AMD Navi 23 ‘NVIDIA Killer’ GPU Rumored to Support Hardware Ray Tracing, Coming Next Year

[Hedge]

you’re the one making things up. Like your previous post about the guy buying a 3090 to play path of exile and turn on RTX.

Poe doesnt even have raytracing. I mean where do you even come up with these things?

Maybe the point is that not that many games support raytracing anyway. Next gen console ports will but not the majority of pc games right now.

 

[Nablaoperator]

https://forums.overclockers.co.uk/posts/34011222

i was more interested in knowing why these 2 games are being called true RT games..
from what i understand RT performance is heavily dependent on geometry complexity and thats why these 2 games are probably not the best benchmarks to conclude the potential of Ampere RT.. but am all ears for an argument that suggests otherwise

 

[KillBoY_UK]

i was more interested in knowing why these 2 games are being called true RT games..
from what i understand RT performance is heavily dependent on geometry complexity and thats why these 2 games are probably not the best benchmarks to conclude the potential of Ampere RT.. but am all ears for an argument that suggests otherwise

As nothing in those 2 RTX games are rastered, all parts of the rendering is calculated from the ray bounces. While most other RTX games are Rastered with some effects ray traced

 

[Nablaoperator]

nothing in those 2 RTX games are rastered

occlusion culling is still done through traditional techniques.. even in Quake 2 RTX i believe.
but anyways, is there any reason to believe that just because other lighting effects are rastered it would cause a delay in pipeline while rendering the final frame?

 

[five8five]

And that would be entirely the point no arguments, no whataboutism, straight up beating Intel.

Definitely, but a double-edged sword though. People hoping for RDNA2 performance clues will be also be upset.

 

[KillBoY_UK]

occlusion culling is still done through traditional techniques.. even in Quake 2 RTX i believe.
but anyways, is there any reason to believe that just because other lighting effects are rastered it would cause a delay in pipeline while rendering the final frame?

What ? you asked why are those two games considered to be "real" RT games over hybrid's . This is a different question entirely but I would really think it be down to the engine of the order of things and weather it was Raster or path traced at its core. You obviously can't calculate lighting until you have scene geometry

 

[gpuerrilla]

true RT games ( pathtraced ) like quake 2 rtx and minecraft RTX show way bigger differences. So the reality is that in raytracing only games differences are that big if not even bigger.

in hybrid rt + raster yes, it’s lower.

Yes, in your context -so depends on your game, but as you can see from the responses they are few and far between (2 games - really?) - but in a nutshell there is a video on this explaining it out there, it will get better over time once the software catches up with the 30 series hardware. AMD get panned for this type of tactic, as per 5700XT which was the typical term 'fine wine' where non-AMD users slate it for "well it doesnt do that out of the box, theres no point in being better six months later". Out of the box the 3080 isnt a huge leap in performance over the 2080Ti no matter how hard you massage the figures to be otherwise.

 

[five8five]

• Navi21 3080 competitor will have 16GB VRAM and likely be as fast as 3080 in most things, but potentially not all. On sale around end November 2020 and priced at $549-$699.
• Navi21 3070 competitor will have 16GB VRAM and 8GB VRAM options and should be faster than 3070 overall. On sale around December 2020 and priced at $449-$599.
• Navi22 full-fat should have 12GB VRAM and be RTX 2080 Super performance or higher. Launch Q1 2021 for $379-$449.
• Navi22 cut-down should have 10-12GB VRAM and be RTX 2070 Super performance or higher. Launch unknown, priced around $299-$379.
  
• Navi23 may have 8GB VRAM and be RX5700 performance with much lower power draw. Launch Q2-Q3 2021, priced around $179-$279.

If these dates are accurate, I can understand why Nvidia have gone with a "fill yer boots strategy".

 

[Nablaoperator]

What ? you asked why are those two games considered to be "real" RT games over hybrid's . This is a different question entirely but I would really think it be down to the engine of the order of things and weather it was Raster or path traced at its core. You obviously can't calculate lighting until you have scene geometry

I consider true RT as something which is capable of occlusion culling, till the time that doesnt happen they are all hybrid to me

Regarding the second question, i was trying to think aloud as to why folks are considering these 2 games as a better benchmark for RT performance, unless they believe that the "rasterized" lighting effects are somehow stalling the RT output (theres no way they'd be bottlenecking RT only stalling looks possible).. i will be needing a proper 101 with examples to follow this line of reasoning.. am just trying to enhance my knowledge of RT and how it would apply to other GPUs as well.. particularly rDNA2

 

[gpuerrilla]

If these dates are accurate, I can understand why Nvidia have gone with a "fill yer boots strategy".

I dont know how this amazes you. The flagship nvidia card was launched first, then the 3090 a week later. All a few weeks after Jensens announcement. The 3070 is not out yet. So why are you gurning at the lower stack AMD cards being December+ ?

NAvi21 being available in November is what they said all along..

 

[Richdog]

I struggle to understand anyone who had a decent time with a 2080ti.

It's pocket change for some, but still throwing money away for what it is.

The 2080Ti was the fastest GPU from the beginning to the end of its generation and still retained 70-80% of its value almost 2 years later. What's not to understand?

 

[Rroff]

occlusion culling is still done through traditional techniques.. even in Quake 2 RTX i believe.
but anyways, is there any reason to believe that just because other lighting effects are rastered it would cause a delay in pipeline while rendering the final frame?

Quake 2 RTX largely still uses the PVS data for culling - though some is done on the shaders for the RTX renderer but that is more incidental.

