AMD’s Ryzen 7 3700X and Ryzen 9 3900X Reviewed: Red Storm Ryzen

After months of rumors, disclosures, and previews, AMD’s 7nm moment of truth has finally arrived. The 7nm, third-generation Ryzen 3000 family represents AMD’s first chance to seize the top of the CPU market in nearly 15 years.

This review assumes you are generally familiar with Ryzen and the Zen 2 architecture; our E3 deep dive on the topic will help you get up to speed if you need a quick refresher. To recap: Today, AMD is launching its Ryzen 7 3000 family, a new top-to-bottom refresh of its previous-generation Ryzen products. These new chips are built on TSMC’s 7nm process node and use a new “chiplet” design that splits off I/O circuitry and DRAM controllers that don’t particularly benefit from die shrinks and keeps them in a unified, common die that all chiplets connect to.

The Ryzen 7 3000 family doesn’t make any dramatic changes to AMD’s core counts or overall product positioning in the low or midrange markets, but it does introduce a new 12-core desktop CPU with a 16-core chip still waiting in the wings. AMD has teed up an extremely aggressive product lineup against Intel and clearly intends to deliver a beat-down against its chief competitor.

The pricing says it all. AMD has positioned its 12-core Ryzen 9 3900X to go head-to-head against Intel’s Core i9-9900K. The eight-core Ryzen 7 3800X will tackle Intel’s eight-core Core i7-9700K, while the $329 Ryzen 7 3700X sits in a class of its own at that price point. Lower-priced parts in the stack will face off against their equivalently positioned Intel counterparts.

Historic Firsts

In the 18 years I’ve covered the IT industry, AMD has never beaten Intel to a node, until now. Nearly 20 years ago, AMD effectively tied Intel when both firms began shipping 180nm parts at very nearly the same time. It never happened again. Thereafter, Intel moved to new nodes months or even years before its competitor, until today.

Even allowing for the fact that Intel’s 10nm and TSMC’s 7nm are considered to be roughly equivalent, AMD is about to have 7nm CPUs and GPUs in-market at volume. Intel currently has a single 10nm Core i3-8121 that ships without a GPU and is only used in a handful of low-end systems.

That’s a clean, solid, AMD win.

The Ryzen 3000 family will also be the first family of desktop CPUs to push a mainstream platform up to 12 CPU cores. AMD initially had plans to nudge CPU core counts above 8 cores with follow-ups to its Bulldozer family, but ultimately decided to focus on improving that architecture’s low-power performance while it built the Zen architecture as a replacement. While both Intel and AMD have offered CPUs with more than eight cores for a number of years, neither company has ever brought such high-count chips to its mainstream desktop platform. Intel reserves them for its HEDT (High-End DeskTop) family and AMD has deployed them as part of its Threadripper product line.

What to Watch For

AMD has made no secret of its ambitions for the Zen 2 architecture. At E3 this year, AMD’s Director of Client Management, Travis Kirsch, declared that he saw no reason for anyone to buy an Intel CPU once Ryzen debuted. That’s setting a high bar for AMD to clear, considering how poorly its CPUs compared with Intel just a few short years ago.

When AMD launched Ryzen in 2017, it forced Intel to shift its entire product lineup, introducing Hyper-Threading on low-end CPUs in the Pentium family and adding additional cores to its Core i3, i5, and i7 CPUs. After nearly 6 years of static core-count configurations, the market has changed significantly in just the past two years.

There isn’t a space where AMD isn’t gunning for Intel, but there are a few places we’re particularly keen to compare, including:

2700X versus 3700X: The 2700X delivered a modest improvement over the original Ryzen 7 1800X, but AMD has pledged at least a 1.15x overall gain over the 2700X. That’s difficult for any company to pull off these days, but we’ve theorized that there must have been some low-hanging fruit left over from the breakneck rush to get the original Ryzen into the market. AMD promised a 1.15x uplift over and above the 2700X, so we’ll be checking to see what the company actually delivers.

Improved Power Consumption: 7nm is supposed to deliver major improvements in overall power consumption and performance per watt. We’ll be checking to see how AMD compares with both Intel and its own previous CPUs on this score.

Overall Performance Leadership: AMD has queued up a 12-core Ryzen to match the Core i9-9900K, and you don’t have to be a CPU engineer to guess how that comparison is going to go. What’ll be interesting is to see how much stronger AMD’s overall single-threaded performance is or isn’t.

1080p Gaming Performance: Whether Ryzen had a 1080p “weakness” largely depended on which games you played and what GPU you used to play them, but some games didn’t run as quickly on AMD’s new Zen architecture compared with Intel. That situation has supposedly improved — we’ll be watching to see how much.

