Reviews

AMD’s new champion CPU

The Ryzen 9 9950X3D is a real performance winner, especially for gamers.

Jon Peddie and Mark Poppin

The Ryzen 9 9950X3D processor uses 2nd Gen AMD 3D V-Cache technology. Testing showed performance gains in both gaming and content creation compared to the previous generation. The new cache design improves thermal resistance, which contributes to higher clock speeds. Unlike previous versions, the cache sits below the processor cores. One complex has 3D V-Cache, while the other balances clock speed and throughput. The processor includes overclocking capability, a change from the previous generation. AMD suggests using Precision Boost Overdrive for mixed workloads. The company also encourages using Extended Profiles for Overclocking (EXPO) DIMMs.

9950X3D champ

AMD says their Ryzen 9950X3D is the world’s fastest gaming processor for gamers who want the best possible gaming experience from their PC and creators who need a top-tier, high-performance solution. Based on our test results, we’re inclined to agree.

“Best damn gaming CPU I’ve ever seen, and I’ve seen a few,” said Mark Poppin, founder of BabelTechReviews and JPR’s chief gaming benchmarker.

Among the many improvements from the last-generation CPU, the formidable Ryzen 7 9800X3D, the new Ryzen 9 9950X3D employs a second-generation AMD 3D V-Cache: AMD’s proprietary 3D V-Cache technology designed to lower latency and increase frame rates. AMD says they have improved this technology to allow for improved thermal resistance, higher clock speeds, and better performance. 

But, we weren’t going to take their word for it. So, it was off to our Mt. Tiburon Testing Lab (MTTL), where we set up two machines, one with the 7 9800X3D and one with the new 9 9950X3D. The motherboards and all else were the same. To demonstrate the real CPU contributions to gaming and creator workloads, we used an RTX 5090 AIB. The results were more than impressive. “It’s a beast of a CPU,” said one of the guys.

And unlike AMD’s 7000 gen, it also beats the 9800X3D in gaming—the former World’s Fastest Gaming CPU. It’s impressive—AMD has solved their “core parking” issues.

The new CPU offers overclocking capability, and for the first time, AMD has created a processor with AMD 3D V-Cache technology offering precise and direct overclocking with the same voltage limits as other 9000-series processors. 

AMD’s Ryzen 7 9800X3D, Ryzen 9 9950X3D, and Ryzen 9 9900X3D use second-gen AMD 3D V-Cache technology, which differs from earlier versions by placing the cache below the processor cores instead of above. In previous models like the Ryzen 5800X3D and Ryzen 7000X3D, the 3D V-Cache was layered on top of the cores. With the new placement, the Zen 5 cores are directly interfaced with the cooling solution, improving thermal resistance by up to 46%. This allows for higher sustained clock speeds, giving the Ryzen 7 9800X3D a 500 MHz faster base clock and a 200 MHz faster boost clock compared to the Ryzen 7 7800X3D. This design also enhances multi-threaded performance, allowing the 9800X3D to fully utilize its power provision.

9950X3D V-Cache
Figure 1. AMD’s 3D V-Cache offers improved thermals, higher clock speeds, and enhanced gaming and multi-threaded performance compared to the previous generation.

The Ryzen 9 9950X3D and Ryzen 9 9900X3D use second-gen 3D V-Cache on one of their two core complexes (CCXs). The other CCX lacks 3D V-Cache, similar to the Ryzen 9 7950X3D and Ryzen 9 7900X3D. This configuration aims to balance latency optimization from the CCX with 3D V-Cache and clock speed and throughput optimization from the CCX without it.

“It’s the best damn gaming CPU I’ve ever seen, and I’ve seen a few,” said Mark Poppin, founder of BabelTechReviews and JPR’s chief gaming benchmarker.

He continues: “I really can’t say enough good things about the Ryzen 9 9950X3D. It’s incredibly powerful for any task, plus it’s the fastest gaming CPU—there’s no compromise. Using a midrange 360 mm AIO cooler, it runs relatively cool under the heaviest benchmark/stress test—under 85 C—while pegging all cores to their limits. And it overclocks with a very reasonable power consumption!”

