Our launch day coverage of the GeForce RTX 4070 includes: NVIDIA Founders Edition, ASUS Dual, Gainward Ghost, MSI Ventus and PNY RTX 4070. Tomorrow we'll have another round of reviews, for cards selling above the MSRP of $600.

Today NVIDIA is launching their fourth GeForce 40 Series graphics card—the GeForce RTX 4070. Just three months ago we saw the release of the $800 RTX 4070 Ti, now the green team's new card comes without the "Ti," for $200 less. Under the hood, the RTX 4070 is very similar to the RTX 4070 Ti. Both cards are based on the NVIDIA AD104 graphics processor and feature 12 GB GDDR6X memory on a 192-bit wide memory bus. Of course you're getting all of NVIDIA's GeForce 40 goodies, like DLSS 3 Frame Generation and improved ray tracing cores. What is different is the number of cores in the GPU—the RTX 4070 has 5,888 cores, which is a pretty hefty 30% difference to the 7,680 cores on RTX 4070 Ti. Clock speeds are lower, too, the RTX 4070 Ti had a rated boost of up to 2610 MHz, whereas the non-Ti is rated for 2475 MHz.

The Gainward RTX 4070 Ghost is a reference-clocked, cost-optimized card that's designed to sell at NVIDIA's MSRP. For this launch, NVIDIA has changed the rules, only reviews for cards priced at $600 can be published today, tomorrow all reviews for the more expensive cards can go live. This creates an incentive for their board partners to not only sample expensive high-end custom designs, but also the bread and butter models that the vast majority of people are buying. For me this of course increases the workload, because many vendors have submitted two samples, but I have to admit I've enjoyed reviewing these affordable models and I'm actually impressed that they are offering good performance, thermals and noise.

Averaged over the 25 games in our test suite, at 1440p resolution, the Gainward Ghost ends up a tiny 2% slower than the NVIDIA RTX 4070 Founders Edition. This is the case for all cards tested today, they are all running slightly lower clocks than the FE, despite all having the same base and boost clock rating, with the same power limit. Binning does play a role in the clocks the NVIDIA Boost algorithm selects, so maybe NVIDIA is putting the best GPUs on the FE, while other AICs reserve those for their factory-overclocked models. Either way, these differences are negligible in daily use, there's no way you'll be able to notice such a tiny difference.

Compared to last generation's RTX 3070, the performance uplift is 25%—a relatively small gen-over-gen gain (RTX 4090: +64%, RTX 4080: +49%, RTX 4070 Ti: +44%). It is still a substantial improvement, especially when you consider other factors like DLSS 3 and efficiency. Compared to AMD's offerings, the RTX 4070 matches the Radeon RX 6800 XT almost exactly, and is 8% behind the RX 6900 XT. AMD hasn't released any Radeon RX 7000 graphics card competitive to the RTX 4070 yet, actually they haven't released any graphics card this year.

These performance numbers make GeForce RTX 4070 an excellent choice for gaming at 1440p. Even smooth 4K is in reach in most titles, especially if you're willing to enable upscaling technologies like FSR or DLSS 2, or DLSS 3 Frame Generation. Just like the RTX 4070 Ti, I've noticed that RTX 4070 does a little bit worse at 4K than at 1440p relative to our other comparison cards. It seems the smaller L2 cache and 192-bit memory bus hold those cards back a little bit.

Just like on other NVIDIA cards, ray tracing performance is better than on AMD's offerings, because NVIDIA is executing more operations in dedicated hardware, and they have had a one GPU generation head start. While the performance differences are not night-and-day, they are still noteworthy. For example, the Radeon RX 6800 XT matches the RTX 4070 in pure rasterization workloads, but falls behind over 20% with ray tracing enabled. While RX 7900 XT offered 30% higher FPS with RT off, turning RT on makes that gap shrink to just 8%. Everybody agrees that ray tracing is the future, they just disagree on how fast it will become an essential part of gaming. If you're in the "RTX on" camp, then NVIDIA's GeForce graphics cards will give you higher framerates than AMD's offerings.

Another important selling point is that NVIDIA's GeForce 40 Series graphics cards support DLSS 3 Frame Generation. The algorithm takes two frames, measures how things have moved in those two frames and calculates an intermediate frame in which these things moved only half the distance. While this approach is definitely not problem-free, especially when pixel-peeping at stills or slowed down video, in real-time it's nearly impossible to notice any difference. As you run at higher FPS and resolution it becomes even more difficult, because the deltas between each frame are getting smaller and smaller. Implementing DLSS 3 is very easy for developers if they already have DLSS 2 support in their game, and NVIDIA is offering a lot of support to developers for integrating these technologies. Another interesting NVIDIA tech is Reflex, which reduces the total gaming latency (which NVIDIA likes to call whole system latency), so you see things earlier on your screen and can react faster, to get more kills or survive for longer.

