The Nvidia GeForce RTX 3070 Founders Edition is here — or at least it will be on October 29, and we can provide a full review today. Following in the footsteps of the GeForce RTX 3080 and GeForce RTX 3090, this is the third of the new Nvidia Ampere architecture GPUs. In terms of performance, price, and features, these new GPUs rank at the top of our GPU benchmarks hierarchy and are some of the best graphics cards right now. Except, they're still almost impossible to find in stock. Will the RTX 3070 suffer the same fate at launch, and how does it stack up against the other current and previous-gen GPUs? That's what we're here to find out.
Fundamentally, the GeForce RTX 3070 will be similar in a lot of ways to the RTX 3080 and 3090 … except, not really, because changes to the core counts, memory capacity, and memory speed can all have a pretty significant impact on performance. Nvidia claimed during the Ampere reveal in September that the RTX 3070 was "faster than the RTX 2080 Ti," which would be an impressive result if it's true. It has more cores and higher TFLOPS ratings, but less memory and bandwidth. Here's the block diagram for GA104 along with the full spec sheet for the Ampere GPUs, along with the RTX 2070 / 2070 Super that the newcomer replaces.
|Graphics Card||RTX 3090||RTX 3080||RTX 3070||RTX 2070 Super||RTX 2070|
|Process Technology||Samsung 8N||Samsung 8N||Samsung 8N||TSMC 12FFN||TSMC 12FFN|
|Die size (mm^2)||628.4||628.4||392.5||545||445|
|SMs / CUs||82||68||46||40||36|
|Base Clock (MHz)||1395||1440||1500||1605||1410|
|Boost Clock (MHz)||1695||1710||1725||1770||1710|
|VRAM Speed (Gbps)||19.5||19||16||14||14|
|VRAM Bus Width||384||320||256||256||256|
|GFLOPS FP32 (Boost)||35581||29768||20314||9062||7880|
|TFLOPS FP16 (Tensor)||142 (285)||119 (238)||81 (163)||72||63|
The RTX 3070 is both a big step up from the previous-gen cards it replaces, as well as a relatively large step down from the RTX 3080. On paper, it has about 32 percent less compute than the RTX 3080, and 41 percent less bandwidth. For comparison, the 2070 had 25 percent less compute than the 2080 but had the same bandwidth. Compared to the RTX 2070, however, the 3070 should be a significant jump in performance. It has over 150 percent more FP32 compute and 67 percent more RT TFLOPS, though it still only has the same 448 GBps of bandwidth. Going into the review, we were very curious to see how the 3070 would stack up against the previous-gen cards in real-world benchmarks, and Nvidia's claims of beating the RTX 2080 Ti seemed unlikely.
It's also interesting to note the massive difference in power requirements compared to the RTX 3080. There's a 100W gap in TDP, and we know from our testing that the RTX 3080 and RTX 3090 can use every bit of their power budgets. This is the first Ampere GPU where we have a basically direct TDP comparison point with the previous-gen Turing GPUs. Both the RTX 2070 Super and RTX 2080 have 215W TDPs, so this will finally be Nvidia's chance to prove that, yes, Ampere actually can be significantly more power efficient than Turing.
Meet the GeForce RTX 3070 Founders Edition
The design and aesthetics of the GeForce RTX 3070 Founders Edition are similar in many ways to the RTX 3080 and 3090 Founders Editions — it's just nowhere near as large. You get the finned radiator look on the main body, with dual fans that include a built-in rim to improve static pressure, all in a dark gray color with silver highlights. There are several noteworthy differences between the 3070 FE and the 3080/3090 FE, however.
First, both fans are on the same side of the cooler (the side that normally faces down in a typical ATX case). There's also no 'X' silver cross on the side that faces up. Instead, a large grooved rectangular piece with a cutout at the back of the card allows the air to still flow through the radiator fins. Nvidia talked a lot about redesigning the 3080/3090 to handle the higher TDP. Obviously, the 3070 isn't in the same category of power and heat, and being a smaller card, it perhaps wasn't as beneficial. The card still runs cool and quiet, as we'll see later.
The other big change in terms of aesthetics is that there's no lighting at all, RGB or otherwise. The GeForce RTX logo on the top of the card is pretty tame compared to some graphics cards, though for some people, this would be a positive change. There's also only a small RTX 3070 logo on the side of the card, which is a bit weird to me as there's a ton of room for a larger logo. All of Nvidia's previous-gen 20-series Founders Edition models at least had a green GeForce RTX logo on the top edge, and the 3080 and 3090 have RGB logos (though the tool to configure the RGB lighting still hasn't been released). The RTX 3070 eschews all such nonessentials and just provides the basics as far as looks go.
