the rest of the reviewer planet.

Note: I'm using a Threadripper 3960X 24C48T since 2019, with Linux. I'll only start looking for a full upgrade by 2029.

 

[Mattzun]

Well, for one, Windows is now an advertisement delivery system so that Microsoft can make an $extra $ buck $.

Ubuntu has no such overhead. APIs, graphics delivery, etc etc. Whatever is needed to make the advertisements work. Ubuntu isn't trying to adware you.

The bloat is getting to the point where Linux can be faster than Windows for games.
That is a truly impressive level of overhead.

 

[m3city]

Zen 5 isn't actually 10% faster in gaming even when you compare Zen 4 using lower speed memory. I'm also not sure why anyone would buy something at a premium price today in hopes the performance would be better tomorrow. These CPUs simply don't deliver the amount of performance and efficiency required to justify their price premium unless you're running AVX512 workloads.

The Zen 5 parts are good and in time will likely supplant Zen 4, but right now they're just not very impressive. This is very much like the sort of uplifts we saw gen on gen while Intel was controlling the market. In this case paying 25% more over the prior generation to get this little just doesn't seem very logical.

Just to clarify my %. Article says 8% in gaming , 12% in apps, so I simplified it to 10%. Power draw 7÷15% less, simplified to 10% again.

Upgrade from 7950 is obviously a waste of money as it's a great CPU. Upgrade from 5000? Sure. And yes - upgrade hoping that perf will get better. It works that way when it comes to fresh GPUs, doesn't it? I do realise that improvements similar to Intel arc are not to be expected, but still.

 

[Amdlova]

Got one 14600T from ebay 210usd maybe make some tests put on tom's hardware forum.
13600T x 14600T aka refresh... the 13600T with 50w power limit make the same multi tread performance of the 9600X at 88w (cinebench r23)

I will try UPDATE the motherboard to run ddr5 (still using ddr4) maybe get another 20% of raw power

 

[thestryker]

And yes - upgrade hoping that perf will get better. It works that way when it comes to fresh GPUs, doesn't it? I do realise that improvements similar to Intel arc are not to be expected, but still.

No it really doesn't work that way unless something is not working right and seeing as AMD hasn't issued any guidance that doesn't seem to be the case here. Intel having to redo their entire driver and hardware philosophy is why it mattered for Arc, but when it comes to CPUs and most GPUs you simply won't see any significant differences from launch onward.

If you want to go through and look I'm fairly certain you'll find very few performance changes in like for like hardware despite these reviews being 15 mo apart (they did change GPUs, but the rest is largely the same):
https://www.techpowerup.com/review/intel-core-i9-13900k/4.html
https://www.techpowerup.com/review/intel-core-i9-14900ks/4.html

 

[acadia11]

Took me 10 years to upgrade from i7-860 to zen 4 7950x , and now I see why my upgrade cycle is so long. I'm kidding of course I've had plenty of other machines in between but seems this 7950x will last me quite a while as well.

 

[usertests]

What you expect from AMD next gen will be better

Just Wait

Disappointment? AMD spoiled you. Excuse me for simplification of numbers, but based on info here, a direct comparison to older sibling gives 10% more perf in gaming/apps with 10% lower power draw. Similar price. Same platform. And when drivers, firmware mature, then you may get even more. Now if I was about to update CPU from sth AM5 then I would have to be plainly dumb to choose 7950 over 9950.

There are clearly issues. Maybe they're primarily the fault of Microsoft Windows, but something is up. It's no coincidence AMD delayed the launches by 1-2 weeks.

 

[neojack]

thanks AMD, at this pace, my 5800X3D will remain relevant for a many more years
feels as good as owning a paid-off car

 

[jg.millirem]

What if a person doesn’t care about gaming and just wants an all around great productivity machine - do the core parking issues matter then?

I was set to wake up early tomorrow to order from Newegg. Perhaps no longer.

 

[cyrusfox]

Sheesh, MLID sure called this one wrong(Larger increase than zen 3 to zen4), I was expecting 15-20% uplift as was leaked, what I am seeing is 6% in Single thread, and 18% in multi thread, with an aggregate being between 8-10% uplift.... With all the bugs and the price premium, I would only get this for crunching and encoding.

 

[braxus]

How much faster is the 9950X vs 7950X for rendering video, Photoshop, and other productivity things? Im not into gaming, so I dont care about results for that.

