Last August, I compared the 3700X to the 2700 (stock and overclocked) and came to the conclusion that upgrading to the 3700X from the 2700 didn’t make a ton of sense. The ~20% single thread performance improvement Ryzen 3000 brings is technologically impressive, but it’s hard to recommend a $330 upgrade for just 20% more performance (or less, especially in games). That isn’t to say that Ryzen 3000 CPUs and Zen 2 CPUs as a whole aren’t great; they are, but that particular upgrade wasn’t very compelling to me, and I instead suggested the 3600 because it provides similar multi core performance to the 2700, greater single threaded performance, similar gaming performance to the 3700X, and comes in at a much lower price.
I have decided to run this test again, however, for a few reasons. Firstly, I originally wanted to test the 1700 but was unable to due to time constraints. Secondly, I want to test some newer (and also less) games. Games like Dota 2, the Division 2, and Total War: Attila, though valid games to test, don’t exactly impart useful data; Dota 2 can run on any CPU, the Division 2 is only worth testing on the in game benchmark, and Total War: Attila barely runs on any CPU. And finally, the original 8 core analysis is actually our most viewed original article and most viewed review, so an update would be very beneficial. This time around, I’ve also decided not to test at stock but overclocked in regards to both the CPU cores and the DRAM.
CPU Specifications
Here are the specifications of each 8 core Ryzen CPU I’ve tested.
Specifications | 1700 | 2700 | 3700X |
Cores/threads | 8/16 | 8/16 | 8/16 |
Base/boost clock | 3.0/3.7 GHz | 3.2/4.1 GHz | 3.6/4.4 GHz |
Process | 14nm GF | 12nm GF | 7nm TSMC (cores), 12nm GF (IO) |
L2+L3 cache | 4+16 MB | 4+16 MB | 4+32 MB |
PCIe support | 3.0 | 3.0 | 4.0 |
TDP | 65 watts | 65 watts | 65 watts |
Launch date | 3/2/17 | 4/19/2018 | 7/7/19 |
MSRP | $329 | $299 | $329 |
The main difference between the 3700X and its predecessors is cache size and frequency. The 3700X has double the L3 cache of both the 1700 and 2700, and the 3700X is 600-700 MHz and 300-400 MHz faster than the 1700 and the 2700 respectively. The 3700X is obviously very different in other ways, but for the purposes of this benchmark, it is the cache size and the frequency that matters the most.
Additionally, Zen 2 has architectural changes (such as the L3 cache increase) that makes it, on average, 15% faster than the original Zen at the same clock speed. This figure is called instructions per clock or IPC. The IPC uplift is usually more important for applications that really take advantage of the full capabilities of a core. Games don’t tend to take advantage of all CPU functions, and this is why the difference between Zen 2 and previous Zen CPUs shrinks in gaming benchmarks. But the L3 cache does sometimes make a big difference; there’s a reason AMD calls it “game cache.”
Another important thing to note is price. This is exactly the reason why it was so hard to recommend the 3700X to 2700 owners. The 3700X came out a little over a year after the 2700 and costs a little bit more, so it’s hard to recommend to someone that they should pay that kind of money just a year later for something just slightly faster. The 1700 is different, however: it’s older and, theoretically, slower than the 2700. For 1700 owners, it could be time to upgrade.
Test Methodology
The following parts were used:
Test Bench Configuration | Part Name |
CPU Cooler | Noctua U12A |
GPU | EVGA GTX 1080Ti FTW3 (~2 GHz core, ~6.4 GHz VRAM) |
Motherboard | ASRock X570 Extreme4 |
RAM | G.SKILL Flare X 3200 MHz CL14 2x8GB |
OS Storage | Samsung 850 EVO 250GB |
PSU | Seasonic Focus 650 Watt Gold |
The U12A was provided by Noctua and the motherboard was provided by ASRock.
For CPU gaming benchmarks, I determine what settings to use when I test the CPU I expect to be the fastest overall (the 3700X in this case). I alter the settings until I achieve 120-144 average FPS in non-competitive titles and around 240 in competitive titles. In other words, the fastest CPU will almost always perform within the 120-144 FPS range in titles like the Witcher 3, or around 240 FPS in titles like Rainbow Six Siege. I test like this because such framerates are both realistic and able to expose CPU bottlenecks. While testing at the lowest possible settings does reveal CPU bottlenecks more consistently, that kind of testing is prone to revealing CPU bottlenecks that most users never encounter; how many reviews have you seen where one CPU was averaging 300 FPS and another 500?