There's not a lot to say about the rest of the card. It has the same triple DisplayPort 1.4a outputs and a single HDMI 2.1 port and can drive four monitors without trouble. The rest of the IO bracket is for the front fan's exhaust, but the bracket is silver instead of the matte black finish used on the 3080 and 3090 FE cards.
The RTX 3070 Founder Edition also uses Nvidia's new 12-pin power connector, but this time, instead of a y-combiner that takes dual 8-pin PEG connectors, it's just a single 8-pin port that's converted to a 12-pin port. Was this really necessary? I mean, I sort of like the extension cable because it's far easier to leave it connected and just detach the 8-pin connector. But then again, I swap GPUs pretty much daily, so my usage is not at all like a typical user. Also, the 12-pin connector is basically the same length as an 8-pin connector and is only a few millimeters narrower, but the orientation is such that putting on an 8-pin connector instead wouldn't have been a problem.
We haven't had time to do a full teardown of the 3070 Founders Edition yet, as we wanted to finish benchmarking before taking it apart. We'll update this section with additional images and commentary in the near future.
Nvidia GeForce RTX 3070 Overclocking
I thought I'd try something new with the RTX 3070 FE overclocking. Rather than manually tweaking settings to try to find something that appeared stable, I used the OC scanner functionality built into MSI Afterburner (4.6.3 beta 2). You can access this by pressing Ctrl+F. I increased the power limit to 109 percent, the maximum allowed, and increased the GDDR6 clocks by 500MHz (15 Gbps effective). Increasing the core voltage didn't appear to help at all, as the GPU appears to run into power limits, so I left that alone. I also tweaked the fan curves to be a bit more aggressive, so the fan will run at 100 percent if the temperature reaches 80C — but don't worry, it didn't come anywhere near that mark.
With those settings in place, the OC scanner generated the above voltage-frequency curve. What you'll immediately notice, hopefully, is that the gains at the far right of the voltage range are very limited, while gains at lower voltages are quite a bit higher. Basically, anything above 1075mV ends up with the same 'stable' result. Overall, the RTX 3070 was only about 3-5 percent faster when overclocked, and a lot of that is thanks to the power limit increase and higher memory clocks. This is similar to what I've seen on the other 30-series Ampere GPUs so far.
Overclocked results for the 3070 are present in all of the charts, but not in the bonus gaming tests. Other RTX 3070 cards will likely hit higher clocks than the Founders Edition, and this is a pretty conservative overclock. However, it's also an overclock that the card can probably maintain for many years without difficulty.
Nvidia GeForce RTX 3070 — 1080p Gaming Benchmarks
The GeForce RTX 3070 is in a bit of an interesting situation as far as resolutions and performance go. On the one hand, it's supposed to be as fast as the 2080 Ti, and we generally don't think people with that sort of GPU are going to play games at 1080p. At the same time, it's a $500 instead of a $1,200 card, and some people prefer higher refresh rate 1080p displays. What I'm getting at is that, even though this still isn't really a card built for 1080p gaming, we're going to go full monty and show all the benchmark results this time. Yeah, that includes RTX 3090 and 3080 at 1080p as well.
So, disclaimers out of the way, let's hit the benchmarks. Our conclusion will focus more on the 1440p and 4K performance than 1080p, and all of the bonus games will only be tested at 4K due to time constraints and intended usage. Still, if you're wondering how the 3070 and other cards handle 1080p, we've got you covered.
Our test PC remains the same as in recent GPU reviews. We'll probably be looking to change things up when Zen 3 launches, based on the early rumors, but the Core i9-9900K remains plenty fast for now. We did do CPU scaling tests on the RTX 3080 and found that the Core i9-10900K was a bit faster in a few instances, but nothing particularly noteworthy.
9 Game Average
We'll skip commentary on the individual games at 1080p, but the overall results are telling. At 1080p ultra, the 3070 FE comes in just behind the RTX 2080 Ti, but it's still 15 percent faster than the 2080 Super, 24 percent faster than the 2070 Super, and 38 percent faster than the 2070 FE that launched two years ago.
1080p medium narrows things down even more, and the RTX 3070 does squeak past the 2080 Ti, thanks to slightly better minimum fps. However, many of the games run into CPU bottlenecks, and the 3080 even comes out ahead of the 3090 at 1080p. Not that you should use either one for 1080p unless it's with ray tracing or maybe VR, or in a game where you could push 360 fps on a 360Hz monitor, maybe.
