Nvidia Post-Volta (Ampere?) Rumor and Speculation Thread

Status
Not open for further replies.
jlippo said:
They did add new features which can be used with rasterization. (Mesh shaders, texture shading stuff etc.)

If they reduce possible limitations of current RT core we might see decent improvement with same amount of RT units and ALUs.
Click to expand...
Current gaming RT implementations are not limited by RT core performance or capabilities, as can be seen from performance comparisons between gaming RT gains on h/w with RT cores against gains on the same h/w in professional RT applications. Thus there really is no reason to improve RT cores specifically, without improving the rest of gaming / hybrid RT pipeline - unless you're only aiming at professional rendering niches or fully path traced games like Q2RTX and Minecraft RTX - which I'm fairly sure will be in the minority during the whole next gen console h/w cycle.
 
DegustatoR said:
Current gaming RT implementations are not limited by RT core performance or capabilities, as can be seen from performance comparisons between gaming RT gains on h/w with RT cores against gains on the same h/w in professional RT applications. Thus there really is no reason to improve RT cores specifically, without improving the rest of gaming / hybrid RT pipeline - unless you're only aiming at professional rendering niches or fully path traced games like Q2RTX and Minecraft RTX - which I'm fairly sure will be in the minority during the whole next gen console h/w cycle.
Click to expand...
I'm more worried about performance pitfalls due use of Cuda cores in cases in which might be better if RT core could handle it by itself.

Currently instancing seems to be limited to single layer before needing to use Cuda cores during tracing etc..

Hybrid pipelines will be the way forward for now and any flexibility that allows RT part not swamping rest of the GPU should be good.
 
Could you get the stated performance increase (70-75%) in Indiana's Big Red 200 if Ampere were in fact at 10nm instead of 7nm?
It is also mentioned that Big Red 200 gained an additional 2 petaflops of performance even though it uses a smaller number of GPUs than the Volta V100 based design. The reason for going with a smaller number of next-generation GPUs is simply because they offer 70-75% better performance than existing parts and by that, we are comparing it with Volta-based Tesla V100 GPUs as no Tesla GPUs based on the Turing GPU architecture, aside from the Tesla T4, exist.
...
In a previous interview, NVIDIA's CEO, Jensen Huang, had confirmed that the majority of the orders for their next-generation 7nm GPU will be handled by TSMC while a small portion will be sent to Samsung for production.
Click to expand...
https://wccftech.com/nvidia-next-gen-ampere-gpu-75-percent-faster-existing-gpus/
 
DavidGraham said:
This didn't age well. XSX supports Machine Learning through RDNA2 CUs now, which means AI upscaling will be a real thing at least on Xbox. Which means this will be thriving very well on PC.
Click to expand...
Assuming anyone creates an AI upscaling system (that doesn't violate any Nvidia patents) on Xbox and it's actually used by developers.
 
DavidGraham said:
This didn't age well. XSX supports Machine Learning through RDNA2 CUs now, which means AI upscaling will be a real thing at least on Xbox. Which means this will be thriving very well on PC.
Click to expand...
Remains to be seen. Using GP SIMDs for AI upscaling can be both not fast enough and not make much sense as this way you're taking away the computational power from actual rendering.
 
DLSS is post-processing. So it has the same limitations.

Microsoft said that the Xbox Series X would need additional 13TFLOPs to archive the same performance with RT. Doesnt sound that RDNA2 is even close to Turing's RT Cores (nVidia said that RT Cores are 10x faster than using the shader cores)...
 
RDNA 2 fully supports the latest DXR Tier 1.1 standard, and similar to the Turing RT core, it accelerates the creation of the so-called BVH structures required to accurately map ray traversal and intersections, tested against geometry. In short, in the same way that light 'bounces' in the real world, the hardware acceleration for ray tracing maps traversal and intersection of light at a rate of up to 380 billion intersections per second.

"Without hardware acceleration, this work could have been done in the shaders, but would have consumed over 13 TFLOPs alone," says Andrew Goossen. "For the Series X, this work is offloaded onto dedicated hardware and the shader can continue to run in parallel with full performance. In other words, Series X can effectively tap the equivalent of well over 25 TFLOPs of performance while ray tracing."
Click to expand...
https://www.eurogamer.net/articles/digitalfoundry-2020-inside-xbox-series-x-full-specs

For me it sounds for the whole "hardware acceleration" part. But english is not my native language.
 
troyan said:
For me it sounds for the whole "hardware acceleration" part. But english is not my native language.
Click to expand...
Yeah you might be right but could be misinterpretation from the source context as well. Could very well be true.
 
DavidGraham said:
This didn't age well. XSX supports Machine Learning through RDNA2 CUs now, which means AI upscaling will be a real thing at least on Xbox. Which means this will be thriving very well on PC.
Click to expand...
ML memes take many forms, shoddy upscaling is but one of them.
 
DavidGraham said:
Yes, but you can bet AI upscaling will take center stage, especially to mitigate the impact of RT.
Click to expand...
I doubt they will be able to do upscaling in a classical sense due to at least 10 bit precision requirements for HDR rendering. 8 or 4 bit ints can be used only for some very specific things in graphics, which can tolerate such low precision.
 
Status
Not open for further replies.

Similar threads

A
NVIDIA Morgan Stanley Webcast: Sony, G80 and 80/65nm
2
Replies
25
Views
13K
Nelsieus
N
Back
Top