Starfield: FSR 3 vs DLSS 3 vs XeSS Comparison

[wolf]

Nintendo likely getting DLSS in the Switch 2 is also a big boon for proprietary upscaling tech in my opinion, so I really do think that the statement "all proprietary upscaling tech will be shit canned" is misleading.

And to quote this part again, it was even more misleading now we know what MS annocuned, DirectSR which makes it easy to add any and all upscalers, gee that sounds familiar. Yeah I don't see DLSS going away now, certainly not by the end of this year and unlikely in the immediate years beyond that.

Yet another reason I don't watch MLID or take anything he says seriously till the players that matter actually announce what they're doing.

 

[phxrider]

I don't see that much of a difference staring at frozen frames, and I'm positive I wouldn't see one with the game in motion. I play at native resolution with everything cranked on my main system because I can at 1440UW, but on my living room system I need to use FSR3 to get the same frame rate @4K, and I don't see anything out of sorts there. Seems like a big rabbit hole for people who have nothing better to do than engage in brand wars.... While the tech is interesting, when it comes to choosing, it seems they all do a more than good enough job of upscaling on the fly to get a better frame rate, and I don't see why one would be a selling point over another.

 

[Miss_Cherry_Bomb]

Nope. Read the review.



Any time there's a game tested without presets, we use the quality slider to set equivalents.

I just did my own testing, and FSR isn't anywhere as bad as the review is selling it as. But hey, no one ever clearly fudges data, ever. And I definitely didn't include visual proof of fudged data on this site with screenshots.

 

[dgianstefani]

I just did my own testing, and FSR isn't anywhere as bad as the review is selling it as. But hey, no one ever clearly fudges data, ever. And I definitely didn't include visual proof of fudged data on this site with screenshots.

Strong claims.

And what do you have as proof? OC temperatures are lower than stock? That's because the OC voltage used for a static all core OC was lower than the stock voltage. Bearing in mind that the 7950X with a 5.1 GHz all core is clocking 600 MHz lower than stock boost algorithm/voltage (and more than that with PBO). Yet you're surprised temperatures are lower? Seems more indicative of a bad understanding of the silicon involved than TPU somehow "fudging" data.

Perhaps @W1zzard can weigh in further.

The 7950X and 14900K CPUs were also reviewed in Sept 2022, and Oct 2023, more than a year apart. That means that an updated CPU test rig was used for the 14900K review. Hence, why the temperatures are different between your two screenshots. Review data is only comparable to that generated on the same test rig, with the same test methodology. Hence why W1zzard retests every CPU/GPU when updating the test rigs, a huge and intensive task, but one that seems to go unappreciated by you, because apparently it's easier to believe that we're somehow "fudging" data, than it is to understand context.

 

[Miss_Cherry_Bomb]

Strong claims.

And what do you have as proof? OC temperatures are lower than stock? That's because the OC voltage used for a static all core OC was lower than the stock voltage. Bearing in mind that the 7950X with a 5.1 GHz all core is clocking 600 MHz lower than stock boost algorithm/voltage (and more than that with PBO). Yet you're surprised temperatures are lower? Seems more indicative of a bad understanding of the silicon involved than TPU somehow "fudging" data.

Perhaps @W1zzard can weigh in further.

Well, if you actually looked at both screenshots, you'd see the numbers on the 7950x review are all lower than the numbers for the 7950x on the 14900k review. 7950x chart says max temps for load 94c and the gaming one says 76c. Whereas the 14900k chart says 96.9 for load on the 7950x and 95.5 for gaming on the 7950x. I marked different parts of the 7950x chart to show that no matter where you look on that chart, the temps are always lower than what is shown on the 14900k chart. Two charts, same 7950x CPU, two different data sets, equals fudging/faking data to me.

 

[dgianstefani]

Well, if you actually looked at BOTH screenshots, you'd see the numbers on the 7950x review are all lower than the numbers for the 7950x on the 14900k review. 7950x chart says max temps for load 94c and the gaming chart says 76c. Whereas the 14900k chart says 96.9 for load on the 7950x and 95.5 for gaming on the 7950x. I marked different parts of the 7950x chart to show that no matter where you look on that chart, the temps are always lower than what is shown on the 14900k chart. Two charts, same 7950x CPU, two different data sets, equals fudging/faking data to me.