The Division 2
Far Cry 5
Final Fantasy XIV
Forza Horizon 4
Red Dead Redemption 2
Shadow of the Tomb Raider
Nvidia GeForce RTX 3070 — 1440p Gaming Benchmarks
Okay, let's move on to the more meaningful tests for the RTX 3070. While the card can certainly still handle 4K at playable framerates, like the RTX 2080 Ti, there are going to be plenty of situations where you'll have to tune and tweak the settings to get a smooth 60 fps. 1440p gaming, on the other hand, shouldn't pose any problem to the card.
9 Game Average
Nvidia says the 3070 is faster than the 2080 Ti, but that's only in select games. At 1440p ultra, it's more like a tie — there was only one game in our test suite where the 3070 won by more than a hair. Still, even matching the 2080 Ti is no mean feat, especially considering the price difference. Of course, now 2080 Ti cards will likely fall to much lower prices on places like eBay, but historically speaking, Nvidia's new architectures tend to improve with time, so the 3070 could end up being a bit faster than the 2080 Ti down the road.
That's interesting to see, considering the rather large memory bandwidth deficit. Overclocking doesn't make much of a difference either, so perhaps the 3070 is mostly held back by the GPU and not the GDDR6 memory. Regardless, this is a good showing for the Ampere architecture. It's as fast as the previous generation 260W TDP card while using 40W less power. And it costs $500, which means plenty of gamers can actually afford it, whereas the $1000-$1200 pricing of the 2080 Ti was out of reach for most people.
There's not much to say about the individual games. Borderlands 3 reaches very playable framerates of 85 fps at 1440p ultra and 153 fps at 1440p medium. Compared to the RTX 2070, it runs 50-65 percent faster — or 35-45 percent faster than the 2070 Super. Not that you'd really need to upgrade from either of those GPUs just yet. However, if you're sitting two generations back, the 3070 is now more than twice as fast as the 1070 and basically double the performance of the Vega 56.
The Division 2
The Division 2 puts the 2080 Ti ahead of the 3070 at all settings, but never by much. At 1440p, 2080 Ti is about 3-5 percent faster than 3070. Meanwhile, the 3070 is 30-35 percent ahead of the 2070 Super, 40-55 percent ahead of the 2070, and 110-135 percent ahead of the GTX 1070. Vega 56 definitely widened the gap relative to the 1070 over time, so the 3070 is 'only' 60-90 percent ahead of Vega 56.
Far Cry 5
Far Cry 5 is getting a bit old, and it definitely hits some CPU limits at 1080p. That starts to fade away at 1440p ultra, but 1440p medium showed relatively weak minimum fps on the overclocked 3070 for some reason. Anyway, we're looking forward to Far Cry 6 next year, so we won't spend any more time analyzing this 2.5 years old game.
Final Fantasy XIV
This is another older game that hits pretty high framerates, and it's one of the few instances where even the stock 3070 leads the 2080 Ti. By less than half of a percent, of course, so it's basically just a tie. This is a less demanding game, so the gaps between the various GPUs aren't quite as large here. We'll be switching to something else (Horizon Zero Dawn, probably) in the near future.
Forza Horizon 4
Forza Horizon 4 is the biggest lead we saw for the 3070 over the 2080 Ti, to the tune of up to four percent at 1440p. It's not much, but it's something. It's another game we're looking to retire from our GPU tests, mainly because of how much of a pain the Microsoft Store is to deal with. Project CARS 3 is the likely replacement, which we'll hit in our bonus games below.
We didn't enable ray tracing or DLSS (1.0) for Metro Exodus, but it's still a very demanding game — we'll check ray tracing and DLSS performance in the bonus round. The 3070 still runs very well at 1440p ultra and enabling RT and DLSS should still result in 60 fps gameplay, but the 2080 Ti ends up being 3-5 percent faster yet again. Elsewhere, the 3070 again leads the 2070 Super by around 30 percent, and is up to twice as fast as the GTX 1070.
Red Dead Redemption 2
Red Dead Redemption 2 runs into CPU bottlenecks with the fastest GPUs, even at 1440p medium, but pushing up to ultra settings spreads things out. While the 3070 leads at 1440p medium, the 2080 Ti ends up five percent faster at 1440p ultra. The 3070 also leads the 2070 by 50 percent, and is close to 2.5 times as fast as the GTX 1070 here. Note that we test using the Vulkan API, which tends to perform best for both AMD and Nvidia GPUs.
