NVIDIA RTX 3070 TI
Mid-range masterpiecee or overhyped dud?
AFTER LAST MONTH’S RTX 3080 Ti left us feeling a little underwhelmed, the RTX 3070 Ti tries to recover. And then stumbles again. They say an elephant never forgets, but what happens when the elephant only comes equipped with 8GB of memory? Sure, it’s faster memory this time, with GDDR6X replacing the 3070’s GDDR6, but that doesn’t feel sufficient for the $100 jump in price—and let’s be real, almost no one will be able to buy the RTX 3070 Ti for $599 anytime soon.
If the market were normal, the 3080 Ti and 3070 Ti likely would have launched at the 3080 and 3070 price points, pushing those cards down to lower tiers. But the pandemic and global shortages coupled with cryptomining—all of which are thankfully starting to taper off—mean Nvidia can sell everything it makes, and then some. The bigger problem stems from getting enough parts to sell, which is why prices are going up.
Nvidia claims these new cards were planned for months in advance, meaning chips were already allocated and these products are in addition to what was already on the market. If correct, that would mean more supply and more cards for people to buy. So far, that doesn’t seem to have mattered—even with the hash rate limiters now in effect on all of Nvidia’s Ampere GPUs (except the RTX 3090). But with China cracking down on miners (see “Tech Talk,” pg. 13), there’s hope things will improve significantly in the coming months.
Performance of the RTX 3070 Ti is good, leading the 3070 by around 10 percent across our test suite, but it seems to depend more on compute and memory capacity than on memory bandwidth. The 3070 Ti offers 7 percent more theoretical compute and 36 percent more bandwidth, meaning the extra speed of GDDR6X mostly goes untapped. With pricing landing halfway between the 3070 and 3080, but performance that tracks closer to the 3070, the RTX 3080 still ends up being the best option in our book.
Power consumption also ends up being 70W higher than the 3070. That’s more than a 30 percent increase, and it’s mostly thanks to the GDDR6X memory. At least the 3070 Ti doesn’t have as much difficulty cooling the memory— there’s less of it, after all—but we can’t help but think vanilla GDDR6 clocked at 16Gb/s would have been a better choice, particularly if Nvidia doubled down on capacity like AMD did with its RX 6800 and above, or like it did with the RTX 3060.
Lest we forget, this isn’t the first time Nvidia has been stingy with VRAM. The GTX 1060 came in 6GB and 3GB variants, while AMD’s competing RX 470/480 offered 4GB or 8GB at slightly lower prices. But those were mainstream GPUs; the RTX 3070 Ti is a high-end product, and it feels like Nvidia’s entire RTX 30-series stack (outside of the 3090) has less memory than we’d like. 8GB became the norm for high-end cards five years ago, and now it’s the minimum we’d recommend for a mainstream product.
Considering you can pick up an RX 6800 with 16GB of memory for less than the cost of the RTX 3070 Ti, on eBay at least, and that the RX 6800 generally performs better—until or unless you enable ray tracing and DLSS—this isn’t a very impressive showing for Nvidia. It’s still slightly faster than 2018’s RTX 2080 Ti and at a lower price, but less memory and higher power use for roughly equivalent performance after nearly three years? Yawn. Wake us up when Lovelace launches (hopefully in 2022), and maybe by then supply and pricing will be back to normal.
Nvidia GeForce RTX 3070 Ti
ELEPHANTS Good performance; proven architecture; maybe more supply.
MICE 8GB feels stingy; big TDP jump; not much faster than 3070.
$600, www.nvidia.com