GeForce RTX 2060 Super, 2070 Super, and 2070 all idle along with fans spinning at 1,500 RPM. As their GPUs warm up, those rotational speeds gradually ramp up.
By the end of our three-loop Metro benchmark, GeForce RTX 2060 Super’s fans spin the fastest, followed by GeForce RTX 2070. Since GeForce RTX 2070 Super is based on a very 2080-like design, its larger thermal solution does a better job of cooling TU104 than the original 2070 and its TU106 chip.
The GeForce RTX 2060 Super’s TU106 does warm up under that card’s dual-slot sink. But it doesn’t get worryingly hot. And there’s clearly still room in the fan curve for more aggressive cooling.
Meanwhile, GeForce RTX 2070 Super roughly matches the operating temperature of GeForce RTX 2070, despite a higher power rating and larger TU104 processor. Nvidia has the 2080’s big sink to thank for these results.
Nvidia does need higher voltages to maintain GeForce RTX 2070 Super’s more aggressive clock rate. But our thermal and power consumption measurements show that TU104 has plenty of headroom. We didn’t see a single example of throttling throughout our testing.
The resource-rich GeForce RTX 2060 Super operates at slightly lower frequencies than the standard 2060, which we’d expect. Its more capable TU106 processor is already running a bit of a fever, requiring faster-moving fans. And since Nvidia doesn’t grace the 2060 Super with souped-up cooling, we just have to live with the trade-offs of topping a hotter chip with the same old heat sink.
The fan curves engage a bit earlier under FurMark. And by the end of our sample period, GeForce RTX 2060 Super, 2070 Super, and 2070 all converge on the same rotational speed. Only GeForce RTX 2060 spins notable slower.
We also see less difference between each GPU’s temperature under FurMark. Vanilla GeForce RTX 2060 holds steady under 70°C, while the other three models we tested flirt with 71-73°C.
GeForce RTX 2060 Super sheds clock rate more noticeably under FurMark, as does GeForce RTX 2070.
MORE: Best Graphics Cards
MORE: Desktop GPU Performance Hierarchy Table
MORE: All Graphics Content
Guess now we wait for Navi to see if they have anything decent to counter.
Unfortunately, all those leftover RT and AI cores make for a giant and expensive to manufacture die. I'm not sure if we'll ever see an RTX card for under $350, but AMD could probably drop the 5700 down well below that if they were really pressed.
As for the value proposition of Ray Tracing itself... That's still really dubious. The overall number of upcoming RTX games is still lower than the overall number of RTX -laptops- on the market. I don't know about anyone else, but I'm not exactly lining up to spend $400+ for somewhat prettier graphics in the 2-3 games I find interesting. I have a feeling that most of the early adopters banking on the promise of future support probably bought an RTX card months ago.
So, cool on Nvidia for flexing their muscles and remaining the strongest, I just don't see it winning over that many new customers.
DLSS has proven to be more or less useless, being little more than a mediocre upscaling method in practice. It might not be bad if it looked or performed better than all other forms of upscaling, but it doesn't. A number of games use other upscaling methods that are not only better, but also don't require any special hardware, allowing them to run on any card while offering better performance than DLSS for a given level of image quality. So, as far as that feature is concerned, the comparisons are even worse now than they were when people didn't know what DLSS had to offer. It's gone from being a potentially useful feature to being a non-feature.
And there are still only a few games released so far that support raytraced lighting effects, and enabling them still kills performance at the resolutions most people are likely buying these cards for. Raytraced effects have the potential to improve visuals, but the current cards don't have nearly enough RT cores to run them well. I suppose the Super lineup helps a bit by making slightly more RT cores available at a given price point, but it's not nearly enough to prevent the performance hit from being substantial relative to the minor improvements to visuals. Perhaps next year's cards will make hybrid raytracing more viable.
Still, I do agree that if two cards otherwise offer similar performance at a given price level, hardware raytracing support is definitely something that can help differentiate one card from another. Enabling those effects might cause a big hit to performance, but at least the option is there. For that reason, I think the 5700 and 5700 XT may have a hard time justifying their launch prices compared to the Super cards, unless they perform significantly better than expected. The pricing seemed a bit underwhelming from the start, and even without the Super cards available, I felt each of those Navi cards should have been priced about $50 lower. The initial 20-series pricing was underwhelming from the start, and AMD didn't seem to be pushing value much beyond that, despite these cards launching the better part of a year later. AMD had the opportunity to build some hype for their new generation of cards through competitive pricing, but appear to have followed Nvidia's lead in price-gouging their new cards, despite not bringing anything really new to the table. Better pricing from the start could have prevented Nvidia from raining on their parade with the Super cards.
And on that note, the Radeon VII is the real loser compared to the Super lineup. Its value was already a bit questionable for anyone not utilizing its 16GB of VRAM for certain professional applications, since the RTX 2080 was already offering more gaming performance and dedicated RT hardware with lower power demands for about the same price, but now the 2070 Super will be offering similar performance along with RT hardware for almost $200 less. That makes the VII a no-go for just about anyone interested in it for gaming, and due to its large amount of expensive HBM2 VRAM, I doubt AMD could drop its price by $200 to compete.
That's another way to look at it, yes. I see it as where 2070 should have been at launch, agreed!
I hope their 7 nm cards bring a return to the Wattage of that era.
Anyway, it's essentially a software feature (although it depends on the tensor cores). So, there's always a chance they'll reverse course.
First, it's much easier to lower prices than to raise them. So, AMD could quickly come down from its launch pricing.
Second, I see the Super cards as pretty much inevitable, no matter what AMD did with pricing. The only real variable was whether and how Nvidia adjusted their pricing. The big question is whose price floor is higher. If it does come to a price war, how much can they each drop?
I've always seen the RTX features as very much about creating enough differentiation to justify higher pricing. However, they're far from the must-haves Nvidia surely hoped. As such, I think they're not yet sufficient to keep the RTX cards out of a price war. Maybe in the next gen, but not now.
It's true and a fair point, but you surely know it wasn't designed as a gaming card. The way to think of it is as a Titan - something of a specialty product that's soon to be surpassed by the next generation. Like the Kepler Titan and the Titan V, it's a professional GPU that they let gamers and prosumers have at a more accessible price point.
Inevitable, absolutely. But when, in what form, and at what price points? Would they have been satisfied to just sell FE or FE+ as "Super" if there was no competition? Or even if they pushed further than that, would the 2060 Super be unlocked to near-2070 levels, or would they still have gimped it slightly - retain the narrower memory interface perhaps? I mean it's all kind of academic at this point but it's an interesting thought exercise. I believe the timing and other decisions were influenced by Navi. With that being said the Super models as-spec'd are a substantial boost and should really make this an interesting fight!
Anyway yeah, the pricing may shift around a bit in the coming months. Having a high price floor is one of the things that bit AMD in the hind with their Vega lineup. This time AMD has both a moderate sized chip and conventional GDDR, though the new process is doubtless more expensive at the moment. The other wildcard is drivers - RDNA is supposedly different enough that I wouldn't be surprised to learn that there's decent room for optimization. That's not really a positive at launch though, but it's something else to chew on.
Indeed, it's got a ton of crunching horsepower for a relatively low price. The FP64 performance in particular is untouchable at that price.