Clearly NVIDIA has had issues keeping up with demand for their 3000 series graphics cards. And while they claim that they pushed back the launch of today's RTX 3070 to shore up supply, given that this is positioned as an RTX 2080 Ti killer for 500 bucks US, I doubt anything they've done in the last couple of weeks is going to help. Unless of course, we fire it up on our trusty test bench and it falls way short of expectations.
It looks like as far as NVIDIA's big gaming performance claims go, the RTX 3070 does adjust fine, matching or beating our 2080 Ti Founder's Edition more often than not. At least in games that take advantage of it's beefier RT cores, and tensor cores, which are focused on real-time Ray tracing and deep learning workloads respectively.
There was a time in the past when we might've called NVIDIA's DLSS deep learning upscaling a performance hack and forced their cards to compete with it disabled. But, there's just no other way to put this. It's gotten really good. When we turned all that stuff off, in Wolfenstein: Youngblood though, and forced a more traditional rendering approach, we did observe a small but measurable four to 5% dip in performance. So, it isn't quite a 2080 Ti, but at less than half the price, I don't think anyone's complaining.
Microsoft Flight Simulator, while CPU-bound at virtually any resolution, revealed a curious twist in our story though. The minimum frame rate is lower than the 2080 Ti, but the average is higher, suggesting that while the GPU is fast enough to keep up, there's a memory bandwidth bottleneck in the mix, thanks to the 3070's narrower 256 bit bus. It still excels in our parallel Nintendo 64 test but Counter-Strike: Global Offensive sees it fall below the RTX 2080 Ti again.
Although, we're still talking 450 versus 470 frames per second. So, it's pretty easy to forgive when you look at the bigger picture.
Productivity is another area where NVIDIA's new Ampere architecture GPUs really shine. Our blender benders on the 3070 went roughly 20 to 30% faster than the 2080 Ti. That is a significant improvement, likely down to the sheer number of cooler cores that NVIDIA has packed into this thing, with Samsung's eight nanometer process. V-Ray and Redshift show us similar results while OctaneBench shows us once again the difference between having RTX on versus off. Those RT cores are where this GPU's main advantage comes into play whenever memory bandwidth is an issue.
Speaking of which, the new SPECviewperf 2020 shows us that while it's great for rendering, and nearly as good for some other tasks that just need to crunch numbers, a lot of high-end workstation users suffer from the constrained memory bandwidth of consumer cards like this one, and performance of our RTX 3070 fell to as low as 80% of what the last gen RTX 2080 could provide.
To be fair to NVIDIA though, that's not what this card is designed to do. It just happens to be a consideration if you're self-employed or working from home and you do need to run these applications during work hours.
If we average all of our results though, we can see that we're hovering right around that magic 100% mark compared to the 2080 Ti, which is extremely impressive. Unless the 3070 has an Achilles heel that we haven't found yet. Let's take a closer look then, at our founder's edition card.
It's got the same flow-through cooler design as the RTX 3080, and RTX 3090, but it's super tiny and has the same fan layout as the RTX 20 series, which isn't to say it's light. I was surprised at how heavy it is for its size. And I was surprised to see NVIDIA's new radical 12-pin power connector, in the same visually jarring position.
I mean, at least the adapter only requires a single eight pin instead of two.
Interestingly, the grill on the I/O plate is indented a little. So it bolts directly onto the cooler rather than tabs on the display outputs, making the design feel much sturdier and therefore more likely to survive a drop.
Our card was we're able to ramp to 1.9 to two gigahertz all day long, with its core sitting at RTX 2080 Ti temperatures. Which is toastier than our RTX 2070 Super. But in fairness, the cooler here is significantly smaller which shows just how efficient NVIDIA's pass-through approaches. When we look at power consumption it was a touch higher over the course of our SPECviewperf run, compared with the RTX 2070 Super, with our peak recorded wattage no higher 253 Watts compared to that card's 224. This is excellent news because the RTX 3080 guzzled far more power than even the 2080 Ti, and the RTX 3090 sees massive power spikes that can cause mere mortal power supplies to just blink off, thanks to their over current protection.
On that subject, you'll have to forgive me for going back to the previous 3000 series launches again here. But there was yet another power debacle where some powder boards would crash due to a flawed power delivery design. Some brands ended up shipping bad boards all the way to customers. But even ones who didn't like EVGA suffered product delays from not having all the testing and info that they needed in time for launch.
Nobody would go on the record for us blaming NVIDIA for this outright. But as far as we can put together, team green just didn't give their partners enough lead time to flesh out their designs, while they themselves enjoy a massive headstart, putting together revolutionary new coolers for their Ampere cards.
I've given NVIDIA the benefit of the doubt on this so far. But with today's move, it's clear that they are openly and unashamedly competing on an uneven playing field against their own partners.
What have they done today, you might ask? Well look around. Where are all the reviews of non-Founders RTX 3070s, that have normal power connectors on them. Not in any NVIDIA-sanctioned reviews at the embargo lift. Why? Well, because their embargo hasn't lifted yet. It's kind of genius, really. NVIDIA goes and creates a big splash and sells out of their Founder's Edition cards first which they've had longer to perfect. Then their partners rush to launch their competing products with less time to prepare, giving NVIDIA a second wave of review coverage, comparing their product, the GPU itself, with their product, the Founder's Edition card, often quite favorably.
I doubt that anything that I say is going to change NVIDIA's behavior. But, it's a crappy way to treat a partner. And I wanted to make sure that at least they feel bad about it.
On the subject of competition, though, AMD's Radeon announcement is tomorrow, which, again knowing NVIDIA is probably why they carpet-bombed the review cycle.
So today, from numbers alone. It was always gonna be easy to recommend the RTX 3070 to anyone in this price range. The good news, for the entire industry though, is that NVIDIA's supply woes could be AMD's big chance to make a splash with the unveiling of their new Radeon 6000 series. We still don't know for sure what that's gonna look like. But what we do know is that between it, 3070's compelling value, and the potential for deals on last gen cards come Black Friday, it is going to be a good season to buy a gaming rig.