The 7950X and 14900K CPUs were also reviewed in Sept 2022, and Oct 2023, more than a year apart. That means that an updated CPU test rig was used for the 14900K review. Hence, why the temperatures are different between your two screenshots. Review data is only comparable to that generated on the same test rig, with the same test methodology. Hence why W1zzard retests every CPU/GPU when updating the test rigs, a huge and intensive task, but one that seems to go unappreciated by you, because apparently it's easier to believe that we're somehow "fudging" data, than it is to understand context.

BTW the reviews detail the testing methodology, including the test rig. So when it's updated, that is mentioned, something you might have known if you'd read the reviews instead of screenshotting two charts more than a year apart and trying to use it as evidence of fraud.

 

[Miss_Cherry_Bomb]

BTW the reviews detail the testing methodology, including the test rig. So when it's updated, that is mentioned, something you might have known if you'd read the reviews instead of screenshotting two charts more than a year apart and trying to use it as evidence of fraud.

I just realised you're a staff member. Are you the one who did the review?

Here's my temps for the 7950x under a full load test with PBO on. I have the same setup and cooler that is mentioned in the 14900k review. Only real difference is I live in Aus, and it's currently 35c where I am. I suggest you find a new tester. Or you can continue to attack people who point out your errors/fake data when testing.

 

[dgianstefani]

Let me guess. You don't have a 4090, so your CPU doesn't need to supply as many frames, therefore doesn't work as hard, therefore is a lower temperature? Sounds about right. Your CPU reaches 95°C because that's literally the thermal goal of the silicon as designed by AMD, to boost until that temperature is reached. Something you'd understand if you knew what you're talking about.

But logic is hard. Jumping to conclusions in an objectively wrong manner, to justify your own beliefs is much easier.

Everyone else is conspiring to falsify data. It's not possible as an alternative that you've just misunderstood something fundamental, and continue to suffer from wilful ignorance after it's been pointed out.

Notice any differences in those setups? That's what more than a year of time leads to in differences between standardised test benches. A 4090 is used in newer reviews instead of a 3080, as one major example. And yes. The 7950X in the 14900K review comparison chart was tested with a 4090 at that time, not the 3080 it reviewed with on launch day in 2022.

 

[Miss_Cherry_Bomb]

Let me guess. You don't have a 4090, so your CPU doesn't need to supply as many frames, therefore doesn't work as hard, therefore is a lower temperature? Sounds about right. Your CPU reaches 95°C because that's literally the thermal goal of the silicon as designed by AMD, to boost until that temperature is reached. Something you'd understand if you knew what you're talking about.

But logic is hard. Jumping to conclusions in an objectively wrong manner, to justify your own beliefs is much easier.

Everyone else is conspiring to falsify data. It's not possible as an alternative that you've just misunderstood something fundamental, and continue to suffer from wilful ignorance after it's been pointed out.


View attachment 338541

Actually, I have two pc's. One is a 14900k with a 4090 and the other is 7950x with 7900 xtx. I've copied your tests and the only ones that are somewhat accurate are the 14900k benchmarks, but they are close enough to make me think margin of error. I've even tried a combination of 14900k and 7900 xtx as well as 7950x with 4090 in my testing. From what I can see, we even have the same motherboard and memory kit for the 14900k. The only difference in pc specs I can see is that my 4090 Is the rog strix, my AMD psu is a corsair 1200w, while my Intel is a seasonic prime TX-1600. My M.2 drives are both WD SN850x 2TB AMD and 4TB Intel, your paste is different as I'm using the paste that came with my coolers. I have two NH-D15 air coolers on my CPU's. Again, I've coppied your testing and can not get temps like you claim on your 7950x testing, even with my 4090 in the system. We even have the same AMD motherboard and memory kit.

But logic is hard. Jumping to conclusions in an objectively wrong manner, to justify your own beliefs about what hardware I have or don't have, or what software I've used or when I did my AMD testing the first time, is much easier. It's not possible as an alternative that you've just misunderstood something fundamental when I told you I have the same hardware that you used in your tests or when I said I copied your testing methodology.

 

[dgianstefani]

Actually, I have two pc's. One is a 14900k with a 4090 and the other is 7950x with 7900 xtx. I've copied your tests and the only ones that are somewhat accurate are the 14900k benchmarks, but they are close enough to make me think margin of error. I've even tried a combination of 14900k and 7900 xtx as well as 7950x with 4090 in my testing. From what I can see, we even have the same motherboard and memory kit for the 14900k. The only difference in pc specs I can see is that my 4090 Is the rog strix, my AMD psu is a corsair 1200w, while my Intel is a seasonic prime TX-1600. My M.2 drives are both WD SN850x 2TB AMD and 4TB Intel, your paste is different as I'm using the paste that came with my coolers. I have two NH-D15 air coolers on my CPU's. Again, I've coppied your testing and can not get temps like you claim on your 7950x testing, even with my 4090 in the system.

