UL releases 3DMark Mesh Shaders Feature Test for DirectX 12 Ultimate

Armenius

Fully [H]
Joined
Jan 28, 2014
Messages
23,694
https://videocardz.com/newz/ul-rele...t-results-of-nvidia-ampere-and-amd-rdna2-gpus

1613076533602.png


The DirectX12 Ultimate API brought four technologies: DX Raytracing, Sampler Feedback, Variable Rate Shading, and Mesh Shaders. UL had already released feature tests for ray tracing and variable-rate shading. The developer has now released a third DX12 test for Mesh Shaders. We have already had a glimpse of a new 3DMark test thanks to Raja Koduri. Intel is currently evaluating its unreleased Xe-HPG gaming GPU, which should debut later this year.

The Mesh Shaders test is relatively short. First, it records a baseline without mesh shading and then compares how much faster is the same scene with the technology enabled.

We have gathered some data from 3DMark featuring the latest graphics cards. Interestingly, NVIDIA Turing GPUs are actually faster in baseline test than NVIDIA Ampere, but the latter see a higher performance increase when mesh shaders are enabled. Overall we are looking at the following average performance increase:
  • NVIDIA Ampere: 702%
  • AMD RDNA2: 547%
  • NVIDIA Turing (RTX): 409%
  • NVIDIA Turing: 244%
Do note that we chose the best or close to best scores that were available at the time of writing. It is by no means an accurate representation of each architecture performance. It is only meant to provide a basic understanding of how fast each DirectX12 architecture might be.

Looks like this could be the solution to video cards hitting a wall in terms of performance when it comes to rasterization.
 

DejaWiz

Fully [H]
Joined
Apr 15, 2005
Messages
20,890
The Radeon RX 6000 series results, being so low, blow my mind. Could be a coding problem with the way 3DMark performs with the 6000 series, though.
 

auntjemima

Supreme [H]ardness
Joined
Mar 1, 2014
Messages
7,273
The Radeon RX 6000 series results, being so low, blow my mind. Could be a coding problem with the way 3DMark performs with the 6000 series, though.
Pretty much. I don't expect them to be on part with the 3080 or 3090, but I would definitely expect them closer.
 

auntjemima

Supreme [H]ardness
Joined
Mar 1, 2014
Messages
7,273
Which timestamp? They only have an article publish date. But oh well, if it was then it was.
How dare you miss that information!!! You OBVIOUSLY are trying to push a narrative!

Or, you know, you missed it, they found it, showed it and that's that. I mean, how could it NOT be nefarious?

Simply astounding.
 

GoldenTiger

Fully [H]
Joined
Dec 2, 2004
Messages
20,365
How dare you miss that information!!! You OBVIOUSLY are trying to push a narrative!

Or, you know, you missed it, they found it, showed it and that's that. I mean, how could it NOT be nefarious?

Simply astounding.
Pssssht, he's clearly a hard core amd fan! /s
 

GoodBoy

[H]ard|Gawd
Joined
Nov 29, 2004
Messages
1,995
26/200 to 28/500 after the AMD driver update? That comes out 6 hours after the article posts?

That's not at all suspicious...

Rename 3dmark.exe to gofast.exe and re-run the test see if the AMD flops... Either they fixed some glaring bug, or the new driver is cheating in the bench... I suppose only way to know for sure would be to use a game that uses this feature.
 

kirbyrj

Fully [H]
Joined
Feb 1, 2005
Messages
27,651
Yes it was. It was updated hours ago according to their timestamp in the comments section.

According to the article, there is new information out there, but the graph posted here and on the website doesn't show the updated info.

Et-Z57qXIAcarP4?format=jpg&name=small.jpg

The chart in the OP and in the article reflects the lower old driver number. Clearly with the later run with the new driver, there is a significant uplift in performance.
 
Last edited:

polonyc2

Fully [H]
Joined
Oct 25, 2004
Messages
19,576
so will switching to Mesh Shaders make the need for DLSS obsolete (with RTX enabled)?
 

socK

Supreme [H]ardness
Joined
Jan 25, 2004
Messages
4,115
26/200 to 28/500 after the AMD driver update? That comes out 6 hours after the article posts?

That's not at all suspicious...

Rename 3dmark.exe to gofast.exe and re-run the test see if the AMD flops... Either they fixed some glaring bug, or the new driver is cheating in the bench... I suppose only way to know for sure would be to use a game that uses this feature.

It's not _that_ unbelievable.

They shaved 3 milliseconds off the render time.

Imagine they went from 51 fps to 60 fps. Same difference. It's pretty big, but not utterly unreal.
 

GoodBoy

[H]ard|Gawd
Joined
Nov 29, 2004
Messages
1,995
yeah the 2 fps change is believeable in the score without the mesh shader enabled. It is +8.6% which is a decent uptick for a driver update. With the mesh shader enabled, it went from 206 to 523 with a new driver for a simple benchmark, this is where it looks like the driver is cheating. That's +253.6%. So very suspicious. 3dMark releasing a second different version of that test would be telling.
 
Last edited:

socK

Supreme [H]ardness
Joined
Jan 25, 2004
Messages
4,115
yeah the 2 fps change is believeable in the score without the mesh shader enabled. It is +8.6% which is a decent uptick for a driver update. With the mesh shader enabled, it went from 206 to 523 with a new driver for a simple benchmark, this is where it looks like the driver is cheating. That's +253.6%. So very suspicious. 3dMark releasing a second different version of that test would be telling.
FPS just starts becoming an increasingly poor metric once you get to very high frame rates because it's not linear. The numbers will begin looking increasingly inflated and volatile as it ramps up, it becomes easier and easier to see very large swings.

Like the difference between 500 fps and 1000 fps is 1 single millisecond. 1000 fps to 2000 fps is a 0.5ms difference.
 

Lakados

2[H]4U
Joined
Feb 3, 2014
Messages
3,613
FPS just starts becoming an increasingly poor metric once you get to very high frame rates because it's not linear. The numbers will begin looking increasingly inflated and volatile as it ramps up, it becomes easier and easier to see very large swings.

Like the difference between 500 fps and 1000 fps is 1 single millisecond. 1000 fps to 2000 fps is a 0.5ms difference.
I’ve been getting a few good reads in about how this interacts with and helps improve ray tracing in games. Something to do with how the buffers interact but it looks like we can expect some nice looking well performing titles in the future.
 

Armenius

Fully [H]
Joined
Jan 28, 2014
Messages
23,694
70-80% of hybrid rendering time is still in rasterization. Decreasing the amount of time it takes to rasterize the image for each frame should bring significant improvements to ray tracing performance in games using mixed rendering, so I'd like to see a ray tracing demo where mesh shaders are being used.
 
Top