3DMark Gets AMD FidelityFX Super Resolution 2 (FSR 2) Feature Test

[btarunr]

UL Benchmarks today released an update to 3DMark that adds a Feature Test for AMD FidelityFX Super Resolution 2 (FSR 2), the company's popular upscaling-based performance enhancement. This was long overdue, as 3DMark has had a Feature Test for DLSS for years now; and as of October 2022, it even got one for Intel XeSS. The new FSR 2 Feature Test uses a scene from the Speed Way DirectX 12 Ultimate benchmark, where it compares fine details of a vehicle and a technic droid between native resolution with TAA and FSR 2, and highlights the performance uplift. To use the feature test, you'll need any GPU that supports DirectX 12 and FSR 2 (that covers AMD, NVIDIA, and Intel Arc). For owners of 3DMark who purchased it before October 12, 2022, they'll need to purchase the Speed Way upgrade to unlock the AMD FSR feature test.



View at TechPowerUp Main Site | Source

 

[oxrufiioxo]

Mine didn't update but the test is there.


Looks ok


I still find FSR2 to be pretty trash but at least in this benchmark it isn't too bad.

 

[Vya Domus]

I hate FSR2 in every game I've tried it in but I also dislike DLSS at anything but the 4k quality setting and even then in some games it's also trash but if I was forced to use one it wouldn't be FSR.

I have a 4K TV and when I set games to 1800P even without any kind of upscaling I often can't tell the difference between native. I think because very few people have 4K displays they don't realize how low the bar is to make the image quality look good on a displays with such a high pixel count.

The most impressive upscaling method I have seen is Epic's TSR, I tested it in Fortnite and at 50% which is reconstructing from 720P at 1440P it looks shockingly good, makes FSR and DLSS look like a joke like they're not even close.

 

[oxrufiioxo]

I have a 4K TV and when I set games to 1800P even without any kind of upscaling I often can't tell the difference between native. I think because very few people have 4K displays they don't realize how low the bar is to make the image quality look good on a displays with such a high pixel count.

The most impressive upscaling method I have seen is Epic's TSR, I tested it in Fortnite and at 50% which is reconstructing from 720P at 1440P it looks shockingly good, makes FSR and DLSS look like a joke.

I use a 65inch LG G2 oled it also looks fine at lower resolutions but I still find DLSS quality looks better than 1800p in most games although if FSR was my only option and I needed more performance the choice would be much harder.

I don't like TSR in most games I've tried it in not even fortnite. The UE5 version of it seems much better than the UE4 version but I still find it pretty meh. In some games it's better than FSR in others it's worse which I already find worse than DLSS.

 

[Vya Domus]

I don't like TSR in most games I've tried it in not even fortnite. The UE5 version of it seems much better than the UE4 version but I still find it pretty meh. In some games it's better than FSR in others it's worse which I already find worse than DLSS.

UE4 had TAA, TSR is UE5 exclusive and as far as I know only Fortnite and Returnal has it, so what you saw in any other games was definitely not TSR but TAA which was a lot worse.

TSR's advantage is at lower resolutions, it absolutely trounces DLSS and FSR in performance mode.

 

[oxrufiioxo]

UE4 had TAA, TSR is UE5 exclusive and as far as I know only Fortnite and Returnal has it, so what you saw in any other games was definitely not TSR but TAA which was a lot worse.

TSR's advantage is at lower resolutions, it absolutely trounces DLSS and FSR in performance mode.

Out of curiosity I booted up fortnite again since I haven't checked it out again since they switched it over to UE5.

I still prefer DLSS quality in motion which wasn't in the game at the ue5 launch vs TSR quality but at lower settings they are definitely comparable and could really go either way at least at 4k.

DLSS seems to be more perfomant as well going from high 70s low 80s with TSR quality vs high 80s mid 90s with DLSS all settings maxed out otherwise with hardware RT on

I definitely agree though it looks much better than anything I've seen FSR do hopefully it gets fsr support so we can make better direct comparisons

And yeah i meant to say I'm not a fan of UE temporal upscaling in general but the UE5 version is much better.

 

[Ruru]

"free update", wait, wut?

edit: Oh, it needs the Speed Way benchmark. Hate this milking as I have the advanced edition already.

 

[oxrufiioxo]

"free update", wait, wut?

View attachment 289826

edit: Oh, it needs the Speed Way benchmark. Hate this milking as I have the advanced edition already.