Shadow of the Tomb Raider
Like Metro, we've tested without RT or DLSS enabled for our primary benchmarks. We'll have RT + DLSS results below in the bonus gaming section. At the highest preset, the 3070 cranks out 115 fps, just a bit behind the 2080 Ti yet again. Do keep in mind that the 2080 Ti FE has a 90 MHz factory overclock, but then so did most other 2080 Ti cards — not many people were willing to pay that much money only to get a stock-clocked card, it seems.
Last, we have Strange Brigade, another game where we use the Vulkan API. It's pretty light compared to some of the other games in our suite, and the 3070 pushes well beyond the 144 fps mark that you'd want for a higher refresh rate 1440p display. Performance is again a bit behind the 2080 Ti, but well ahead of the 2070 Super and more than double the GTX 1070 and Vega 56.
Nvidia GeForce RTX 3070 — 4K Gaming Benchmarks
Based on what we've seen already, the 4K results are likely to favor the 2080 Ti just slightly, and the 3070 will need to run many games at medium to high settings to get 4K performance above 60 fps. That's basically the same as the 2080 Ti, with a much more palatable price tag. Whether you'll be able to buy one later this week is another matter.
9 Game Average
We'll just focus on the overall picture at 4K once again, as the individual results are below and mostly tell the same story. At 4K ultra, the 3070 averages 68 fps across our nine-game suite. That's good, but four of the games are below 60 fps, while two of the games run at more than 100 fps, somewhat skewing the results. 4K60 is within reach of the 3070, just not at maxed-out settings in all games. The 4K medium results meanwhile range from a low of 73 fps in Metro Exodus to a high of 140 fps in Strange Brigade, and every game breaks 60 fps.
Looking at the competition, the 2080 Ti is about three percent faster than the 3070, so again, the memory capacity and bandwidth deficit didn't really factor into things as much as we expected. The 3070 is also 35 percent faster than the 2070 Super, 53 percent faster than the 2070 FE, 130 percent faster than the GTX 1070, and 95 percent faster than the Vega 56.
If you're running a card that's three or more years old, upgrading to the 3070 should provide a significant boost to performance. If you have anything more recent — at least in the high-end category of GPUs — you can probably stick with whatever you have for a couple of more years.
The Division 2
Far Cry 5
Final Fantasy XIV
Forza Horizon 4
Red Dead Redemption 2
Shadow of the Tomb Raider
Nvidia GeForce RTX 3070 — Bonus Round 4K Gaming
As we did on the GeForce RTX 3080 and GeForce RTX 3090 reviews, we've run a bunch of additional benchmarks at 4K and maxed out settings, with ray tracing and DLSS enabled where applicable. Well, except for 3DMark Port Royal, which runs at 1440p by default, but who's counting? We're only testing a subset of GPUs here, basically RTX 2080 and above — AMD GPUs can't run these same benchmarks in many cases since they don't support ray tracing or DLSS yet, and they're not going to support DLSS ever, most likely.
That's a big problem, actually, as DLSS 2.x really does work well and makes Nvidia's RTX cards far more capable of running many games at higher resolutions. Is 4K with DLSS the same as 4K native? No. But if 4K DLSS looks similar and performs much better, it's the clearly superior choice that most gamers would want to use. Even if AMD's RX 6800 XT and other Big Navi GPUs can match the RTX 3080 in standard performance metrics, not having a comparable alternative to DLSS 2.x is going to be a concern.
There may only be 20 or so games with DLSS right now, but Unreal Engine 4 has apparently made it quite easy to switch on DLSS — there are a few demos we're looking at here that use Unreal Engine, and even the solo developer effort Pumpkin Jack managed to enable ray-traced shadows, reflections, diffuse lighting, and DLSS thanks to Unreal Engine. Considering how many games use variants of Unreal Engine, we'll likely see even more DLSS enabled games in the coming years.
15 Game Average
Looking at the combined results of all 15 games/benchmarks, the results are very close to what we saw with our primary test suite. The RTX 3070 ends up 28 percent faster than the RTX 2080 FE, and one percent slower than the RTX 2080 Ti. It's also 25 percent slower than the RTX 3080 and 34 percent slower than the RTX 3090. Average fps is just above 60 for the 15 tests, but individual results will definitely dip below that mark in some games.
3DMark Port Royal
We're going in alphabetical order, so 3DMark Port Royal is the first in our list of bonus 'games.' It's quite demanding, averaging just 38 fps at native 1440p rendering, and the 3070 is about 29 percent slower than the 3080 here. It's also 8 percent slower than the 2080 Ti, which is a larger gap than we've seen in any actual games.
