AMD Ryzen 9 7900X and 7700X Review: The Dawn of ZEN 4

After nearly two years, AMD is back with their newest Desktop processor based on the Zen 4 microarchitecture. This round of Ryzen CPUs comes with a laundry list of changes. First, we have the switch from 7 nm to TSMC’s 5 nm process node, DDR5 memory, and the next generation PCI Express 5.0. The flagship CPU using the new Zen 4 architecture, code-named Raphael, is a 16-core, 32-thread CPU with a maximum boost speed of 5.7 GHz, yes, you read that correctly, and we’ll elaborate more on this later. The new Ryzen 9 7950 X has a suggested release price of $699, which is $100.00 less than the Ryzen 9 5950 X released in November 2020, nearly two years ago. We have also included a list of SKUs with specifications and pricing below.

Today we have the 12-core, 24-thread Ryzen 9 7900 X with a suggested E-tail price of $549 and the Ryzen 7 7700X (SEP $399) to run through our testing suite. We also include the i9-12900K (16-core @ $564) and the Ryzen 9 5950X (16-core @ $549) for comparison to align core count and pricing.

Specifications and Features

AMD’s Ryzen CPUs still use the same general layout with an I/O die (IOD) and up to two Core Cluster Dies (CCD) per CPU. All the CPUs are unlocked; they all have SMT (Simultaneous Multithreading) and still follow the same core count progression through the CPU lineup of six, eight, twelve, and sixteen cores at the top end. One thing that will stand out throughout this review is the clock speeds these new CPUs can reach on light loads and maintain with heavier loads.

Process Node Improvements

With Zen 4, AMD has moved to TSMC’s 5-nanometer process for the CCDs and 6-nanometer for the redesigned I/O die. With the new 5n process node, the CCD size has shrunk from almost 81 mm² to 70 mm² but has more than a 50% increase in transistor density, moving from 2.09 to 3.4 Billion transistors per CCD. The IOD size is marginally less, but we see a dramatic increase in transistor density from 2.09 to 3.4 Billion. The transition to 5 nm also gives the Ryzen 7000 series a 40% gain in performance per watt when averaged out over their TDP range.

The new node also brings some incredible clock speed improvements over Zen 3. The all-new Ryzen 9 7950 X can boost a couple of threads up to a whopping 5.7 GHz and, according to AMD, maintain a 5.2 GHz all-core boost on heavier loads. We can verify that the Ryzen 9 7900 X hit 5.65 GHz in SuperPi, a single-threaded benchmark, and during Cinebench R23, the all-core boost floated between 5.2 and 5.3 GHz across the board. This is very impressive compared to the Ryzen 9 5950 X with a single and all-core boost of 5.0 and 4.5 GHz, respectively. The Ryzen 7 7700 X didn’t quite reach the same single-core speed of 5.65 GHz and came in only 50 MHz behind, while the all-core boost was similar.

Ryzen 7000 Series IOD

The new Ryzen 7000 series CPUs also bring a redesigned 6 nm IOD with an eye to the future, including a DDR5 DRAM controller and PCIe Gen 5.0. AMD has committed to the new AM5 socket until the year 2025, so these inclusions are necessary to remain competitive in the future. We also have an enhanced Infinity Fabric for improved data transfer between the IOD and CCD.

RDNA2 Graphics

The new IO die includes two RDNA2 compute units for the first time, and the 7000 series of Ryzen CPUs no longer require a discrete graphics card in a workstation environment. The included iGPU is by no means a powerhouse, but it does support HDMI 2.1, Display Port 2.0, and USB C with 4K60 resolution.

PCI Express Gen 5.0

Starting with Zen 4, AMD is the first to include PCIe Gen. 5.0 technology on its AM5 platform. It wasn’t that long ago that PCIe Gen 4.0 was introduced, which doubled the transfer bandwidth over PCIe Gen 3.0. Even now, there aren’t many add-ons that take advantage of the extra bandwidth aside from storage. Graphics cards still haven’t reached that point and likely won’t for some time. The PCIe 5.0 lanes are laid out with 16 lanes dedicated to graphics with a total bandwidth of 128 GB/s and eight lanes for storage. The remaining four lanes are for the chipset downlink but run at PCIe Gen 4.0 only.

