Monday, March 2nd 2026

Microsoft Shader Execution Reordering Brings 90% Performance Increase on Intel Arc B-Series, 80% on NVIDIA "Blackwell" GPUs

Microsoft recently updated its Agility SDK to version 1.619, bringing DirectX Shader Model 6.9 alongside some new DirectX 12 improvements. However, Microsoft's latest product demo about Shader Execution Reordering (SER) now confirms a massive performance uplift across several GPUs, with up to a 90% improvement on Intel Arc B-Series and more than 80% from independent sources' benchmarks on NVIDIA "Blackwell." With SER, the API gives applications the ability to dynamically sort rays for highly optimized parallel execution, improving performance by a large margin. In Microsoft's own testing, Intel's Arc B-Series GPUs, which include "Battlemage" discrete GPUs and Xe3-based integrated GPUs in "Panther Lake," managed to achieve a 90% framerate increase in the technology demonstration, suggesting that ray tracing performance still has some tricks up its sleeve for optimization.

Meanwhile, the company also tested the NVIDIA GeForce RTX 4090 with SER, scoring a 40% improvement over the default ray sorting in the previous execution model. Independent testing from Osvaldo Pinali Doederlein on X showed that the GeForce RTX 5080 "Blackwell" GPU scored about an 80% improvement running this demo, which gives confidence that games implementing this technology will provide gamers with a massive performance boost once implemented. Microsoft has built this D3D12RaytracingHelloShaderExecutionReordering demo with a minimum demonstration of the SER technology, so anyone can test their own hardware and the performance improvement.
Finally, it is important to note that this is just a technology demonstration, not real-world game testing. In real-world game testing, ray sorting using SER technology will definitely yield an improvement. However, given the nature and complexity of modern game engines and their ray tracing implementations, the improvement could be significantly smaller. This will remain the case until a game has been tested with the new technology update from the Agility SDK v1.619, which will likely require a few months of implementation and testing before being incorporated into an actual game. Only then can we test and see the real-world impact.
Sources: Osvaldo Pinali Doederlein on X, Microsoft, VideoCardz
Add your own comment

60 Comments on Microsoft Shader Execution Reordering Brings 90% Performance Increase on Intel Arc B-Series, 80% on NVIDIA "Blackwell" GPUs

#26
Jtuck9
ToTTenTranzThat said, be very cautious about these big numbers in performance gains. These "90%" numbers are often achieved in "lab results", or conditions that are ideal to show off the gains by purposely putting a huge bottleneck in the issue being solved. I wouldn't assume any game is going to get a 28-30% performance boost from Shader Execution Reordering, let alone these 90% on Arc GPUs.
They should get into the "nutritional product" game...
Posted on Reply
#27
remekra
ToTTenTranzAMD doesn't have hardware support Shader Execution Reordering in RDNA4 or earlier. They have Out-of-Order Memory which is a different, hardware-only approach.
I'm not sure AMD can do anything in their driver to gain some performance out of SES.


That said, be very cautious about these big numbers in performance gains. These "90%" numbers are often achieved in "lab results", or conditions that are ideal to show off the gains by purposely putting a huge bottleneck in the issue being solved. I wouldn't assume any game is going to get a 28-30% performance boost from Shader Execution Reordering, let alone these 90% on Arc GPUs.


Just look at the performance gains being touted in the days of Xbox's "full RDNA2" marketing campaign. Neither did VRS Tier 2 bring a 20% performance boost in any game, nor did Sampler Feedback Streaming ever "multiply the effective memory and bandwidth by 2.5x". Otherwise we'd all be rocking 8GB GPUs without running out of VRAM at 4K.
SER was already implemented in CP2077, Alan Wake 2 and Indiana Jones. Not sure if there will be additional performance with it, just because it's going to be implemented on DX level rather than NvApi level. So on Nvidia cards those games and new ones will run basically the same as they were, Intel will get a boost, AMD will get simply the fact that game with SER can be launched with their GPU but no performance boost.

There is also a neat demo to show off PT and SER is also included there:



It's the NVAPI HitObject codepath and ReorderThreads setting. If I disable it, frames drop to around 60-70. Same if I disable ReorderThreads, so SER will not make a difference if the hardware does not support it. It will just allow the game to still work on a GPU that does not have it.
Posted on Reply
#28
AGlezB
We've seen a wave of games, mainly in UE5, where it seems clear the devs have no idea how to optimize the graphics pipeline so I don't expect they'll be able to take advantage of SER directly. Odds are it will be added as another feature to the engine (enabled by default) and do more harm than good.

I wish I was more optimistic about this but Nanite and Lumen reinforced something I already knew: give a hammer to a monkey and everything becomes a nail. :(
Posted on Reply
#30
Tomorrow
Oh look. Another DX addon that requires game specific integration and thus will be seldomly used by today's devs who are accustom to working only with high level languages.
Bjorn_Of_IcelandSo it will definitely improve? Exclusive for RDNA5 users of course
Exclusive to 40/50 series too. And here i though Nvidia supported all features back to 20 series like certain people claim in TPU.
Posted on Reply
#31
mechtech
So dx12 been out over 10 years now, about the same as RT.

