Everything we know about the Nvidia GeForce RTX 3070, including specs, performance, price, and release date.
Nvidia just announced the GeForce RTX 3070 and other GPUs that use the new Ampere architecture, and the great news is that pricing hasn’t radically changed from the current RTX 20-series Super cards. The RTX 3070 will pick up where the RTX 2070 Super left off, at the $499 price point. What’s more, it should offer better performance than the RTX 2080 You. Ampere has arrived, and it’s going to kick some butt over in the best graphics cards and GPU hierarchy. Here’s what we know.
We’ve covered most of the underlying Ampere architecture here. Here, the focus is specifically on the upcoming GeForce RTX 3070. We were afraid of what Nvidia might do with pricing, and we were excited to see what it would do with performance enhancing features. The fears, it turns out, were mostly unfounded, and the specs are mouthwatering.
The Nvidia GeForce RTX 3070 will continue as the ‘mainstream high-end’ offering, priced at $499. That’s a big step up from the old GTX 970’s $330 price point, but not too much worse than the GTX 1070’s initial $449 launch price. Whether we’ll be able to buy RTX 3070 cards at $500 any time in 2020 is debatable of course. Prepare yourself for some price gouging if you want one sooner than later.
The main GPU core now packs a whopping 5888 CUDA colors. That’s far more than the outgoing RTX 2080 You, and with higher clockspeeds it should end up delivering better performance. For less than half the price. You may now commence to party.
It’s not just the shader cores getting a substantial boost in performance. The RT cores are 1.7 time as potent as in the previous generation, which means that for ray tracing calculations the 3070 still delivers slightly more performance (40 RT TFLOPS) than the RTX 2080 You (34 RT TFLOPS). The 3rd gen Tensor cores meanwhile are four times as fast per core as the Turing Tensor cores, so FP16 performance ends up being 163 TFLOPS. That’s about 50% faster than the RTX 2080 You, and over 2.5 times the Tensor performance of the RTX 2070.
The only major unknown right now is the memory speed. Nvidia says it will use GDDR6 memory, where the RTX 3080 and RTX 3090 will use faster GDDR6X memory. We assume the VRAM will clock in at 16 Gbps, but it might only be 14 Gbps. We’ll get confirmation in the coming days.
We also don’t know if the RTX 3070 will use the same GA103 chip as the RTX 3080, only with fewer SMs and memory channels enabled, or if it’s a completely separate chip. It doesn’t really matter, as either way you would get the same performance. (Nvidia has used two or more different GPU cores for previous graphics cards.)
One important aspect of the RTX 3070 is that, unlike the RTX 3090, it keeps with a relatively tame power rating of just 220W. Sure, that’s more than the 2070’s 175W, but it basically matches the 2070 Super’s 215W TDP.
The combination of more shader cores, faster RT cores, and faster Tensor cores should prove very potent. Nvidia provided the above slide during the Ampere reveal on September 1, but didn’t specify which GPUs were used. Regardless, the raw numbers suggest RTX 3070 is going to be a great upgrade over anyone still using a Pascal GTX 10-series (or earlier) GPU.
We don’t have the fine-grained details on how things have changed with the Ampere architecture. We know shader core counts have basically doubled, but did that double performance as well? In practical applications and not just theoretical performance? We’ll have to wait and see.
Theoretically, RTX 3070 has 43% more compute power than the RTX 2080 You, for both the CUDA cores and Tensor cores. It also has 18% more ray tracing performance. And it does that while using 12% less power, at less than half the cost. With games like Cyberpunk 2077 set to double down on the use of ray tracing effects, having at least an RTX 3070 is probably a good idea.
There’s one area where it doesn’t beat the RTX 2080 You, however: memory bandwidth. The 2080 Ti had 11GB of 14 Gbps GDDR6 memory on a 352-bit bus. Even if we’re right on the 16 Gbps speed of the 3070’s memory, that would be 17% less total bandwidth because of the 256-bit bus. It should make a huge difference, particularly at 1440p and lower resolutions, but there may be a few edge cases where the 2080 Ti can keep up.
Something else to look forward to is even better DLSS. As in, 8K DLSS using upsampled 4K content. Okay, you probably aren’t going to buy a $500 GPU to power a $3000+ TV, but you could. Maybe. More likely, 4K (and 5K) gaming using DLSS will now easily break the 60 fps barrier. The 2080 Ti can already do that, so an even faster GPU will potentially be great if you happen to have a 4K 120Hz display.
Nvidia hasn’t provided an exact date, but say the GeForce RTX 3070 will be available in October. Given we know the launch dates for the RTX 3080 (September 17) and RTX 3090 (September 24), it’s reasonable to expect the 3070 to land in the middle of the month.
Nvidia tends to launch major new GPUs on Thursdays, so that means October 15, but it could also come a week earlier or later. Again, whether or not you’ll actually be able to buy a card on the launch date, for the $499 launch price, is another matter entirely.
Historically speaking, popular new AMD and Nvidia GPUs often sell out ‘immediately’ for the first month or two, and price gouging is common. Our best advice is that if you really want to buy an RTX 3070, you should pre-order, but at the same time we don’t like pre-ordering anything. Wait for the reviews to arrive, verify that performance and everything else works as expected, and then click the buy button. Or wait for it to be in stock so you can click the buy button.
We don’t have actual hard performance data yet, and we won’t until some time in October. However, everything on paper makes the GeForce RTX 3070 look like an easy recommendation. Performance has increased substantially over the previous generation cards, and price has remained the same.
If you’ve been skeptical of ray tracing, and whether it’s even necessary for games, now it really doesn’t matter. You can get higher ray tracing performance than Nvidia’s first generation hardware, and at the same time you’ll get a major boost in performance for traditional rendering modes. That’s our expectation at least; stay tuned for the full review next month.
We understand why a lot of people skipped the Turing generation of GPUs. They were expensive, and not really that much faster than Pascal. Besides, if you already bought a GTX 1070, GTX 1080, or GTX 1080 You, there wasn’t a real need to upgrade. Skipping a generation in hardware — and skipping the first generation of any new technology — is rarely a bad idea. But now, based on what we’ve seen of the specs, features, and pricing? This is going to be a very tasty card.
Nvidia has thrown down the gauntlet with the GeForce RTX 3070. What remains to be seen is whether AMD can match or even beat Nvidia when it comes to the next generation GPUs. Big Navi might end up being just as ‘big’ as Ampere. We should know more by the time these cards launch in October.
Tom’s Hardware is part of Future US Inc, an international media group and leading digital publisher. Visit our corporate site.
World news – THAT – The Nvidia GeForce RTX 3070: RTX 2080 Ti Performance at $499