DDR5 Memory and AMD EXPO

The IO die of the Ryzen 7000 series also houses the DDR5 DRAM controller and supports up to 128 GB of DDR5 in a dual channel configuration. The “official” supported speed is JEDEC 5200, but according to AMD, the “sweet spot” is at 6000 MHz, and we can say the IMC had no issues running at 3000 MHz.

For those familiar with the previous generation, Ryzen 5000 series, the “sweet spot” was 3600 MHz for the DDR4 RAM speed. Optimal performance was achieved by setting the Infinity Fabric, Memory Controller, and RAM in a 1:1:1 ratio. For most CPUs, this was around the 1800 MHz mark (3600 MHz DDR); better samples could squeeze out another 100 MHz but were sometimes unstable. With ZEN 4, this has changed. AMD recommends leaving the Infinity Fabric on AUTO in the BIOS and setting the memory and memory controller at a 1:1 ratio. During our testing, this worked very well, and the Gigabyte Aorus Master we had for testing seemed to default to a 2000 MHz Infinity Fabric speed.

With the new platform, the very first boot after platform assembly will take an extended period of time. According to AMD, the system may appear unresponsive for possibly two to three minutes. This is due to “first boot memory training” and will revert back to the typical 5-10 seconds for subsequent restarts. As the BIOS matures this first/initial boot time will be reduced.

With the new platform, we are being introduced to the AMD EXPO, short for extended profiles for overclocking. Much like Intel’s XMP, the AMD EXPO DDR5 is optimized for AMD Ryzen to deliver better performance and lower latency. Any vendor can also adopt EXPO free if they publish documentation listing all timings and testing protocols. Availability coincides with the AM5 platform release with speeds up to DDR5 6400.

Ryzen 7000 IPC Gains

AMD’s engineers worked hard at improving the efficiency of the front end of the Ryzen CPU. In a nutshell, by improving the branch prediction and expanding cache storage, there are a lot fewer “misses” that need to start over, making it more efficient. The L2 cache, for instance, has been doubled to 1 MB, the Op cache is 68% larger, and they’re predicting two branches per cycle. According to AMD, this has resulted in a 13% IPC increase averaged over 22 different types of loads. This IPC gain, combined with increased operating frequencies, has given the Ryzen 7000 series a +29% improvement in single core performance compared to the previous generation.

Raphael also includes power-efficient AVX-512 support in the floating point unit. AMD’s approach to this is what they call “double pumped” AVX-512 capability on a 256-bit data path. This approach does have a slight impact on performance but prevents the thermal and frequency compromises of other implementations and gives the new Ryzen CPU a performance boost in AI inferencing workloads.

Below is a list of the new Ryzen 7000 unlocked CPUs with specifications and pricing; keep in mind this is MSRP.

AMD 600 Series chipsets and LGA Socket 1718 (AM5)

With ZEN 4, we also have a new socket, LGA 1718. LGA sockets are nothing new, but it’s been a while for a mainstream AMD CPU/Motherboards to be configured this way. Moving from the 1331 PGA to a 1718 LGA allows nearly 400 more contacts to communicate with the motherboard and peripherals, and no more CPUs are stuck to the bottom of your cooler when you remove it. AMD says that a compatible AM4 cooler will fit the new socket, but there’s a caveat here. Your AM4 cooler will work using the AMD backplate or hold-downs to secure it. The AMD backplate holds the socket latch mechanism in place, so the backplate itself can’t be removed. We ran into issues initially as our cooler has its own backplate and standoffs that go through the motherboard. We managed to make it work using various parts, but this isn’t likely to happen for most people.