But logic is hard. Jumping to conclusions in an objectively wrong manner, to justify your own beliefs about what hardware I have or don't have is much easier. It's not possible as an alternative that you've just misunderstood something fundamental when I told you I have the same hardware that you used in your tests.

In case you've forgotten, your original claim was that our tests were flawed because the 2022 temperature results were different from the 2023 ones, and this was apparently conclusive proof of tampering with data despite the two benches using completely different hardware setups with vastly different graphics performance and wattage.

Now you're shifting the goalposts again to a different focus.

Believe whatever you want to believe. I've detailed enough here to prove to other readers that our data is not, in fact, flawed, so the misinformation you tried to spread has been countered.

 

[Miss_Cherry_Bomb]

detailed enough here to prove to other readers that our data is not, in fact, flawed, so the misinformation you tried to spread has been countered.

From what I can see with what you've shared, you're using the same stuff to test the 7950x today as you were back in 2022, My argument is also that I can't get temps as high as you with the same hardware and software you used in the newer of the benchmarks.

"I've detailed enough here to prove to other readers that our data is not, in fact, flawed, so the misinformation you tried to spread has been countered."

And who controls the flow of information here? Pretty sure it's the staff member who will say and do whatever to keep their job and get paid.


Who keeps insulting users who call them out?

Who keeps coming back after they claim to have "detailed enough here to prove to other readers that our data is not, in fact, flawed"?

 

[W1zzard]

Well, if you actually looked at both screenshots, you'd see the numbers on the 7950x review are all lower than the numbers for the 7950x on the 14900k review. 7950x chart says max temps for load 94c and the gaming one says 76c. Whereas the 14900k chart says 96.9 for load on the 7950x and 95.5 for gaming on the 7950x. I marked different parts of the 7950x chart to show that no matter where you look on that chart, the temps are always lower than what is shown on the 14900k chart. Two charts, same 7950x CPU, two different data sets, equals fudging/faking data to me.

Thanks for your report. I looked into it, sorry for taking so long to get back to you

The old test system used a RTX 3080, which was much more CPU limited than the RTX 4090 in the new setup. Gaming power was tested at 1080p in both cases.

If you look at the FPS results, the 7950X is running 118 FPS in one review and 183 FPS in the other.

This greatly increases the CPU load (by +55%: 100% / 118 FPS * 183 FPS = 155%), so it's expected that temperatures are higher. This effect scales roughly proportionally with the bottleneck, i.e. CPU speed

I also vaguely remember that CP was upgraded around that time to make better use of more cores, not sure how much of an effect this has, if any.

Are you wondering why some Intel CPUs are running at lower temperatures than expected? That's because they will run into their power limiter after a few seconds, which clocks the CPU down to lower power consumption, which lowers heat output. I'm intentionally testing steady-state conditions as that's how we are using our computers. No gaming session lasts for <60 seconds

Here's my test data for the newest 7950X run

Retry. StdDev too high: 0.4 °C / 0.4 % (Min: 93.0 °C, Max: 94.1 °C, Avg: 93.7 °C)
Retry. StdDev too high: 0.1 °C / 0.1 % (Min: 94.1 °C, Max: 94.3 °C, Avg: 94.2 °C)
Retry. StdDev too high: 0.1 °C / 0.1 % (Min: 94.1 °C, Max: 94.3 °C, Avg: 94.2 °C)
Retry. StdDev too high: 0.5 °C / 0.5 % (Min: 94.1 °C, Max: 95.8 °C, Avg: 94.5 °C)
Retry. StdDev too high: 0.5 °C / 0.5 % (Min: 94.1 °C, Max: 95.8 °C, Avg: 94.6 °C)
Retry. StdDev too high: 0.1 °C / 0.1 % (Min: 94.1 °C, Max: 94.4 °C, Avg: 94.2 °C)
Retry. StdDev too high: 0.0 °C / 0.0 % (Min: 94.1 °C, Max: 94.3 °C, Avg: 94.2 °C)
Retry. StdDev too high: 0.1 °C / 0.1 % (Min: 94.3 °C, Max: 94.4 °C, Avg: 94.3 °C)
Retry. StdDev too high: 0.0 °C / 0.0 % (Min: 94.1 °C, Max: 94.3 °C, Avg: 94.2 °C)
Retry. StdDev too high: 0.1 °C / 0.1 % (Min: 94.1 °C, Max: 94.4 °C, Avg: 94.2 °C)
Average: 94.2 °C
Min: 94.1 °C
Max: 94.3 °C
StdDev: 0.0 °C
StdDev: 0.0 %
Ambient: 23.3 °C
Normalized: 95.9 °C

