AMD Ryzen 9 3900X vs. Ryzen 7 3700X vs. Intel Core i9-9900K: Which is the best gaming CPU?

Note: This article was first published on 7 July 2019.

Ryzen continues to get better

It's time to say hello to Ryzen 3000. Now more than ever, AMD is looking to reaffirm its commitment to gamers, and the company is stressing all sorts of improvement for third-generation Ryzen and Zen 2, including IPC (instructions per cycle) boosts, clock speed improvements, and more cache and cores. There are over 2.3 billion gamers across the globe, generating over 2 million hours of content on Twitch, which is why AMD wants to make a big play for their wallets. 

However, Ryzen has never been solely about gaming performance. In fact, first-generation Ryzen even struggled somewhat in games, especially at 1080p. Instead, Ryzen's appeal was always about the great balance it offered between games and content creation, and at an attractive price to boot. With Ryzen 3000, AMD is shoring up both these aspects, but with a stronger emphasis on single-threaded performance. That should get many gamers excited, since gaming performance mostly revolves around precisely that. More importantly, improved single-threaded performance will also translate into better performance metrics across the board, since each core can now do more work.

Ultimately, Ryzen 3000 also pivots around the more advanced 7nm process. This enabled up to 29 percent smaller CCX sizes, better performance per watt, twice the cores in the same package, and higher frequencies at the same voltage as the previous generation. 

Making way for chiplets

The biggest paradigm shift at the heart of Zen 2 is probably the use of small 8-core chiplets built on TSMC's 7nm manufacturing process. Each chiplet has two CPU core complexes, or CCXes, that house four cores and their dedicated L3 cache. 

Next, each Ryzen 3000 chip also features a central I/O die built on Global Foundries' 12nm process. These are linked to the chiplets via the second-generation Infinity Fabric interconnect, where the I/O die functions as the central hub for all communications going off-chip. It houses all the processor's PCIe lanes as well, in addition to the memory channels and Infinity Fabric links to other chiplets. The new Infinity Fabric comes with a couple of major updates, including support for PCIe 4.0, an increased bus width from 256-bit to 512-bit, and power efficiency improvements of up to 27 percent. 

With the first generation of Infinity Fabric, the frequency was also coupled to the DRAM frequency, which meant that sometimes they were both limited by the nature of the interconnect clock. Now, AMD has decoupled the Infinity Fabric clock from the main DRAM clock, introducing both a regular 1:1 ratio or a 2:1 ratio that cuts the Infinity Fabric clock in half. If you're out shopping for RAM, what this means is that you want to aim for what AMD says is the performance sweet spot at DDR4-3733, which is the point at which the Infinity Fabric is tied to memory clock at a 1:1 ratio. Beyond that point, even if DRAM frequency is super high, the slow Infinity Fabric frequency may still limit any performance gains from the faster memory.

Obviously, this helps with memory overclocking as well, since the 2:1 ratio effectively lets you ramp speeds up without being constrained as much by the Infinity Fabric clock. 

Furthermore, splitting up the processor into chiplets allows for better yields, improved compute scalability, higher density, and additional flexibility in terms of layout. Furthermore, the central I/O die means that all the cores can now access the cache at approximately the same latency. 

The Ryzen 9 3900X and 3950X both have two chipsets, while the other 6- and 8-core processors have just one. However, the I/O is the same throughout, so all the chips have access to the same 24 PCIe 4.0 lanes and dual-channel memory. But to cram the chiplets into the AM4 package, AMD moved from traditional solder bumps to copper pillars (the I/O die still uses solder bumps though). The copper pillars are more compact and narrower, allowing for a higher density and a more uniform height. 

In comparison, AMD's AM4 socket started off by hosting a quad-core 28nm chip built on a single monolithic die. Today, AMD says every functional IP block on Ryzen 3000 has been shifted relative to its socket pinout location in previous Ryzen generations, but the company has still managed to maintain full socket compatibility with AM4.

Double the L3 cache

Another major change is the size of the L3 cache, which has doubled on Zen 2. Each chiplet, or Core Chiplet Die (CCD) as AMD sometimes calls it, now has 32MB of L3 cache, which works out to 4MB per core, an upgrade that was directly enabled by the smaller 7nm process that increased the transistor budget inside each chiplet. According to AMD, game performance is pretty closely tied to the L3 cache size, so the larger capacity should directly translate into improved performance in game. In general, games based on older APIs or those that are more sensitive to the CPU should see the best benefit. The larger L3 cache also reduces effective memory latency by reducing the number of calls that have to be made to the main system memory.

