• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

AMD Ryzen 9 9900X

Cheeseball

Not a Potato
Supporter
Joined
Jan 2, 2009
Messages
1,981 (0.34/day)
Location
Pittsburgh, PA
System Name Titan
Processor AMD Ryzen™ 7 7950X3D
Motherboard ASRock X870 Taichi Lite
Cooling Thermalright Phantom Spirit 120 EVO CPU
Memory TEAMGROUP T-Force Delta RGB 2x16GB DDR5-6000 CL30
Video Card(s) ASRock Radeon RX 7900 XTX 24 GB GDDR6 (MBA) / NVIDIA RTX 4090 Founder's Edition
Storage Crucial T500 2TB x 3
Display(s) LG 32GS95UE-B, ASUS ROG Swift OLED (PG27AQDP), LG C4 42" (OLED42C4PUA)
Case HYTE Hakos Baelz Y60
Audio Device(s) Kanto Audio YU2 and SUB8 Desktop Speakers and Subwoofer, Cloud Alpha Wireless
Power Supply Corsair SF1000L
Mouse Logitech Pro Superlight 2 (White), G303 Shroud Edition
Keyboard Wooting 60HE+ / 8BitDo Retro Mechanical Keyboard (N Edition) / NuPhy Air75 v2
VR HMD Occulus Quest 2 128GB
Software Windows 11 Pro 64-bit 23H2 Build 22631.4317
Question for CPU/GPU reviews in general, would it be possible to add PugetBench Davinci Resolve to the test suite? I would love to see the performance metrics for that and how it improves vs older generations. Trying to decide if I wait around another generation or just need to jump.

I currently have a Ryzen 7 2700 and know it would show some improvement bumping up to a 9900X (or 7900X or 5800X3D), but I am not sure about how much improvement to expect and to then decide if it is worth the relatively huge cost one of these newer models would present (especially AM5 models as a new board and RAM would also be required). I have a Radeon RX 6800 XT GPU and it was a big jump up from my old GTX 1070, but now I am looking to improve the CPU bottleneck I have created. I game and edit video about the same amount - video editing is the priority for me though. I am an admitted AMD fan, so not really considering Intel. Lol
Since you're currently using a 2700 (AM4), with the appropriate BIOS update for your motherboard you can always upgrade to 5800X3D or even 5950X if you value productivity over gaming instead moving onto the newer AM5 platform. That way you only need to upgrade the CPU (and possibly your CPU's cooler) and keep the rest of your current parts.
 
Joined
Jun 20, 2024
Messages
363 (2.59/day)
I always develop my own tests, but yeah I'll look into it for next rebench
Out of interest, did you test the 9900/9950 with any CPU cores disabled?

Seeing as the single CCD parts win in some cases, it would be interesting to see if disabling some cores improves the 9950 Vs the 9700 and 9900 Vs the 9600. I'd assume with the cores disabled, the package retains it's higher TDP / PPT ratings.
Yeah, AMD have added the core parking stuff from the X3D chips but it would be interesting to see if them forced off would make any difference - maybe some form of redemption for owners of they want to squeeze out a few more FPS.
 

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
27,719 (3.70/day)
Processor Ryzen 7 5700X
Memory 48 GB
Video Card(s) RTX 4080
Storage 2x HDD RAID 1, 3x M.2 NVMe
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 10 64-bit
Out of interest, did you test the 9900/9950 with any CPU cores disabled?
I did not, so many things to test, so little time
 
Joined
Apr 21, 2021
Messages
250 (0.19/day)
System Name Silicon Graphics O2
Processor R5000 / 180MHz
Cooling noisy fan
Memory 384 MB
Storage 4 GB
Case the one with the old logo and proud of it ;)
Software IRIX 6.5
I always develop my own tests, but yeah I'll look into it for next rebench
Adding DaVinci Resolve Studio would be great for both CPU and GPU benchmarking, especially since it also supports up to 8 GPUs for those who really need it. For CPU benchmarking, it's also a quite unique application that actually supports AMD's Smart Access Video, aka the knock-off version of Intel Quick Sync Video.

However, from my personal experience as a Resolve user since the days before da Vinci Systems were acquired by Black Magic Design, I have to say that Pugetbench is sadly not really that great since it focuses way too much on small render jobs to test various format conversions, simple effects/overlays and scaling images than actually mimicking tasks of an actual editing or finishing process. Not to mention that it lacks tests for quite a lot of Resolve's features, like audio mixing, working with timelines/edits/conforming, more complex effects/motion graphics/restoration tools, or even decent color grading benchmarks with masks and secondaries - the thing Resolve was originally developed for - are almost non-existent.
Although I doubt that TPU has the man-power and expertise to cover all those sections in detail, having something that isn't just the Nth render of the same RAW shot would already add more variety when it comes to benchmarking.

