AMD Radeon RX Hardware Acceleration

[debraspicher]

@slizgi It was discovered in another thread that doing an unofficial sandbox of Photo rendered 14-20% performance increase on NVIDIA cards. Perhaps you might want to try with this method on your machine and see what difference benchs it gives you? Not tenable for production use, but @Nomad Raccoon had suggested UWP architecture interferes with GPU performance in a significant way.

Here is the post with the instructions. An attractive thread title, I know.

 

[slizgi]

5 minutes ago, debraspicher said:

@slizgi It was discovered in another thread that doing an unofficial sandbox of Photo rendered 14-20% performance increase on NVIDIA cards. Perhaps you might want to try with this method on your machine and see what difference benchs it gives you? Not tenable for production use, but @Nomad Raccoon had suggested UWP architecture interferes with GPU performance in a significant way.

 

I did not know about that, will read and try and check it out today! Thank you, @debraspicher ! However, even if it will work on AMD same way, we still be underperformed about 80% compare to NVIDIA.

[debraspicher]

Sharing tests I ran in another thread using a 5600XT using the "new" benchmark version 20300 in the current beta. First results are MSIX sandboxed install. Second results are run from the unzipped MSIX package with 7zip to a desktop folder and running Photo.exe from within that folder using Administrator Privileges. My NVIDIA 3080 12GB still sees a 18-20% difference, but with the AMD, it's still a noticeable change.

Hardware:

Quote

Microsoft Windows 10 Home (Build 19045)
AMD Ryzen 7 5800X @ 3.8Ghz (-30 all core +200mhz PBO); Mobo: Asus X470 Prime Pro
32GB DDR4 (3600Mhz)
Monitor 1&2 4K @ 150%

AMD 5600XT Results

Variance: 5% ~ 6%

Sandboxed (2.0.3.1670 Beta Benchmark Version 20300)

Unofficial Unsandboxed (2.0.3.1670 Beta Benchmark Version 20300)

[slizgi]

4 hours ago, debraspicher said:

Sharing tests I ran in another thread using a 5600XT using the "new" benchmark version 20300 in the current beta. First results are MSIX sandboxed install. Second results are run from the unzipped MSIX package with 7zip to a desktop folder and running Photo.exe from within that folder using Administrator Privileges. My NVIDIA 3080 12GB still sees a 18-20% difference, but with the AMD, it's still a noticeable change.

Yea, but it is a very insignificant change on AMD. The problem in this topic is not about sandboxed and unsandboxed (it is some type of issue I agree) but here is a bigger problem. I believe the problem is caused by AMD - Radeons have only 5-15% of the performance they should have in this app, and because of that GPU acceleration is basically useless - more waiting for effects and results on heavier files.

[debraspicher]

8 hours ago, slizgi said:

Yea, but it is a very insignificant change on AMD. The problem in this topic is not about sandboxed and unsandboxed (it is some type of issue I agree) but here is a bigger problem. I believe the problem is caused by AMD - Radeons have only 5-15% of the performance they should have in this app, and because of that GPU acceleration is basically useless - more waiting for effects and results on heavier files.

Yes I tried to brush something in the app. Oooof. It is also very laggy compared to before. A different experience. I was adamant not to do AMD GPUs during my last upgrade, because it doesn't matter who is to blame. It doesn't work with Affinity. Still, AMD is a common, though not as common as Nvidia, GPU manufacturer.

[SrPx]

In the graphic apps world... it is not only CUDA cores... (though now HIP tech from AMD seems to be doing better than open CL, but it's too new: nvidia seems always will keep the earlier implementation and decades of working with software companies) . I use Blender for 3D, and OptiX (nvidia tech for nvidia hardware) is really fast, it reduces crazily the already short GPU render times, there's simply no contest there, for AMD. If  you see  the benchmarks comparison, it's unreal.