In fact, AMD is so confident of the impact it has on game performance that it is labeling its L3 cache AMD GameCache instead. What's more, the company says that doubling the cache size has a similar effect as installing faster memory, so AMD is simply keeping data on chip rather than going off-die, which is supposedly more efficient. 

Elsewhere, floating point performance has been doubled as well by moving up to two 256-bit floating point units (FPUs) that support AVX2 instructions. All this, together with an increase in integer resources and load/store resources and the following changes to branch prediction and micro-op cache, should produce an IPC uplift of roughly 15 per cent. 

Better branch prediction and doubled micro-op cache

Ryzen 3000 uses a new TAGE, or Tagged Geometry, branch predictor, which is able to make selections with improved accuracy and granularity, in order to increase throughput by reducing stalls from branch mispredicts. This is on top of the existing Perceptron predictor for the first stage, so Ryzen 3000 actually uses a double-stage branch predictor. In addition, it is better able to handle deeper branch histories, helped along by larger branch-target buffers (BTB) and a doubling of the L1 BTB and nearly doubled L3 BTB. At the simplest level, a BTB basically contains predictions about whether the next branch will be taken or not.

And if you were sensing a theme that everything is getting larger, you'll only find confirmation of that with the larger micro-op cache. The size has doubled from 2K entry to 4K entry, so it will accommodate more decoded operations and also increase throughput by preventing the re-decode of operations. To further help this along, the dispatch rate from the micro-op cache to the buffers has been increased to up to 8 fused instructions. AMD first added a micro-op cache to its Zen architecture in 2016, and it means the CPU doesn't need to keep fetching from other caches to implement frequently used micro operations, or detailed low-level instructions used to implement more complex instructions. 

Windows 10 optimizations

Finally, AMD has added OS-level optimizations to the Windows 10 May 2019 update.  

Where the cores are relative to the L3 cache plays a big impact on the overall performance of the processor as well. If the OS has topology awareness and understands where the cores and cache are, the process scheduler can allocate a certain number of threads to one CCX, or one cluster of four cores, before even spawning or migrating threads onto a secondary or tertiary CCX. This improves performance because the cores have direct L3 cache access to the nearby cache or the neighboring cache, and grouping the threads together on the nearest CCX enables the lowest latency access to its resources. 

That said, not all tasks will behave in this manner. There are times when you might want to have a second thread spawn on a different chiplet, as far away as possible, in order to allow the CPU to maintain high performance without having to deal with regions of high power density. This helps with turbo performance across multiple threads. 

Another thing that's important is clock speed selection. With previous Windows builds, if you wanted to ramp the clock speed from low to high, or just pick any new speed at all, it could take about 30ms. But with UEFI CPPC2, or Collaborative Power and Performance Control, which hands over control to the processor's firmware, this can be reduced to around just 1 to 2ms. This change is particular useful to brief workloads that produce short bursts in clock speeds, such as webpage rendering, web browsing, and application launches. 

Precision Boost 2

Precision Boost 2 is an opportunistic boost algorithm that drives loaded cores to the highest possible frequency until a limit, such as socket power, VRM thermal limit, VRM current limit, or maximum clock speed, is encountered. Once the limit is reached, the processor will cut back on boost and dither the frequency until the situation changes, a process that occurs every 1ms for greater responsiveness. 

That said, there is no change in algorithm between the Ryzen 3000 and Ryzen 2000 series, so you can refer to what I've already written about Precision Boost 2. In a nutshell, Precision Boost 2 ditches the "single-core" versus "all-core" boost model that Precision Boost used in favor of a more granular approach. So instead of switching sharply between "single-core" and "all-core" boost scenarios, Precision Boost 2 enables a far more graceful and effective utilization of each core. 

However, considering that the top Ryzen 3000 chip currently available has 12 cores now, compared to just 8 cores on the Ryzen 7 2700X, it's pretty impressive that AMD is claiming a 4.6GHz boost frequency on the Ryzen 9 3900X that is the same as the peak frequency on the Ryzen 7 2700X, which has 50 per cent fewer threads. 