Generally speaking, I'd actually liked it if TPU was a bit more open and more detailed about their custom benchmarks. For example, reading the description of the Adobe Premiere and Aftereffects benchmarks, I'm always wondering if @W1zzard just mixed up the benchmark description of the two text blocks. Not to mention that reading the "Cost per Frame" headline is always quite funny, considering it's "Cost per FPS".
 
Joined
Dec 28, 2012
Messages
3,837 (0.89/day)
System Name Skunkworks 3.0
Processor 5800x3d
Motherboard x570 unify
Cooling Noctua NH-U12A
Memory 32GB 3600 mhz
Video Card(s) asrock 6800xt challenger D
Storage Sabarent rocket 4.0 2TB, MX 500 2TB
Display(s) Asus 1440p144 27"
Case Old arse cooler master 932
Power Supply Corsair 1200w platinum
Mouse *squeak*
Keyboard Some old office thing
Software Manjaro
"Considering that the 7900X suffered the same fate, and AMD made it sellable with heavy discounts, this is probably the future of the 9900X, too, so don't give up all hope."

Man, like, there HAS to be a reason AMD keeps doing this. Are initial sales that good on new AMD parts? Do they need to keep their average MSRP prices up for some shareholder-related reason? Maybe I'm just ignorant of how business works but I feel like there could've been easy wins here.
Easy answer: AMD knows fanbois exist. They can release these parts for silly prices, sell to all the fanbois that want them, while reserving most production for higher margin professional chips. Once that market is saturated for that generation, they switch production to these consumer chips and lower the price to where it should be.

It's worked wonders for their balance sheet.
 
Joined
Jun 20, 2024
Messages
363 (2.59/day)
I did not, so many things to test, so little time
Yeah I get that. It's not until you look at the numbers you get the chance to think "Hmmm, interesting....".

It would be interesting if disabling a CCD on either the 9950/9900 works better as a performance solution for certain tasks (like games) Vs the 2 CCD default Vs the equivalent 9700/9600 product Vs the AMD Windows software managed core parking / thread management.
That might indicate an area where actually some performance has been left on the table - potentially you could maybe even get AMD to actually look in to that and provide a response.
I haven't seen any other publication look in to this disparity.

Easy answer: AMD knows fanbois exist. They can release these parts for silly prices, sell to all the fanbois that want them, while reserving most production for higher margin professional chips. Once that market is saturated for that generation, they switch production to these consumer chips and lower the price to where it should be.

It's worked wonders for their balance sheet.
Not just fanboys - those twitch / YouTube gamers / "influencers" have gotta have the latest stuff, especially if they are sponsored, so that they can talk about it and how amazing it is, especially when you buy via their link in the video description...
 
Joined
Sep 17, 2019
Messages
488 (0.26/day)
Yeah I get that. It's not until you look at the numbers you get the chance to think "Hmmm, interesting....".

It would be interesting if disabling a CCD on either the 9950/9900 works better as a performance solution for certain tasks (like games) Vs the 2 CCD default Vs the equivalent 9700/9600 product Vs the AMD Windows software managed core parking / thread management.
That might indicate an area where actually some performance has been left on the table - potentially you could maybe even get AMD to actually look in to that and provide a response.
I haven't seen any other publication look in to this disparity.


Not just fanboys - those twitch / YouTube gamers / "influencers" have gotta have the latest stuff, especially if they are sponsored, so that they can talk about it and how amazing it is, especially when you buy via their link in the video description...
ANNND there is more to this. You have to look at the financials to see how much they really make in a quarter.
The short answer is if you strip all of the market/accountant speak... as a big as a company as they are are, AMD is not making that much NET profit. The Lost Swede puts up a very nice chart on this.
 
Joined
Apr 21, 2021
Messages
250 (0.19/day)
System Name Silicon Graphics O2
Processor R5000 / 180MHz
Cooling noisy fan
Memory 384 MB
Storage 4 GB
Case the one with the old logo and proud of it ;)
Software IRIX 6.5
It would be interesting if disabling a CCD on either the 9950/9900 works better as a performance solution for certain tasks (like games) Vs the 2 CCD default Vs the equivalent 9700/9600 product Vs the AMD Windows software managed core parking / thread management.
That might indicate an area where actually some performance has been left on the table - potentially you could maybe even get AMD to actually look in to that and provide a response.
I haven't seen any other publication look in to this disparity.
From what I've seen/read so far from AnandTech, Computer Base, and JayzTwoCents, it looks like core parking was/is AMD's attempt to increase performance in games on dual CCD CPUs, and also reduce latency penalties by hopping fewer times between CCDs in other applications. However, AMD doesn't seem to have an official reaction to that mess yet, and the explanations those folks got from their AMD contacts seemed to be a little different each time as well.
It would be probably quite interesting to test the CPUs on Windows 10 (w/o PPM driver) and on Linux as well for comparison.
 