If was only that, but for example, In Davinci Studio (the pro version, as the free is mostly cpu based), used massively these days by freelancers, medium cards from nVidia have same or greater performance in benchmarks and real life tests, than higher rank mainstream AMD GPUs, at least in the non pro cards. And pro cards are super overpriced, for what they are, so many freelancers don't go for a quadro or Radeon Pro, as they need 12 or 16GB of VRAM, but won't use 2k or 5k dollars, euros or whatever for a card, when a 3060 - 3070 kind of makes the deal. Besides, most apps are more than fine (some often just with a 3050 8gb) with the low-mid range from nvidia. Is when your projects are next level that you need sth better (8k video editing but also with a very complex workflow, in Davinci, or GPU rendering HUGE scenes in 3D in Blender or any other GPU accelerated 3D package, etc).

The thing is... while most creatives go for nvidia (I do, but the mid-low segment), a lot of users in Affinity.. I'm sure are just people who like a bit of photography here and there. Many in their mid 20s early 30s are gamers. And many of them have gone for the best deals for their home pc, having an AMD cards in it. Or, are using the kids' PC -who just play games with it and some school work, or a grandma/grandpa "inherited" one of those, or whoever in any age group and circumstances got a great deal from a gamer who's buying sth better,  all of them using what they have to enjoy their photography or design, painting hobby.  So... not a bad idea to ensure full support for AMD cards (imo).
I see a major problem with laptops (and some desktops with no discrete cards) using only integrated intel cards.

[DylanGG]

I just upgraded to a ASUS Zephyrus G14 with an RX 6700s GPU and was wondering why it didn't seem any quicker than my previous GTX 1060 laptop. I really hope this is addressed sometime soon. This new GPU should be a huge upgrade. But it's not due to this issue.

UPDATE:
I ran the benchmark on my new PC vs. old PC. The old PC's GPU is running way faster on NVIDIA. This AMD driver issue has to get sorted. My new processor is running twice as fast, but that new GPU should be at least twice as fast too. Bummer.

[DylanGG]

Super disappointing. Had I known, I would've just gotten something with an RTX 3060 instead... I have a feeling this won't ever get resolved by AMD. 😑

[ziplock9000]

1 hour ago, DylanGG said:

Super disappointing. Had I known, I would've just gotten something with an RTX 3060 instead... I have a feeling this won't ever get resolved by AMD. 😑

I have a feeling Serif wont use another API that works just fine on AMD cards that competitors already use either.

Remember, competing products work just fine. Serif have just chosen to use the API that has poor performance. Seemingly even with V2.x of the suite

The issue is mostly with them not AMD who provide a valid, working API.

Again, let me repeat: Competing products work fine.

 

[DylanGG]

I contacted AMD and a representative told me that they're looking into this issue and that they would update me when they have more information.

[theelijahmacdougall]

On 1/20/2023 at 4:17 PM, DylanGG said:

I contacted AMD and a representative told me that they're looking into this issue and that they would update me when they have more information.

Any updates on this? I'm coloring a comic book at the moment, and whole conditions aren't horrible if I use designer for too long it nearly grinds to a hault. 

Got to a point yesterday where I couldn't get the program to switch from Lasso to Bucket without seconds of delay. The team can point at AMD but they can also stop being so hardheaded and make their products usable for the other large number of people with AMD cards, just saying. 

[DylanGG]

11 hours ago, theelijahmacdougall said:

Any updates on this?

Not yet, the moment I hear anything back I'll be sure to post here. I agree AMD users deserve functional GPU support.

[Ian R]

Glad I found this thread.

Baited breath...

[DylanGG]

I emailed the AMD rep, he told me the issue is still under investigation, no updates yet. When he has any info he'll make sure to update me.

[slizgi]

7 minutes ago, DylanGG said:

I emailed the AMD rep, he told me the issue is still under investigation, no updates yet. When he has any info he'll make sure to update me.

They have it "under investigation" since a very long time, I had some email exchange with them too about this, still nothing serious tho. They improved real-time runtime OpenCL compilation a bit (like 8x better times from ~1400 ms to ~150 ms), but it did not change the scores in benchmark as far as I tested it. So I do not know.

