AMD Ryzen 9 9950X3D benchmarked in gaming, and productivity vs 7950X3D

By CEO - Neowin.net Neowin ·
View full version

Hot on the heels of the AMD Radeon RX 9070 and 9070 XT graphics cards" release and review of the latter earlier this month, AMD is back with another new bit of hardware, this time the 9950X3D desktop CPU, which is a follow-up and successor to the 7950X3D.

With that out of the way, now the disclosure: AMD lets us keep the hardware they provided for the review, but this does not influence our findings.

Sayan Sen contributed to this review, and also provided the benchmark graphics.

For our benchmarks, we used six self-bought games, or where possible, trials with access to the benchmark. They are:

  • Final Fantasy XV
  • Assassin"s Creed Odyssey
  • Shadow of the Tomb Raider
  • Cyberpunk 2077
  • Hitman 3
  • Far Cry 6

We mainly focused on games that are fairly CPU-heavy, spanning across various engines.

hwinfo.com provided a commercial license of HWiNFO, and UL Solutions provided us with Professional (commercial use) licenses for 3DMark, and Procyon. We also used 7-Zip and Cinebench 2024.

Introduction

The Zen 5 family of desktop processors, under which the 9950X3D falls, started releasing last July, and are based on the same AM5 socket design of the 7000 series Zen 4 generation. They support DDR5 memory using the A620, B650(E)/X670(E), or B850(E)/X870(E) chipsets. The 9950X3D, which AMD says is a "Gaming and Content Creation Processor" and sits right at the top end of the Zen 5 launch, is officially releasing today, and will go on sale tomorrow, March 12 at retailers.

Below are the specs of both the 9950X3D and 7950X3D, on which our comparison will be based:

AMD Ryzen 9 9950X3D (Zen 5) 7950X3D (Zen 4)
Cores/Threads 16/32
Max Boost Up to 5.7 GHz
Base Clock` 4.3 GHz 4.2 GHz
L2 Cache 16 MB
L3 Cache 128 MB
TDP / cTDP 170 W 120 W
Default Socket Power 200 W 162 W
Max Current (EDC) 225 A 180 A
Max Current Thermally Limited (TDC) 160 A 120 A
Tjmax 95 C 89 C
Processor Graphics AMD Radeon Graphics
2 Cores @ 2000MHz
Boost Algorithm Precision Boost 2
Recommended Cooler 240 - 280 mm liquid (or equivelent)
Max Memory Speed DDR5-5600 DDR5-5200
Max Memory capacity 192 GB 128 GB
Price (MSRP/SEP) $699

Although on paper there does not seem to be a significant difference between the old and new in the above table, this is because most of the changes are actually architectural. For example, the 2nd generation AMD 3D V-Cache Technology, which is included on the 9950X3D is reversed, meaning the 3D V-Cache layer is now placed under the processor cores, which results in cooling solutions having an almost direct contact to the hottest part of the CPU. The higher socket power also results in longer boosts, and there"s even support for higher clocked DDR5 memory.

Building the beast

Test system

  • Computer ATX Open Motherboard Rack (Amazon)
  • AMD Ryzen 9 9950X3D (Amazon)
  • Gigabyte X870E AORUS MASTER (BIOS N10) (Amazon|Newegg)
  • be quiet! Pure Power 12 M 850W (Amazon|Newegg)
  • Thermalright Phantom Spirit 120 CPU Air Cooler (Amazon|Newegg)
  • Thermal Grizzly Carbonaut (38x38x0,2mm) (Amazon|Newegg)
  • 2x 16GB G.SKILL Trident Z5 Neo RGB Series DDR5 6000MT/s CL28-36-36-96 (Amazon|Newegg)
  • Samsung 990 Pro 1TB (PCIe Gen4) (Amazon|Newegg)
  • NVIDIA GeForce RTX 5090 Founders Edition (NVIDIA)
  • Windows 11 Pro Build 26100.3194 (Neowin Deals)
  • dGPU driver: Nvidia driver 572.47
  • Adrenaline Edition: 24.12.1

Reference System

As an Amazon Associate when you purchase through links on our site, we earn from qualifying purchases.
When you purchase through links on our site, we may earn an affiliate commission.

We initially planned on using the same memory kit (DDR5-6000 CL36) on both platforms to keep it equal but decided against it since using higher latency memory on the 9950X3D would be artificially hamstringing its performance by not taking advantage of the memory-specific optimizations in the new architecture.

We did try the AMD-supplied DDR5-6000 CL28 on our Ryzen 7950X3D but it was not fully stable, with it going into memory training time and time again, so we dropped that idea.

By the way, a single self-purchased 5090 FE was used between both systems, (I do not have two of these cards!) this was to see how the CPUs complimented the same dGPU in some of the benchmarks. Game Mode was also enabled in Windows 11, this was also requested of AMD in our reviewers guide.