Another first with the Raphael launch is PCIe 5.0 (32 GT/s transfer rate) compatibility from the CPU. We end up with various PCIe standards between the CPU and the 600 series chipset, including PCIe 4.0 and PCIe 3.0 lanes. We’ll attempt to break this down as clearly as we can.

The new Ryzen 7000 CPUs have 28 dedicated PCIe Gen 5.0 lanes; 16 are dedicated to graphics in either an x16 or x8/x8 format. We also have eight PCIe 5.0 lanes for NVMe storage devices. That brings us to the X670E and X670 chipsets which are comprised of two separate chips named Promontory 21 (PROM21) and supplied by ASMedia. One PROM21 is connected to the CPU with four PCIe 4.0 lanes as normal and the second is connected to the first. From the combined chipsets, we have an additional 20 PCIe lanes. These lanes are broken down into 12 PCIe 4.0 and 8 PCIe 3.0 enabling massive connectivity on the AM5 platform. Depending on how the vendor lays out the motherboard, we’ll also see native support for USB 3.2 Gen 2×2 (20 Gbps) through the chipset.

As you can see below, the B650E chipset that uses a single PROM21 chip, offering less connectivity for peripherals, but the CPU layout remains unchanged. Comparing the two chipsets, we can see the USB connectivity is cut in half, and the same goes for the PCIe Gen 3.0 lanes dedicated to SATA storage. The B650 chipsets PCIe Gen 4.0 lanes are also reduced from twelve to eight.

At this point, the only difference we’ll likely see between the “E” and non “E” motherboards will be the price and the vendors’ choice of layout, for example, one X16 PCIe Gen 5.0 slot VS two PCIe Gen 5.0 X16 slots which could both share the bandwidth and run eight lanes each. We’re also assuming the “Overclocking” oriented motherboards with robust power systems will also carry the “E” designation.

Meet the Ryzen 9 7900X and Ryzen 7 7700X

Here’s a slideshow of the sample kit and packaging, some close-ups of the CPUs’ top and bottom, plus a comparison of the Ryzen 7000 and the Ryzen 5000. Overall, the IHS shape has changed from a simple square to something resembling a turtle or a beetle. The PCB appears to be thicker when compared to the last generation, and the obvious contact pads versus pins. We’ve also included a picture of the new socket with its array of 1718 pins. I find the design reminiscent of the Threadripper HEDT CPUs with that dividing line down the center.

This slideshow requires JavaScript.

Here we have CPU-Z captures of the Ryzen 7900X and 7700X at their stock settings with EXPO enabled. The second image shows the boost speed of the Ryzen 9 7900X at 5.67 GHz while running the SuperPi benchmark.

Special Thanks

At this point, we would like to thank AMD for arranging a Gigabyte motherboard and G.Skill DDR5 to facilitate our testing.

Starting with the motherboard, We received Gigabyte’s X670E AORUS Master. The AORUS series is aimed squarely at enthusiasts and sits at the top of Gigabyte’s lineup. The X670E AORUS Master is an E-ATX motherboard and comes fully decked out with RGB lighting and a beefy power delivery system.

Now for some highlights of the X670E AORUS Master:

• AMD Socket AM5: Supports AMD Ryzen 7000 Series Processors
• Unparalleled Performance: Twin 16+2+2 Phases Digital VRM Solution
• Dual Channel DDR5: 4 SMD DIMMs with EXPO & XMP Memory Module Support
• Next Generation Storage : 2 PCIe 5.0 x4 and 2 PCIe 4.0 x4 M.2 Connectors
• Fins-Array III & M.2 Thermal Guard III: To Ensure VRM Power Stability & 25110 PCIe 5.0 M.2 SSD Performance
• EZ-Latch Plus : SMD PCIe 5.0 x16 Slot & M.2 Connectors with Quick Release & Screwless Design
• Hi-Fi Audio with DTS: X Ultra: ALC1220 CODEC
• Fast Networks : Intel 2.5GbE LAN & Intel Wi-Fi 6E 802.11ax
• Extended Connectivity: DP, HDMI, USB-C with DP Alt Mode, Dual USB-C 20Gbps, and Upcoming GIGABYTE USB4 AIC Support
• Q-Flash Plus: Update BIOS Without Installing the CPU, Memory, and Graphics Card

For more details, you can check Gigabyte’s website.

