When Ryzen 3000 launched in early July, we found it to be quite a decent improvement over the previous 2000 series and first generation 1000 series thanks to its much higher IPC and slightly higher clock speeds as well as offering 12 (soon to be 16) cores on the same AM4 socket AMD has been using since 2017. Despite all the motherboard compatibility confusion and launch bugs, Ryzen 3000 looks to be a very compelling upgrade option.
But who exactly should upgrade to Ryzen 3000? 2700(X) and 1700(X)/1800X owners who just want that extra single threaded performance might end up paying as much as or more than they did a year or two ago for an 8 core CPU today. Does it really make sense to buy another 8 core CPU just for the single threaded improvements? Or should these users just keep what they have already?
Today we’re going to find out whether or not that IPC and clock speed gain is really worth it for first and second generation Ryzen 8 core owners. I’ve tested 19 games and some synthetic benchmarks to highlight the differences between the 2700 and 3700X.
8 CORE RYZEN CPUS OF THE PAST
AMD has launched a new Ryzen series about once every year, and here you can see what has changed between each 8 core Ryzen CPU across almost 3 years.
7nm TSMC (cores), 12nm GF (IO)
Overall, they differ in many ways but not every way. These are all low power, 65 watt SKUs (despite the 3700X having that X) around the $300 price point. Despite having the same TDP, every single successive 8 core has added 200-400 MHz to its clock speed thanks to newer processes. The 3700X represents the largest shift with a doubling of L3 cache from 16 to 32 MB, the addition of PCIe 4.0 support on compatible platforms, and a return to the $329 price point that was set for AMD’s cheapest 8 core in 2017.
On paper, some of these improvements on the 3700X are quite impressive: doubling of L3 cache, doubling of PCIe bandwidth (provided you can utilize 4.0), and much larger clock speed gains than the 2700 brought, all in the same 65 watt TDP. There’s also the things that I didn’t list on that table like the doubled floating point performance (important for productivity). Again, though, just 8 cores for the same price as 2 and a half years ago.
Though Ryzen 2000 brought CPU prices down all across the board, Ryzen 3000 represents a return to the pricing structure of the original Ryzen 1000 CPUs, though instead of an 8 core at $500 we have a 12 core. Normally we would expect the next 65 watt 8 core from AMD to be cheaper than their previous, but this is not the case with the 3700X. We’ll have to see whether or not it’s actually worth it.
The 3700X was provided to our CPU reviewer, Alex, who loaned it to me after he finished his Ryzen 3000 review. Thanks Alex, and AMD who provided the 3700X to him in the first place.
The following parts were used:
Test Bench Configuration
CPU thermal interface
IC Graphite Thermal Pad (larger size)
EVGA GTX 1080Ti FTW3
ASRock X470 Taichi
G.SKILL Flare X 3200 MHz CL14 2x8GB
Samsung 850 EVO 250GB
Seasonic Focus 650 Watt Gold
I’d like to thank Noctua for providing the U12A for this test bench, and Jim for sending me his spare 1080Ti.
While I wanted to test the 1700, 2700, and 3700X, I only had enough time to get testing done on the 2700 and 3700X. I did, however, test the 2700 at both stock and overclocked (core and RAM) to represent the benefit of overclocking on the 65 watt unlocked SKUs. I did not overclock the 3700X for reasons I will explain later. I may test the 1700 later (I already have some overclocking data collected) but in the interest of getting this thing out the door we’re just going to have the 2700 vs. the 3700X.
This test bench, despite using a last generation GPU, should be more than enough to reveal the performance differences between our CPUs. Remember, the 2080 Super is only about a few percent faster than the 1080Ti and the 2080Ti isn’t even 30% faster than the 2080 Super in most games. If you’re concerned that we’re GPU bottlenecked here, don’t be.
Throughout my testing for the 3700X, I targeted 100-144 FPS for most titles and up to 240 FPS for our competitive titles like Rainbow 6 Siege and Dota 2. Now, I could have cranked all the settings to the lowest possible and set the resolution to 720p in order to reveal the most extreme differences possible, but realistically most people aren’t looking for more than 144 FPS in typical games and more than 240 FPS in competitive titles.