Yeah it's only free if you bought the DLC which was like 5 bucks at launch.

 

[Ruru]

Yeah it's only free if you bought the DLC which was like 5 bucks at launch.

4EUR from Steam. I guess I have to get that, but still sucks that they milk with later parts of the benchmarks even after having the full version.

 

[oxrufiioxo]

4EUR from Steam. I guess I have to get that, but still sucks that they milk with later parts of the benchmarks even after having the full version.

While I don't disagree if it means better version of benchmarks I'm ok with it.

I've been a huge fan of their work and it's always my go to for stress testing, fan profile setup, amd generalized performance comparisons.

 

[kapone32]

Out of curiosity I booted up fortnite again since I haven't checked it out again since they switched it over to UE5.

I still prefer DLSS quality in motion which wasn't in the game at the ue5 launch vs TSR quality but at lower settings they are definitely comparable and could really go either way at least at 4k.

DLSS seems to be more perfomant as well going from high 70s low 80s with TSR quality vs high 80s mid 90s with DLSS all settings maxed out otherwise with hardware RT on

I definitely agree though it looks much better than anything I've seen FSR do hopefully it gets fsr support so we can make better direct comparisons

And yeah i meant to say I'm not a fan of UE temporal upscaling in general but the UE5 version is much better.

The best implementation and most non invasive was Sapphire's tool in their software package.

 

[Dr. Dro]

4EUR from Steam. I guess I have to get that, but still sucks that they milk with later parts of the benchmarks even after having the full version.

Being fair, they've also raised the price by 5 bucks for new customers with each of these upgrades - Time Spy, Speed Way, etc. - I just find it somewhat ironic that they've locked it behind the one upgrade that most Radeon users didn't purchase, Speed Way is too demanding for RDNA 2 GPUs (including the likes of 6900 and 6950 XT).

 

[oxrufiioxo]

Being fair, they've also raised the price by 5 bucks for new customers with each of these upgrades - Time Spy, Speed Way, etc. - I just find it somewhat ironic that they've locked it behind the one upgrade that most Radeon users didn't purchase, Speed Way is too demanding for RDNA 2 GPUs (including the likes of 6900 and 6950 XT).

It's a pretty neat benchmark though and now maybe with FSR2 Radeon users can actually run it. The implementation seems like one of the better ones to me.

This was probably the plan all along... Release it with fsr so Radeon users will buy it.

 

[Dr. Dro]

It's a pretty neat benchmark though and now maybe with FSR2 Radeon users can actually run it. The implementation seems like one of the better ones to me.

That's interesting. My CPU's delivery has been delayed since they are having stock issues, so I can't try it for myself just yet.

With 4 GB, my laptop's 3050 has no chance of running it

 

[Braegnok]

FSR 2.2 doesn't quite match DLSS. For image quality, it's very, very close. Still, you do lose some detail when using upscaling, whether DLSS or FSR.

If your GPU-limited in most games the simplest way to crank up your framerate in that case is to slash down the render resolution. If you do this the old-fashioned way, by simply lowering your resolution, your going to end up with a blurry, pixelated mess, but using these smart upscalers you can really retain a high amount of image quality.

My 7900 XTX, running Cyberpunk maxed-out struggles to produce framerates,.. so personally FSR 2.2 is an evolution rather than a revolution for AMD's upscaling technology. It's a solid step forward, and it makes an already-good upscaler even better.

I'll take the 68% and lose some detail vs native.

 

[ratirt]

Companies started to race which up scaling tech is better. I only hope they did not forget to race with better GPUs with a decent price as well.
I have not had a chance to use the FSR2 yet. Good that it is there just in case my GPU craps out with a game.

Assassin's creed Valhalla 8K res with 6900xt FSR enabled. Not bad. I do understand that type of usage for any upscale tech.

 

[finalzone]

Being fair, they've also raised the price by 5 bucks for new customers with each of these upgrades - Time Spy, Speed Way, etc. - I just find it somewhat ironic that they've locked it behind the one upgrade that most Radeon users didn't purchase, Speed Way is too demanding for RDNA 2 GPUs (including the likes of 6900 and 6950 XT).

Speed Way runs around 40 frames a second on a Radeon 6950 XT (Sapphire Nitro + Pure version) at 4K. The FSR 2 demo is basically a Speed Way without real-time ray tracing (the specific Nvidia method implemented by 3DMark). I honestly think that version is good without implementing that particular method of real-time ray-tracing.

