There are two things you should know about the RTX 3070. First, this is a fabulous GPU by any measure. Nvidia has been claiming that this $500 GPU could match its $1200 RTX 2080 Ti Founder’s Edition. That claim has been demonstrated as broadly true with the release of the Founder’s Edition earlier this week — the RTX 2080 Ti can still pull ahead by a whisker in some 4K games, but in the vast majority of titles, the two are neck-and-neck.
Second: As great a GPU as the RTX 3070 is, you may still want to hold off a bit before purchasing a new card. It may not be able to buy one in any case — Nvidia has warned would-be customers to expect limited supply through the end of this year and into 2021. Separately from this, AMD will launch its own Big Navi cards in a matter of weeks. It may be prudent to see how the two compare before pulling the trigger.
GA102 and the MSI RTX 3070 Gaming X Trio
The MSI RTX 3070 Gaming X Trio is built around Nvidia’s GA104 GPU, while the RTX 3080 and RTX 3090 use the larger GA102 processor. Nvidia has been building multiple die at the high end for its recent launches as opposed to fusing off parts of the same design to create lower-end parts. When Turing launched, the RTX 2080 and RTX 2070 each used a different GPU design (TU104 and TU106, respectively). This time around, the RTX 3090 and RTX 3080 share a common GPU core (GA102), while RTX 3070 is built around its own unique chip (GA104).
Like its Turing and Pascal predecessors, the RTX 3070 retains the same 8GB RAM capacity Nvidia has used in this price segment since 2016. The company had plans to launch high VRAM variants of the RTX 3070 and RTX 3080, but those plans have been canceled or at least put on hold until overall availability improves.
The RTX 3070 packs 5888 cores, 96 ROPs, 14Gbps GDDR6, and a 1.725GHz boost clock. The GPU’s massive number of cores — more than 2.5x as many as the RTX 2070 — is why Nvidia is so confident in its ability to match the RTX 2080 Ti with a much cheaper card. The chip is a 17.4B transistor-design and its built on Samsung’s 8nm process. Computationally, the RTX 3070 has ~30 percent less compute power than the RTX 3080, and 41 percent less memory bandwidth. In addition to using slower RAM (16Gbps, down from 19Gbps on the RTX 3080), the RTX 3070 also has a smaller, 256-bit memory bus.
That’s the GA104 GPU itself. What has MSI brought to the table with the Gaming X Trio?
The Gaming X Trio uses a tri-axial cooler design, with additional bracing included to strengthen the GPU and prevent bending during transport.
I’ve seen more than one GPU shipped with inadequate packaging around it in the last few years, and I suspect the underlying problem is that people don’t appreciate just how heavy these cards have gotten. MSI’s strengthening bracket is a nice touch.
RGB support is provided via MSI’s “Mystic Light” system, and the default RGB is quite pretty if you like that sort of thing. The RGB lighting is designed to be controlled from MSI’s Dragon Center application, which functions as an all-in-one stop for controlling RGB, overclocking, and managing various system functions.
For those of you concerned about the power cabling situation, the MSI RTX 3070 Gaming X Trio uses twin 8-pin connectors, not the specialized Nvidia cable.
Our competitive selection against the RTX 3070 is a bit more limited than I’d like — I don’t have an RTX 2080 Ti to compare against. We’ll be comparing the RTX 3070 against the just-launched RTX 3080, the older RTX 2080 (non-Super), and the Radeon VII from AMD. The RTX 2070 Super and the RTX 2080 perform quite similarly, so the RTX 2080 does double-duty representing both SKUs.
I chose the Radeon VII over the 5700 XT because AMD’s highest-end consumer GCN product often outperformed the newer RDNA chip last year, even if it was only by a few percent. The fact is, AMD doesn’t have a great GPU to compare in this bracket at the moment. The 5700 XT is currently selling for ~$390, which is quite a bit less than the RTX 3070’s $500 baseline MSRP, and the Radeon VII isn’t on the market at the moment.
For this review, I’ve decided to put AMD’s overall best foot forward. In a matter of weeks, we’ll have competitive figures from the Radeon 6800 and 6800 XT, and will be able to give you a much better estimation of what Big Navi does and doesn’t bring to the table. Regardless, the AMD figures are here for reference. Until Big Navi debuts, Nvidia is competing against itself. If you want to estimate RX 5700 XT performance, assume it’s close enough to the Radeon VII that you’d never actually notice in reality, but factually a bit slower if you ran the numbers.
All testing performed on an Asus Maximus XII Hero Wi-Fi with 32GB of DDR4-3600 installed. Intel’s Core i9-10900K was used for all testing, with Windows 10 2004 and the latest set of updates and patches installed.
Ashes of the Singularity: Escalation shows the RTX 3070 pulling ahead of the RTX 2080 by 1.2x at 1080p, though the gap actually shrinks a bit as we go up. Ashes doesn’t bring GPUs to their knees quite the way it used to.