I’m aiming for a more practical than scientific angle in this analysis. It would be interesting to know that the 3700X does 400 FPS in one title while the 2700 is only doing 300, but that tells us literally nothing on whether or not either of them are going to be different in a typical gaming situation. If they both do 144 FPS in a game, even though the 3700X could widen the gap at more CPU bound settings, then it’s practically a tie. Or if the 3700X does 144 FPS and the 2700 only does 120 FPS then that’s a notable, albeit small, performance difference. I believe this kind of testing methodology is very well suited to showing what you may or may not give up by switching to a theoretically slower CPU.
I tested 19 games, both new and old, some synthetic stuff like 3DMark, and Blender. This should give us a good idea on what AMD has improved on and what they haven’t, and it’ll also tell us where games are heading when it comes to improving performance (single thread vs. multi thread). But first, overclocking.
Ryzen’s biggest strength, expecially for its 65 watt SKUs, is overclocking, which is available for all Ryzen CPUs and on every single chipset but A320. Not only can you overclock the cores, but also the RAM. I tried overclocking all of my CPUs and the Flare X I paired them with just to see how much performance you could squeeze out of these CPUs.
I decided to stick to 1.35 volts on both CPUs since anything more can overwhelm even a really good heatsink as well as potentially damage your CPU. I was able to get 4.05 GHz on the 2700, which is only on par with the 2700X’s all core turbo and short of the 2700X’s 4.35 GHz single core boost. On this particular 2700, going beyond 4.1 GHz was a challenge, and even 4.1 GHz required almost 1.45 volts. Still, that’s almost a 20% overclock considering the 2700 typically boosts to around 3.4 GHz
The 3700X, in this regard, severely disappointed with its measly 4 GHz overclock, lower than the 2700 managed. Not only that but it’s also lower than the all core turbo on rendering applications like Blender (the 3700X will run around 4.2 GHz there) and only was on par with the stock 3700X in Prime 95 which is an AVX workload. It’s even worse when you compare games which, thanks to their lighter core usage, saw clock speeds between 4.3 and 4.35 GHz.
To say the least, overclocking the 3700X is pointless at best. I just decided to leave it at stock, where it was able to hit 4.4 GHz at very brief moments in lightly threaded tasks, 4.3 to 4.35 GHz in gaming, 4.2 GHz in Blender, and 4 GHz in Prime 95. I have heard complaints that Ryzen 3000 never hits its rated boost clock, but in my experience it does so frequently, provided you have updated Windows 10 which is required in order to hit these turbos. I’d imagine these complaints also come from either a lack of understanding on what AMD’s boost is or issues with firmware where the CPU is actually not boosting correctly (which is definitely possible).
Onto RAM overclocking. The RAM I used was Samsung B die equipped Flare X 3200 MHz CL14, which is basically the best thing you can get for Ryzen at the moment. The 3700X struggled to overclock the RAM much higher than it came in stock, though I suspect for lack of good firmware on the X470 Taichi since this 3700X was able to achieve good RAM overclocking on Alex’s (our CPU reviewer) system. The 2700, on the other hand, did quite well with a modest overclock and latency tightening.
A quick disclaimer, though, a 20% overclock will not always translate into 20% more performance. You will see that the overclocked 2700 is usually only slightly ahead of or just ties the stock 2700 in gaming even though there is a clear performance difference in synthetic and rendering tasks. This is just simply down to Ryzen having bottlenecks other than clock speed. I did have to sanity check myself on this just to make sure I wasn’t wrong, but it does seem like overclocking does not get you much more performance in gaming with the Ryzen 2700, at least not in the way I tested.
To determine thermal performance, I put each CPU under a Prime 95 torture test which should tell us the worst case scenario for these CPUs, especially since I used a carbon pad and not thermal paste which should lower temperatures by a few degrees.