 

[pavle]

The last image (pool table) with FSR2 looks blurry and where it's not blurry, it's blocky.

 

[human_error]

I just find it somewhat ironic that they've locked it behind the one upgrade that most Radeon users didn't purchase, Speed Way is too demanding for RDNA 2 GPUs (including the likes of 6900 and 6950 XT).

Not at all. The original runs really well in 1440p on my 6900XT (~40fps), and 27fps in 4k. Considering this is the high-end benchmark for 3D mark I consider those reasonable speeds.

The new version is hitting 32FPS without FSR, and 62 FPS at 4k, so a 94% improvement in framerate. At 1440p I was hitting 160FPS with FSR2 so exceeding my monitor's refresh rate.

 

[Dr. Dro]

Not at all. The original runs really well in 1440p on my 6900XT (~40fps), and 27fps in 4k. Considering this is the high-end benchmark for 3D mark I consider those reasonable speeds.

You do realize that is an RTX 3070's performance on that benchmark, right?

Speed Way runs around 40 frames a second on a Radeon 6950 XT (Sapphire Nitro + Pure version) at 4K. The FSR 2 demo is basically a Speed Way without real-time ray tracing (the specific Nvidia method implemented by 3DMark). I honestly think that version is good without implementing that particular method of real-time ray-tracing.

So, you see, like I replied to the other dude: it's a 3070's performance level. It's just too much for the architecture when you need to resort to a thrice-refreshed, fully enabled, top tier silicon on the best cooling, best aftermarket design and giving it a blank check for power and thermal budget to perform like a cutdown GA104. Sorry chief, RDNA 2 may have many strengths, RT performance isn't one of them.

I'll never understand the mind of the Hardware Vendor #3 fan - they'll be arguing that *insert ridiculously expensive, ultra overengineered fringe halo design that can chug 500W+ of power and has completely off the curve performance here* does something about as well as *mass market NVIDIA GPU*, yet they'll be the very first to raise their pitchforks against "NGreedia" when the tables are turned.

End of the day, it's why NVIDIA pulls stunts like Ada - AMD fans don't seem to be willing to demand the company to change and improve, the ends justify the means... eventually people who hold no allegiances get sick of it and just buy a GeForce card to preserve their sanity, it's literally my story and I had been one of the most avid AMD fans ever in the very recent past. I just can't take it seriously anymore.

 

[human_error]

You do realize that is an RTX 3070's performance on that benchmark, right?

Doesn't matter how other cards perform on it. I am merely highlighting that they can run the toughest benchmark out there at 4K, which you said was too demanding. I'd say it was too demanding if they couldn't run it at all, or were close to single digits. The toughest benchmarks are meant to bring cards to their knees (honestly these days I feel benchmarks aren't challenging enough).

No-one is claiming RDNA has top tier RT performance (especially the way Nvidia suggests it be implemented), as its strength lies in raster performance. I don't know why you're going on a rant about AMD fanyboys in a discussion about an FSR2 benchmark release.

 

[Dr. Dro]

Doesn't matter how other cards perform on it. I am merely highlighting that they can run the toughest benchmark out there at 4K, which you said was too demanding. I'd say it was too demanding if they couldn't run it at all, or were close to single digits. The toughest benchmarks are meant to bring cards to their knees (honestly these days I feel benchmarks aren't challenging enough).

No-one is claiming RDNA has top tier RT performance (especially the way Nvidia suggests it be implemented), as its strength lies in raster performance. I don't know why you're going on a rant about AMD fanyboys in a discussion about an FSR2 benchmark release.

Yeah, it doesn't matter until you see that you're comparing it with a 3070 of all things. There's no rant, I'm simply pointing out the hypocrisy. If such a halo product that is so far ahead of the pack has trouble keeping up with a middling SKU in the Ampere product stack, then my point of "Speed Way is too heavy for the RDNA 2 architecture" is well warranted, normal cards such as the 6700 XT (3070 competitor) will simply not run it decently at all. At that point i'm pitching the #1 6700 XT score on HWBot vs. average 3070s stuffed in dusty gamers' PCs to prove my point:

u-9600`s 3DMark - Speed Way score: 2498 marks with a Radeon RX 6700 XT

The Radeon RX 6700 XT @ 2950/2150MHzscores getScoreFormatted in the 3DMark - Speed Way benchmark. u-9600ranks #347 worldwide and #7 in the hardware class. Find out more at HWBOT.

hwbot.org

That's not half the score I got with my power-limited 3090 on launch day with the launch driver for this benchmark... You're happy with 40 fps, great, folks who bought 3070's for half of the 6950 XT Nitro+ Pure's MSRP are doubly so... who am I kidding, anyway. One only got sold to miners and the other is a literal Hardware Vendor #3 moment.

