NVIDIA Turing-Based GeForce GTX 1660 Ti Launched At $279 (hothardware.com)

MojoKid writes: NVIDIA has launched yet another graphics card today based on the company's new Turing GPU. This latest GPU, however, doesn't support NVIDIA's RTX ray-tracing technology or its DLSS (Deep Learning Super Sampling) image quality tech. The new GeForce GTX 1660 Ti does, however, bring with it all of the other GPU architecture improvements NVIDIA Turing offers. The new TU116 GPU on board the GeForce GTX 1660 Ti supports concurrent integer and floating point instructions (rather than serializing integer and FP instructions), and it also has a redesigned cache structure with double the amount of L2 cache versus their predecessors, while its L1 cache has been outfitted with a wider memory bus that ultimately doubles the bandwidth. NVIDIA's TU116 has 1,536 active CUDA cores, which is a decent uptick from the GTX 1060, but less than the current gen RTX 2060. Cards will also come equipped with 6GB of GDDR6 memory at 12 Gbps for 288GB/s of bandwidth. Performance-wise, the new GeForce GTX 1660 Ti is typically slightly faster than a previous gen GeFore GTX 1070, and much faster than a GTX 1060. Cards should be available at retail in the next few days, starting at $279.

Comments

Re:Aaaaannd they gimped it with 6gb of ram

[Tough Love]

Mid-range cards are similar to their high-end counterparts but with the expensive parts stripped off. RAM is expensive, and we don't really need more than 6GB now, so it makes perfect sense.

But it does mean that "future proof" is a term that does not apply to this card. Very shortly, 8 GB will be a minimum even for low end cards. The current #1 bestseller on Amazon is a Radeon 580 with 8GB, $190.

The more I think about it, the more I see 6GB as a killer misfeature for a midrange card. Soon there will a bunch of premium games with texture assets targeted at 8GB cache. Not fitting in cache can actually be more of a problem for performance that cut down processing units or lower clocks.