While it may look like the 3700X is quite competitive on thermal performance, you need to consider the fact that it is a 65 watt SKU like the 2700. Even though they’re consuming about the same amount of power, the 2700 is far cooler and only becomes comparable when overclocked. This is almost certainly due to Ryzen 3000’s unique die configuration: the IO die on one side and the CPU chiplet on the other, stowed away in a corner. This is highly undesirable for high thermal performance since not having the CPU chiplet in the center causes heat to spread unevenly and inefficiently. Though AMD’s TDP definition is entirely based on thermal performance, it seems like the massive disparity here is not enough to classify the 3700X as a 95 or 105 watt SKU. Don’t get me wrong, the 3700X is very power efficient, but when AMD focuses on the ‘T’ in “TDP” it seems very odd to put these CPUs in the same class.
For my rendering benchmarks, I just went with Cinebench Release 20 and Blender’s official benchmark.
CINEBENCH RELEASE 20
The 3700X gets to stretch its legs in Cinebench, whether its using all or just one of its cores. It is nearly 20% ahead of the overclocked 2700 and almost 40% ahead of the stock 2700 in both multi and single core results.
BLENDER OFFICIAL BENCHMARK
Again the 3700X absolutely thrashes the 2700, with it completing the benchmark in 86% of the time the overclocked 2700 needed to finish. Compared to the stock 2700, it finished the benchmark in just 73% of the time.
I have three synthetic benchmarks here: Ashes of the Singularity Classic (yes I know it’s a game but it’s basically a synthetic benchmark now), 3DMark Time Spy Extreme, and PCMark 10.
ASHES OF THE SINGULARITY CLASSIC
The 3700X is strangely potent in Ashes, being almost 30% faster than the overclocked 2700. Considering the 2700 didn’t benefit much from 20% higher clock speeds except on the minimums, I think the difference must be down to something the 3700X does much better than its predecessor, something that Ashes really utilizes. Perhaps its the improved cache or floating point performance.
3DMARK TIME SPY EXTREME
We see more of the same in Time Spy Extreme where the 3700X just keeps hammering the last generation 2700, even when it’s overclocked.
Finally, we have PCMark 10 and it’s nothing unusual, just the 3700X continuing its beat down on last year’s cheapest 8 core CPU.
Overall, the 3700X’s performance over the 2700 is quite impressive in these few benchmarks but keep in mind synthetic benchmarks are designed to scale well with clock speed and IPC. Rendering is also something Ryzen just does really well, even on CPUs of the same core count; it’s not what I would describe as typical performance. We’ll see what I mean in the gaming benchmarks.
I wish I could have tested just one more game to get to a nice, round 20 but my time and especially patience to test more games was running out (for Metro Exodus in particular) during the week I had the 3700X so I decided this was it. 16 of these titles are your typical AAA games that you would want to run at up to 144 FPS and at the end I put three e-sports and competitive titles that I targeted up to 240 FPS on.
Most of these titles were released within a year or two ago but some of these titles are considerably older like the Witcher 3 and Total War: Attila. It’s just nice to know if Ryzen 3000 has improved performance on titles that were released before Ryzen even existed.
All of these titles were tested at 1080p and in game settings were turned up or down to reach my framerate targets. Some of these games, however, do not have quality presets, and in those titles I just either maxed out the settings or put everything on low; so, the 3700X isn’t always hitting 144 FPS in every single one of these titles, but wherever possible I was making sure it could get well above 60 FPS.
ASSASSIN’S CREED ODYSSEY
Having just seen the productivity and synthetic benchmarks, you may be surprised to see that the 3700X is not really that far ahead and even loses to the overclocked 2700 on 1% lows. Unfortunately, Zen 2’s amazing performance in those tests doesn’t really translate well into most games, and lack of software and firmware optimization makes gaming a little painful at times for Ryzen 3000, especially in minimum FPS performance, though I expect this to improve in the future.
I’ll be honest, I’m not entirely sure whether or not BFV is a competitive title, but I stuck it with the normal games anyways since the 3700X could barely get over 144 FPS on the low preset. I doubt you’ll be playing this game at 240 FPS at 1080p, at least not with Ryzen and/or a 1080Ti.