 

[human_error]

The days of buying GPUs for the sake of high benchmark scores are well behind me. For games, the performance I get is fantastic. For benchmarks like this I use it for tuning my overclocks and fan profiles, like many others. It also lets me compare against like for like systems to see if I'm getting the most out of my kit. I'm glad they've released the FSR benchmark as it can showcase the uplift vs quality trade-off. I have only used FSR in one game so far, and that was because the visual output of FSR was better than the TAA or no AA options. I'm massively disappointed in where AA has gone these days where we have crappy temporal AA adding Vaseline all over my screen or nothing so FSR is at least giving me an anti-aliased option that is clear and maintains detail.

I have no idea what you mean by hardware vendor 3 moment - is this a new meme I've missed?

 

[Deleted member 185158]

Yeah, it doesn't matter until you see that you're comparing it with a 3070 of all things. There's no rant, I'm simply pointing out the hypocrisy. If such a halo product that is so far ahead of the pack has trouble keeping up with a middling SKU in the Ampere product stack, then my point of "Speed Way is too heavy for the RDNA 2 architecture" is well warranted, normal cards such as the 6700 XT (3070 competitor) will simply not run it decently at all. At that point i'm pitching the #1 6700 XT score on HWBot vs. average 3070s stuffed in dusty gamers' PCs to prove my point:

u-9600`s 3DMark - Speed Way score: 2498 marks with a Radeon RX 6700 XT

The Radeon RX 6700 XT @ 2950/2150MHzscores getScoreFormatted in the 3DMark - Speed Way benchmark. u-9600ranks #347 worldwide and #7 in the hardware class. Find out more at HWBOT.

hwbot.org

That's not half the score I got with my power-limited 3090 on launch day with the launch driver for this benchmark... You're happy with 40 fps, great, folks who bought 3070's for half of the 6950 XT Nitro+ Pure's MSRP are doubly so... who am I kidding, anyway. One only got sold to miners and the other is a literal Hardware Vendor #3 moment.

That's not even a valid score. Looks like tesselation is disabled, but I'm not entirely sure. I didn't look at it that hard, but some of these benchmarks allow tesselation disabled and some do not allow it.

It would be worth investigating the accuracy of ALL 3D settings when doing these comparisons.

But yes, RDNA2 is awful for SpeedWay benchmark. Really sad honestly.

 

[Dr. Dro]

The days of buying GPUs for the sake of high benchmark scores are well behind me. For games, the performance I get is fantastic. For benchmarks like this I use it for tuning my overclocks and fan profiles, like many others. It also lets me compare against like for like systems to see if I'm getting the most out of my kit. I'm glad they've released the FSR benchmark as it can showcase the uplift vs quality trade-off. I have only used FSR in one game so far, and that was because the visual output of FSR was better than the TAA or no AA options. I'm massively disappointed in where AA has gone these days where we have crappy temporal AA adding Vaseline all over my screen or nothing so FSR is at least giving me an anti-aliased option that is clear and maintains detail.

I have no idea what you mean by hardware vendor 3 moment - is this a new meme I've missed?

In general, agreed, yeah. The benchmark is nice, even if people who have DLSS will gravitate towards using it, I don't think I've seen a single game where it is worth picking it over DLSS when I have access to both.

As for the vendor #3 thing, yeah, it's a meme, and I hate NVIDIA for it... although it's more than an earned kick in the nuts sometimes. Source is this presentation for Streamline from GDC 22, stems mostly from AMD's unwillingness to write a plugin for it despite NV making it possible for them to do so (it is open source and designed to be cross-IHV)

That's not even a valid score. Looks like tesselation is disabled, but I'm not entirely sure. I didn't look at it that hard, but some of these benchmarks allow tesselation disabled and some do not allow it.

It would be worth investigating the accuracy of ALL 3D settings when doing these comparisons.

But yes, RDNA2 is awful for SpeedWay benchmark. Really sad honestly.

Well, it'd not help the case in the slightest... you can see average validated 3070 scores on 3DMark's website are over 1000 points higher than that and average between 38 to 41 fps