This feature is also designed to reward the use of more capable cooling solutions, since it utilizes available thermal headroom to drive higher average frequencies. If you want to make the most use of it, ditch the stock cooler. 

Precision Boost Overdrive and Auto OC

AMD is also bringing Precision Boost Overdrive (PBO) to its mainstream chips, which you may remember from AMD's Threadripper processors. PBO expands power limits like package power tracking, thermal design current, and electrical design current to match that of the host motherboard. These expanded limits are communicated by the motherboard BIOS, and will vary according to board designs. The processor will then manage these new and higher limits by itself. 

This means that the CPU will draw more power and hold higher average voltages over time. The extra power delivery allows more cores to achieve a higher effective frequency for a given workload over a longer duration of time. However, PBO does not raise the maximum frequency of the processor beyond the OEM specification unless the user chooses to do so, which is a new feature of PBO on Ryzen 3000. 

You can now opt to enable a maximum boost clock offset of up to an additional 200MHz. This is on top of the previously mentioned PBO features, and is a separate toggle from enabling PBO. However, AMD says that it is best used in conjunction with PBO for the best result. This new feature, called Auto OC by AMD, will tell the processor to manage a maximum boost clock that is up to 200MHz higher than default. This can be configured in 25MHz intervals through AMD Ryzen Master or the motherboard's BIOS. 

That said, Auto OC is not intended to be a guarantee of a higher boost clock on any number of cores, and things like frequency and boost duration are still dependent on firmware-managed limits, even though they may be higher with PBO enabled. It is better thought of as a toggle that opens the door to possibly better single- and multi-threaded performance, all without the hassle of manual overclocking. There's also the benefit of down clocking when the processor is idle, something that's not possible with manual overclocking. 

There's also one final caveat. PBO does drive the processor and platform beyond AMD's official specifications, just like manual overclocking, so you must consent to a voided warranty before proceeding. 

Test setup

The configurations of the test setups we used are listed below. All results were obtained with the Ryzen Balanced power plan and Precision Boost Overdrive disabled. For the Core i9-9900K, multi-core enhancement was turned off on the ASUS motherboard. 

Ryzen 9 3900X/Ryzen 7 3700X

• Thermaltake Water 3.0 360 Riing RGB Edition with Thermal Grizzly Kryonaut
• ASUS Crosshair VIII Hero (Wi-Fi)
• 2 x 8GB G.Skill Trident Z Royal DDR4-3600 (Auto timings: CAS 16-16-16-36)
• NVIDIA GeForce RTX 2080 Ti Founders Edition
• Samsung 860 EVO 500GB M.2 SATA SSD
• Windows 10 Home (64-bit)

Intel Core i9-9900K

• Thermaltake Water 3.0 360 Riing RGB Edition with Thermal Grizzly Kryonaut
• ASUS ROG Maximus XI Extreme
• 2 x 8GB G.Skill Trident Z Royal DDR4-3600 (Auto timings: CAS 16-16-16-36)
• NVIDIA GeForce RTX 2080 Ti Founders Edition
• Samsung 860 EVO 500GB M.2 SATA SSD
• Windows 10 Home (64-bit)

[hwzcompare]

[products=668796,668798,650200]

[width=200]

[caption=Test CPUs compared]

[showprices=1]

[/hwzcompare]

Here's a list of the benchmarks used:

• PCMark 10
• SPECviewperf 13.0
• Cinebench R15
• Cinebench R20
• Handbrake 1.2.2
• Blender Benchmark
• POV-Ray 3.7
• 3DMark
• Ashes of the Singularity: Escalation
• Deus Ex: Mankind Divided
• Far Cry 5
• Metro Exodus
• Middle-earth: Shadow of War
• Shadow of the Tomb Raider
• Tom Clancy's The Division 2

PCMark 10 Extended

PCMark 10 Extended assesses the performance of systems in a variety of workloads, including basic computing tasks, productivity applications, digital content creation, and gaming. Compared to PCMark 8, it also adds in new test metrics, such as app startup times, which quantifies how long it takes to launch a variety of real-world apps, and a rendering and visualization workload to simulate professional graphics and engineering applications. In addition, existing workloads have been updated to reflect modern usage.