Leading up to this launch, I've noticed a lot of discussion around the 12 GB VRAM size of RTX 4070. While I agree that 16 GB would be better, I disagree with people who say that 12 GB is already too small, or obsolete. There are a few (badly coded) games out there that use a ton of VRAM, especially at 4K, but the vast majority of titles won't even get close to such VRAM usage numbers. In our whole test suite not a single game saw a meaningful performance hit with 12 GB, not even at 4K—and RTX 4070 is fundamentally a 1440p card. You'll also have to consider that making a 16 GB card isn't just "let's add another 4 GB memory chip," but you also need to put additional signal traces on the PCB, and widen the memory controller inside the GPU, so that it can talk to all these chips in parallel. I don't think anyone would be willing to pay $700 for a 16 GB RTX 4070, would you? On the other hand, AMD does offer 16 GB VRAM on the Radeon RX 6800 XT and 6900 XT, which could make them an option for those who want to focus on VRAM future-proofing.

The Gainward Ghost is sold at MSRP and comes with a fairly basic design. For example, you're getting a plastic backplate, instead of one made from metal—a great compromise in my opinion. Our apples-to-apples cooler comparison test clearly shows that the Ghost has the weakest thermal solution of all the cards tested today, even weaker than the Founders Edition. This means that the fans have to run faster (=louder) to achieve the same temperatures as on other cards. While temperatures are low enough with 72°C, they are still a bit higher than on other cards tested today. Also, the noise levels are higher, to achieve these temps. With 36 dBA, the card isn't "loud" in any way, but it's not nearly as quiet as competing models, and the Founders Edition is quieter, too. Given the cooler's capabilities I still think that Gainward picked the optimum balance between temperatures and noise, and didn't go overboard with the fan speed to reach some thermal targets. Just like on all other cards tested today, idle-fan-stop is included with the card, for the perfect noise-free experience during idle, desktop work and Internet browsing.

Being able to run at these speeds with such a small cooling solution is an amazing feat, thanks to the stunning energy efficiency of NVIDIA's Ada architecture. With just 200 W, the power supply requirements are minimal, any beige OEM PSU will be able to drive the RTX 4070 just fine, so upgraders can just plop in a new graphics card and they're good to go. Performance per Watt is among the best we've ever seen, slightly better than RTX 4070 Ti and Radeon RX 7900 XTX; only the RTX 4090 and RTX 4080 are even more energy-efficient. Surprisingly, the Gainward Ghost consumes 5 W more than the FE in idle, maybe due to the RGB lighting element. In gaming on the other hand, the power draw is slightly lower than the FE.

Gainward has chosen to not allow any manual power limit increases on their card, no idea why. They are not alone though, PNY and MSI have locked their cards down, too. Manual overclocking is still possible of course, but having a bit of additional headroom for OC could come in handy for advanced users. NVIDIA's Founders Edition allows increasing the power limit up to 220 W, ASUS allows up to 216 W. Unlike the FE, Gainward is using an 8-pin power input, just one, which is a good alternative to the 16-pin, especially for people wanting to avoid the adapter. The 8-pin cable can also be bent more tightly, people are less scared of plugging it in the wrong way, and I suspect it's also a few cents cheaper to integrate, so no complaints from me. In theory, the 16-pin could deliver up to 600 W, but all RTX 4070 cards are capped at around 200 W, unable to use more power, even if it was available, so there's no performance, boost or power difference.

The Gainward GeForce RTX 4070 Ghost comes at the NVIDIA base price of $600, same as the GeForce RTX 4070 Founders Edition, and the other cards tested today. Compared to these, the Gainward card does have the weakest cooler, which means buyers will have to compromise a little bit on heat and noise. While it's not a big difference, I still feel that people will opt for other models, especially if availability is good and pricing really stays at MSRP. We'll know more in two days. On the other hand, Gainward's card is clearly less expensive to build, which means they could be the first to go below the MSRP if sales are slow. I totally would consider a $589 Gainward Ghost—performance matches the other models and everything else is "close enough." Also, I've been hearing rumors that NVIDIA is incentivizing these MSRP cards, so once such a program ends, prices might go up and the Ghost might be the only card left at MSRP.

$600 is an alright price given the current GPU pricing landscape, but is still $100 more expensive than the launch-price of RTX 3070 and RTX 2070. At that price point the RTX 4070 is clearly the best option, even though it's not standing out spectacularly, like it would had it been $500 instead. There's no way I'd buy RTX 3080 10 GB for $550 if I can get the RTX 4070 for $600. The selling points here are DLSS 3 and energy efficiency. Strong competition comes from the RX 6800 XT at $510. It offers a larger framebuffer of 16 GB and similar rasterization performance, but slower ray tracing and worse efficiency. Spending $200 more for an $800 RTX 4070 Ti is out of reach for people shopping with a $600 budget, just like Radeon RX 7900 XT.

NVIDIA knows these numbers, too, and that's exactly why the RTX 4070 is positioned like that. It's the best choice in its segment no doubt, with good performance and impressive efficiency. Still, graphics cards have become extremely expensive in the last years, and people are frustrated, and looking elsewhere. Consider $500 will buy you a PlayStation 5, a complete gaming system that's ready to go, which will run 4K and has every single game hand-optimized for the platform, without any compatibility issues, shader compilation, stutter, or other hoops to jump through.