There’s little difference between the 3700X and 2700 on high settings, with the gap measuring only around 10%. There is also some run-to-run performance variation here; I only ran each benchmark once (again, time constraints with the 3700X) and sometimes the overclocked 2700 falls behind the stock 2700. Since each real gameplay benchmark isn’t always quite the same, you will see results like this in games where overclocking the 2700 didn’t really do anything and the performance just happens to be lower. If I had more time to run these games multiple times per CPU, we would likely see the results we expected, with the overclocked 2700 on top of the stock 2700.
On low settings, the gap is a little wider between the 3700X and overclocked 2700 but it’s still not a massive gap at 15% higher FPS, and the overclocked 2700 was able to hit my target of 144 FPS.
CIVILIZATION VI: GATHERING STORM
The 3700X does manage to get away from the 2700 in Civ VI’s Gathering Storm AI test, and accomplishes the test in just 86% of the time it took the overclocked 2700 to finish. Given you might have a dozen or more factions in a game of Civ VI and considering turns take longer to process as the game progresses, the 3700X’s improved performance here is very important.
On the other hand, when testing graphics the 3700X is pretty much tied with the overclocked 2700, although it does manage a significant lead over the stock 2700. But considering Civ VI doesn’t really benefit from an FPS higher than what you consider playable (30 or 60 FPS), this lead isn’t all that great. An interesting thing to note here is that Civ VI is one of very few games that bottlenecks on the stock 2700’s low clock speed, as overclocking the 2700 yields a large performance increase.
FAR CRY 5
While the 3700X doesn’t offer much on average framerates, we do see a rather large boost in the minimums in Far Cry 5. The 3700X is clearly a much more consistent CPU, but I would have liked to see equally larger average FPS gains too, especially since 60 FPS is a totally fine result for 1% minimums.
FORZA HORIZON 4
Similar to Far Cry 5, the 3700X isn’t that much faster on average but takes a significant lead on minimums.
GRAND THEFT AUTO V
GTA V was one of the titles AMD was bragging about when it came to performance because they claim it runs as well on Ryzen as the 9900K. Although GTA V is an old game on an old engine, it still is a pretty popular title and actually runs pretty well on most hardware.
At maximum graphics (except for all advanced graphics) there’s not a huge difference between the 3700X and the 2700, overclocked or not, though again we see a more of an improvement in the minimums rather than the average.
At minimum graphics, the 3700X does have a larger margin against the 2700 but it should be noted that when approaching 200 FPS in GTA V, performance is subject to really noticeable lag spikes that the 1% minimum framerate did not catch. It’s not stuttery as these spikes are only occasional but it’s still pretty annoying. I also believe there is effectively a cap on increasing FPS around 180-200 FPS, where FPS simply just doesn’t get higher on any hardware or settings. Thus aiming for more than 144 FPS is not recommended in GTA V; even though the 3700X is decently faster it’s actually not to its benefit.
The 3700X actually barely loses to the overclocked 2700 in Hitman 2, though it should be noted that it was only just barely. We’re also still CPU limited with the slightly faster overclocked 2700, so maybe there’s some room for optimization on the 3700X.
RESIDENT EVIL 2
Even though we’re hitting above 144 FPS at maximum settings in Resident Evil 2, we’re not seeing much of a CPU bottleneck here on any of the CPUs. The 3700X still wins, but not by very much, not even on the minimum FPS.
SHADOW OF THE TOMB RAIDER
In both the cave and city scenes in the beginning of the game, the 3700X is somewhat ahead of the 2700 which seems to struggle especially in the city, perhaps due to all the NPCs there. I wouldn’t say the 2700 was rendered unplayable though, even if overclocking didn’t really do much for performance.
TOM CLANCY’S THE DIVISION 2
Unfortunately I didn’t get the memo that you’re supposed to benchmark using the built in benchmark and not within the actual game itself in the Division 2 so here we are with these strange results. The Division 2 clearly has a great deal of performance variation depending on factors outside of the player’s control and this makes it a hard title to test with actual gameplay, though I think we can get a general idea of how these CPUs differ. The 3700X is clearly faster than the 2700.