Joined
May 3, 2018
Messages
2,881 (1.21/day)
Confused about your tuned PBO + curve optimiser results vs KitGuru's He tuned his so it used same power as stock 9900X, got higher clocks and better performance. You show massive power increase, and one would be mad to do such a thing for the pathetic gains. It uses more power than 7950X, so why would you even bother.

Did you run the full curve optimiser which take well over an hour to run, or did you just set a single offset?

I don't think the lack of NPU is really a downside, but maybe it was considered taking the price tag.
Software is starting to leverage NPU, eg Topaz photo AI apps. I'm sort of annoyed Strix gets a good NPU. Even Arrow Lake gets the 11 TOPS NPU from Meteor Lake, with a possible refresh getting another bump. I think at this sort of money of around $900 in Australia, just getting 12 cores is a rip-off. Much more inclined than ever to wait for Arrow Lake and see what the 14700 replacement brings to the table.
 

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
27,719 (3.70/day)
Processor Ryzen 7 5700X
Memory 48 GB
Video Card(s) RTX 4080
Storage 2x HDD RAID 1, 3x M.2 NVMe
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 10 64-bit
Adding DaVinci Resolve Studio would be great for both CPU and GPU benchmarking, especially since it also supports up to 8 GPUs for those who really need it. For CPU benchmarking, it's also a quite unique application that actually supports AMD's Smart Access Video, aka the knock-off version of Intel Quick Sync Video.

However, from my personal experience as a Resolve user since the days before da Vinci Systems were acquired by Black Magic Design, I have to say that Pugetbench is sadly not really that great since it focuses way too much on small render jobs to test various format conversions, simple effects/overlays and scaling images than actually mimicking tasks of an actual editing or finishing process. Not to mention that it lacks tests for quite a lot of Resolve's features, like audio mixing, working with timelines/edits/conforming, more complex effects/motion graphics/restoration tools, or even decent color grading benchmarks with masks and secondaries - the thing Resolve was originally developed for - are almost non-existent.
Although I doubt that TPU has the man-power and expertise to cover all those sections in detail, having something that isn't just the Nth render of the same RAW shot would already add more variety when it comes to benchmarking.

Generally speaking, I'd actually liked it if TPU was a bit more open and more detailed about their custom benchmarks. For example, reading the description of the Adobe Premiere and Aftereffects benchmarks, I'm always wondering if @W1zzard just mixed up the benchmark description of the two text blocks. Not to mention that reading the "Cost per Frame" headline is always quite funny, considering it's "Cost per FPS".
Can you create a real-life-oriented project workload for my testing? That would save me a lot of time, because I'm a noob with Resolve. No plans to show multiple results for it, but a single aggregated result would still be extremely cool

"Cost per Frame" is a simplification because "Cost per FPS" can be logically challenging for many readers because of "per .. per" .. new addition for this week's reviews BTW, how do you like? I know they can be a bit misleading because the cheapest CPUs have such a big advantange

You show massive power increase, and one would be mad to do such a thing for the pathetic gains. It uses more power than 7950X, so why would you even bother.
My goal is to get a feel for the maximum performance that can be reached with PBO. Not claiming what I show is the absolute maximum that you can find with a week of tweaking, but it should be within 1%.
No doubt, if you lower the limits and trade performance for efficiency, you can gain a lot of efficiency, but in my experience that's not what the vast majority of readers want to know.
 
Joined
Oct 12, 2022
Messages
21 (0.03/day)
Regarding the reliability of the results, why do you test both platforms on DDR5-6000, which is optimal for AMD but clearly underperforms Intel's platform against DDR5-7000 and above which Intel can handle without any problems? Have you done comparison tests with DDR5-6000 and much faster for Intel?
 