Battlefield V was the first game to release with ray tracing and DLSS support. RT is only used for reflections, and this is DLSS 1.0 — meaning you can't even use DLSS with higher-end Nvidia GPUs at less than 4K. Still, we enabled ultra ray tracing and DLSS, and that's enough to average 64 fps on the 3070. Like Port Royal, the 3070 is about 25 percent slower than the 3080 and 7 percent slower than the 2080 Ti, and it's only 15 percent faster than the RTX 2080.
Bright Memory Infinite is a benchmark for a game that's currently in development. I have no idea what the game is about, but it looks absolutely amazing. There are multi-bounce reflections, refractions, caustics, and more at the highest settings, and even with DLSS in performance mode (4x upscaling from 1080p), the RTX 3070 only manages 38 fps. Using the high preset doesn't help much, but the good news is that the game supports DLSS 2.0, so running at 1440p with 4x upscaling ends up being a far better result. Interestingly, while the 3070 is 28 percent behind the 3080, it's also 17 percent faster than the 2080 Ti at maxed-out settings. Considering the number of RT effects being used, future games may also favor the 30-series by a wider margin.
It's a different developer, but Boundary otherwise has a lot in common with Bright Memory Infinite. Both use Unreal Engine, support a bunch of RT effects, DLSS 2.0, and will eventually be full games on Steam. Maybe. I'd say Bright Memory Infinite currently looks better, or at least the benchmark does, but we'll have to see what the final games are like — and if they're available in English (both devs are Chinese). As for performance, Boundary is perhaps slightly more demanding, depending on the settings used. With DLSS in performance mode, the 3070 averages 39 fps, but minimums are only 23 fps.
Call of Duty: Modern Warfare
Call of Duty: Modern Warfare (2019) doesn't support DLSS and only uses ray tracing for shadows. It makes a difference, but it tends to be less noticeable than some of the other RT options. Also, it still drops performance quite a bit. In our benchmark sequence, which uses the single player mode, performance is just a bit shy of 60 fps. In multiplayer, especially frantic shootouts, performance can be quite a bit lower. Most competitive players will turn down a lot of the effects in this game, though playing at 1440p or 1080p should also boost framerates a lot.
Up to now, Control remains the best implementation of ray tracing and DLSS effects in a shipping game. It has reflections and transparent reflections, diffuse lighting, and ambient occlusion. We tested in both DLSS Quality (1440p) and DLSS Performance (1080p) modes. The 3070 comes out slightly ahead of the 2080 Ti in both cases, and is about 35 percent faster than the RTX 2080.
Can it run Crysis Remastered? At 4K with maxed out settings, including ray tracing, the answer is no — your PC cannot handle it. Even the RTX 3090 gets just 41 fps with ray tracing enabled, while the 3070 manages a rather pathetic showing of 24 fps, and that's still faster than the 2080 Ti. Crysis Remastered lives up to its progenitor's legacy, though I'm not convinced the ray tracing effects even look that good. It's also using some weird hybrid ray tracing solution that runs on top of DirectX 11. Even without ray tracing, the 3070 plugs along at 47 fps, just 1 fps behind the 2080 Ti. If you want to play Crysis Remastered with RT enabled, plan on running at 1440p or 1080p.
Sort of the opposite end of the spectrum from Crysis, Doom Eternal rips and tears along at an impressive 115 fps on the 3070. This is also a game that tends to like memory bandwidth, so it's no surprise that the 2080 Ti ends up 12 percent faster than the 3070. The 3070 does beat the 2080 FE by 37 percent, at least.
Sorry, Death Stranding, no all-caps for you! But DLSS in Quality mode looks and runs great. The 3070 gets over 100 fps, and even the 2080 gets 86 fps. Norman Reedus would be proud of the little guy (little relative to the 3090 at least).
Microsoft Flight Simulator is another game that likes lots of VRAM, so the 2080 Ti posts a larger lead over the 3070 again. The 3070, meanwhile, is only 13 percent faster than the RTX 2080 this time. Also, it can't break 30 fps in our test sequence, which consists of the autopilot bringing in a plane for a landing at my local airport. This is a brute of a game to run at ultra settings, needless to say.
Horizon Zero Dawn
There are no fancy RT or DLSS effects in Horizon Zero Dawn, but it does look quite nice. The 3070 and 2080 Ti are basically tied in average fps, but the 3070 has lower minimum fps — probably due to having 8GB VRAM instead of 11GB. A few tweaks should get 4K performance above 60 fps.