[Patrick Connor]

Dear Thread Followers.

In the Windows betas builds of the Affinity range, we have recently reworked the memory handling in hardware acceleration which may benefit AMD cards. Users of AMD (Navi) Graphics cards who now have Version 2 of our software, we would be grateful if you can sign up to the beta and let us know in the Beta Memebers Area if you find it has had any positive or negative effect, as is already being reported elsewhere.

[ziplock9000]

16 minutes ago, Patrick Connor said:

Dear Thread Followers.

Users of AMD (Navi) Graphics cards who now have Version 2 of our software, we would be grateful if you can sign up to the beta and let us know in the Beta Memebers Area if you find it has had any positive or negative effect, as is already being reported elsewhere.

Can you confirm any fixes for this will never come to version 1 of the software, even though hardware acceleration was actually part of that version?

[Patrick Connor]

4 minutes ago, ziplock9000 said:

Can you confirm any fixes for this will never come to version 1 of the software, even though hardware acceleration was actually part of that version?

I can confirm that. There are no plans to patch V1, sorry

[DylanGG]

The benchmark scores are substantially better and the performance is also definitely smoother in 2.1 on my RX 6700s.

However, I still feel that my old GTX 1060 was faster, even though the score was lower. Not sure. Would love to hear what everyone else is experiencing.

[kkoukos]

Hi,

I just tested out the latest v.2 beta compared to the latest v.2 stable (2.0.4) on a Radeon RX 6900 XT. There is a huge difference in the benchmark result. More than 18000 for v.2.1 while only marginally more than 1000 for 2.0.4. I also monitored the GPU activity in parallel and there is also a noticeable difference in the GPU utilization. When it comes to testing it with real data i don't see a huge difference (as of 18x) from 2.0.4 but it feels smoother and seems to better utilize the GPU when using filters (in both destructive and non-destructive modes). Huge difference in responsiveness was noticed in the Lens Blur filter, which i believe is GPU accelerated in 2.1 while it wasn't in the stable version (2.0.4). And that is actually an indication that GPU acceleration does a great job. So overall i believe the developer team has made a great job.

[eqyizo]

If benchmarks mean anything to actual usage, things should work much more as expected. I certainly hope so. Now the better question is: can I run Spotify at the same time and not have Affinity crash my video card driver constantly? I'm scared to test that one, since one of those times my computer refused to turn on for 3 hours afterward. I thought the app had bricked my whole PC. After that, I turned on WARP rendering and never looked back. Perhaps this update is hope for the future. We shall see.

[nitro912gr]

23 hours ago, eqyizo said:

If benchmarks mean anything to actual usage, things should work much more as expected. I certainly hope so. Now the better question is: can I run Spotify at the same time and not have Affinity crash my video card driver constantly? I'm scared to test that one, since one of those times my computer refused to turn on for 3 hours afterward. I thought the app had bricked my whole PC. After that, I turned on WARP rendering and never looked back. Perhaps this update is hope for the future. We shall see.

Nothing can brick a PC, just update your drivers and try again. Probably just a bad moment of some software conflict.

[eqyizo]

3 minutes ago, nitro912gr said:

Nothing can brick a PC, just update your drivers and try again. Probably just a bad moment of some software conflict.

It's an issue with the 6800 gpu and affinity. It happened when hardware acceleration was on, and using another app with acceleration crashed the driver more often. Using Affinity and Spotify at the same time crashed the driver constantly (and affinity would crash half the time too, losing some work depending on the restore point). There is a forum post with others having the same issue, all with the same card. And yes, drivers are updated all the time. So I'm hoping this update fixes that conflict. Like I said, we'll see.

[nitro912gr]

Wtf really? Man this is why I never win the lottery, I never had problems with AMD GPUs in my life, probably spent all my luck there

[Vasto7]

Just wanted to report that I just installed V2.1.0.1799 and saw a very large improvement in benchmark results with my AMD Radeon RX5700.  Thank you to Affinity Developers!!