Joined
Apr 17, 2021
Messages
558 (0.43/day)
System Name Jedi Survivor Gaming PC
Processor AMD Ryzen 7800X3D
Motherboard Asus TUF B650M Plus Wifi
Cooling ThermalRight CPU Cooler
Memory G.Skill 32GB DDR5-5600 CL28
Video Card(s) MSI RTX 3080 10GB
Storage 2TB Samsung 990 Pro SSD
Display(s) MSI 32" 4K OLED 240hz Monitor
Case Asus Prime AP201
Power Supply FSP 1000W Platinum PSU
Mouse Logitech G403
Keyboard Asus Mechanical Keyboard
Thanks for the best review on the internet as usual, Wizzard! Good fun reading it all. Cheers :)
 
Joined
Apr 21, 2021
Messages
250 (0.19/day)
System Name Silicon Graphics O2
Processor R5000 / 180MHz
Cooling noisy fan
Memory 384 MB
Storage 4 GB
Case the one with the old logo and proud of it ;)
Software IRIX 6.5
Can you create a real-life-oriented project workload for my testing? That would save me a lot of time, because I'm a noob with Resolve. No plans to show multiple results for it, but a single aggregated result would still be extremely cool
I can definitely look into it, but I'm more of a Jack-of-all-Trades kind of freelancer working in post-production, not a filmmaker or producer. Resolve is sort of my Swiss Army knife for work. I'm not an expert in any specific area of the software, but I know a lot about all of its parts to keep projects together and finish them while herding cats other freelancers.
That also means I don't have many useful assets that I can give away freely. Most of my clients are even fairly annoying when it comes to footage for things like my personal show reels, etc. It's probably not that difficult to put something together from stock footage, but that's nothing that can be done overnight since a lot of stock footage is quite awful for color grading workflows.
However, I'm not sure how to automate all of it, because some things should be measured in the UI, while other stuff like proxy conversions or renders can be run via console commands, similar to what Puget does with their timed LUA script.
"Cost per Frame" is a simplification because "Cost per FPS" can be logically challenging for many readers because of "per .. per" .. new addition for this week's reviews BTW, how do you like? I know they can be a bit misleading because the cheapest CPUs have such a big advantange
After reading through the comments of the Ryzen 9000 reviews, I now get why you try to keep some things a certain way in your reviews. ;) But overall I do like it. Although, as someone living in the Eurozone, pricing in US$ is often more hilarious than useful to me personally.

---

Back to the R9 9900X: Mindfactory sold less than 10 CPUs on the first day and dropped the price by 8% on day #2 after some Austrian and Czech online retailers went below MSRP almost instantly.
Let's hope Dr. Lisa Su does us all a favor and finally gifts AMD a competent marketing department as little present for her 10th anniversary as CEO. :rockout:
 

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
27,719 (3.70/day)
Processor Ryzen 7 5700X
Memory 48 GB
Video Card(s) RTX 4080
Storage 2x HDD RAID 1, 3x M.2 NVMe
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 10 64-bit
Considering that the actual output is not relevant, i.e. if it looks broken, any free material should work fine. I like to use movie trailers for these tasks

Resolve is sort of my Swiss Army knife for work
I think that actually makes you more qualified than a movie producer, because you are aware of different workflows that might be unknown to others

I'm not sure how to automate all of it
I'm the master of automation, so need to worry about that ;)

as someone living in the Eurozone, pricing in US$
me too, but USD is the standard pricing, especially for a globally oriented website like TPU that's written in English. Reporting and maintaining multiple price points is such a pain, I feel that our time is better invested doing other things
 
Joined
Apr 21, 2021
Messages
250 (0.19/day)
System Name Silicon Graphics O2
Processor R5000 / 180MHz
Cooling noisy fan
Memory 384 MB
Storage 4 GB
Case the one with the old logo and proud of it ;)
Software IRIX 6.5
Considering that the actual output is not relevant, i.e. if it looks broken, any free material should work fine. I like to use movie trailers for these tasks
The issue is that you can't properly key and color grade 8-bit h.264 clips with 4:2:0 sampling. Not to mention that camera RAW is computationally much more intensive. However, I found a workaround for that issue thanks to licensing agreements written in Bavarian-English. :roll:

Since the original idea of creating my own little DaVinci Resolve benchmark is mainly based on the fact that I need something to evaluate the performance difference of the upcoming Resolve 19 release between Windows and Linux, I'll take a more serious look at it during the next few weeks once I've got a bit of downtime. I'll drop you a DM, in case it turns out to be workable, or just DM me a contact email where I can reach you.
Small warning ahead, you'll probably need between 50GB and 80GB free space on your benchmark drive, maybe a bit more if you bench with a studio license and not the free one.
 