AMD not only offers overclocking, but seems to be encouraging it—quite a switch from past precautions and warnings. The Ryzen 9 9950X3D and 9900X3D are presented as overclockable processors This means direct overclocking is enabled, with the same clock and voltage limits as standard Ryzen 9000 processors. 

AMD says Precision Boost Overdrive (PBO) is the best path for gaming and mixed-use case scenarios. It offers the best power efficiency, says the company, alongside a blend of multi-threaded performance (nT) and single-threaded performance (IT) uplift. Direct (manual) overclocking can give an increased performance in specific and targeted scenarios, such as high uptime in all core use cases. 

Curve optimizing steps can be set via Curve Optimizer and Curve Shaper features. Curve Optimizer values will vary among CPUs, but in general, the user can do a search of values and run their workloads of interest to ensure stability and benefit.

In AMD’s Core Complex Die (CCD) is a group of CPU cores and their associated cache memory. Each CCD can include up to eight CPU cores along with L3 cache. Multiple CCDs can be combined on a single processor to increase the core count and performance.

AMD suggests the way to start is to do something like -10 to -15 Curve Optimizer steps on available CCDs and then go from there, looking at stability for boot and their workloads of interest. The platform allows up to -50 steps, but not all CCDs will support that level of margin reduction. 

A good quick start is to use Ryzen Masters Derive Function (this can be done on a per-core basis but will take longer) to automatically derive values and use those directly or as a ballpark/baseline and then tweak from there.

Curve Shaper, which allows fine-grained control of the curve at defined temperature and frequency bands, can be used to improve performance/stability in targeted use cases. (These new controls can be found in BIOS but will be added to Ryzen Master in the future.) 

AMD advises overclockers to use Extended Profiles for Overclocking (EXPO) DIMMs and load the EXPO profile for those DIMMs to get that memory OC benefit. 

The 9800X3D and 9950X3D share identical settings. Therefore, we didn’t see any reason to overclock either one. Lowering the voltage by setting the PBO to -30mV allows the CPU to run cooler.

The only real requirement is low-latency DDR5 6000 MHz (to achieve a Command Rate of 1T). Any typical user who just sets EXPO to optimize the DDR5 automatically will get ~98-99% of the maximum performance without overclocking anything.

But for the tweakers who insist on overclocking everything, using DDR5 8000 MHz (with a Command Rate of 2T), they will definitely be able to squeeze out a bit more performance. Unlike the last-gen 7000 series, overclocking is possible and even practical (for the tweakers), although AMD officially doesn’t recognize it, and it could void a warranty (as noted when you enter the Overclocking section of an AMD BIOS).

With all that to think about and set up, we dove in and started running tests. The total details of all the tests can be found here, and the summary results are shown in the following charts.

Games

We ran 15 games through 19 benchmark scenarios on the new 9 9950X3D and the previous-generation 7 9800X3D processors and got the results shown in Figure 2.

Figure 2. Game benchmark results for AMD processors.

The new Ryzen 9 9950X3D showed an average 6% improvement of game performance, which surprised and delighted us.

We also ran seven traditional synthetic 3DMark benchmarks and saw an average of 2% improvement and as much as 47% in frames per second (fps) for Time Spy Extreme.

9950X3D 3DMark
Figure 3. 3DMark benchmark scores.

9950X3D 3D Mark fps
Figure 4. 3DMark benchmark fps.

AMD advertised the new CPU as not only good for gaming but also good for digital content creation, so we also ran the suite of SPEC benchmarks.

9950X3D SPEC 4.0
Figure 5. SPEC benchmark scores (fps and score).

Once again, the results were impressive, averaging 36% and hitting a max of 93%.

All the buzz today is about AI and the AI PC, so we ran UL’s Procyon to test office productivity and AI image processing.

9950X3D Procyon
Figure 6. UL’s Procyon Office productivity results.

Procyon AI image generation
Figure 7. UL’s Procyon AI image-generation results.

Ai text generation average
Figure 8 UL’s Procyon text-generation results.

AIDA64 is diagnostic and benchmarking software used to test and analyze various aspects of computer hardware and performance. It provides detailed information about system components and includes tools to assess the performance and stability of a system. 

Figure 9. The AIDA64 memory test results, higher is better.

AIDA64 CPU
Figure 10. The AIDA64 CPU test results, higher is better.