TOTAL WAR: ATTILA
I know it’s been a very long time since anyone has ever seen this game in a benchmark suite, but I promise it’s here for very good reasons: it’s the most recent Total War title other than Three Kingdoms to use their historical game engine, it was pretty CPU bottlenecked when it came out, and it also happens to be my current favorite of the series.
As you can see, Attila has not gotten better with time or higher end CPUs. The 3700X is uncomfortably close to its predecessor and it’s pretty much a wash. It was at least decently playable but you definitely shouldn’t expect a nice performance boost in this game with a new CPU. While it is technically a CPU bottleneck, a more apt term for what’s happening here is an engine bottleneck; despite having more CPU resources available, Attila simply isn’t able to utilize them for whatever reason.
TOTAL WAR: THREE KINGDOMS
As I said for Attila, this is the first Total War game since then to use their historical engine, which is different from the engine that is used in their Warhammer games, hence Three Kingdoms does not have DX12 while the Warhammer games do. Attila is in this analysis primarily because Three Kingdoms is, so let’s see how things have improved since 2015.
Unlike Attila, Three Kingdoms was able to push over 100 FPS on the 3700X and 2700, so that’s nice. The 3700X does have a notable lead over the stock 2700 in this title but when overclocked the 2700 is just behind it, making it not much of a lead unless you don’t overclock.
THE WITCHER 3
In the Witcher 3, I tested in Novigrad, which is a massive city, and the 3700X does pull ahead slightly of the 2700 though it is not a very significant lead. At lower settings the 3700X does pull ahead even more but I decided not to test on anything lower considering the framerates everywhere else ended up being over 200 and the visual quality was really bad.
Testing in Velen produced a very odd result all around, which makes me wonder if the Witcher 3 engine struggles to utilize additional CPU resources at higher framerates. The 3700X and the 2700 are still in the same ballpark, however.
WOLFENSTEIN II: THE NEW COLOSSUS
Wolfenstein II is simply incredibly optimized. The 3700X managed 271 FPS on the second most intensive graphics preset and the 2700 still managed above 200 FPS. However, it should be noted that unless you’re using a 240 Hz display, these extra frames aren’t really going to be a big deal for most people. The lead the 3700X has sure looks impressive, but it’s unlikely you will be able to tell the difference between any of these CPUs. On the other hand, this might bode very well for the 3700X in future Vulkan titles.
Dota 2, our first competitive title, has two different APIs you can use, but considering Vulkan performs worse you should always pick DX11 if you’re playing on Windows. Regardless of API, however, the 3700X and 2700 aren’t too far from each other.
GEARS OF WAR 4
If Gears of War 4 isn’t a competitive title, then I apologize, but I’ve never played it and it performs very well so I’m pretty sure it is. Anyways, Gears 4 scales very well with faster CPUs and the 3700X was able to lead by a large margin. Since Gears 4 is a competitive title, this extra performance is welcome even if you’re not using a 240 Hz panel since the more frames you have, the lower the input lag is. I do regret not testing low settings on the 2700 since the 3700X did manage 240 FPS at that preset; I imagine the gap would be at least as significant as it is at high settings.
TOM CLANCY’S RAINBOW SIX SIEGE
Finally, we have Rainbox Six Siege, and this is where the 3700X shines, even better than in Gears 4. The 3700X has a massive 30% lead over the overclocked 2700 and in a competitive title those 30% extra frames are going to matter. It also landed right around 240 FPS which is perfect for a 240 Hz monitor. When looking at Gears 4 and Rainbow Six Siege (and considering results we’ve seen in games like CSGO which I could not test myself) the 3700X makes a ton of sense for competitive gamers.
So, with all that data, we can get an idea of how much faster the 3700X is than the 2700. In our synthetic and rendering benchmarks, we see a large difference of 20-30% between the 3700X and overclocked 2700 and around a 40% difference between the 3700X and the stock 2700. Despite having only improved clock speeds and IPC, the 3700X definitely has come out on top in these benchmarks.
On the other hand, in gaming, the 3700X was merely 10% faster than the overclocked 2700 and 15% faster than the stock 2700 on average. Though the 3700X performed much better than the 2700 in our competitive shooters and Wolfenstein II, on the whole it’s not that much faster. Increased clock speed and IPC did not impress most of the games I tested, sadly.