Metro Exodus was already demanding without enabling its ray traced global illumination. Even with DLSS enabled (DLSS 1.0, sadly — I'd love to see 4A Games release a patch that supports DLSS 2.1 instead!), the 3070 only manages 45 fps, about 6 percent behind the 2080 Ti.
Project Cars 3
We looked at Project CARS 3 at launch, and we'll probably replace Forza Horizon 4 with the game in the near future. It's newer and more demanding, though it's missing a few features like TAA. It's also one of the games where the 3070 comes out a few percent ahead of the 2080 Ti for a change.
Shadow of the Tomb Raider
The shadows in Shadow of the Tomb Raider don't really look that much better with ray tracing, but it was at least a somewhat clever marketing schtick. Performance takes a pretty big hit, even with DLSS 1.0 enabled, and the 2080 Ti again gets a relatively large win over the 3070.
Last on our list, we have Wolfenstein Youngblood, another game that uses the Vulkan API and the only game we're testing that uses VulkanRT. Performance is very high, even with DLSS in quality mode, and the 3070 comes out ahead of the 2080 Ti by 6 percent — and 9 percent with DLSS performance mode.
GeForce RTX 3070: Power, Temperatures, and Fan Speeds
The first two RTX 30-series Ampere GPUs pushed power ratings to the highest levels we've ever seen for a 'reference' design. RTX 3070 dials things back to relatively sane levels of just 220W, basically matching the 215W TDP of the RTX 2080 and RTX 2070 Super.
It's a bit surprising that Nvidia only uses a single 8-pin PEG connector to feed the 12-pin connector, though, since the 8-pin is rated for 150W. Add in 75W from the PCIe slot, and it's basically right at the limit — the 2070 Super and 2080 both had an 8-pin plus 6-pin PEG configuration. At stock, the GeForce RTX 3070 Founders Edition should be within spec, but overclocking might have some oddities to investigate further.
For these tests, we run Metro Exodus at 1440p ultra without ray tracing or DLSS. We also use FurMark running at 1600x900 in a window using the stress test mode. We'll look at power, GPU clocks, temperatures, and fan speeds. We also measured noise levels of the RTX 3070, from a distance of 15cm. At idle, from that distance, our PC's noise floor is 46.0 dB (or 34 dB from where my head is relative to the case on the floor). After running Metro for 15 minutes, the 3070 FE noise levels were barely above that — 46.8 dB. Needless to say, the fans are doing an excellent job of cooling without making a lot of noise.
Clock speeds, temperatures, fan speeds, and power are all interrelated. Drop the clocks, and you reduce the power and temperature. Raise the fan speed, and you reduce the temperature and power use, potentially allowing for higher clocks. It's a four-way balancing act, and utilities like EVGA Precision X1 and MSI Afterburner allow you to customize your particular card (within limits).
Power use is just a bit below the rated 220W during Metro Exodus testing, while FurMark comes in at 224W. That's for stock operation. Overclocked with the power limit raised by 9 percent, power use reached 238W during Metro and 245W with FurMark.
Incidentally, that also pushes the card beyond the 225W theoretical limit of an 8-pin connector and the PCIe slot. Our PCIe slot measurements did go just above 75W during the overclocked tests, reaching 78.8W, while the maximum from the PCIe slot was 70W at stock. Meanwhile, the 8-pin connector had a peak power draw of 178.4W — even stock saw the 8-pin hit a peak of 165.3W, but we're not particularly worried about that since the power cable has two 8-pin connectors.
Average GPU clocks for the RTX 3070 Founders Edition are above the advertised boost clock, but what's interesting is that they're also slightly below the measured clocks for the RTX 3080. The boost clock is officially 1730 MHz on the 3070 and 1710 MHz on the 3080, but the 3080 ran at 1876 MHz average compared to 1863 MHz average on the 3070. FurMark flips the tables, with the 3070 averaging 1595 MHz and the 3080 getting 1576 MHz — both below the rated boost clock but above the base clock (which Nvidia doesn't actually list these days).
With the overclocked settings, the 3070 also averaged nearly 2GHz, coming in just 37MHz short. It periodically boosts above 2000 MHz but won't maintain those clocks — at least with the Founders Edition and a 220W + 9% (240W) power limit. Some third-party cards will very likely stay above 2GHz while gaming.
The RTX 3070 FE at stock runs pretty cool, reaching a peak temperature of 69C after a couple of minutes of gaming. The average temperature during our Metro testing was 66C, but note that the longer you game, the higher the average will get. When overclocked, the peak temperature and average temperatures were basically unchanged, but that's because we applied a custom fan curve that was a bit more aggressive. In FurMark, clock throttling kicks in to keep temperatures in check, with a peak stock result of 68C and an overclocked result of 70C, with average temperatures of 66.5C and 68C, respectively.