CPU-Z is a lightweight system utility primarily used to provide detailed information about a computer’s hardware, particularly the CPU, memory, and motherboard. While not a benchmarking tool in the traditional sense, CPU-Z offers several tests and monitoring features:

• Processor information
• Memory information
• Motherboard information
• Graphics information
• Benchmarking and stress testing
• Monitoring

CPU-Z tests
Figure 11. CPU-Z test results.

CPU-Z reporting
Figure 12. CPU-Z reporting mechanism.

While CPU-Z doesn’t provide in-depth synthetic benchmarks like AIDA64 or Cinebench, it is a valuable tool for identifying hardware components and performing quick performance checks.

The four popular DCC benchmarks are Blender, Cinebench, KeyShot and V-Ray.

9950X3D Blender
Figure 13. Blender ray-tracing samples per minute, higher is better.

9950X3D cinebench
Figure 14. Cinebench benchmark results.
9950X3D Keyshot
Figure 15. KeyShot scores, higher is better.
9950X3D V-Ray CPU
Figure 16. V-Ray CPU test results, higher is better.

Handbrake, a codec tester, is a good measure of video streaming and decoding performance. 

9950X3D Handbrake
Figure 17. Handbrake, encoding 4.36GB DVD to MP4 in seconds, lower is better.

Most of the popular codec settings were used: H.264 (x264), 60 fps, CFR – AAC, Stereo, Decomb.

Last but not least is JetStream 2.2. It is a JavaScript and WebAssembly benchmark suite focused on the most advanced Web applications. It rewards browsers that start up quickly, execute code quickly, and run smoothly.

9950X3D Jetstream
Figure 18. JetStream test scores, higher is better.

In office and DCC apps, the 9950X3D wins in all benches except for spreadsheets.

9950X3D system spec
Table 1. Test system and software.

AMD gave us an extensive table of product data, which is close to mind-numbing, and since we’re all sadists here, we’re going to subject you to it too.

9980X3D product specs
Table 2. AMD’s product overview specifications. (Source: AMD)

These new processors should be available now and likely will begin showing up in OEM and ODM systems this fall, if not sooner.

AMD clearly has another winner, and as it should be, the best of them all so far.

Final thoughts

Among the many improvements from the last-generation CPU, the formidable Ryzen 7 9800X3D, the new Ryzen 9 9950X3D employs a second-generation AMD 3D V-Cache: AMD’s proprietary 3D V-Cache technology designed to lower latency and increase frame rates. AMD says they have improved this technology 

The Ryzen 7 9800X3D is this generation’s eight-core CPU built especially for gaming—it came out four months ago.

The eight-core Ryzen 7 7800X3D was the last generation’s CPU built for gaming. However, the 7800X3D was faster (only in gaming) than AMD’s former flagship, the 7950X3D, because the flagship was a Dual-CCD and had issues with latency and core parking.

The last generation’s 7800X3D was the fastest CPU for gaming in the world—until the new 9800X3D was released.

Moving the cache and fixing the software issues with core parking now allows the 9950X3D to match or beat the world’s formerly fastest gaming CPU of 4 months ago, while absolutely shredding it in workstation/pro/creative tasks.  

Epilog

Try as you might, something will always bite you in the ass—ancient proverb

Google researchers found a security vulnerability, EntrySign, affecting all AMD Zen processors from Zen 1 to Zen 4. The flaw allows attackers with local administrator privileges to bypass cryptographic verification and install custom microcode updates. The issue arises from AMD’s use of AES-CMAC as a hash function, which is not designed for secure hashing. Researchers found AMD had used an example key from NIST documentation since Zen 1, enabling forged signatures and arbitrary microcode modifications. These changes can alter CPU behavior until a system reboot. 

AMD has released mitigation and firmware updates to address a high-severity vulnerability that can be exploited to load malicious CPU microcode on unpatched devices. Attackers with local administrator privileges can exploit this weakness, resulting in the loss of confidentiality and integrity of a confidential guest running under AMD Secure Encrypted Virtualization-Secure Nested Paging (SEV-SNP).

LIKE WHAT YOU’RE READING? INTRODUCE US TO YOUR FRIENDS AND COLLEAGUES.