As a side note, based on the limited testing I’ve done with a 1700 overclocked to 3.85 GHz, it would slot in between the overclocked 2700 and stock 2700, and I imagine the stock 1700 would be a tad beneath the stock 2700. If there’s demand for 1700 data to be added to this analysis, I’ll gladly finish testing it and update.
So, with all that data, I can finally make a call on whether or not the new Ryzen 3000 CPUs are worth it for those who already have the slightly older 2700 and 2700X. I think overall, you’re wasting your money buying a 3700X since the performance improvements are rather marginal compared to our previous generation CPUs, whether it’s in gaming or rendering or synthetic benchmarks.
Don’t get me wrong, 10-30% is a decent uplift in performance, but for $330 when you might have paid $330 or $300 just a year or two ago for the same core count is just too much. If you mostly game then a 3600 makes more sense because it’s almost as fast as the 3700X for $130 less; losing two cores is a downgrade, but if you plan to upgrade frequently and don’t do anything very core intensive, it’s not a terrible idea. If your primary concern is productivity and rendering work, a 3900X is better because you can get 50% more cores for just $170 more.
The improvement in IPC and clock speed simply isn’t enough to make the 3700X a much better CPU than the previous 8 core Ryzens, especially in gaming. It’s just the case that you don’t really need a extremely high end, bleeding edge CPU to get good gaming performance, even at framerates that push high end monitors to their limit. So, we see even the stock 2700 not being that much slower than the 3700X despite the latter boasting 15% higher IPC and over 20% higher clock speeds depending on the load.
Of course, for competitive gaming, the 3700X is able to push much higher framerates, but it’s not like the 2700 couldn’t deliver a playable experience. If 240 FPS means a great deal to you, then the 3700X does the job really well in these competitive titles, but I imagine 180 FPS is enough for some or perhaps even most people. Maybe I’m wrong on this since I don’t play these sorts of games, however.
When it comes to normal AAA gaming, though, I don’t think most people are looking for framerates in the 200s. They’re wanting high quality graphics, lots of NPCs, and a good render distance, for example. Those last two are certainly CPU intensive, but a CPU that gets a higher framerate isn’t always a CPU that handles CPU intensive tasks particularly better than others, as is evident in Total War and GTA V. While a faster CPU is obviously nicer because it gives more consistent performance (you don’t want your 90 FPS at 1440p or 4K to transform into 60 or 50 because you walked into a CPU intensive area), it’s not clear that either a 3700X will prevent that from happening or that the 2700 cannot prevent that from happening.
If I had $330 to spend on upgrading my PC and I already had a 2700(X) and something like an RX 580 or a GTX 1060 or maybe an older GPU like a Fury X or 980, it would make much more sense to go ahead and just buy a faster GPU, such as the 5700 (for just $20 more) or the 1660Ti or something last gen like a Vega 56 or 2060. Even for those that already have a 1080Ti and could only afford to upgrade their 2700(X), keeping what they have is just a much better idea than getting something new, even if they have to have the best 1080p performance possible. Even for people who are using Ryzen primarily for rendering and productivity (which the 3700X does much better than the 2700), they’re still only getting about 30% better performance at best for $330. Instead of buying a new CPU, why not buy more solid state storage, a better boot SSD, more RAM, a better monitor, or whatever? 30% is significant, but is it $330 significant?
So, in conclusion: if you own a previous generation Ryzen 8 core, even if you don’t overclock, I don’t imagine the 3700X is going to blow you away for $330. If you mostly game, just settle for a 3600, losing those extra cores won’t lower your framerate at the moment, or just keep what you have (especially if you plan on not upgrading for a long time). If you’re more of a productivity and rendering kind of person, then either pay the $500 for a 3900X, the $750 for a 3950X, or just keep what you currently have. Ryzen 3000 is great, but it’s not a massive improvement for single threaded performance, a metric that is getting more and more marginalized even in gaming. From now on, when you upgrade, you should focus on getting more cores and not just faster cores.