James Archer
I knew the RTX 5070 was tricking me. Parked next to the extravagant silliness of the two-grand RTX 5090, this £539 / $549 graphics card looked like a very agreeable deal, offering all the same DLSS 4 and Multi Frame Generation as its bigger, pricier brothers. Also, an upgrade to the RTX 4070 Super, a GPU that could handle 4K without looking too out of place in a premium 1080p rig. Tragically, though, the RTX 5070 breaks a sacred covenant, a mutual understanding between PC owners and parts makers that’s held strong for decades: if you buy a new version of a thing, it should be faster than the old version of that thing. Look past the MFG illusion, and far too often, it isn’t.
Rifling through the specs doesn’t turn up any clear reasons for this lack of performance gain; the RTX 5070’s CUDA core count comes in at 6144, putting it merely in-between the RTX 4070 (5888) and the RTX 4070 Super (7168), but both base and boost clock speeds have been raised to compensate. VRAM, meanwhile, remains at 12GB, but with a switch from GDDR6X to GDDR7 that confers significantly more bandwidth. Then there’s the power rating, which both the delayed Founders Edition and this here triple-fan PNY GeForce RTX 5070 OC set at 250W – 50W more than an entry-level RTX 4070 Super. It looks, and sounds, like a decent generational update.
And yet…
Nvidia GeForce RTX 5070 review: 4K benchmarks
While the RTX 5070 does make the cut as a 4K contender, surpassing 60fps in most Ultra-quality games without upscaling, the only time it produced a visibly faster framerate was in Cyberpunk 2077. Elsewhere, it only outperforms the RTX 4070 Super by a scant handful of single-digit frames per second, and repeatedly loses out to the non-Super RTX 4070 Ti in the process.
Obviously, very few potential buyers are going to be upgrading directly from last year’s RTX 4070 Super to the RTX 5070, and the latest model does at least make for a meaty improvement on the RTX 3070. Trouble is, that improvement has already been made in the last generation, and for prices that – even if you’re lucky enough to find an RTX 5070 in stock below £600 – are likely drop much faster, especially where the secondhand market is concerned.
It’s also worth addressing a specific claim, made by Nvidia during their original RTX 50 series announcement, that the RTX 5070 delivers “RTX 4090 performance” with MFG. This is, depending on how generous you want to be, either lacking in context or complete tosh. While it’s true that, for instance, 4x MFG can drag the RTX 5070 to 71fps in a fully path-traced Alan Wake 2, that’s only equal or faster than the RTX 4090 if you deny the older GPU its own tools. Sure enough, it only needs DLSS 3’s old-timey, 2x frame generation to average 82fps on otherwise identical settings. In short, the only way the RTX 5070 is truly moving at the same pace as the RTX 4090 is if they’re in the same delivery van.
You could still consider MFG to be the RTX 5070’s party trick, though at 4K specifically, it’s not as much of a difference-maker as it is on on burlier cards like the RTX 5070 Ti and RTX 5080. In both Cyberpunk and Alan Wake 2, the RTX 4070 joins its 40-series rivals in being unable to churn out playable path tracing performance on DLSS Quality; in fact, in the latter, it’s narrowly the slowest of the lot. As a result, when 4x MFG kicks in, it’s producing a higher number in a framerate counter but isn’t actually helping the games to run and smoother or more responsively than on the DLSS 3-limited 40 series models. As far as your PC knows, it’s still only running at 26/21fps, and all the AI-generated frames in the world won’t get rid of the sluggish feel that aiming and camera control will suffer in those conditions.
Nvidia GeForce RTX 5070 review: 1440p benchmarks
1440p is a more comfortable environment for the RTX 5070’s frame gen tech, though its problems with overfamiliar performance remain. If anything, it’s more bothersome here, as the higher framerates make it even harder to tell a naked-eye difference between this and its predecessors.
In Assassin’s Creed Mirage, the RTX 4070 Super even finishes ahead, if only be a single frame, while even the games that the RTX 5070 does relatively well in (Cyberpunk 2077, F1 24) only see an uptick of 10% or so. There is a simultaneous gap-narrowing with the RTX 5070 Ti, suggesting you don’t need to stretch to the Ti version for quality Quad HD, and again, anyone upgrading from the RTX 3070 will enjoy a sizeable speed boost. Still, these benefits could apply to the RTX 4070 Super just as much as the RTX 5070.
The same can’t be said, in fairness, for frame gen performance. Although these tests suggest the RTX 5070 is in fact slightly worse at dealing with path tracing/full ray tracing than those pesky RTX 40 GPUs, unlike at 4K, it is able to produce enough pre-generation frames to support a workable application of 4x MFG.
I’m not saying this is an adequate replacement for traditionally rendered frames, but at least at this resolution, you can kind of see where Nvidia is coming from. Spared from the need to compensate for sub-30fps framerates, frame gen is free to simply shine up the visual smoothness of already adequately-running games, as it should be.
Still, it’s also hard not to peer over at those towering RTX 5070 Ti and RTX 5080 bars, wondering why the standard RTX 5070 couldn’t just do a little more in the conventional rendering department.
Nvidia GeForce RTX 5070 review: 1080p benchmarks
The drop to an even less demanding screen rez does little to raise the RTX 5070’s value proposition. Yet again it’s haunted by the Ghosts of XX70 GPUs Past, especially in Assassin’s Creed Mirage, where it repeats its embarrassing loss. It does gain a few extra frames over the RTX 4070 Super in Shadow of the Tomb Raider, but then that’s already running so fast that you’d need a 300Hz monitor and downright inhuman observation skills to perceive them in action.
That all means we must return to the world of maxed-out lighting effects and AI frame generation in search of a consolation prize. Like at 1440p, the RTX 5070 starts off inexplicably worse than the RTX 4070 Super when path tracing in involved, though in both games it still forms itself a solid enough base to apply 4x MFG without input lag coating your PC innards in treacle.
You’ll have to forgive me for not sharing these results with the kind of bubbly enthusiasm that multicoloured bar charts deserve. I know it’s not like the RTX 5070 has regressed to the point that it’s a bad GPU in the general sense – get it at RRP, like this PNY model costs, and it’s still a fine multi-discipline graphics card that will comfortably play anything outside of the most brutalising 4K settings. And what I was saying earlier, about the RTX 4070 Super going cheaper faster? Reader, I’m afraid that was conjecture. Right now, these are still on sale in the £600-£700 range, and while plenty of board partner RTX 5070s will surely fill the same space, if these two GPUs cost the same then there’s no compelling reason to stick with the older one. There’s maybe an argument in favour of the RTX 4070 Super’s power efficiency – I measured it peaking at 219W during my tests, versus 251W on the RTX 5070 – but for all its shortcomings, I would rather have MFG than a 30W saving.
Even so, it’s also completely fair and reasonable to expect some kind of meaningful, not wholly AI-reliant improvement from each new generation. The RTX 5070 simply does not deliver, and I’m honestly a little worried that if we simply shrug and accept that, frame generation really will be the only way that future GPUs bother to upgrade “performance”. Even when, as this very card’s 4K path tracing results show, such an approach is inadequate when the games side of the industry continues to ask more and more of your hardware.
This review is based on a retail unit provided by the manufacturer.
ranzware
Add comment