The test hardware was open plan, and our BIOS version was N10, which was supplied by AMD itself, as this version is not on the official Gigabyte support web page. AMD also suggested we use Adrenaline Edition 24.12.1 which we did so we could verify their scores against ours. Resizeable BAR was enabled (default BIOS setting) as was MultiCore Enhancement (MCE), which was on Auto by default, showing it as "Enabled". On a similar note, Precision Boost Overdrive (PBO) was set to "Advanced" on our AMD system, which is also the default setting.

Essentially, we ran everything on the default settings since that was the idea, as the average user won"t go into BIOS immediately to play around with the settings.

I resisted using thermal paste again because of the mess, but also because I have had great temperatures and results with Thermal Grizzly"s Carbonaut thermal pad with my i7-14700K test system.

It really, really works!

The highest temperature we recorded with our 9950X3D in our tests was 83.4C in Cinebench 2024, compared to 85.2C with the 7950X3D which uses thermal paste (installed in the summer of 2023).

When handling any of the hardware, I always used an anti-static band. I even ordered a cable that let me power on, reset, and see the status of HDD and Power LEDs, so I didn"t have to use a screwdriver on the pins.

Benchmarks

The following benchmarks were done with NVIDIA GeForce RTX 5090 Founders Edition in the system.

We start with 3DMark"s Fire Strike Ultra which is a DirectX 11 synthetic gaming test.

As you can see in the chart below, aside from the default physics test, we also used custom settings and measured the CPU performances. We turned the resolution down from the default (4K) to 720p to minimize the GPU workload since we are trying to tax the CPUs" and not the GPU. This is generally the simplest way to do it as we have discussed earlier. AMD"s new 9950X3D is around 19% faster than the 7950X3D in the physics tests.

We must say that we are impressed with how well UL has designed the physics test since there is hardly anything to separate the 4K and the 720p results. We definitely expected a small yet significant difference between the resolutions considering how much less GPU-heavy 720p is compared to 4K but that"s not the case.

The 720p graphics test is certainly revealing. In this, we see that the new 9950X3D is 15% faster than the 7950X3D. At default(4L) settings, this graphics test would be tied as then the 5090 would be tested.

Following that, we have Time Spy Extreme which is a DirectX 12 4K synthetic test. We run a similar test scenario here.

The gap between the two chips closes down here. In the physics tests, the new Ryzen 9 9950X3D comes out about 16% faster than its predecessor, and in the 720p graphics test, it is 12% better.

Moving on to the more modern 3DMark DirectX 12 tests, we have Steel Nomad and Steel Nomad Lite.

Interestingly, on the Steel Nomad 720p graphics test, the performance differential was exactly the same as we found on 720p Time Spy Extreme graphics, the 9950X3D was 12% better than the 7950X3D. Meanwhile, on Steel Nomad Lite, the gap between the two 16-core X3D CPUs was 15.5%.

Unlike Time Spy and Fire Strike, Steel Nomad does not have any standalone physics tests, and thus we relied on low-resolution graphics tests.

After looking at the rasterization performances, we next look at Speed Way which tests ray tracing.

Even though ray tracing is stressful for the CPU, it is still the GPU that does most of the heavy lifting, as we see a rather insignificant 2.1% improvement with the 9950X3D even at 720p.

Finally, we have the 3DMark CPU profile test which sees how well a CPU scales across cores and threads.

At its best, the Ryzen 9 9950X3D shows up to 32% better performance but generally, the advantage is around 22% over the 7950X3D. Due to how the AMD chipset driver handles the CCDs (Core Compute Die) in gaming-like workloads like this test, it does not scale at all above 16 threads.

Following synthetics, we move on to real-world games.

2016"s Final Fantasy XV, based on Luminous Engine, is the oldest game in our suite. Here, we saw almost identical performance between the two CPUs.

Up next, we have Assassin"s Creed Odyssey based on AnvilNext 2.0. This is from 2018 and the 9950X3D was 7% and 7.5% better in averages and minimums, respectively.

2018"s Shadow of the Tomb Raider, built on the Foundation Engine, showed close to a 14.5% uplift over the 9950X3D in the averages. The 95% lows were 17% better on the new X3D.

Please note that here, we dropped down to 720p as we observed near 100% utilization of the gaming CCD (the one with the 3D V-cache) even at 1080p and wanted to see how the game would behave if we further increased the load on the processor.

On Cyberpunk 2077 that"s based on the Red engine, the Ryzen 9950X3D was 11% better than in average framerates and 12.75% better on the minimums.

2021"s Hitman 3 based on the Glacier 2 engine was 9.3% ahead on the 9950X3D vs the 7950X3D, which is substantial and not totally insignificant.

Finally, we come to Far Cry 6 which saw the biggest gain in our entire suite. The average FPS on the new 9950X3D was 45% better than on the 7950X3D. The minimum FPS differential was even higher at 51.2%.

Far Cry 6 is based on the Dunia engine and is known to be quite single-thread bound. What this tells us is that other single-thread heavy games, the likes of Counter-Strike 2, will probably also see big improvements on 9950X3D. This was also one of the titles in AMD"s new Application Compatibility Database which is meant to boost the X3D performance on certain games.