Even Q2 RTX doesn't have full scene, per pixel, rendering via RT but that isn't so important at a game level.

Mixed techniques tends to be inefficient as even partial ray traced features often have much of the overhead associated with a fuller implementation.

 

[queamin]

AMD Big Navi Leak: If Nvidia Ampere Wins, it’s a Pyrrhic Victory

Executive summary:

• Navi22 full-fat should have 12GB VRAM and be RTX 2080 Super performance or higher. Launch Q1 2021 for $379-$449.
• Navi22 cut-down should have 10-12GB VRAM and be RTX 2070 Super performance or higher. Launch unknown, priced around $299-$379.
  
• Navi23 may have 8GB VRAM and be RX5700 performance with much lower power draw. Launch Q2-Q3 2021.
• Priced around $179-$279.

This worry's me, Navi22 full-fat Launch Q1 2021 and Navi23 looks very late Launch Q2-Q3 2021, if Nvidia gets their 3060/50 out late Oct/Nov time.

Navi22 cut-down should have 10-12GB VRAM and be RTX 2070 Super performance or higher. Launch unknown, priced around $299-$379.
I hope this is launched this year.

 

[Besty]

'fine wine'

Jay1Centz looks at his free £1,650 3090 card, drinks Kool-aid, turns to the camera and says with a tear glimmering in his eye:

"Nvidia has an amazing year on year track record of improving launch performance of all it's hardware, i got no worries at all guyz."

 

[gpuerrilla]

That guy is not worth it if you want an impartial opinion on tech components. Many other tech channels that didnt receive sample cards are highlighting how hard it is to get a card, so if you have the blinkers on its business as usual.

Reality is the availability is still ****-poor no matter how hard shillcents ignores it.

 

[humbug]

AMD Big Navi Leak: If Nvidia Ampere Wins, it’s a Pyrrhic Victory

Executive summary:

• Performance of RDNA2 will not be bandwidth-hungry, GDDR6 is enough to fuel it. This could be due to a new cache pool.
• Navi21 3080 competitor will have 16GB VRAM and likely be as fast as 3080 in most things, but potentially not all. On sale around end November 2020 and priced at $549-$699.
• Navi21 3070 competitor will have 16GB VRAM and 8GB VRAM options and should be faster than 3070 overall. On sale around December 2020 and priced at $449-$599.
• Navi21 will have a 32GB, 256-bit model that competes well with the Nvidia Quadro A6000 with full and uncut professional features. Launch Q1 2021 and priced at $1999-$2999.
  
• Navi22 full-fat should have 12GB VRAM and be RTX 2080 Super performance or higher. Launch Q1 2021 for $379-$449.
• Navi22 cut-down should have 10-12GB VRAM and be RTX 2070 Super performance or higher. Launch unknown, priced around $299-$379.
  
• Navi23 may have 8GB VRAM and be RX5700 performance with much lower power draw. Launch Q2-Q3 2021, priced around $179-$279.
• Ray Tracing should be as good or better as Turing.
• There will likely be some pseudo-DLSS algorithm.
• Machine learning will play a big part in the new strategy
• The driver bugs with RDNA1 in later 2019 were real and due to hardware-level issues
• AMD were pleasantly surprised how bad the 3090 improvement is vs the 3080, as it removed a big performance concern from their minds. They were expecting it to be far better.

On sale around end November 2020 and priced at $549-$699.

That sort of crap, where they claim to know the prices and give you a huge margin between the lowest and highest price, they don't know #### all and are taking a wild educated guess, its unlikely to be less than $550 and unlikely to be priced higher than the $700 3080, so $550 to $700, well we could have all guessed that.

Navi21 3080 competitor will have 16GB VRAM and likely be as fast as 3080 in most things, but potentially not all.

Not as fast or just as fast... well yeah! again, he has no idea.

For weeks he has been banging on about having RDNA2 details that he didn't want to divulge yet, was this it?

I have had the feeling for a while this guy was making this crap up, like some politician being deliberately convoluted and vague saying nothing at all.... now i'm connived of this guy is fake.

 

[Nablaoperator]

Quake 2 RTX largely still uses the PVS data for culling - though some is done on the shaders for the RTX renderer but that is more incidental.

Even Q2 RTX doesn't have full scene, per pixel, rendering via RT but that isn't so important at a game level.

Mixed techniques tends to be inefficient as even partial ray traced features often have much of the overhead associated with a fuller implementation.

What actually causes this:
Bandwidth issues
Cache misses
RT cannot run concurrently with other lighting calculations
RT lighting cannot be merged with other lighting calculations while applying values to the triangle
or something else

Also, unrelated to the above, was wondering if 2 mirrors set-up in parallel (one facing the other) can crash RT by running into a potentially unstable solution (just one of those contrived thought experiments )

 

[melmac]

Also, unrelated to the above, was wondering if 2 mirrors set-up in parallel (one facing the other) can crash RT by running into a potentially unstable solution (just one of those contrived thought experiments )

It wouldn't cause any problem now because developers limit the number of Rays and the number of bounces to improve performance.

And by the time Ray Tracing is at a stage where hardware is powerful enough to run Ray Tracing without limits, it won't be a problem.

 

[humbug]

RDNA1 is defiantly bandwidth starved, any GPU overclock, or underclock makes little difference, barely anything. Any memory overclock scales almost 1:1.

 

[gpuerrilla]

RDNA1 is defiantly bandwidth starved, any GPU overclock, or underclock makes little difference, barely anything. Any memory overclock scales almost 1:1.

Cache that.