It looks like the extra cores on the 12-core Ryzen 9 3900X helped it out slightly here, it was still a mere 2 percent ahead of the 8-core CPUs. The higher 5.0GHz turbo boost clock of the Core i9-9900K also didn't seem to help it against the Ryzen 7 3700X, and both turned out really similar overall scores.

That said, the Ryzen 9 chip's lead seemed to come mostly from the Digital Content Creation workload, where it beat out both octa-core chips. 

SPECviewperf 13.0

SPECviewperf is used to assess the 3D graphics performance of systems in professional applications. Each individual workload, called a viewset, represents graphics and content from an actual real-world application. SPECviewperf actually runs a total of eight different viewsets, but we’ve picked the four which have the greatest performance variation across CPUs display here.

The new SPECviewperf 13.0 incorporates new models and raycasting for volume visualization. Select viewsets have also been updated with new models and fresh application traces. 

The Intel Core i9-9900K still had a slight lead in these workloads, so it seems like the tasks here still prefer higher boost clocks. 

Cinebench R15/R20

Cinebench R15 is a better indicator of multi-threaded performance because of its ability to utilize up to 256 threads to evaluate a processor’s performance in a photorealistic 3D rendering. We ran both single-core and multi-core benchmarks to evaluate single-threaded performance and multi-threaded scalability here.

The newer Cinebench R20 is even more demanding, featuring increased workload complexity, higher memory use, and the latest rendering engine from Cinema 4D R20. Under the hood, the R20 engine also supports AVX, AVX2, and AVX512 instruction sets and the benchmark now supports up to 256 render threads.

Unsurprisingly, the 12-core Ryzen 9 3900X raced ahead to a large lead. It was a good 39 percent faster than the Core i9-9900K in Cinebench R15's multi-threaded benchmark, and a solid 44 percent quicker in R20. The Intel chip still managed to hold its own in the single-threaded benchmarks in both Cinebench R15 and R20 though, even though R20 still favored the Ryzen 9 3900X overall. It looks like AMD has really narrowed the gap in terms of IPC performance, and in Cinebench R15, the Ryzen 9 3900X was just under 3 percent behind the Core i9-9900K.

Handbrake 1.2.2

Handbrake is a video transcoder that converts videos into a format for use on PCs and portable electronic devices, and is a good indicator of a processor’s video encoding capabilities. YouTube content creators, Twitch streamers, and other video creators will be most interested in this performance metric. For this benchmark, we used a 1.7GB .mkv file. 

The results were predictable here, although it's pretty impressive that the Ryzen 7 3700X was ahead of the Core i9-9900K, especially since it also has 8 cores and a lower boost clock. 

Blender Benchmark

Blender likes having many cores as well, and the open-source software has been used for modeling and to create effects in movies. The new Blender Benchmark offers the option between Quick and Complete runs, and the numbers seen here are from the Quick benchmark, which puts the CPUs through scenarios like the BMW and Classroom demos. 

The 12-core Ryzen 9 3900X is way ahead here as expected, but the Ryzen 7 3700X really stands out as it also beats the Core i9-9900K. 

POV-Ray 3.7

The POV-Ray built-in benchmark also favors having more cores, and both Ryzen processors skipped ahead of the Core i9-9900K. 

Gaming benchmarks

We only ran our gaming benchmark suite at 1080p, since that's where the CPU is more of a factor. At the higher resolutions that are more graphically intensive, the performance differences start to even out. 

3DMark

The synthetic 3DMark benchmark tests graphics and computational performance at different resolutions, starting at 1080p and going all the way up to 4K. A series of two graphics test, one physics test, and then a combined test stresses your hardware in turn to assess its performance.

We’ve also teased out the Physics and CPU scores for the Fire Strike Ultra and Time Spy Extreme tests and compiled them in a separate table to give a better idea of how each processor performed.

Gaming performance was surprisingly even in terms of overall scores, but a look at the Physics and CPU numbers provides a better idea of CPU performance. Both the Core i9-9900K and Ryzen 7 3700X were pretty even for the Fire Strike Ultra Physics scores, but the Intel chip had a slight lead in terms of the Time Spy Extreme CPU numbers. 