In most of our gaming results, we exceeded what AMD had suggested in the review guide and that was because we used different settings that were even more CPU-focused.

Wrapping up our game testing, we move to productivity. First up we have UL Procyon Office Productivity benchmark.

Despite being around 23 to 26% faster than the 7950X3D in Word, Excel, and Outlook, the overall score was only 9% better on the 9950X3D. That is because we saw a very substantial 18% regression in PowerPoint output on our test. This does not match AMD"s guidance as it did not see a performance loss on its newer chip.

On Cinebench 2024, the 16-core Zen 5 9950X3D part is 24% faster. This is a rendering test and thus stresses all available threads.

Next up, we tested a couple of AI workloads starting with Procyon"s Computer Vision. For those wondering, computer vision is the ability of machines to extract and analyze information from images or videos.

We saw a massive 51% boost in the Computer Vision test. It looks like Zen 5"s AI-specific optimizations and improvements are paying off.

Next up, in text generation, we saw anywhere between 2.6% to 20% improvement on the Ryzen 9 9950X3D.

Last but not least, we also benchmarked 7-Zip compression and decompression, which measures in GIPS (billion instructions per second). The 9950X3D was 10% better in decompression and 14% better in compression.

With all the game and productivity app results out of our way, we now turn our attention to power efficiency and value.

For power efficiency, we only measure gaming efficiency as the X3D chips are advertised first and foremost for gaming.

The new 9950X3D is a mixed bag here with it showing excellent performance-per-watt (for wattage we considered the CPU package power) in titles like Far Cry 6 with a 57% advantage. However, there is more than one instance where it loses too. In Final Fantasy XV, for example, it was 23.3% behind the 7950X3D and in Cyberpunk 2077, that gap is 22%.

Although we did not individually measure efficiency figures in productivity tasks, it is noteworthy that the 9950X3D recorded its highest power usage in Cinebench 2024 as it consumed an average of ~182 watts with a peak of 200 watts. The 7950X3D, meanwhile, was much more efficient with an average draw of ~131 watts and a maximum of ~139.5.

Credit where it"s due, the new 9950X3D, despite guzzling down more power, manages to outdo the 7950X3D in the cooling department as it peaked at 83.4 C compared to 85.2 on the 7950X3D, in spite of the 9950X3D using a slightly worse (air) cooling solution versus liquid with the 7950X3D.The effect of the new 3D V-cache arrangement is evident here.

Moving on, after a mediocre showing in gaming power efficiency, we check out if the value prospect on the X3D chip fares better, and it certainly does, although, not by a massive margin.

Again, Far Cry 6 put on the best showing here with a 45% better score. In most of the other games, it was around 10 to 15%.

Following gaming, we also test the value of the chips in productivity tasks. Computer Vision with a 51% performance-per-dollar uplift was the most impressive. This is a phenomenal performance gain and if you are someone looking at such tasks, then the 9950X3D will absolutely be worth the upgrade.

Other workloads saw less than 30% gains and typically around 20%. The PowerPoint number stands out as a failure.

Conclusion

AMD thoroughly impressed us with its RX 9070 XT graphics so much so that we gave it a 10 out of 10. That is not quite the case with the 9950X3D, though that"s not to say that it"s a bad part and you shouldn"t buy it. However, you aren"t looking at any massive performance uplift here gen-on-gen outside of scenarios like single-threaded games and certain AI applications. So a jump from a 7950X3D to a 9950X3D might be a bit of an (expensive) stretch for most.

However, if you are on a lower-core count part and are in need of more cores, the 9950X3D should serve you very well. At the end of the day, this is still one of the fastest CPUs money can buy and it is still on AM5. So, as a Ryzen 7000/9000 series CPU owner, you can simply jump on it with a firmware update (assuming your motherboard VRM section is good enough for a 16-core 9950X3D).

For those on AM4-based Ryzen 5000 or older who haven"t made the jump to the AM5 socket yet, the 9950X3D will offer an enormous performance improvement if you can justify the platform upgrade cost.

Meanwhile, people on Intel, especially 11th Gen and older, can also consider AMD"s new X3D flagship for some serious performance jump.

The Ryzen 9 9950X3D gets an 9 out of 10 from us as there is very little to pick on as a flaw. Its efficiency aspect, as we see above in the performance-per-watt chart, is one area we hoped AMD would have done a bit better; although the company should still be comfortably ahead of Team Blue in this department, Intel could end up catching up soon if AMD does not practice caution.

AMD Ryzen 9 9950X3D: Amazon US || Newegg US || Amazon UK

As an Amazon Associate when you purchase through links on our site, we earn from qualifying purchases.
When you purchase through links on our site, we may earn an affiliate commission.

 

Report a problem with article
Next Article

AMD makes CPU changing easy as you won't need to reinstall Windows 11/10 any more

Previous Article

Apple is reportedly working on a 'dramatic' UI overhaul for iOS, iPadOS and macOS