Next is G.Skill with a Trident Z5 neo 2 x 16 GB DDR5 6000 Cl 30-38-38-96. The G.Skill Trident Z neo is made with hand-picked, tightly screened SK Hynix memory ICs, solid aluminum heatsinks, and high-performance PCBs. For more information on this G.Skill TridentZ neo memory, follow this link to G.Skill’s website.

Test Setup and Results

Here we take a slightly different approach to CPU testing with ours based on many Hwbot.org benchmarks since that is what we are known for, overclocking and benchmarking. We also use real-world testing with Cinebench, Blender 3.3, Corona, and 7Zip to give readers a good idea of the product’s general performance.

AMD Ryzen 7000 Test System

12th Gen Intel Test System

AMD Ryzen 9 5950X Test System

AMD Ryzen 9 7900X and Ryzen 7 7700X Performance Testing

CPU Tests

• AIDA64 Engineer CPU, FPU, and Memory Tests
• Cinebench R20 and R23
• Corona Benchmark
• Blender 3.3 Benchmark, We use the sum of all three tests
• SuperPi 1M/32M
• WPrime 32M/1024M
• 7Zip

All CPU tests were run at their default settings with XMP/EXPO enabled unless otherwise noted.

Gaming Tests

We have updated our gaming tests and dropped them down to four games for CPU reviews. Our testing is done at 1080p as it’s more CPU-dependent, but the titles we use to cover both CPU-heavy and GPU-bound titles. All game tests were run at 1920×1080 with all CPUs at default settings unless otherwise noted. Please see our testing procedures for details on in-game settings.

• Shadow of the Tomb Raider – DX12, “Highest” preset
• Far Cry 6 – DX12, Ultra preset, VSync Off
• F1 2021 – DX12, Very High defaults, TAA, and x16 AF, Bahrain track, show FPS counter
• Metro: Exodus – DX12, Ultra defaults
• UL 3DMark Fire Strike (Extreme) – Default settings

AIDA64 CPU, FPU, and Memory Tests

Below are a couple of shots of the AIDA64 cache and memory benchmark results for the Ryzen 9 7900X and 7700X. As you can see, the number of cores seems to slightly impact memory bandwidth, with the 12-core 7900X taking the lead over the 8-core 7700X. The new Raphael CPUs using DDR5 dramatically improve the memory bandwidth while the latency only takes a slight hit compared to the previous generation, 5950X. The overall winner is Intel’s Alder Lake by a reasonably large margin in the bandwidth category.

In the AIDA64 CPU tests, we see that the Ryzen 9 7900 X takes the lead over the Alder Lake rival in most tests. The 7700 X also puts up a good fight considering it has half the cores of the i9-12900K. The PhotoWorx test, always a blue team favorite, is Intel all the way.

Moving on to the AIDA64 FPU tests, the 7900X leads every test across the board by significant margins except SinJulia, where the 5950X eeked out a narrow lead. Again, the Ryzen 7 7700X put on a good show besting the i9-12900K in every test.

Real-World Tests

Last year in our i9-12900K review, AMD lost its hold over Cinebench, even if it was by a narrow margin. Cinebench had always been a strong benchmark for AMD, and today the crown has changed hands once again thanks to the Ryzen 9 7900X. As you’ll see later in the review, Intel still has AMD in the IPC race for Cinebench, but the new Ryzen CPUs maintain a +5.2 GHz all-core boost during these benchmarks, pushing them ahead of their rivals. As you can see below, this fantastic speed gave AMD an easy win across all our real-world benchmarks.