Test Setup

Our test configuration is complex enough to merit a chart.

Originally, our plan was to test all AMD CPUs on the same platform, but that hit a snag related to some of the early motherboard UEFIs we had available and the tight schedule for this review. While AMD provided our kits with a fairly reasonable testing window, the sheer volume of hardware we had to test and retest has been substantial. Nvidia’s 430.86 WHQL driver and a Gigabyte Aorus RTX 2080 was used for all game testing.

We elected to use 16GB of DDR4-3600 for the Ryzen 3xxx CPUs, but limited older CPUs to 16GB of DDR4-3200. This was partly a practical decision — we simply didn’t have time to validate the faster clocks on every CPU — but also, we wanted to compare what has been a top-end configuration on the various chips we’ve been using against the advantage of equipping the Ryzen CPUs with faster RAM.

All systems were tested using Windows 10 May 2019 Update with all security patches installed. All security patches were installed and the latest UEFIs were used for each motherboard. All of the results presented in this review are new, nothing has been pulled from previous coverage. Some of our benchmark versions have been updated as a result, and we’ve changed our Handbrake encoding test — results are no longer comparable to previous iterations of the benchmark.

Note: Our 9700K results are a bit lower than expected in non-gaming benchmarks. This is under investigation. Gaming benchmarks with the chip were quite good. The issue is not thermal and all UEFI settings appear to be correct.

Power Consumption Testing

Our power consumption testing was handled a bit differently than our mainstream testing. All testbeds were outfitted with 32GB of DDR4-3200 for this test. We’ve historically used Prime95 for power consumption testing, but Prime95 has recently released an update that somewhat changed the power consumption of both Intel and AMD CPUs. Previous iterations of the application offered the option to test medium-length FFTs and positioned this as the most power-hungry application in the benchmark. The newest version of the test has a new option for small FFTs. Power consumption increases on both AMD and Intel CPUs when tested with this new version of the program, but it increases far more on Intel CPUs.

For this reason, we tested power consumption in both versions of the program, as well as in Cinebench R20 during the multi-core scene render.

Our results are embedded in the slideshow below.

Actually, one of our graphs isn’t sized properly for the slideshow, so it’s going right below. Everything else is embedded in the slideshows below.

The Blender 1.02Beta benchmark offers a range of scenarios and scenes for comparison — and Ryzen chews through all of them. Even before we hit the 12-core Ryzen 3900X, the Ryzen 7 3700X is 1.13x – 1.16x faster than the 2700X and 1.05x – 1.1x faster than the Core i9-9900K. The Ryzen 9 3900X is 1.34x – 1.4x faster than the Core i9-9900K.

In the older Cinebench, the 9700K struggles to pass the Core i7-8086K in multi-threaded and falls well behind in single-threaded code. The Ryzen 7 3700X is 1.04x faster than the 9900K in multi-threaded code and 1.04x slower in single-threaded. The 3900X matches the 9900K in single-thread and beats it by 1.54x in multi-threaded.

The Ryzen 9 3900X is not actually the only CPU on this graph, it only looks that way. In Cinebench 20, the 3700X maintains a 1.21x gain on the 2700X in multi-threaded and a 1.06x improvement in single-threaded. It takes a comfortable lead against the Core i9-9900K in multi-threaded as well, outpacing it by 8 percent. The Ryzen 9 3900X delivers what can only be described as a "whuppin," smashing the Core i9-9900K by 1.58x.

The popular Wii emulator Dolpin.emu contains a CPU benchmark to test the JIT compiler. Dolphin has always preferred Intel CPUs, but Zen 2 closes the gap with a 23 percent improvement on the Ryzen 7 2700X. The Core i7-9700K scores a surprising win, here — sometimes, not having Hyper-Threading affords a small advantage.

Historically, Intel has dominated 7zip's compression test, while AMD won the decompression benchmark. The 3700X is 1.4x faster than the 2700X in compression and 1.2x faster at decompression. It fully matches the 9900K in the former, while surpassing it by 10 percent in the latter. Then the Ryzen 9 3900X arrives and obliterates the playing field, with scores 1.34x and 1.68x higher than the Core i9-9900K.

In Handbrake, the 3700X takes 1.14x less time to render H.264 than the 2700X and 1.27x less time to render in H.265. It's about 1.08x faster than the 9900K in both cases. The Ryzen 9 3900X is nearly a third faster than the Core i9-9900K in both H.264 and H.265.

