• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Discussion Nvidia Blackwell in Q1-2025

Page 185 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
MrMPFR said:
OMM SW emulation = RTX only. Zero DXR 1.2 support by Radeon.
Click to expand...
Official info via Microsoft:
  • OMM = AMD support is planned on future hardware platforms. Please contact your AMD relations manager for more details
    • Intel = Intel is actively evaluating OMMs and looks forward to supporting the feature in the future
  • SER = AMD support for SER is available in this preview driver
 
basix said:
Official info via Microsoft:
  • OMM = AMD support is planned on future hardware platforms. Please contact your AMD relations manager for more details
    • Intel = Intel is actively evaluating OMMs and looks forward to supporting the feature in the future
  • SER = AMD support for SER is available in this preview driver
Click to expand...
I was talking about RDNA4. SER without reordering = useless.

Guess gfx13's and Xe3P's DXR 1.2 compliance is so much of a given that I forgot to mention it lol
 
Sure. But the statement clarifies that OMM is not planned to be supported by RDNA4. But there should be SER support (in some form), altough currently without any actual reordering.

But why having SW API support when not doing anything with it? SW based reordering (emulation) in the future? Same could be done for OMM (see Nvidia).
With SW emulation you do not get HW based efficiency gains. But you might catch the low hanging fruits (let's say the first 50% of the HW based potential).

Edit:
If reading the docs, SER consists mainly of two things:
- MaybeReorderThread (not supported by AMD preview driver)
- HitObject, seems to be supported by RDNA4
HitObject improves the flexibility of the ray tracing pipeline in general. First, common code, such as vertex fetch and interpolation, must no longer be duplicated in all closesthit shaders. Common code can simply be part of the raygeneration shader and execute before closesthit shading. Second, simple visibility rays no longer have to invoke hit shaders in order to access basic information about the hit, such as the distance to the closest hit. Finally, HitObject can be constructed from a RayQuery, which enables MaybeReorderThread and shader table-based closesthit and miss shading to be combined with RayQuery
Click to expand...

I think this is the latest AMD preview driver. The limitation with MaybeReorderThread still persists (MaybeReorderThread call on RDNA4 will simply yield a NOP instruction) and will likely stay like this.
 
Last edited:
basix said:
Sure. But the statement clarifies that OMM is not planned to be supported by RDNA4. But there should be SER support (in some form), altough currently without any actual reordering.

But why having SW API support when not doing anything with it? SW based reordering (emulation) in the future?
Click to expand...
Indeed it supports it but doesn't impact perf at all.

Why would AMD do that. They're in the business of dropping support for old cards fast and offering bare minimum support.

~80% perf gain on 5080. L2 + mem BW roughly the same between 4090 and 5080. 4090 looks like the outlier so something else is going on with 4090. Any guesses?
https://twitter.com/x/status/2028288491579211808
 
jpiniero said:

Looks like OEMs were indeed going to get updated Blackwell with 3 GB chips. This one talks about the 5070 Laptop 12 GB.
Click to expand...

The fact that the 5070 Laptop launched with 8GB VRAM in 2025 was already pretty ridiculous. It released with the same amount of VRAM as the GTX 1070 Laptop which launched 9 years earlier with a chip that has 1/3rd the amount of transistors.

As a comparison, 9 years prior to the GTX 1070 Laptop, Nvidia was launching the 8700M GT which had 256MB or 512MB VRAM. They went from 16x or 32x more VRAM to freezing capacity altogether.
 
I mean, you can release supers with just more VRAM and at this point nobody would complain nor argue about that fact (price and availability is another matter). But I think they'd be able to raise clocks with some extremely minor amount or indeed the same with SM's as well.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top