Pi and Prime-Based Tests

Next up are the Pi and Prime number-based tests. SuperPi has always been an Intel favorite, but that is no longer the case, thanks to the new ZEN 4 architecture. The new Ryzen CPUs come with incredible single-core boost speeds, with the 7900X reaching nearly 5.7 GHz and the 7700X in the 5.6 GHz range, easily walking them past the i9-12900K. Wprime, on the other hand, loves cores and threads. Combine that with the high all-core boost speed of the 7900X, and the i9 didn’t stand a chance, losing the Prime battle by ~30%.

Gaming Results

We’ll let the gaming results speak for themselves. In all the games aside from Metro Exodus, the Ryzen 9 7900X and the 7700X easily outpaced the i9-12900K by 20 FPS. It appears that we have a new generation of top gaming CPUs at least for a month or two. Raptor Lake is just around the corner, and leaks so far look pretty impressive.

Next up, we have 3DMark Fire Strike Extreme, a DX11-based test; UL says the graphics are rendered with detail and complexity far beyond other DX11 benchmarks and games using 2560×1440 resolution. As you can see below, the overall and graphics scores were reasonably close, but both Ryzen 7000 CPUs came out ahead. The physics test is strictly CPU-dependent; typically, thread count is the most significant factor, and here is the 7900X leading the pack.

IPC Testing

With the new ZEN 4 CPUs, AMD claims up to a 13% improvement in IPC (instruction per clock) across 22 different workloads. We have our own range of tests to compare IPC across generations and platforms, which we feel has a real-world approach comprised of commonly used software. All CPUs are set to the same clock speed of 4.0 GHz and performed on one thread without any hyper-threading for these tests.

As we can see below, Raphael offers 7-8% gains in both Cinebench benchmarks compared to previous generations, but they’re still about 8% behind the Alder Lake CPUs. The WPrime test, on the other hand, shows a vast IPC gain of +20% compared to the 12900K and the 5950X. The 7Zip test shows minor improvements but barely a percentage point over the last generation of Ryzen.

Power Consumption and Temperatures

AMD doesn’t use Power levels like Intel. Instead, they use Precision Boost 2; The Ryzen processors have a sophisticated set of sensors designed to analyze conditions inside your PC to squeeze out as much speed as possible. Precision Boost 2 has a set of parameters, including the type of workload and number of cores involved, power consumption from the processor socket, the motherboard’s power circuitry, and a maximum CPU temperature of 95°C. This mechanism will continue to increase the frequencies until one or more of those limits have been reached and then level out the core speeds at that point.

We used the EK Predator CLC, a very effective all-in-one 360 mm cooler for testing. We then subjected the CPUs to the AIDA64 stability test, Corona benchmark, and Blender 3.3 benchmark. The CPUs run at default speeds and then repeated the tests with our overclocked settings for comparison.

Interestingly, the 5.5 GHz all-core overclock at 1.28 V produced better results in all of these benchmarks performance-wise but used considerably less power and ran cooler than stock operation. This is typical of auto settings in the BIOS as they tend to offer more voltage than necessary for most clock speeds. Even at stock, the Ryzen 9 7900X used nearly 100 W less power than the i9-12900K in the AIDA64 stability test and Blender benchmark. Overall, the temperatures were controlled well, with the EK 360 staying in the mid-eighties on top of the 7900X and 7700X.

Overclocking the Ryzen 9 7900X and Ryzen 7 7700X

In our first attempts at overclocking the Ryzen 9 7900X, we used AMD’s Ryzen Master software. I have never been a huge fan of software overclocking but utilizing the curve optimizer plus and offset to the voltage for stability did yield some decent results. It maintained the great single-core boost and bumped up the multicore performance to nearly the same levels as our 5.5 GHz all-core overclock.

It wasn’t difficult; I just set the per-core curve optimizer in the software and let it run, which did take a long time. Then some testing was needed to stabilize it as some loads would crash. In the end, a slight 0.025 V offset was added, and the system ran relatively stable. The power usage and temperature were up slightly from stock but still within limits. My biggest issue with the software is that each time I tried to reset the overclock Ryzen Master had set, the system would fail to POST after the required reboot. At that point, I had to clear the CMOS and reset the BIOS. This was the only way to get the system to start again.

Next, we took the conventional approach and overclocked the CPUs using the BIOS. The Ryzen 9 7900X and 7700X ended at the same 5.5 GHz all-core overclock using 1.28V. Trying for 5.6 GHz kept failing on more demanding tasks, and adding voltage didn’t seem to help as the temperatures were well into the nineties, and it was apparent we were at the end of our overclocking journey.

We also tried a bit of memory overclocking trying to see if the IMC could handle speeds above 6000 in a 1:1 ratio. We ended up with 6400 MHz at 1:1 with slightly relaxed timings of CL32-39-39-102 using the stock RAM voltage. These settings passed 20 minutes of AIDA64’s Cache and Memory stability testing, so it’s relatively stable. We include the results of the Cache and Memory benchmark for comparison to the 6000 MHz stock settings.

Conclusion

The new Ryzen 7000 CPUs are here, bringing with them a move to a new 5 nm process node and additional performance with significant power savings. We also see a massive boost to the single/lightly-threaded results due to higher clock speeds. Operating in the plus 5.5 GHz range is a big step in the Ryzen evolution, something that was barely possible with extreme cooling on previous generations. With this boost in speed and a noticeable IPC gain over previous generations, they were able to wrestle the performance crown away from Intel in the majority of our benchmark suite.

Zen 4 brings with it DDR5, the next step in RAM technology, bringing higher densities and much faster speeds. We’re also being introduced to AMD EXPO; this should make overclocking RAM on AMD as easy as setting the profile and rebooting, much like Intel’s XMP. We say overclocking since AMD’s top “officially supported” speed is 5200 MHz, so any speed above this is considered overclocking. We did reach 6400 MHz RAM speed, and thanks to the revamped IO die, the system maintained a 1:1 ratio with the new DDR5 IMC.

AMD is also trying to keep up with Intel and use the latest standards by introducing PCIe Gen 5.0. The new PCI express standard effectively doubles the bandwidth, gigatransfer, and frequency rates over PCIe 4.0 for substantially faster data transfer rates. Since there aren’t any PCIe 5.0 devices available at the time of this writing, we cannot test the performance. However, PCIe Gen. 5.0 NVMe storage devices will be available very soon with some impressive speeds.

Adopting the new Ryzen CPUs requires an upgrade to DDR5 as DDR4 is no longer compatible. A new LGA 1718 motherboard will be on that list as well, and as usual, there will be a variety of options available to suit every need and budget. The AM5 platform remains compatible with most AM4 coolers, but as I mentioned earlier, that’s only if the stock backplate is used. The backplate is not removable on the new motherboards since it secures the new LGA latching mechanism.

Looking at the Ryzen 9 7900X from a performance per dollar perspective and based solely on the suggested retail price, the 7900X would win hands down. With an MSRP of $549.00, it’s on a level playing field with the i9-12900K on Amazon for $584 and the Ryzen 7 5950X ($549.00 on Newegg). If you’re in the market for a new system and the upgrade to DDR5 is within the budget, an upgrade to the Ryzen 7000 series would be a good move. Whether you choose the 12-core, 24-thread 7900X, or the 8-core 7700X, AMD has committed to this platform until 2025, offering a drop-in upgrade path in the future.

In the end, the new Ryzen 7000 CPUs have managed to shift the tide back to the “Red Team” in terms of performance. They also bring a new architecture with new features and a decent IPC improvement. With the addition of DDR5 and PCIe 5.0, AMD is now back on a level playing field with Intel and blazing a path into the future with some impressive clock speeds.

Shawn Jennings – Johan45

Related Articles:

• Intel Core i9-12900K and i5-12600K Review
• Intel Core i9-11900K and i5-11600K Review
• Intel Core i9-1900K and i5-10600K Review
• Ryzen 9 3900X and Ryzen 7 3700X review
• Intel Core i9-9900K CPU Review
• AMD Ryzen 5 3600X CPU Review