Additionally, I test using actual gameplay as often as possible. I only use built in benchmarks if it accurately reflects in game performance. I will note when I have used a built in benchmark; if I do not specify assume I have used actual gameplay to measure performance. Though testing performance in game is realistic, it is also prone to being quite variable or even inconsistent even within the same session. To offset this, I run each game at least 3 times and then take the median result.
I used OCAT to record frametimes which I converted into framerates. I also used MSI Afterburner to overclock the GPU and to provide additional statistics as I was benchmarking. There were some background applications while running benchmarks: task manager, file explorer, and game launchers (though only one is ever open at a time).
On the subject of frametimes, we use the 99th percentile (provided by OCAT) instead of 1% lows for our minimum FPS calculation. The 99th percentile measures the lowest possible value for a framerate to be considered within the top 99% of framerates, where as a 1% low averages the bottom 1% framerates. Calculating using 1% lows is a major mistake because this range of data is often highly inconsistent and unnoticed in gameplay. One random lag spike would factor into the 1% low, for example, but not into the 99th percentile.
Every possible piece of software (OS, drivers, games, etc) was updated when testing began on March 6. Some required game updates occurred over the testing period, but these did not impact performance in any noticeable way.
Overclocking
Let’s start with the 3700X. With Ryzen 3000, AMD basically made obsolete the old method of changing the multiplier and manually adjusting voltage. Out of the box, most Zen 2 based Ryzen CPUs will boost to the maximum it possibly can. However, AMD does allow you to adjust how aggressively the CPU will boost; this is called Precision Boost Overdrive or auto overclocking. I removed all power and current limits and set the 3700X to boost 200 MHz higher than it normally does. Now, it never actually went above its rated 4.4 GHz boost, but auto overclocking might have helped it boost higher under load or sustain its boost for longer. The RAM was overclocked to 3600 MHz, which is the most optimal speed for Zen 2 based systems to have. However, tightening timings were almost impossible with such a high frequency, and so they remained at what the XMP profile had by default.
My 2700 was able to achieve a 4 GHz overclock at 1.35 volts. It’s not the greatest overclock, but it is running on a pretty safe voltage that ensures there won’t be any silicon degradation, and it is also quite easy to cool with the U12A. As for the RAM, it didn’t clock nearly as high as it could with the 3700X, becoming stable only at 3333 MHz. I did have some wiggle room for tightening timings, however; TRDCRD, TRCDWR, and TRP were all brought down by 1 to 13 and TRAS was brought down by 6 to 28.
My 1700 could do 3.85 GHz at 1.4 volts, which is not particularly great even for a CPU on the 14nm GlobalFoundries node, though this is not unexpected given that was bought the very day it launched. The memory wasn’t great at overclocking either, it could only maintain the stock 3200 MHz speed and the timings were tightened to the same degree seen on the 2700, with the exception of the TRAS which could only be reduced by 2 down to 32.
I believe the X570 motherboard I am testing on is partially responsible for the poor RAM overclocking on the 2700 and 1700; on a platform more focused on first and second gen Ryzen (something with the X370 or X470 chipset), it’s possible that I would have been able to achieve better results. But the core overclock is consistent with results I’ve had on other boards. Even if I had more wiggle room for the memory, I don’t believe it would have made too much of a difference.
Applications and Synthetic Loads
This section includes the Time Spy Extreme Physics test, the PCMark 10 Express benchmark, the Cinebench R20 benchmark, and the Blender BMW benchmark. These workloads are generally reflective of real world applications even if they are sometimes considered to be synthetic workloads. We should get a good idea of how these CPUs perform when all or most functions of a core (or all cores) are being utilized.
Time Spy Extreme Physics
PCMark 10 Express
I would have tested the normal or extended version of PCMark 10, but an Nvidia driver issue causes the benchmark to crash on a specific part of the test. The Express benchmark still suffices, however.
Cinebench R20
Blender BMW
Application Performance Summary
Combined, here’s what our performance difference looks like among these CPUs. It is a convincing lead for the 3700X, especially in Cinbench R20 single thread where the 3700X was 30% faster than the 1700. However, the 1700 and 2700 aren’t too far behind.