This also explains why you're seeing >95°C in those results, despite the 95°C limit on 7950X (it's all mentioned on the temps page, too)

Here's my temps for the 7950x under a full load test with PBO on

Same 95°C maximum that I'm reporting

AMD Ryzen 9 7950X Cooling Requirements & Thermal Throttling

High temperature seem to be an issue on the new Ryzen 7000 processors. We're pairing a Ryzen 9 7950X with a $10 stock cooler, a Noctua air-cooler and a 420 mm AIO to get a feel for what the differences are like in terms of °C, MHz and performance in both applications and games.

www.techpowerup.com

This article might help explain the mechanics behind AMD's temp vs clock behavior

 

[Miss_Cherry_Bomb]

Thanks for your report. I looked into it, sorry for taking so long to get back to you

The old test system used a RTX 3080, which was much more CPU limited than the RTX 4090 in the new setup. Gaming power was tested at 1080p in both cases.

If you look at the FPS results, the 7950X is running 118 FPS in one review and 183 FPS in the other.

This greatly increases the CPU load (by +55%: 100% / 118 FPS * 183 FPS = 155%), so it's expected that temperatures are higher. This effect scales roughly proportionally with the bottleneck, i.e. CPU speed

I also vaguely remember that CP was upgraded around that time to make better use of more cores, not sure how much of an effect this has, if any.

Are you wondering why some Intel CPUs are running at lower temperatures than expected? That's because they will run into their power limiter after a few seconds, which clocks the CPU down to lower power consumption, which lowers heat output. I'm intentionally testing steady-state conditions as that's how we are using our computers. No gaming session lasts for <60 seconds

Here's my test data for the newest 7950X run

Retry. StdDev too high: 0.4 °C / 0.4 % (Min: 93.0 °C, Max: 94.1 °C, Avg: 93.7 °C)
Retry. StdDev too high: 0.1 °C / 0.1 % (Min: 94.1 °C, Max: 94.3 °C, Avg: 94.2 °C)
Retry. StdDev too high: 0.1 °C / 0.1 % (Min: 94.1 °C, Max: 94.3 °C, Avg: 94.2 °C)
Retry. StdDev too high: 0.5 °C / 0.5 % (Min: 94.1 °C, Max: 95.8 °C, Avg: 94.5 °C)
Retry. StdDev too high: 0.5 °C / 0.5 % (Min: 94.1 °C, Max: 95.8 °C, Avg: 94.6 °C)
Retry. StdDev too high: 0.1 °C / 0.1 % (Min: 94.1 °C, Max: 94.4 °C, Avg: 94.2 °C)
Retry. StdDev too high: 0.0 °C / 0.0 % (Min: 94.1 °C, Max: 94.3 °C, Avg: 94.2 °C)
Retry. StdDev too high: 0.1 °C / 0.1 % (Min: 94.3 °C, Max: 94.4 °C, Avg: 94.3 °C)
Retry. StdDev too high: 0.0 °C / 0.0 % (Min: 94.1 °C, Max: 94.3 °C, Avg: 94.2 °C)
Retry. StdDev too high: 0.1 °C / 0.1 % (Min: 94.1 °C, Max: 94.4 °C, Avg: 94.2 °C)
Average: 94.2 °C
Min: 94.1 °C
Max: 94.3 °C
StdDev: 0.0 °C
StdDev: 0.0 %
Ambient: 23.3 °C
Normalized: 95.9 °C

This also explains why you're seeing >95°C in those results, despite the 95°C limit on 7950X (it's all mentioned on the temps page, too)


Same 95°C maximum that I'm reporting

AMD Ryzen 9 7950X Cooling Requirements & Thermal Throttling

High temperature seem to be an issue on the new Ryzen 7000 processors. We're pairing a Ryzen 9 7950X with a $10 stock cooler, a Noctua air-cooler and a 420 mm AIO to get a feel for what the differences are like in terms of °C, MHz and performance in both applications and games.

www.techpowerup.com

This article might help explain the mechanics behind AMD's temp vs clock behavior

It was more about how you reached 95.9c in CP2077 in a sub 26c room, when my temps only reach 81c in CP2077 in an Australian summer that averaged 43c where I am using the same cooler, and have used the same as and newer updates than you. I don't have air con, only a 30cm desk fan, and I need that more than my pc does. I also can't reach 90c in gaming even when I copy your testing methodology in a much hotter room. Before dgianstefani threw their hat into the ring, I just figured it was a typo, their attitude made me believe it was deliberate. TBH, when it gets to 35c by 10am it can get a little hard to think and make a good point, but that's on me.