Here you can better see the results of our fan speed customization with the overclocked 3070. It settled in at about 1600 RPM when overclocked, compared to 1300 RPM at stock. Also, note that the GTX 1070 is getting old and can be a bit of a curmudgeon, and the fans appear to be working extra hard to keep it running at around 75C. A good deep cleaning of the radiator fins and fan blades would likely help — something to keep in mind if you have an older GPU that's overheating or running loud.
In general, the GeForce RTX 3070 Founders Edition performs well, and the new cooler and fan design get the job done. Considering this is a much smaller card than the 3080, the thermal and noise results, in particular, are welcome. You could definitely fit the card into a smaller case, though we'd still recommend something with a decent amount of airflow. If you want a cooler-running RTX 3070, third party cards will definitely fill that role — some companies are basically taking their RTX 3080 coolers and slapping them on a 3070, which will provide ample cooling for the lower-power GPU.
Nvidia GeForce RTX 3070 Founders Edition: Potent, but Let's Wait and See
The GeForce RTX 3070 Founders Edition is everything we expected. It's a lower power card with a smaller footprint, and it basically trades blows with the previous generation king of the hill, the RTX 2080 Ti. Two years later, and $500 now potentially gets you the same performance as the old $1,200 GPUs. If there's one constant in the world of GPUs, it's the ever-increasing performance at any given price point. But we're in the midst of a lot of GPU stuff, and without seeing what AMD's Big Navi brings to the table, it's impossible to give a final verdict for the RTX 3070.
AMD will spill the beans on Big Navi tomorrow, but we don't expect the hardware to arrive for review until mid-November or later. That's fine because, in all likelihood, almost no one is going to be able to buy an RTX 3070. Just like the RTX 3080 and RTX 3090 sold out within minutes (or seconds) of launch, we expect the initial batches of RTX 3070 will go fast. Historically speaking, that's what always happens for new high-end cards. RTX 2070, 2080, and 2080 Ti sold out at launch; GTX 1070, 1080, and 1080 Ti also sold out at launch; and so did GTX 970 and 980. The question isn't so much whether the new GPUs will sell out, but rather how long they'll be harder to purchase.
Launching in the midst of the COVID-19 pandemic, plus the holiday shopping season, is only going to exacerbate things. That's why Nvidia's CEO already said supply likely wouldn't keep up with demand until 2021. Hopefully, the supply of the RTX 3070 is better than it's bigger siblings — it should be since it's a smaller chip that will likely yield better. Plus, Nvidia knows it sells more $500 GPUs than it does $700 and $1000+ GPUs, so it almost certainly ordered more GA104 chips than GA102.
Besides AMD's incoming RX 6000 series, you can also count on Nvidia releasing RTX 3060 or RTX 3060 Ti or something similar in the not-too-distant future. It's basic silicon manufacturing: The GA104 chip has up to 48 SMs, 46 of which are enabled on the RTX 3070. While there will be some chips that are fully functional and could run with all 48 SMs, only selling 46 SM chips means overall improved yields. But inevitably, there will be a bunch of chips that can't meet the 46 SM requirement, so there will be another graphics card that's a step down from the 3070, and probably one more that's a step down from there. That's where the RTX 3060 Ti and RTX 3060 come into play. What we don't know is exactly when they'll launch or how much they'll cost, but we'd bet on a $399 replacement for the RTX 2060 Super, and probably a $349 or $299 replacement for the RTX 2060. It all hinges on how well AMD's RX 6000 does.
Speaking of which, early indications of Big Navi and RX 6000 are promising. There's a good chance AMD will have something that comes pretty darn close to the RTX 3080, or maybe even exceeds it, and then there will be lower-tier parts trickling down from there. The catch is that most of the rumors and purported leaks are for games running without ray tracing. AMD will support DirectX 12 Ultimate, which includes ray tracing and VRS (variable rate shading), as well as VulkanRT. What we don't know is how fast AMD's chips will be in such workloads. The general opinion is that AMD won't match Ampere when it comes to RT performance, which could prove to be increasingly important, what with the next-gen consoles adding RT support, not to mention games like Cyberpunk 2077.
While ray tracing is arguably going to become even more important in the coming years, I think the bigger problem for AMD is DLSS. AMD has made no mention of anything equivalent to the Tensor cores present on Turing and Ampere RTX GPUs. For any game that supports DLSS — and the list is growing — Nvidia will probably offer a superior experience. Given a choice between 4K DLSS running at 75 fps or 4K native with TAA running at 45 fps, as an example, I'd take the DLSS option every time.