Why weren’t improvements like this made years ago?
Posted on Reply
#32
makehuugo
This is good news, really love to see technology evolve🔥
Posted on Reply
#33
AGlezB
mechtechSo dx12 been out over 10 years now, about the same as RT.

Why weren’t improvements like this made years ago?
A decade ago nobody cared about RT so it wasn't a problem that needed solving. It took a combination of newer architectures, more VRAM and performance-enhancing techniques like DLSS for RT to become mainstream and for enough people to start caring about it. And if you ask me that only hapened in the last 2-3 years, despite nVidia's doing everything they could to make RT a thing, to the point of penalizing youtube tech channels that didn't prioritize RT in their performance comparisons.
Xzibit in Nvidia blacklist Hardware Unboxed for unfavorable RTX raytracing reviews, LTT & JTZ rally around HUBNvidia and Reviewer drama TechSpot: The ugly side of Nvidia: A rollercoaster ride that shows when Big Tech doesn't get it A different tale about ray tracing iXn9O-Rzb_M 3GyaSfOi6fs Nvidia has started…
Posted on Reply
#34
mrnagant
For SER and AMD not supporting reordering, is that just like super new capability on the hardware side? Says the 4090 saw a 40% improvement, but the 5080 a 80% improvement. That seems to imply that there is something big that is different between the two architectures. Like does the 4090 support SER but no the reordering, so that 40% improvement is just brute force?
Posted on Reply
#35
LastDudeALive
mrnagantFor SER and AMD not supporting reordering, is that just like super new capability on the hardware side? Says the 4090 saw a 40% improvement, but the 5080 a 80% improvement. That seems to imply that there is something big that is different between the two architectures. Like does the 4090 support SER but no the reordering, so that 40% improvement is just brute force?
The Blackwell Architecture summary (www.techpowerup.com/review/nvidia-geforce-rtx-5090-founders-edition/2.html) does say that SER is improved by 2x. So the numbers line up exactly lol.
Posted on Reply
#36
mechtech
AGlezBA decade ago nobody cared about RT so it wasn't a problem that needed solving. It took a combination of newer architectures, more VRAM and performance-enhancing techniques like DLSS for RT to become mainstream and for enough people to start caring about it. And if you ask me that only hapened in the last 2-3 years, despite nVidia's doing everything they could to make RT a thing, to the point of penalizing youtube tech channels that didn't prioritize RT in their performance comparisons.
Xzibit in Nvidia blacklist Hardware Unboxed for unfavorable RTX raytracing reviews, LTT & JTZ rally around HUBNvidia and Reviewer drama TechSpot: The ugly side of Nvidia: A rollercoaster ride that shows when Big Tech doesn't get it A different tale about ray tracing iXn9O-Rzb_M 3GyaSfOi6fs Nvidia has started…
Well MS is separate and doesn’t even make gpus. Just thought it it’s their api and software and their resources this would stay ahead and not lag behind?
Posted on Reply
#37
commission3r
HereticbarWhat's wrong with them being "fake" as long as it looks good?

I know a girl with two of those, doesn't mean to say I'm joyful about it
Posted on Reply
#38
Vayra86
remekraSER was already implemented in CP2077, Alan Wake 2 and Indiana Jones. Not sure if there will be additional performance with it, just because it's going to be implemented on DX level rather than NvApi level. So on Nvidia cards those games and new ones will run basically the same as they were, Intel will get a boost, AMD will get simply the fact that game with SER can be launched with their GPU but no performance boost.

There is also a neat demo to show off PT and SER is also included there:



It's the NVAPI HitObject codepath and ReorderThreads setting. If I disable it, frames drop to around 60-70. Same if I disable ReorderThreads, so SER will not make a difference if the hardware does not support it. It will just allow the game to still work on a GPU that does not have it.
LOL this is PT?

It looks ridiculous, lighting wise.
Posted on Reply
#39
AGlezB
mechtechWell MS is separate and doesn’t even make gpus. Just thought it it’s their api and software and their resources this would stay ahead and not lag behind?
By my count it's been like 3 decades since MS stood ahead of anything. All they've done since is trying to get their hands in every pie by imitating what the rest are doing, often with disastrous results.
Advertising in the start menu of Win 11 is the only recent thing they haven't copied from someone else (AFAIK) and we all love MS for such a great idea, don't we? 🤮
Posted on Reply
#40
mouacyk
Vayra86LOL this is PT?

It looks ridiculous, lighting wise.
One can have perfect lighting, but if the scene or materials are unrealistic, yeah it will look ridiculous. Care to elaborate?
Posted on Reply
#41
RejZoR
ToTTenTranzAMD doesn't have hardware support Shader Execution Reordering in RDNA4 or earlier. They have Out-of-Order Memory which is a different, hardware-only approach.
I'm not sure AMD can do anything in their driver to gain some performance out of SES.