We compiled Qt using Microsoft Visual Studio 2019 Community Edition. The Core i7-8086K, 2700X, and 9700K were closely grouped together, with the Core i7-8086K leading of the three. The 3700X, 9900K, and Ryzen 9 3900X were at the fore, with the Ryzen 9 3900X compiling in 60 percent the time of the 9900K. The Ryzen 7 3700X compiles in 70 percent the time of the 2700X.

Corona Render has always been Intel-friendly and the 9900K steals a rare win off the 3700X here, by a narrow 2 percent. The beleaguered Core i7-9700K can't seem to catch a break — again, we're looking into it.

Maxwell Render has historically been an Intel-friendly rendering test, but that's a relative metric today. The 3700X is 1.24x faster than the 2700X, 1.13x faster than the Core i9-9900K, and the Ryzen 9 3900X brings a 1.63x advantage down on top of Intel like the fist of an angry god.

Our examination of storage performance is perforce limited, but CrystalDiskMark's read and write tests give a quick idea of what PCIe 4.0 storage is capable of. We've used a 970 EVO drive on the Intel systems and the same Phison controller on AMD systems in PCIe 3.0 to show that the Samsung solution keeps up on that interface — but opening up the throttle still pays major dividends.

In this case, there is a sequential write difference between the 970 EVO and the Phison, even in PCIe 3.0 mode, but it's not nearly as large as the boost from enabling PCIe 4.0.

Non-Gaming Performance

The non-gaming performance of the Ryzen 7 3000 family is, in a word, excellent. The Core i9-9900K has been dethroned in virtually every case by a CPU that costs 65 percent as much. If you’re willing to drop $500, the Core i9-3900X doesn’t beat the Core i9-9900K so much as it nukes it from orbit. The price/performance advantage AMD now offers for $500 is the same size as what Threadripper has offered over Xeon — but for a fraction of the price.

Now that we’ve covered non-gaming performance, let’s take a look at gaming. We tested eight titles in 1080p, 1440p, and 4K, using high-end visual settings that match the options someone with a top-end GPU is likely to pick. Results are in the slideshow below. Each slide can be clicked to open it in a new window if desired.

Gaming Benchmarks

In Ashes, the 9700K takes pole position, with the other Intel CPUs not far behind. The 3700X is 1.08x faster in 1080p than the 2700X. The gaps narrow as resolution rises, but the 9700K wins overall, with the AMD chips still a bit behind Intel at 4K.

DXMD shows the not-uncommon scenario of complete gridlock in a GPU-bound game. This is an effective tie across the board.

Hitman is a game where the 3700X picks up a solid 1.16x over the 2700X at 1080p. Intel retains a technical advantage (and the 9700K notches up another win), but the gap begins to collapse at 1440p and is gone by 4K. Gamers seeking to exploit every single frame may still prefer Intel in games like this, but the distance between them is fast disappearing.

By our 4K setting, Metro is being rendered at 8K internally, which is why it turns into a 30fps title. Intel's narrow 3 percent lead at 1080p is within a technical margin of error. Past that, the game is completely GPU-bound.

Shadow of War is another game where the Ryzen 7 3700X picks up a few frames on the 2700X, but doesn't quite close the gap with Intel. Just keep in mind, the difference between 130 and 139 fps is literally invisible to the naked eye.

In RotTR, AMD picks up a solid 1.08x over the 2700X, with the Core i9-9900K less than 4 percent ahead. The Core i7-8086K and the Core i9-9900K can't quite open a 5 percent lead. At 1440p the gap between the players is negligible; at 4K it's nonexistent.

The Ryzen 7 3700X delivers an astonishing 1.2x improvement in Warhammer II, effectively matching both the 9700K and 8086K. It keeps pace with both Intel CPUs at 1440p before the tyranny of 4K lays all silicon low.

SotTR just isn't particularly CPU-bound, even at 1080p. The gap between the 9700K and the Ryzen 7 3700X is just 2.5 percent. At higher resolutions, the game is completely GPU-bound.

Gaming Benchmark Conclusion:

Intel still has an advantage at 1080p and the 9700K is a surprisingly strong player in gaming, but there’s no denying that third-generation Ryzen CPUs cut the gap between themselves and Chipzilla. The geometric mean of our 1080p test results for the 9700K is 115fps, compared with 109fps for Ryzen 7 3700X and 101fps for the Ryzen 7 2700X. At this point, Intel’s advantage in 1080p is down to 6 percent. A margin of error of 3-5 percent is not considered unusual in benchmark testing, and while we’re not stating that Intel’s performance advantage is simply due to margin of error — it’s too consistent for that — there’s just not a lot of “there”, there.

In this reviewer’s opinion, Ryzen’s “weakness” in 1080p gaming has always been a bit overblown. With Zen 2 in market, it’s gone from a minor point to a non-issue.

Power Consumption

Finally, we turn our attention to power consumption. As many of you are aware, the behavior of Intel CPUs can vary a great deal depending on how a motherboard manufacturer has programmed their boards and what settings the motherboard has implemented. In our case, the two Intel motherboards we tested appear to implement the chip manufacturer’s intended thermal and current limits — but this has a definite impact on how our Intel CPUs behave under load.

After a relatively short period of time (8-20 seconds, typically), the Core i9-9900K, 9700K, and 8086K will all yank back hard on the metaphorical throttle. The Ryzen 7 2700X, 3700X, and Ryzen 9 3900X do not behave in this fashion. Where the 9900K throttles back before its even halfway through a Cinebench R20 multi-core render, the Ryzen CPUs maintain full clock and power draw the whole way through.

Prime95 29.4b8 power consumption. The Ryzen 7 3000 CPUs idle very high — much higher than their Intel equivalents, and higher than the Ryzen 7 2700X. This could improve with later UEFI revisions or might be the result of the relatively high-power PCIe 4.0-equipped chipset.

In Prime95 29.4b8, the Intel CPUs burst to between 190W – 205W for relatively short periods of time before settling back to lower wattages. This is somewhat different than their behavior in Prime95 29.8b5.

In Prime25.8b5, Intel CPUs will hit higher wattages and then throttle back more quickly. This behavior on our Core i9-9900K was consistent over reboots and multiple tests. When throttled back to a 140W target, the Core i9-9900K draws less power than any other CPU we tested. So we decided to toss a more realistic workload into the mix.

In Cinebench R20 we still see the CPUs burst and then pull back, but they do so less aggressively and turbo clocks are held for a longer period of time. None of the Intel CPUs we tested, however, would hold its turbo clock for the entire length of time it took to finish the CB20 render in multi-threaded mode. This behavior is adjustable in UEFI — we simply used the defaults set by Asus for its motherboards.

AMD’s CPUs do not behave in this fashion, but the fact that they hit higher wattages doesn’t mean they use more total energy. The Ryzen 7 3700X and Ryzen 9 3900X both complete the CB20 render more quickly than the Core i9-9900K, which cuts down on the total power consumption. The Ryzen 7 3700X is also a significant improvement on the 2700X, drawing approximately 80 percent as much wall power in all cases.

Conclusion: Ryzen Shines on 7nm

In the spring of 2005, AMD launched its dual-core Opteron and Athlon 64 X2 processor lines and kicked off what would later be known as its own golden age — an era when it strongly challenged Intel across desktops, servers, and workstations. In the 14 years between then and now, Advanced Micro Devices has never even come close to an equivalent moment.

Not until now.

When Intel launched the Core i9-9900K, we noted that both it and the 2700X were excellent products, but that AMD had a death grip on the performance-per-dollar category. That’s no longer the case. Dollar-for-dollar, the Core i9-9900K is annihilated in multi-threaded applications by the Ryzen 9 3900X and slightly exceeded in overall non-gaming performance by the $329 Ryzen 7 3700X.

Intel has no easy short-term answer here. There are rumors of a 10-core desktop on the horizon, but 14nm TDPs at high clock and core count are not friendly to anyone. AMD already has a 16-core desktop chip coming in September, and while it may not scale as well as its Threadripper equivalent due to limited memory bandwidth, it’ll scale well enough. Like the concept of a fleet in being, the existence of that 16-core CPU is tangible proof that AMD has gas in the tank and products it can introduce to drive performance higher.

Zen 2 is a tremendous victory for AMD. It may not be an absolute victory — Intel maintains a narrow lead in 1080p gaming, and the 9700K makes an argument for itself on those grounds, if you’re obsessed with squeezing out every last frame of performance — but if you were to look back at test results in 2005, you’d find that AMD didn’t win literally every single one of them back then, either.

We’d like to see if AMD can bring idle power on the Ryzen 7 family down at all, and the CPU’s relatively high single-thread power consumption is a little puzzling, but the Ryzen 7 3700X and 3900X’s execution efficiency is, in a word, excellent. AMD has more than delivered on the improvements that it promised.

We’ll have more to say on Ryzen 7, the X570 platform, Intel, and the overall performance of these new CPUs in the days ahead. Stay tuned.