NVIDIA Brings DX-R To Pascal GPUs with new driver

Oh crap my bad I didn't realise the 20 series are affordable now, oh snap! they're not, nevermind.

 

affordability is so subjective.

 

It should be selectable in the games that support RTX according to nVidia. I too launched my game and found out I can't select it. So instead; I went to download the tech demo to see if I can get my RTX fix that way. I found out you need the Windows 10 - 1809 Update installed.

I am downloading that now, I want to see if the option withing Metro Exodus will be there when I am done. It will take me a while, very slow internet.

 

You're missing my point. DXR touches nearly every aspect of a graphics card. It builds a BVH representation of the entire scene and calculates the ray intersects in some degree of parallel to the raster environment. So you need geometry pipeline, you need ALUs, you need high speed memory, you need complex scheduling, you need basically everything on a graphics card. So a "DX-R accelerator card" would essentially just be a graphics card. It would also have a lot of the same problems that SLI/Crossfire have now as current methods of DXR essentially borrow data from the raster environment. So you'd need to do some crazy scheduling in order to send that data to and from the "accelerator card" perfectly to avoid stalls. This why SLI/Xfire are dying in general - the effort to schedule all the inter-frame dependencies aren't worth the payoff.

Also going back to your previous post - the idea of an accelerator card being a "nail in the coffin to RTX" is kind of redundant/unncessary. It basically misses the point of what RTX is. RTX is a collection of hardware/software that accelerates DXR. Any kind of 3D accelerator card would also feature it's own software based raytracing/intersect libraries and various hardware implementations to accelerate them.

There is definitely a line where something becomes unaffordable. When the CEO of the company selling the product issues a statement saying that sales are declining due to the price, I think we've crossed that line.

 

I mean you can say that about anything, everything is subjective blah blah blah, overpriced is overpriced!

 

Yeah but if someone released a DX-R accelerator card similar to the PhysX cards back in the day at an affordable price, I'm talking like $200 - $300 dollar mark, you really think people would be considering Ngreedias latest cards, that was my point, which I think you seemed to miss.

 

if i thought it was overpriced for my budget i wouldn't have bought one.

 

You got mugged mate but that's just my subjective opinion.

 

people were paying thousands of dollars for 1080ti not one year ago.
and now you complain about pricing?

 

I'll try this another way. Why isn't this same argument made with geometry? Why don't we have geometry acceleration cards that only render the geometry stage of the game that are $150-200 add-in boards? Or "Shading Accelerators" that only specifically accelerate the increasingly complex pixel/vertex and compute shaders?

 

luxary items mate, they are only unaffordable to people who don't want to buy them.

 

here go again! Nvidia always shows "news" techs, remember HairWorks, Gameworks, "our lastest cards works better", the reallity is they setup drivers to make old cards run worse, if people analist performance metrics they will see, i know AMD Rays will work better comparing graphics of same range with Nvidia, example: GTX1080 vs Vega64

grettings

 

Okay except the definition of affordable/unaffordable specifically mentions "the average person" - so this idea of re-framing around a subjective view point of an individual is moot. In this case I think we can narrow "average person" down to "average gamer" and I'd be willing to bet a 2080Ti that the average gamer finds the 2080Ti unaffordable.

 

Well the problem I see is the size of the cards, there's only so much room and currently a single card isn't capable enough to do Raytracing at decent framerates especially when we're talking 4k, an accelerator card could bridge that gap, and my point is they did it for PhysX so why not DXR?

 

the price point was on point though, its quite faster than the 1080ti and has more/better features. IF they sold it for cheaper than a 1080ti they would be shooting themselves int he foot. after all the average gamer can still buy a 1080ti at any time.
i bet the average driver wouldn't purchase a sports car either but they sell those as exorbitant prices too. there is a budget for everyone and holding a grudge because it isn't in your budget isn't a downside to something.

 

Biggest difference here is that the raytracing and dlss is running through tensor cores themselves whereas the 10xx series does not have any.

Would be interesting to see how Dlss and ray tracing peformance difference between 2080ti and the 1080ti, straight from the gpu themselves, but no way on the rtx, its running on tensor cores, 1080ti does not, Huge difference here.

