Nvidia GeForce RTX 3080 spotted in drivers and benchmarks

[BigBANGerZ]

I don't believe a 3080 class GPU will get anywhere near the current 2080Ti pricing, the only way that would happen is initial pricing gorging due to supply issues or Nvidia fundamentally not intending to sell that many and looking to reap the highest margins.

3080 is also within possible Big Navi performance, they would open themselves to potential embarrassment from AMD if they price it incorrectly. Sure they could drop prices but then they expose themselves for the price/margin gorgers that they are to their own customer base no less.

[lumireleon]

OKAY.I would wait for intel to show off their HPG, who knows what Raja is upto after knowing he will make them at TSMC 6nm

[Zhaoman]

I don't believe a 3080 class GPU will get anywhere near the current 2080Ti pricing, the only way that would happen is initial pricing gorging due to supply issues or Nvidia fundamentally not intending to sell that many and looking to reap the highest margins.

3080 is also within possible Big Navi performance, they would open themselves to potential embarrassment from AMD if they price it incorrectly. Sure they could drop prices but then they expose themselves for the price/margin gorgers that they are to their own customer base no less.

I really hope you're right! I hope AMD can deliver with Big Navi and help push down prices for mainstream GPUs like they did with Zen and CPUs. I picked up a Ryzen 2600 for £112 a few months ago; a 6 core-12 thread CPU probably unthinkable if Intel were still milking it. GPU pricing is perhaps within the realms of being non-gouging up to the £300 mark (i.e. 5700 and XT) and then Nvidia just runs away with it from £300-£1200+ with no competition. It's sad what the market has become. Given the performance gain rumours about Ampere and Big Navi, and hopefully with a bit of healthy competition across the whole stack, my hope is that £1000+ 2080 becomes mid-range level of performance for a mid-range price in the not too distant future. And by mid-range I mean at least sub-£300 if not sub-£250. Is that too much to ask? Honestly not too much to ask at all if you consider the price of 5600XT/2060 currently and rumoured IPC gains from both camps. I am still expecting to be disappointed though.

[kalniel]

, my hope is that £1000+ 2080 becomes mid-range level of performance for a mid-range price in the not too distant future. And by mid-range I mean at least sub-£300 if not sub-£250. Is that too much to ask? Honestly not too much to ask at all if you consider the price of 5600XT/2060 currently and rumoured IPC gains from both camps. I am still expecting to be disappointed though.

Nah you'll be happy. That's about Xbox series X level performance so they can make cheaper chips in that range already, just a case of manufacturing enough of them so perhaps there in a year or so.

[Rubarb]

As much as i don't like AMD drivers I do hope big navi or what ever it's called trashed this..... not because i'm an amd fan but it will make prices fall

[Quartz]

So the 3080 Ti and the 3080 have the same PCI IDs? Doesn't smell right.

[Xlucine]

I wonder whether AMD have any technology along the lines of Nvidia's DLSS 2 in the pipeline? That would definitely improve the image quality of upscaled games on consoles.

Yep, been available for a year now:
https://hexus.net/tech/news/graphics...-ris-examined/

[kalniel]

Yep, been available for a year now:
https://hexus.net/tech/news/graphics...-ris-examined/

RIS is a non-ML approach (which I prefer in general), but RDNA2 is also bringing ML inference acceleration and resolution scaling was given as an example use of that. (See the Xbox series X hot chips slides)

[MrJim]

Yep, been available for a year now:
https://hexus.net/tech/news/graphics...-ris-examined/

Thanks for the link, that's interesting. I couldn't find a comparison of RIS with DLSS 2 though, which is a shame (there was a Techspot article comparing RIS to DLSS 1, but DLSS 1 was a bit rubbish I think).

[ik9000]

Thanks for the link, that's interesting. I couldn't find a comparison of RIS with DLSS 2 though, which is a shame (there was a Techspot article comparing RIS to DLSS 1, but DLSS 1 was a bit rubbish I think).

DLSS1 was very rubbish. DLSS2 is much better. I'm not sure it's an apples with apples comparison RIS to DLSS in terms of how they work, but I might be wrong. I thought DLSS needed an internet connection - or have I got that wrong?

[CAT-THE-FIFTH]

DLSS1 was very rubbish. DLSS2 is much better. I'm not sure it's an apples with apples comparison RIS to DLSS in terms of how they work, but I might be wrong. I thought DLSS needed an internet connection - or have I got that wrong?

Most of the comparisons are games with AA such as TAA,which essentially softens the native image. What DLSS is doing is general upscaling on a per scene basis(this is what the machine learning part of it is being used for),but using edge detection to add extra sharpening to parts of the image.

[ik9000]

Most of the comparisons are games with AA such as TAA,which essentially softens the native image. What DLSS is doing is general upscaling on a per scene basis(this is what the machine learning part of it is being used for),but using edge detection to add extra sharpening to parts of the image.

But I thought it could only do that for scenes where the GPU recognised it had done the computation before hence it relied on cloud server or something. This all based on one YouTube video a while back so could well be barking up the wrong tree. I'd be lying if I said I'd given it my full attention

[kalniel]

But I thought it could only do that for scenes where the GPU recognised it had done the computation before hence it relied on cloud server or something. This all based on one YouTube video a while back so could well be barking up the wrong tree. I'd be lying if I said I'd given it my full attention

The ML models are trained/created using cloud-compute by Nvidia (effectively finding ways to emulate the look of a supersampled reference set) then downloaded (as part of drivers) to your GPU since running a model based on given inputs is a lot quicker than training it.

There's no realtime uploading and creating models on the fly.

[CAT-THE-FIFTH]

It appears memory is a real problem for Ampere:
https://www.igorslab.de/en/nvidia-am...l-in-the-dark/

It's consuming a ton of power and is clocked to the very edge! I wonder if Micron screwed up??

[ik9000]

interesting, yeah I just read that too. Could this be another GTX480 debacle? On the upside in times of recession it could reduce your winter heating bill

[CAT-THE-FIFTH]

https://videocardz.com/newz/nvidia-g...-card-pictured





https://twitter.com/GarnetSunset/sta...16731378704391

3060 400
3070 600
3080 800
3090 1400

The prices have gone up over Turing for the top model.

The RTX2080TI was $999~$1199. RTX2070 was $499~$599.RTX2060~$350. Only the RTX3080 has the same price as the RTX2080.

Edit!!

What if the RTX3090 is a dual GPU card? The GTX590 and GTX690 were both dual GPUs? Would make the size make more sense!