Games
This time, I’m only testing these CPUs in 15 games rather than 20, but that’s not actually a bad thing. Besides making my job easier, cutting out 5 specific games should make the results much more meaningful. My reasons for dropping some titles are generally these: it runs well on every CPU, it runs poorly on every CPU, it’s old and not popular anymore, it’s a pain to benchmark in game, and/or a sequel was made and it’s just better for testing. None of these titles should be controversial choices, I hope.
Apex Legends
I might stop benchmarking Apex Legends immediately since it’s so well optimized it ran perfectly on each CPU. All of the Ryzens hit the 144 cap and had 99th percentile minimums within the margin of error, effectively making this a tie. If this title didn’t have the 144 FPS cap, it would have been much more interesting for testing.
Battlefield V
Civilization VI: Gathering Storm
Civilization VI: Gathering Storm was tested with the built in AI benchmark, which tests the average turn time. In games like Civilization, it’s much more important to test the turn time rather than the FPS.
Control
Counter-Strike: Global Offensive
Far Cry 5
Forza Horizon 4
Forza Horizon 4 was tested with the built in benchmark and I also took the results from the game rather than OCAT since Forza Horizon 4’s built in benchmark measures the performance of the GPU and CPU separately. The actual average framerate is usually different from these results (about ~125 FPS in across every CPU), but it’s interesting to note the CPU render and CPU simulation performance since these are derived totally independently of the GPU, thereby ignoring any sort of GPU bottleneck.
Gears of War 5
Gears of War 5 was tested with the built in benchmark for two reasons: testing within actual gameplay was difficult and preliminary results from testing in game correlated closely with the results from the built in benchmark. So, I believe using the built in benchmark in Gears of War 5 is alright for testing.
GTA V
Hitman 2
Rainbow Six Siege
Shadow of the Tomb Raider
Total War: Three Kingdoms
Witcher 3
Wolfenstein Youngblood
Gaming Performance Summary
As expected, the margin between these CPUs is much smaller in games than in applications. The 2700 is actually really close to the performance of the 3700X here, at almost 90% of the performance. The 1700 is a little behind, however. By the way, this chart doesn’t include the Civilization VI test because it measures solely in seconds, not FPS. The 2700 and 1700 respectively were 89% and 85% as fast as the 3700X.
Conclusion
Let’s start with applications. Overall, the 2700 was about 83% as fast as the 3700X and the 1700 was 79% as fast. I was actually surprised how well the 1700 kept up given its deficit in clock speed and memory speed, but it really wasn’t bad at all. In some cases the 3700X was quite a bit ahead of the 1700, but when we look at the average it’s pretty clear that the 3700X isn’t quite in a league of its own. That being said, I’m sure there are going to be a few applications where Zen 2 does far better than Zen as well as Zen+. On the whole, however, it seems like the 1700 is still just fine.
Games are a similar story, but there are a few exceptions. Both the 2700 and 1700 gain ground in gaming and prove to be quite capable for the task. Titles both old and new ran well on the older Ryzen CPUs, though it should be noted that the 2700 and 1700 couldn’t always hit our framerate targets when the 3700X almost always did (and when the 3700X didn’t, it was usually a GPU bottleneck). Some games did not run so well on the second and first generation CPUs, however: CSGO, Shadow of the Tomb Raider, Forza Horizon 4, and Far Cry 5 saw these CPUs perform worse against the 3700X than on average.
There might be a case for 1700 owners to upgrade to the 3700X, if they’ve been on the original Zen CPU for a while and they play some of these games where the newer Zen 2 CPU just performs better. Ryzen 3000 does bring more than just better performance, too: higher power efficiency, PCIe 4.0, longer support, better boost technology, etc.
The case for 2700 owners is still weak, though, because the 2000 series isn’t even 2 years old yet and it still receives similar support to the 3000 series. If gaming performance is the primary concern, then surely upgrading to the 3600 or 3600X is the better choice, and I’ll be evaluating that soon. It might sound crazy to give up 2 cores, but if gaming is your main interest, then the sacrifice is probably worth it. It’s not like 6 core CPUs are about to become obsolete, right? Though, that is what they said about 4 core CPUs in 2017, so maybe I’m wrong on this one.