 

One day your grandson/daughter can have a Ti 10000 for just $1 000 000. All thanks to people like you, more money than brains. The only reason this new price model is being used are exactly because of people like you. "It's new, let's throw more money at it, wheeeee, I got the latest tech! respect me! whooo" If you only knew how much more money they save as the processing nodes shrink, they get more GPU's / waffer manufactured, they save money, while you, with more money than sense, pay these high amounts when you should have gotten it for the same amount or cheaper.

You are damn right that the average peasant is pissed. We could always afford mid-range GPU's but they aren't remotely near the mid-range mark anymore.

The previous generation products should become cheaper while the latest generation replaces them. That way, the average gamer can still keep buying the new cards while the ones below average can have good performance at a low price.

Just to add, with inflation, a TITAN today should cost the consumer around $750. Take a guess what the price isn't?

 

Nvidia has rubbed up against the physical reticule limit of processes in the past so I don't think that really invalidates my question of why they didn't offload geometry/shading/etc. That's not to mention that 7nm, a double shrink, gives ample space regardless.

PhysX cards were simple math accelerators and Nvidia did eventually bring it on board the GPU because having it off the GPU doesn't make much sense. It complicates the scheduling, introduces stalls, provides almost no benefit. DXR is the same way. The BVH instance is using all the same geometry the raster is - so why are we building it twice - once on the GPU and once on the "accelerator card"? And if we're not building twice - we're just sending that data back and forth between the main card and slave card - what's the performance penalty of that? What's the scheduling complexity of that by the developer? By the driver? Does it provide a 1-1 performance benefit of having it on board? - We already know it doesn't for SLI/Xfire. So now we have a $300 add in board that increases the complexity of development and/or introduces a performance penalty vs having it on-board and you're only going to sell it to a vanishing subset of people because in a generation you get another ~300mm2 (7nm) to play with due to a die shrink. Not to mention advancements in intersect algorithms, RT cores, denoising (requiring less rays) etc - all to make 4K@60 with the feature a reality.

By Nvidia's own admission the price wasn't on point. The $/fps value gain over the 1080Ti is significantly worse than any prior generation. They sold the promise of features, 21 games when the cards launched - six months later, we have 6 actual titles with the technology. The 680 was 20% faster than a GTX580 at the same MSRP, with a bunch of value-add features and lower power envelope. Nvidia didn't go bankrupt. No one was saying "man nvidia really shooting themselves in the foot here"

I'm glad you're enjoying your RTX card and I don't want to take away anything from you on that. I don't think you were "mugged" like stereoman, I think you valued the features/performance it offered for the dollar more than me and others. That's great and I hope it's a great card for you. That being said it's pretty clear from the backlash/vitriol surrounding the release that the card doesn't offer great performance per dollar. From an objective $/performance metric it's clearly way worse than previous generations of Nvidia cards. I don't really think there is an argument here - some people aren't going to care about that, but in general I personally feel the card didn't meet the requirements of what I'd consider "affordable" and this is coming from someone who owned a GTX690 and basically every Ti/top tier model of video card Nvidia sold since the 6800Ultra.

 

thanks, it was only about 100 more dollars than the 1080ti strix so i felt it was a very good price for the card.
i just find it very ironic that something as complicated as RTX is supposed to be a cheap addition to the lineup according to people like stereoman. we literally paid for their research on the tech. which i quite like. i would love to see more games with it, and that means more people should buy it. I have a feeling the choice to allow DX-R on older cards despite the abysmal performance it will have is to make it easily accessible to indie devs. as far as the games that were promised to have it i have no interest in most of them so that i don't care about.
they may have said that to make gamers happy but look at the TITAN RTX's price and the incredibly small gains it gets over the 2080ti.
there are people who will buy that despite its cost and without competition this is what we get.

 

hah you don't even know me and you attack for no good reason at all?
YOU DON'T HAVE TO BUY ANYTHING. next time you go to the store stick to soups because it seems that with your budget that is all that fits in your price range.