1660Ti Reviews Thread

[BFG10K]

https://videocardz.com/80100/nvidia-geforce-gtx-1660-ti-review-roundup

Roundup above.

It's a 1070 which is much cheaper and uses a bit less power, making it the first decent Turing. If I had still had my OG Titan I'd probably upgrade to this.

It was also a good move they didn't include any of those utterly useless Tensor cores with that "Doubly-Laughable-Super-Sampling".

 

[Hitman928]

https://www.overclockers.com/msi-gt...-turing-without-ray-tracing-and-tensor-cores/

There's one not in the roundup link.

 

[pauldun170]

Pretty much what everyone expected.

 

[Mopetar]

AnandTech review: https://www.anandtech.com/show/13973/nvidia-gtx-1660-ti-review-feat-evga-xc-gaming

 

[LTC8K6]

All that Tensor core talk and it doesn't even have them...

 

[coercitiv]

Surprise-surprise, the leaner Turing chuip delivers a decent price/perf proposition.

 

[LTC8K6]

Surprise-surprise, the leaner Turing chuip delivers a decent price/perf proposition.

Vega 56 does too.

 

[jpiniero]

All that Tensor core talk and it doesn't even have them...

Looks like nVidia swapped out the Tensor cores for FP16 cores.

 

[coercitiv]

Looks like nVidia swapped out the Tensor cores for FP16 cores.

If we are to compare die size it still looks odd, assuming the 284 mm² figure from TechPowerup is true. That's quite close to the 314 mm² of GP104.

 

[tviceman]

This is what the entire Turing lineup should have been. 33% more efficient and 40% faster than the Pascal chip it replaces with sane die sizes.

RTX came a generation too soon.

 

[ozzy702]

This is what the entire Turing lineup should have been. 33% more efficient and 40% faster than the Pascal chip it replaces with sane die sizes.

RTX came a generation too soon.

^ This. RTX should have come on a mature 7nm process. Hopefully the 3000 series cards are more well balanced and at better price points. NVIDIA took a big hit this year and that should play into their roadmap.

 

[happy medium]

So I read a few reviews, this is what im seeing.

1660ti is......
30% faster than a gtx 1060 6gb
20% faster than a rx 590
5% slower than a Vega 56
2% faster than a gtx 1070

*gains 11% more actual gaming performance when overclocked
*Power consumption 120 watts
*is quiet
* runs cool
* better price performance than gtx 1060 or 2060 and the fastest card per price performance.
* super card for 1080p, great card for 1440p

On newegg and other online shops have at least a handful of cards at $279

Nvidia will sell millions of these.

 

[coercitiv]

A look at Turing after fitness training.

 

[NTMBK]

Nice card. This is basically the mainstream card to recommend.

 

[crazzy.heartz]

It's the perfect sub $300 card. Hope it remains at this price point.

 

[mohit9206]

Nice card. This is basically the mainstream card to recommend.

Its not really mainstream. Mainstream would be something between 200 and 230 so probably 1660 non ti.

 

[IEC]

This is a better value, but Nvidia's best competition is still itself. A used GTX 1070 8GB is the better buy around $200 for many, especially if you have need/want for the 8GB vRAM. I got 1070 Extreme Gigabyte cards with 2.5 years warranty remaining for just over $200 each a few months prior. That $200 threshold is where you start losing the mainstream buyers.

If you need new, the recent sales at Newegg and Best Buy of Vega 56 cards at $249.99 AR is also a reasonable alternative, especially with 3 game bundle. I will be watching for more sales, because at that price Vega is actually attractive, warts and all.

For compute purposes, the RTX 2060 card offers better perf/$ and perf/W - so I doubt many will choose this card over RTX for F@H, BOINC, etc.

It'll be interesting to see if the 1660 non-TI can successfully fill the mainstream void.

 

[IntelUser2000]

I agree with IEC. The pricing is better compared to GTX 1070 right now, but that card has been available for 2.5 years. Its still much more expensive compared to the GTX 1060. Newegg.com and .ca have the price increase proportional to its performance increase.

In Canada, the RTX 2060 costs $70 more. If I really wanted a new card it actually seems worth it. Why wouldn't I want a card that has new features I could play around with for fun and is faster? But if you were shopping around, there was sometime in 2017 where the GTX 1060 could have been had for less than today nearly 1.5 years later. It's just overall stagnation.

There are rumors of a GTX 1180. I wonder if the GTX series is a fallback plan or something.

 

[SteveGrabowski]

So in the techpowerup testsuite aftermarket GTX 970 w/o OC is roughly equal to RX 570 and aftermarket GTX 1660 Ti is about 54% faster than RX 570 at 1440p. So in 4 1/2 years a roughly 55% gain in performance for $50 less with 30W lower power consumption. Kind of meh IMO. I'll pass. This card would need to be $239 or less to excite me. Hopefully Navi will deliver there.

 

[FiendishMind]

Doesn't this prove that the RT and Tensor cores in the first lot of Turing GPUs aren't responsible for the larger die sizes at all?

 

[coercitiv]

There are rumors of a GTX 1180. I wonder if the GTX series is a fallback plan or something.

At first glance the 284 mm² area of TU116 suggest there's more in that chip than we see in 1660Ti.

 

[LTC8K6]

Doesn't this prove that the RT and Tensor cores in the first lot of Turing GPUs aren't responsible for the larger die sizes at all?

It looks like the Tensor cores were replaced with FP16 units...

https://www.gamersnexus.net/images/media/2018/gpus/2080-ti/arch/sm-architecture-block-diagram.jpg

 

[FiendishMind]

It looks like the Tensor cores were replaced with FP16 units...

https://www.gamersnexus.net/images/media/2018/gpus/2080-ti/arch/sm-architecture-block-diagram.jpg

Even so, I mean look at the spec differences between TU106 and TU116, the difference between the die sizes isn't much more than the difference in CUDA core count and TU106 has a larger memory bus width to boot.

Also if the Tensor cores were doing double duty as FP16 units then they weren't nearly as useless as they may have seemed.

 

[IntelUser2000]

Also if the Tensor cores were doing double duty as FP16 units then they weren't nearly as useless as they may have seemed.

The ALUs used for the tensor operations are probably similar to the ones used for 16-bit half precisions operations so its likely just repurposed.

Mobile GPUs supported 16-bit precision compute to improve perf/watt, and desktop GPUs started using it. There are games that use the half precision format for certain aspects already.

 

[linkgoron]

At first glance the 284 mm² area of TU116 suggest there's more in that chip than we see in 1660Ti.

It has 50% more transistors than the 1060, for 40% more performance.

Anyway, I'd say it's essentially what was expected. Similar performance to the three year old 1070 for $100 less (2GB less memory). Perf/watt is great. However, as with the whole Turing line, the price is just too high. Worse perf/$ than the 580 and 570, similar to the current 1060. If you've already bought a card for $300+ in the last three years - this is essentially a side-grade. Vega 56 at $275 (and especially $250 AR) is IMO a better buy, but I'm not sure AMD can make any profit with that price.

Spoiler