Deus Ex: Mankind Divided has the dubious distinction of having the worst MSAA mode I’ve ever seen, as far as its impact on performance. I’ve kept it around as a benchmark mostly for this reason. The RTX 3070 holds its 1.2x improvement over the RTX 2080 at 1080p but extends the lead to 1.26x at 4K. This is the first time we’ve seen a GPU clear the 30fps average at 4K with MSAA enabled, so that’s a serious achievement for the RTX 3080.
At 4K, the RTX 3080 is 1.31x faster than the RTX 3070 and 1.4x more expensive. We expect price-performance curves to begin to bend out of a 1:1 curve at these price points, and the ratio isn’t bad — provided either of these GPUs are available for MSRP. The Radeon VII fully matches the RTX 2080 here, but it’s clearly a last-gen card.
The gaps between the GPUs are a little smaller here, but we’re using somewhat lower detail levels, which means the CPU is in play a little more than typical. The RTX 3070 continues to impress. While $500 is a great deal to spend on a GPU, the high performance of the RTX 3070 at $500 implies good things about the GPUs that will follow farther down the stack.
AMD’s Radeon VII would have benefited somewhat from DX12 here, but it wouldn’t be enough to change the big picture. Nvidia is nothing if not consistent here, with regular bands between each GPU.
I threw Strange Brigade into the mixture to see what a Vulkan title might look like. Nothing much to see here, except an interesting performance by the Radeon VII, which lags the RTX 2080 by about 17 percent in 1080p, roughly the same amount at 1440p, and just 3.4 percent in 4K. Every Nvidia GPU loses much more performance from 1440p – 4K than 1080p – 1440p, implying this may be a quirk of the engine.
At the risk of sounding boring, you’ve seen this graph already. Remember that the RTX 2080 is also standing in for the RTX 2070 Super — the gap with the standard RTX 2070 would be larger, and the performance per dollar gains are considerable.
No surprises here.
Final Fantasy XV doesn’t run well on the Radeon VII — this is a test where we suspect RDNA would turn in better results. With Big Navi launching soon, it’s not a big deal either way.
Three different Metro Exodus graphs here, to highlight three different takeaways. First, we have Ultra ray tracing enabled at Extreme Detail on the RTX 3070 versus the RTX 2080. This is a worst-case scenario, with 200 percent supersampling active — and the RTX 3070 still manages to turn in a 1.31x higher framerate at 1080p. Interestingly, the gap is smallest without ray tracing enabled — and here, the RTX 3070 isn’t all that much faster than the RTX 2080. Not sure why the RTX 3080 threw an odd, lower result in 4K — I confirmed the result in the performance log, but I’ll run the test again and update this graph if needs be.
Big Picture Takeaways
At $500, the RTX 3070 is an objectively great GPU. As I expected, it solves all of the problems I initially had with Turing. Ray-tracing support is beginning to show up in games, and the higher-end cards in the family are now powerful enough to enable it at ultra-quality without needing to bother with tricks like DLSS 1.0. (DLSS 2.0 is much nicer). With Turing, I wasn’t comfortable recommending the family as a long-term investment into ray tracing given that the performance hit for enabling it could be 60-80 percent.
But most of all, with Ampere, Nvidia has returned to a Pascal-ish GPU pricing model. When it launched Turing, the RTX 2080 was 1.28x faster than the GTX 1080 on average, and cost about 1.28x more. The RTX 3070, meanwhile, is about 1.25x faster than the RTX 2080, while costing about $200 less than that GPU did at launch.
If you know you’re Team Green forever, and you’ve got $500 to throw at a GPU upgrade, this is a great card to choose. Nvidia powers the majority of gaming PCs, which means game developers will build their titles to target whatever amount of VRAM Nvidia GPUs offer. It’s going to be interesting to see if AMD’s 16GB cards can offer performance advantages, but 8GB cards aren’t going to be outdated in a year or two. MSI’s version of the GPU doesn’t put a lot of english on the ball, but there’s no need to fix what isn’t broken, and this GPU most emphatically isn’t.
The RTX 3070 is the true successor to the value proposition Nvidia debuted with the GTX 1080, assuming one can snag a GPU at MSRP. Whether that’s going to be possible is anyone’s guess, and the impact of bots has been ugly enough this year that I feel obligated to leave a bit of a question mark on this claim.
If you are willing to wait and see what AMD will bring to the fight, I recommend doing so — it’s always a good idea to see what the competition has in store — but only if $579 is still within your price range. Either way, the RTX 3070 is a huge leap forward for gaming, and a great value for gamers that can afford it.
GDDR6, Other GPU VRAM Pricing Expected to Jump Next Quarter
GDDR6 memory prices are expected to rise through Q3 2021. Older GDDR5 may finally be calming down, however.
GPU Pricing, Availability Are Moving in the Wrong Direction
GPU prices are moving again, but not in the direction we'd prefer.
Intel Shows First High-End Arc Benchmarks, Alleged Pricing Also Leaks
The Intel Arc Alchemist A750 Limited Edition card is targeting 60fps at 1440p "high" settings.