Joined
Oct 14, 2022
Messages
33 (0.04/day)
System Name Dreki (https://pcpartpicker.com/b/ZDCLrH)
Processor AMD Ryzen 7 2700
Motherboard MSI B450 Gaming Pro Carbon AC
Cooling ID Cooling SE-224-XT Basic
Memory Corsair Vengeance LPX 4x16GB DDR4-3200 (Running @ 2133 Because of MOBO Issues - Documented on PCPP)
Video Card(s) MSI Radeon RX 6800 XT Gaming X Trio
Storage WD SN770 2TB NVME SSD
Display(s) Viewsonic VP2768 27" 1440p Calibrated Display
Case Cougar MX330
Audio Device(s) Zoom AMS-24
Power Supply Corsair RM750X (2017) 750W
Mouse Cougar Minos X3
Keyboard Cougar Attack X3
Software Windows 10 Home, Davinci Resolve Studio, ON1 Photo Raw 2023
I always develop my own tests, but yeah I'll look into it for next rebench
That would be awesome! I wish I could take the Premiere results and translate that to Davinci, but they operate quite a bit differently - Davinci generally being more GPU based than Premiere. I look forward to more comparisons and tests down the line!
 
Joined
Jan 29, 2013
Messages
120 (0.03/day)
Very nice work Wizzard
I really trust TPU opinions and results & share them here with my collogues in Iraq
 
Joined
Jun 2, 2017
Messages
8,988 (3.31/day)
System Name Best AMD Computer
Processor AMD 7900X3D
Motherboard Asus X670E E Strix
Cooling In Win SR36
Memory GSKILL DDR5 32GB 5200 30
Video Card(s) Sapphire Pulse 7900XT (Watercooled)
Storage Corsair MP 700, Seagate 530 2Tb, Adata SX8200 2TBx2, Kingston 2 TBx2, Micron 8 TB, WD AN 1500
Display(s) GIGABYTE FV43U
Case Corsair 7000D Airflow
Audio Device(s) Corsair Void Pro, Logitch Z523 5.1
Power Supply Deepcool 1000M
Mouse Logitech g7 gaming mouse
Keyboard Logitech G510
Software Windows 11 Pro 64 Steam. GOG, Uplay, Origin
Benchmark Scores Firestrike: 46183 Time Spy: 25121
After reading through this thread, I came to a realization. AMD Ryzen CPUs are like nothing that has been in the market. Just think of the sheer number of CPUs that are available to the average user that are discernible in performance improvements. It is obvious that AM5 will follow the same path as AM4. I have argued that the change from 7000 to 9000 is very similar to 1700 to 2700. If we remember that it was a nice IPC bump and 200 mhz increase in clock speed that made those sweet. Then we got 3000 series and the CPU performance of the 3300X made it unobtanium for like 6 months. It also meant that a 3600 was faster in Gaming than a 2700x. The change to 5000 brought 5.0 Ghz but the IPC was not as high so it did not feel as discernable but that depended on what chip you had. I have already written on how a 5900X feels as smooth as butter and a 5950X is super snappy. The truth is that X3D chips are that good at Gaming. That is the narrative agianst these chips as Gaming is what the narrative uses but the IPC change is real as the power/performance proves. Unfortunately for AMD, AM5 boards have been tooo expensive. The Godlike is still $2000 where I live and well assigneD X670E boards are still north of $400 where I live as well. Thankfully good DDR5 is now $129 for 32GB of 30 timing Team units. I only wish the APUs were about $150 cheaper as those are the future for me.
 
Joined
Oct 12, 2022
Messages
21 (0.03/day)
@W1zzard I reiterate my question, do you have tests to confirm that DDR5-6000 does not limit the full capabilities of the Intel platform? For AMD it is the optimal sweet spot but Intel can run on much higher clocked memory.
 
Last edited:

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
27,719 (3.70/day)
Processor Ryzen 7 5700X
Memory 48 GB
Video Card(s) RTX 4080
Storage 2x HDD RAID 1, 3x M.2 NVMe
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 10 64-bit
@W1zzard I reiterate my question, do you have tests to confirm that DDR5-6000 does not limit the full capabilities of the Intel platform? For AMD it is the optimal sweet spot but Intel can run on much higher clocked memory.
It limits both of course, but it's a reasonable common ground imo
 
Joined
Oct 12, 2022
Messages
21 (0.03/day)
It limits both of course, but it's a reasonable common ground imo
No let's be serious, in this case Intel has objectively worse test conditions here. In my opinion both platforms should have their optimal memory or if you want to leave one set then for example tuned 7000MT/s asynchronously would be good for AMD which should be equivalent of 6000/6400 1:1 and would also be fair for the Intel platform.
 
Top