• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

I want an RTX Addon Card!

Associate
Joined
22 Aug 2005
Posts
542
I think it would be a good idea for nvidia, to make a addon card that handles RTX seperatly, so we could get support for raytracing on the cards we have atm

If the 2080 are not going to be a massive leap performance wise on the 1080ti, I think i'd rather stick with what I have, and get support for rtx with a simple and affordable rtx addon card, cost would be I dunno £150 ish, I wouldnt want to pay more than £200 for it though

I'm thinking something like the physx cards used to be (maybe it could even handle physx aswell)

slap the card in, and you'd get RTX on your 1080 etc for a lot cheaper than buying a new card (maybe it could even work with amd cards?)

I think it would help get widespread support for rtx a lot quicker, as its going to take a long time for a lot of people to get RTX cards with them costing so much

I love the technology, but I cant see it getting much support very soon, if they don't do something like this
 
From my reading , I've gathered that RT cores and Tensor cores are apart of the die so its doubtful you'll get any separate cards . physX was destined seperatly where as here its all integral
 
I'd buy for £150-£200 an add on card that could give 2080ti RTX features to a 1080ti. They won't do it though because they want the new series to sell.

I guess it will all depend on the 4k performance of the 2080, If its not much better than the 1080 then I cant see them selling many 2080's and releasing an addon card for RTX would be a much faster way of getting widespread support for RTX
 
No chance, they want to milk the 20 series as much as possible. Also to stop people using it with AMD cards and save themselves £100s...
 
Last edited:
Ok so you get a add on card, how are you suggesting you connect it to your existing card ie: a 1080ti. Well cannot use standard SLI bridge, nor a HBSLI bridge, as neither if those have anyway near the throughput needed. Could use the PCIE bus, but even that probably isn't fast enough, bear in mind the speed the RTX chip talks to the various cores of the chip, all being on one piece of silicon and all that. Even NVIDIA needed to use their NV link for fast enough speed transfer for multi GPU and of course you can't do that on a 1080ti.

Anyway that I expect is the reason we never saw or will now never see an add on card.
 
Depends on how hybrid their RT implementation is - in theory all an addin card would need is a copy of the world geometry and information relevant to that such as lighting values and material properties and only needs to hand back to the main card the results relevant to screen space - gets a bit more complex if you've intersections that are across processor boundaries but I'd assume there are ways to avoid that.
 
Those who want this add on card have fallen for the RTX hype. It's gonna end up just like PhysX did.
Well at least then we could just opt out of buying the add on card when it becomes defunct. :)

I agree there isn't the bandwidth for real time raytracing on a seperate card, and NVIDIA have too much "business sense".
 
The cost of an add-on card will be more thsn the price increase from Pascal, and you will will 50% slower in regular games.

But in any case, basically impossible because RTX core rely heavily on the CUDA cores and a tight integration between regular rasterisation And raytracing
 
It is a shame we can't make better use of those extra PCIE slots. If the industry got together I'm sure they could come up with something to benefit everyone. I always get the feeling that we never quite get the best we can out of the power our PCs have.
 
Well would it really stretch the bandwidth between cards?

Let’s say for example that the average person has a monitor of 60hz and the gpu limited itself to that. A 1080ti would generally best that by some margin so there is spare time between frames so to speak.

the cpu has to send the gpu each set of data for each frame, if there is spare time between frames, could it also not round trip data to a ray tracing card that can process the scene to pass back to the gpu for rasterisation.

I also thought that in terms of programming, you make high level raytracing api calls and it’s upto the hardware subsystems to decide how to implement the api call. What would prevent a raytracing card being present to run the RT calls and pass it back the info in the same way a sound card would intercept the sound api calls ?
 
Well would it really stretch the bandwidth between cards?

Let’s say for example that the average person has a monitor of 60hz and the gpu limited itself to that. A 1080ti would generally best that by some margin so there is spare time between frames so to speak.

the cpu has to send the gpu each set of data for each frame, if there is spare time between frames, could it also not round trip data to a ray tracing card that can process the scene to pass back to the gpu for rasterisation.

I also thought that in terms of programming, you make high level raytracing api calls and it’s upto the hardware subsystems to decide how to implement the api call. What would prevent a raytracing card being present to run the RT calls and pass it back the info in the same way a sound card would intercept the sound api calls ?

The thing is, the raytracing data is used to act directly on individual pixels, a soundcard is outputting sound via its own output, a physx card is passing object movement data back to the CPU, its only raytracing that needs direct access.to the framebuffer so needs as little latency as possible.

Ray tracing and rasterisation happen in parralel and are then blended also partly using the tensor cores. Its all interdependent on the three working together at effectively the same time for the final pass.
 
There's a reason Ageia got bought out by NVIDIA. RTX is dependent on Tensor cores, so a separate discrete solution isn't likely to work very well and would also still be expensive. I'd also be surprised if many wanted this, the last thing most builds need is more clutter.
 
I think it would be a good idea for nvidia, to make a addon card that handles RTX seperatly, so we could get support for raytracing on the cards we have atm
You're right it would, but it will never happen because it would just encourage people to buy AMD (or Intel) GPUs instead of Nvidia ones.

To put it in perspective, PhysX cards used to be standalone cards that worked with Nvidia/ATi/Intel/etc until Nvidia bought out the company that made them.
 
It's simply won't make any (business) sense for Nvidia to do this, as the standalone price of such card will be expensive leading to them sitting on shelves, an also an alternative way for the consumers opting out of paying the high RTX tax on the new gen Nvidia cards.

Nvidia don't want people to grab a Ray-tracing cards to use with 1080ti, they want people to ditch their 1080ti and drop £1200 on the RTX2080ti instead.
 
To put it in perspective, PhysX cards used to be standalone cards that worked with Nvidia/ATi/Intel/etc until Nvidia bought out the company that made them.
Even the Nvidia graphic cards could works as PhysX card running with a ATI card as primary at hardware level, but Nvidia chose to disallowed it/blocked it at software level, pretty sure it is the same logic as to why they don't let their desktop GPUs run/support Adaptive Sync despite it should definitely support it considering their GPUs have no problem using Adaptive Sync (which they still called Gsync) on some of the gaming laptops.

If Nvidia did make a dedicated Ray-tracing card, we can bet that it will only be compatible using with Nvidia cards, even IF the Ray-tracing card IS compatible with AMD/Intel GPU.
 
Back
Top Bottom