Ashes of the Singularity: Escalation

Ashes of the Singularity has long been the poster child for the performance benefits a low-level API like DirectX 12 can bring. It is based on the Nitrous engine and can be extremely punishing thanks to the huge number of onscreen units and the sheer level of detail accorded to each unit. However, the CPU does become the limiting factor at lower resolutions and settings. 

Results were a mixed bag here, with the Core i9-9900K coming out on top by a small margin at High settings, but falling behind again at Crazy settings. 

Deus Ex: Mankind Divided

Mankind Divided still clearly prefers the higher clock speeds on the Intel processor, and it was ahead of the Ryzen 9 3900X by a good 18 percent. That advantage shrank at Ultra settings though, narrowing to just 5 percent. 

Far Cry 5

The same goes for Far Cry 5, where the Core i9-9900K led the Ryzen 9 3900X by 25 percent, even at Ultra settings. It looks like the game also doesn't really scale that well with more cores, since the Ryzen 7 3700X actually beat the Ryzen 9 3900X by a hair. 

Metro Exodus

In Metro Exodus, the Core i9-9900K still had a small 6 percent advantage over the Ryzen 9 3900X. At High settings, that lead was slightly larger at just under 9 percent. As in Far Cry 5, the Ryzen 7 3700X was also quicker than the 12-core Ryzen 9 3900X by a small margin. 

Middle-earth: Shadow of War

Shadow of War sees the Intel chip take the lead as well. At Ultra settings, the Core i9-9900K was ahead by 7 percent. Again, it seems like the game doesn't scale well with more cores, and there was virtually no difference between the Ryzen 7 and Ryzen 9 chip. 

Shadow of the Tomb Raider

The Ryzen 7 and Ryzen 9 processors turned out very similar performance in Shadow of the Tomb Raider as well. At Ultra settings however, the Core i9-9900K was faster than the AMD chips by roughly 13 percent. 

Tom Clancy's The Division 2

Running the game at Ultra settings helps narrow the performance gap somewhat, where the Core i9-9900K was ahead of the Ryzen 9 3900X by just under 3 percent. 

Temperature and power consumption

Both AMD processors drew more power than the Intel Core i9-9900K, but the Intel chip still ran 10°C hotter than the Ryzen 9 3900X. 

A great third outing from AMD

AMD has pretty much knocked it out of the park with its third-generation Ryzen products. The Ryzen 9 3900X clearly delivers excellent multi-threaded performance and very competitive gaming results, so it's a clear choice for both gamers and content creators. If you're a Twitch streamer that wants to consistently deliver a good quality 60fps stream, the Ryzen 9 3900X is shaping up to be a great choice.

However, it's the Ryzen 7 3700X that is really the standout performer. At just S$498, it consistently leads the Core i9-9900K in multi-threaded workloads. It's not that far behind in terms of gaming performance either, and it's possible to make the argument that the trade-off is pretty worth it. At the moment, the Core i9-9900K is retailing for a whopping S$758 on Lazada (that's the cheapest listing I could find on the site), a hefty premium that is probably not justified by the 10 percent or so lead that it has over the Ryzen 7 3700X. 

The Core i9-9900K is still the king of raw gaming performance but the new Ryzen 7 3700X and Ryzen 3900X offer unparalleled value and bang for your buck.

Similarly, the Ryzen 9 3900X costs S$798, only slightly more than the Core i9-9900K. That's quite some insane value right there, for only slightly lesser gaming performance and a huge lead in content creation tasks.

While it seems like the Core i9-9900K may still be king if all you care about is raw gaming performance, that's often not the case today. Plenty of people want to be able to stream stuff on Twitch or render video, in addition to playing games. For folks like these, the new Ryzen 7 3700X and Ryzen 9 3900X step up to the plate admirably. I can't say that the new Ryzen 3000 chips dethrone the Core i9-9900K as the best gaming CPU, but I can say that they are the best and most value-for-money chips when it comes to both gaming and content creation, which is exactly what AMD was shooting for. 

As processors with over eight cores start to become the norm, multi-threaded performance, and not just in-game frame rates, will start to become an increasingly important metric, and it's in this area that AMD absolutely shines. 

	Performance	Features	Value	Overall
AMD Ryzen 9 3900X	8.5	8.5	8.5
AMD Ryzen 7 3700X	8.5	8.5	9.0