That said, be very cautious about these big numbers in performance gains. These "90%" numbers are often achieved in "lab results", or conditions that are ideal to show off the gains by purposely putting a huge bottleneck in the issue being solved. I wouldn't assume any game is going to get a 28-30% performance boost from Shader Execution Reordering, let alone these 90% on Arc GPUs.


Just look at the performance gains being touted in the days of Xbox's "full RDNA2" marketing campaign. Neither did VRS Tier 2 bring a 20% performance boost in any game, nor did Sampler Feedback Streaming ever "multiply the effective memory and bandwidth by 2.5x". Otherwise we'd all be rocking 8GB GPUs without running out of VRAM at 4K.
Oh I know that. 80% boost of a specific execution stack within GPU may only translate to maybe 10% of real world performance difference on framerate level. And it may further be dependent on what graphic card is used. Usually such boosts are more favorable with lower end cards that are already bottlenecked internally than the flagship model that just churns things out with hardly any effort.

So, Shader Execution Reordering is sort of like programmable shader execution over which you have control in driver where AMD has Out of Order Execution where GPU can reorder tasks, but only in whatever logic they baked into the GPU itself? I have to admit I'm not aware on how AMD works internally in such detail. I don't think it was even ever talked about anywhere until this news.
Posted on Reply
#42
R-T-B
TomorrowAnd here i though Nvidia supported all features back to 20 series like certain people claim in TPU.
I want a quote of someone claiming this. Holy strawman batman.
Posted on Reply
#43
TheinsanegamerN
R-T-BI want a quote of someone claiming this. Holy strawman batman.
What they'll point to is MFG over 2x requiring the hardware on Blackwell and claim "see, nvidia holds things back too 11!11!1!!".
Posted on Reply
#44
Visible Noise
DenverCorrection, it works terribly even on the most expensive hardware available, and you have to lower the resolution and use fake frames on top to pretend it works well.
Clue for the clueless.

All frames are fake.
Posted on Reply
#46
Jtuck9
RejZoRDLSS 4.5 doesn't work natively for graphic cards even as recent as RTX 3000. Sure it looks better visually, but has much higher performance impact compared to RTX 4000...
Which cards can feasibly path trace?
Posted on Reply
#47
LastDudeALive
RejZoRDLSS 4.5 doesn't work natively for graphic cards even as recent as RTX 3000. Sure it looks better visually, but has much higher performance impact compared to RTX 4000...
It does work natively, Nvidia has FP8 (for Ada and Blackwell) and FP16 (for Turing and Ampere) versions. Exact same model, exact same result. The slowdown comes from the slower hardware in Turing and Ampere, FP16 models run slower than FP8, but it has no affect on the output. That's like saying a game doesn't "work natively" just because it runs slower on your i7-8700K than your 7800X3D.
Posted on Reply
#48
RejZoR
LastDudeALiveIt does work natively, Nvidia has FP8 (for Ada and Blackwell) and FP16 (for Turing and Ampere) versions. Exact same model, exact same result. The slowdown comes from the slower hardware in Turing and Ampere, FP16 models run slower than FP8, but it has no affect on the output. That's like saying a game doesn't "work natively" just because it runs slower on your i7-8700K than your 7800X3D.
This is quote from NVIDIA itself:
Please note that GeForce RTX 20 and 30 Series GPUs lack native FP8 support, so Models M and L carry a heavier performance impact for these GPUs. Those users may therefore prefer to remain on Model K (DLSS 4.0), for an optimal balance of performance and image quality. With the NVIDIA app’s DLSS...
You need FP8 for native support for DLSS 4.5. Using FP16 isn't NATIVE compute of DLSS 4.5. It's a fallback mode that allows using of DLSS 4.5 on older generations with severe performance penalty but same output. Your comparison of two different generations of CPU's makes no sense here because it's not just a thing of different raw performance components, yours is comparing CPU's that don't support AVX512 and one that does. It will still work with some fallback code that doesn't use AVX512. But will be dramatically slower in comparison.
Posted on Reply
#49
Onasi
RejZoRYou need FP8 for native support for DLSS 4.5. Using FP16 isn't NATIVE compute of DLSS 4.5. It's a fallback mode that allows using of DLSS 4.5 on older generations with severe performance penalty but same output.
…you keep using the word “native”. I don’t think it means what you think it means. Does the new model work? Out of the box? Officially supported by NV? Yes? Then it’s native. It’s slower, yes, which they never denied and, as your very quote suggests, actually warned of ahead of time. But it still runs NATIVELY. Non-native would be the hacked implementation of FSR4 on RDNA 3, as an example.
Posted on Reply
#50
Arcturio
TomorrowOh look. Another DX addon that requires game specific integration and thus will be seldomly used by today's devs who are accustom to working only with high level languages.
Confusing comment.
High level languages will be updated to support it.
Posted on Reply
Add your own comment
Nov 18th, 2026 18:36 PST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts