Ryzen 9000 Review: Productivity Focus Benchmarks

All of AMD's Ryzen 9000 CPUs have launched and Gordon is here with the testing results.

Subscribe to our PC hardware podcast The Full Nerd: @thefullnerdpodcast

=============
Follow PCWorld!
=============

This video is NOT sponsored. Some links may contain affiliate links, which means if you buy something PCWorld may receive a small commission.

Timecodes:
00:00 - Intro
02:10 - Testing methods
10:34 - Cinebench multithread results
13:49 - Cinebench single thread results
16:49 - Blender results
19:59 - Handbrake results
26:13 - Premiere Pro results
36:12 - DaVinci Resolve results
42:08 - Photoshop results
47:06 - Microsoft Office results
50:49 - Geekbench results
54:43 - Steel Nomad results
56:55 - Cyberpunk 2077 results
58:43 - F1 2023 results
1:01:07 - Rainbow Six Siege results
1:04:21 - Total War: Warhammer III results
1:06:57 - Pricelist
1:19:43 - Final throughts

#amd #Ryzen9 #zen5

[JBrinx]

Wow, those AMD 9000 chips are very boring. They don't die from voltage issues at least, but it seems like 7000 series is the way to go for most

[JBrinx]

I would like to see some Linux Gaming testing. I think the results would be quite interesting for this series

[kevinerbs]

Hi Gordon.
Would consider try doing some mGPU tests to get rid of the GPU bound parts at 4K of your test in gaming & include the 3D V-cache cpus?
I can give a list of games that support, it's only like 7 to maybe 10 games, but I think you should revisit the gaming test with that idea.

1. Ashes of the singularity = Natively supported for All cards even different vendors
2. Chasm the rift = Vulkan
3. Deus EX mankind Divided (DX12 AFR must have single card set in control panel For S.L.I systems)
4. Rise of the Tomb Raider
5. Shadow of the Tomb Raider
6. Quake 2 RTX = (Vulkan)
7. Red dead Redemption 2 (Vulkan)
8. Strange brigade (Vulkcan)
9. Sniper elite 4

[brunogm]

It would be nice to test with 7900xtx, since other possible systemic issue are NV driver/gpus to explain the diff from expectation

[actually]

VBS, you're referring to Memory Integrity setting on Windows right? Did AMD's review guideline suggest having that enabled? I remember you mentioning your numbers being like 10% off from AMD numbers during your live stream: I wonder if VBS caused that? FWIR enabling it comes with a hefty performance hit.

As for PBO, PBO auto, AFAIK has always been PBO off and your Cinebench multi-core numbers for 9700X looks like it's still power-starved, corroborating my memory of how PBO auto behaves. Last time I played around with PBO was on 5900X 3 years ago, and for that, I had to set PBO to on and then set the PPT, TDC, EDC settings manually to see a performance difference, IIRC.

[Antagon]

Sign of really good CPU is ;when you have to take a microscope behind couch cushions during full moon to find any performance improvements.

[CyberTechVR]

Hi Gordon I understand that PBo is overclocking afterall .. did you overclocked bathe Intel chips too

[RayG]

And you please do a desktop Linux based review with Ubuntu because apparently there's an issue with windows, especially with gaming

[tuckerhiggins]

Wendell found that if you run Cyberpunk as an administrator, you gain 10fps on Zen 5. Ive seen tons of weird inconsistencies, so it seems like Windows just isnt working well for Zen5 yet

[marksaxon]

Great work! We have to change our expectations on these gen to gen updates. The days of huge leaps gen to gen are over unless some new chip design comes along. The 7000 series is the sweet spot and the 9000 is just a waste. It's not a waste if someone is coming from 1000 to 3000 series but no one should update from 7000 series. You can say this about smartphones and Apple computers too. Waiting every 3-4 gens is the way to go.

[-INFERNUS-]

Just curious what happened to Keith? Did he leave PC World? I haven't seen him in a long time. Hope all is well. 🙂

[catalin.]

10:35 Why with PBO ON/AUTO, why not test the Default mode?

[IrocZIV]

They said the memory controller was improved, I'm wondering if that means it can handle 4 DIMMs any better. I need 128GB and running it slow is no fun.

[leod]

Gordon, Team: Thank you very much for your amazing work, I love not only the results but also the insights and thoughts when analyzing each test. I have a few comments and recommendations regarding the testing methods though, which hopefully you can read and think about them.

1) I know you were testing CPU parts here but testing the fastest memory each platform supports, even if they differ, is important because the embedded memory controller is tested too. With all the issues surrounding the Intel chips, I don't know which are the Intel recommended specs for RAM modules but I'm sure it's not 6000 MT/s. I understand AMD supports 5600 MT/s although I understand that AMD provided a benchmark guide which suggested using 6000 MT/s, which allow the memory controller to use an Auto:1:1 ratio with the Infinity Fabric Clock (FCLK), unified memory controller clock (UCLK), and memory clock (MEMCLK).

Also, because these are productivity tests, I think it's more appropriate to test the CPUs using four DIMMs, which is what most people use when building a workstation. Using four DIMMs also impact the maximum frequency the memory controller can operate on.

2) Using the Performance power profile on Windows hurts performance most of the time. There's no benefit on having all the CPU parts active at all times. In fact, because all components are active, it makes them use more power and produce heat -- If the parts of the CPU which are not needed are disabled, there would be more room for the CPU to boost higher because the amount of dissipated heat would be lower.

Also, both AMD and Intel tune power and scheduling settings which are only available when using the Balanced power profile on Windows. This is particularly important when testing AMD 3D V-cache chips but it also applies to Intel CPUs with a hybrid architecture and AMD multi-chiplet CPUs, even the non-X3D ones.

Balanced is the default profile, the one that most people use (definitely, the one that everyone should use).

3) The best way to benchmark CPUs on the same platform is by re-formatting the disk and performing clean installs every time you change the CPU. This is especially true if you're testing X3D CPUs (which is not the case here) but it also applies to multi-chiplet CPUs. Multi-chiplet CPUs require special drivers and settings on Windows which are only configured during the Windows installation and AMD drivers installation phases. If you put a single CCD CPU first, those special settings and additional drivers won't be set/installed, hugely impacting how scheduling works when you switch CPUs and put a multi-chiplet CPU to perform the benchmarks without performing a full OS reinstallation.

If you're not going to re-format each time, make sure you test the multi-chiplet CPUs first, then the ones with a single CCD for more consistent results as you'd make sure that the optimizations and required drivers for multi-chiplet parts are installed on.

4) Not sure why you guys decided to use the SSD in PCIe Gen 4 mode when you can use it in Gen 5 mode...Bearing in mind that you're making productivity tests, storage access might be a bottleneck on some scenarios. Please, reconsider setting the SSD to PCIe Gen 5 in the future, and use an M.2 port connected directly to the CPU, not one connected to the chipset.

5) Until Intel releases their final BIOS update, I wouldn't include any Intel CPU benchmark. At this moment, those values are misleading. Performance results will surely vary and people may make uninformed purchases if they base their decisions on what's shown here. This is something I've been asking to all review outlets I follow: by stating that you won't include Intel chips on the charts, it will put pressure on Intel. It's time that someone forces them to stop playing with potential and current customers. They have to resolve the issue once and for all, and stop releasing temporary fixes that only buys them time and makes people buy things that after some time are damaged or performance is degraded, not honoring what the spec sheets initially stated.

[impuls]

Since ALL the AMD Chips are run with overclocking and WARRANTY out the window, you could atleast give the Intel part 7200Mhz ram. Amd cant run more than 6200 and shows little improvement over 6000 so thats fine. I feel that would give a more apples to apples comparison since you are trying to showcase all products in a best case scenario which I appreciate.

[Roman]

lol actually it's very easy to understand there is no reason at all to buy Zen5% over Zen4 even at the prices you showed, I saw most of the CPUs you've shown in the price comparison for way cheaper at times, those "street prices" went up a little before the Zen5% reviews.
IMO even Zen4 is still overpriced and the best CPU that you can buy for gaming and regular PC use (without productivity or work PC) is the 5700x3d from Aliexpress for under 150$ when there is a sale and coupons IMO that's the best bang for the buck CPU.
I'm still on a 9900k and don't feel any need to upgrade, according to time spy my PC is a bit faster than a 7700x with a 4070 system, my system has been manually tuned a few years ago so had that performance for years, stable, efficient and quiet.

[billkillernic]

New Cinebench 24 scores dont make sense to me single core 14900K gets 233 (using its power core and it has 8 of them with HT that makes 16 threads) thats the same as the 9700x which is 232

So since 9700x also has 8 cores and 8 (SMD) extra threads totals at 16 threads.

So on multi thread the 9700x gets 1100
The 19400 gets 2107

since intels power cores seem to be on par with AMDs 9000 cores that suggest that intel only using the power cores would get about 1100 as well, which means the res 1000 point came from the 16 efficiency cores.

And HERE is the weird thing now lets see the 9950x its two 9700x essentially, to each chimplet so to speak makes 1125 in multicore

That means that the 16 efficiency cores are only about 125 points behind that the ryzen 9000 full 8 cores with SMD ??

Either something is wrong or the new cinebench 2024 is bugged... because the way I see it the 14900 should have a smaller multicore score.

[raminpro]

INTEL MANY CPU RUN ABOVE 1.6 VOLT, ,,must be under 1.2

[foxser]

Do you think most people will enable PBO? Why are the only results with PBO on? You said it's because of a time limit that you didn't do both but if there's rime you might come back and revisit without PBO on but of your going to do one and revisit it seems the logical way to do it would be to start with default settings. That would be like testing Intel chips with only XMU auto OC on. Sure it's easy for users to do but if you're only providing one set of numbers to start I'd think you would want to start with default performance. It's on the manufacturers to make sure their products are getting their performance out of the box on release.

[amigatommy]

I have seen reports that windows is currently more of a problem than usual.