 

[YSCCC]

skimmed through Jayz review last night, looks nice for MT performance especially for the power it draws without enabling PBO being around 200W mark yet outperforms the 14900k by 2-3k marks in CB R23, gaming looks respectful but not ground breaking, all of these achieved while not going to full fledged oven mode, kind of overpriced though, but at this point of time if one wants TOTL working performance with great gaming performance you have basically no choice, the X3D likely will excel in gaming but for rendering or video editing it likely suffers

 

[Mattzun]

thanks AMD, at this pace, my 5800X3D will remain relevant for a many more years
feels as good as owning a paid-off car

If you aren't planning to buy a 4090 or 5080 for 1080P low gaming (e-sports), the 5800x3d is likely to last at several years no matter what AMD and Intel do.

The 5800X3D is already maxing out a 4080 at most resolutions - it won't matter if the 9800X3D is twice as fast as a 5800X3d at 1080P on a 5090 if you have a lesser card or play at 4k.

Even a 5800X (not 3d) is 95 percent of the speed of a 7800x3d when gaming at 4k on a 4090

 

[jeremyj_83]

Got one 14600T from ebay 210usd maybe make some tests put on tom's hardware forum.
13600T x 14600T aka refresh... the 13600T with 50w power limit make the same multi tread performance of the 9600X at 88w (cinebench r23)

I will try UPDATE the motherboard to run ddr5 (still using ddr4) maybe get another 20% of raw power

In water is wet news a 2023 CPU with 14c/20t at lower power levels is able to score the same MT performance as a 2024 CPU with 6c/12t at a higher power level.

 

[Amdlova]

In water is wet news a 2023 CPU with 14c/20t at lower power levels is able to score the same MT performance as a 2024 CPU with 6c/12t at a higher power level.

Yes allow 50w power the cpu can clock high 3.2ghz p-cores and 3.4ghz e-cores
96w power
13500T 4.0ghz
13600T 4.1ghz
14600T 4.3 (4.4 I think) boosting 5.1ghz two cores

14600T at 35w - 96w turbo has same power level as 12900!

@jeremyj_83 (you can find the 13500T on ebay low as 145usd) (Cheap way to upgrade)

 

[Amdlova]

9600X

Integer Math	90,330 MOps/Sec
Floating Point Math	60,463 MOps/Sec
Find Prime Numbers	263 Million Primes/Sec
Random String Sorting	37,100 Thousand Strings/Sec
Data Encryption	16,899 MBytes/Sec
Data Compression	334,062 KBytes/Sec
Physics	2,201 Frames/Sec
Extended Instructions	28,599 Million Matrices/Sec
Single Thread	4,512 MOps/Sec

13600T 65W ddr4 (3600)

	Integer Math	103,982 MOps/Sec
	Floating Point Math	77,226 MOps/Sec
	Find Prime Numbers	96 Million Primes/Sec
	Random String Sorting	43,329 Thousand Strings/Sec
	Data Encryption	22,986 MBytes/Sec
	Data Compression	402,902 KBytes/Sec
	Physics	1,542 Frames/Sec
	Extended Instructions	24,558 Million Matrices/Sec
	Single Thread	3,977 MOps/Sec

 

[cyrusfox]

How much faster is the 9950X vs 7950X for rendering video, Photoshop, and other productivity things? Im not into gaming, so I dont care about results for that.

See Page 3, many graphs go over that, for instance here is Photoshop, Going from 9549 → 10658 (11.6% uplift)

Next one goes over batch processing with better results (8265 →9760 - 18% uplift which is the same Multithreaded uplift).
For office its a mixed bag, Outlook is a regression, Excel shows minor improvement
For video encoding its 10.3% uplift (Except for audio encoding where its 5-6%[single thread heavy on audio, and single thread uplift is only 6%...]) When encoding can make use of AVX512 it can accelerate past the limitations of the 6/18% ST/MT uplift and achieve 30%+ uplift.

 

[NinoPino]

@PaulAlcorn
According to this video from JayzTwoCents it seems that the parking problems are related and fixable by a wrong BIOS setting (CPPC Preferred Cores) and the missing of the 3D vcache (amd3dvcachesvc) windows service.
Changing the BIOS option to "driver" instead of "auto" and a successive driver cleaner followed by reinstallation he fixed the problem.
The performance impact seems to be relevant.

 

[mitch074]

About the core parking issue : it's ultimately a kernel responsibility to move threads off a core before it's parked. If Windows allowed control of stuff like core affinity per process, parking the cores for games would be unnecessary.
Linux does that very well. It was already possible to improve gaming performance on a Zen+ 2700X by parking half the cores (I know I did), why is Windows unable to do that ?
I bet AMD worked this limitation around by adding a kernel driver, and I wouldn't be surprised if removing that driver once installed caused a bug in the kernel - it wouldn't be the first.
Considering how Microsoft is currently making sure to milk their users out of their last cent, improving performance or fixing performance bugs is waaaaay down the line for them.