Right now, there are at least 19 games that support DLSS (Anthem (1.0), Battlefield V (1.0), Bright Memory (2.0), Control (2.1), Death Stranding (2.1), Deliver Us the Moon (2.0), F1 2020 (2.0), Final Fantasy XV (1.0), Fortnite (2.1), Justice (1.0?), Marvel's Avengers (2.1), Mechwarrior 5 Mercenaries (2.0), Metro Exodus (1.0), Minecraft RTX (2.0), Monster Hunter World (2.0), Pumpkin Jack (2.0), Shadow of the Tomb Raider (1.0), Watch Dogs Legion (2.1), and Wolfenstein Youngblood (2.1) — I might have missed a couple). DLSS 2.x integration appears to be quite easy, relatively speaking, since it doesn't require game-specific training from Nvidia's supercomputer, and support from Unreal Engine makes it almost silly not to support DLSS.
Besides those already available (or available tomorrow, if you count Watch Dogs Legion), at least 14 more games are coming that support DLSS 2.x (Amid Evil, Atomic Heart, Boundary, Call of Duty Black Ops, Cyberpunk 2077, Edge of Eternity, Ghost Runner, JX3, Mortal Shell, Mount & Blade 2: Bannerlord, Ready or Not, Scavengers, Vampire the Masquerade — Bloodlines 2, and Xuan-Yuan Sword VII). Several of those games aren't even on my radar, but a handful are right at the top of most 'anticipated games' lists.
Let me put it another way: If you give me an option of basically equivalent image quality, with 50-70 percent higher performance, what games would I not want to support that? Maybe pixelart indie games, because it could throw off the aesthetic (and they already run fast enough), but everything else? Yes, please, I'd like some DLSS 2.x. There are games where DLSS lets the RTX 2060 perform more like a 2080 Super, and the 2080 Super ends up looking like an RTX 3080. I didn't think DLSS 1.0 was all that great, but DLSS 2.0 changed things. It's not that it always looks better (sometimes it does, sometimes it doesn't), but it almost always runs a lot faster.
It's possible AMD will have some DLSS alternative that works equally well available at some point. I'd really love to see that happen because TAA tends to be way too blurry, and FidelityFX/CAS, while good, isn't quite at the level of DLSS 2.x. But then AMD will have to get developers to utilize its DLSS alternative, which puts it about two years behind Nvidia's RTX and DLSS efforts. For games that don't support ray tracing or DLSS, it's easier to see how AMD can win the coming battle. But pricing is still likely to be key.
On that point, we don't expect AMD to severely undercut Nvidia. It will simply try to offer an attractive option. The reason is that Zen 3 looks ready to clean Intel's clock come November. Even the Ryzen 9 5950X only uses two 7nm Zen 3 chiplets, each of which measures just 75-80mm square. That's an $800 chip that only requires at most 160mm square of TSMC N7 manufacturing. Sure, CPUs are different than GPUs, but TSMC is basically at capacity for how many wafers it can produce each month. Given the size of Zen 3, and an estimated size of 500-536mm square for Navi 21? AMD will want to charge quite a bit for the GPUs, as much as it can reasonably manage. That or it would just produce more Zen 3 chips, which will probably sell like hotcakes.
The bottom line is that we can't declare a winner right this moment. Nvidia's Ampere RTX 30-series GPUs are potent, and the RTX 3070 brings new levels of performance to the $500 market. We expect to see 30-series parts push down into the $300-$400 range in the coming months as well. AMD's Big Navi is more of a wildcard since we don't quite know what to expect in terms of ray tracing performance or DLSS alternatives. AMD may have as many as four Navi 2x GPUs launching in the next month or two (or three or four), also with prices ranging from perhaps $250 up to $600 or more.
If you're already set on going with Nvidia and don't want to spend more than $500, you can try to pick up an RTX 3070 on Thursday. If you're willing to spend a bit more money, we'd argue the added VRAM, bandwidth, and performance of the RTX 3080 means it's the better option at $700 — not that you can find RTX 3080 in stock, but you can keep trying. For the undecided, we suggest waiting to see what happens with Big Navi, and of course, those who prefer AMD GPUs will want an RX 6000 regardless of how it stacks up.
Whatever your GPU choice, it's an exciting time after about two years of waiting for the next big thing to arrive. Of course, there will be even faster options next year, so if you already have a capable GPU, just enjoy it until it's no longer sufficient. Maybe by then, we'll be talking about RTX 6070 and RX 9000.