Nvidia Reportedly Working on GTX 1660 Ti with 1536 CUDA Cores

[WYP]

Nvidia's sub-RTX Turing graphics card.



Read more about Nvidia's rumoured GTX 1660 Ti and its leaked specifications.

 

[NeverBackDown]

If it's $250 roughly with 1070 level performance that'd be pretty impressive.

 

[JeffDee]

If it's $250 roughly with 1070 level performance that'd be pretty impressive.

I don't think it's that impressive, and I would be VERY surprised if the price tag of the rumoured 1160ti is 250$ UNLESS AMD has Navi on the table.

Naming it a "Ti" would let them market it as premium over a regular card, and I'd be surprised if this card was priced under 300$.

Which also would put a regular non-ti 1160 in a very terrible place price/performance wise. A regular 1160 would be just slightly better than previous 1060? for about the same price? Very cool Nvidia, nice technological progress!

Now, if the 1160ti was 229$ with regular 1160s( 6GB ) 199$ we'd be back to TRUE Mid range pricings and that's where there would be a reason to be excited.

 

[Peace Ð]

It would be a strong recommendation if it at least supported DLSS DLSS is the real game changer, not RayTracing...

 

[Piskeante]

this is a good movement IF the price is on the spot.

RTX 2060 FE is more or less a 1070ti. Nowadays, customs RTX 2060 at 1845mhz boost are ahead of GTX 1070ti.

i actually own an MSI RX 580 gaming X 8gb and almost all games run at ultra detail (heavily oced, btw 1490mhz core 2250 mem) at almost 60fps (not always to be honest, but i will definetely want to see better image quality than 60fps. i'm perfectly fine with 50.

i feel that this RTX 1660 ti will be the level of performance of a GTX 1070 without a doubt. Now, the price should be around 250€-279€ for Europe. In USA, prices are way lower, so expect around 229$.

i don't think that this card has a lot of future. it's diffficult to recomend buying this card if you have a GTX 1060 6gb or an RX 580 8gb. The increase in performance would not make it a bargain IMO.

The bang for the buck right now, is , without a single doubt, the RTX 2060. Custom cards can get (heavily oced tbh) to the performance of a 2070 stock. That's the deal.

 

[grec]

I personally seriously doubt this card is coming, I think more cuts/VRAM configs of the RTX2060 could easily fill in the space in the market immediately below the $350 mark, Turing hasn't really improved over Pascal's price/perf and besides a select few games can't really improve on it much at all while still on the same node with a wider arch so I don't see a reason for them to go much beyond Pascal rebrands for the bottom of their stack, otherwise it'll be essentially a GTX1060 but with more expensive memory, a probably more expensive die(Especially if it's unique to filling in this gap), and barely any performance/feature change.

 

[AlienALX]

I personally seriously doubt this card is coming, I think more cuts/VRAM configs of the RTX2060 could easily fill in the space in the market immediately below the $350 mark, Turing hasn't really improved over Pascal's price/perf and besides a select few games can't really improve on it much at all while still on the same node with a wider arch so I don't see a reason for them to go much beyond Pascal rebrands for the bottom of their stack, otherwise it'll be essentially a GTX1060 but with more expensive memory, a probably more expensive die(Especially if it's unique to filling in this gap), and barely any performance/feature change.

I would say it's coming. What else would they do with all of the dies with failed Tensor and RT cores?

 

[grec]

The RT & Tensor cores are part of the SM, generally you can't fuse them off, they'd have to be disabled in software similarly to how they disable DP cores(Or they'd have to spin a completely new die with a different SM design to Turing), besides there's only a single RT core and 8 Tensor cores per SM, the chances of one of those getting a defect but the rest of the SM being fine is slim. Essentially, there's no real economic benefit to disabling RT/Tensor cores as it's unlikely you'd get any sizable amount of chips with just those specific relatively small parts of an SM not functioning, it'd be purely for product line segmentation, which isn't something they seem to want, nor do I see a reason for them to want it. Sure, then they can keep RTX as a premium feature set, but I don't see why they'd respin a TU106 die for that, especially when the RTX2060 is already a heavily defected die by current 14nm yield standards.