 

[W1zzard]

It was more about how you reached 95.9c in CP2077

What graphics card do you have? Which resolution did you test? Max settings in CP, RT off? What FPS are you getting in that setting?

 

[Miss_Cherry_Bomb]

What graphics card do you have? Which resolution did you test? Max settings in CP, RT off? What FPS are you getting in that setting?

Might be quicker to tell you what cards I don't have, I hoard them and still have non 3d cards TBH. But I tested it at 1080p 180Hz and 1440p 240Hz and I get roughly the same FPS as your chart shows but lower temps. As of an hour ago, I've tested it with a 2080ti a 5600xt & 5700xt a 4090 and a nitro xtx card. (I even threw in a GTX 580 1.5GB and a HD6850 1GB because I like to have fun with numbers, and I was bored, but not because of any numbers you posted or in an attempt to match your data, just for fun.) I always use max settings with RT off, no upscaling ever.

I have used your benchmarks as a buying guide for years and have been able to copy all your results to within a margin of error countless times before, just not this time with the temps, which is why I thought it was a typo and didn't think anything else of it until that staff member. Your OC chart for the 7900 xtx nitro+ is the reason I bought one after I bought my 4090. The only difference when replicating your OC and UV tests for the nitro+ are that I get higher FPS than you showed, but I chalk that up to silicone lottery. (Posted chart for point of reference.) I can match everything with my 7950x tests, but I can't reach the same temps you show in your game testing.

 

[dgianstefani]

I have used your benchmarks as a buying guide for years and have been able to copy all your results to within a margin of error countless times before, just not this time with the temps, which is why I thought it was a typo and didn't think anything else of it until that staff member. Your OC chart for the 7900 xtx nitro+ is the reason I bought one after I bought my 4090. The only difference when replicating your OC and UV tests for the nitro+ are that I get higher FPS than you showed, but I chalk that up to silicone lottery. (Posted chart for point of reference.) I can match everything with my 7950x tests, but I can't reach the same temps you show in your game testing.

I just did my own testing, and FSR isn't anywhere as bad as the review is selling it as. But hey, no one ever clearly fudges data, ever. And I definitely didn't include visual proof of fudged data on this site with screenshots.

And who controls the details and data you're speaking of? Who can go make changes and edits to their website and charts After the fake data is pointed out? Who can edit and change the site, and delete and edit all comments about data that disproves them? In short, who controls the flow of information here? Pretty sure it's the staff member who will say and do whatever to keep their job and get paid.

Can we stop trying to rewrite history just because you or a moderator have edited your posts to be more reasonable after the fact please.

You directly accused TPU of faking data and then covering it up, multiple times. The fact that you yourself edited your posts many times and a moderator also stepped in does not change that. W1zzard has confirmed that the issues you raised were simply due to the CPUs in mention being tested in two different test rigs, at different times.

I asked you what proof you had for your strong claim, and when you didn't have any, said you had jumped to conclusions and didn't know what you were talking about. Hardly "attacking you", more about defending the integrity of TPU.

 

[W1zzard]

roughly the same FPS as your chart shows but lower temps

What temps do you see with 7950X + RTX 4090 @ 1080p ? What CPU power usage?

 

[tabascosauz]

Might be quicker to tell you what cards I don't have, I hoard them and still have non 3d cards TBH. But I tested it at 1080p 180Hz and 1440p 240Hz and I get roughly the same FPS as your chart shows but lower temps. As of an hour ago, I've tested it with a 2080ti a 5600xt & 5700xt a 4090 and a nitro xtx card. (I even threw in a GTX 580 1.5GB and a HD6850 1GB because I like to have fun with numbers, and I was bored, but not because of any numbers you posted or in an attempt to match your data, just for fun.) I always use max settings with RT off, no upscaling ever.

I have used your benchmarks as a buying guide for years and have been able to copy all your results to within a margin of error countless times before, just not this time with the temps, which is why I thought it was a typo and didn't think anything else of it until that staff member. Your OC chart for the 7900 xtx nitro+ is the reason I bought one after I bought my 4090. The only difference when replicating your OC and UV tests for the nitro+ are that I get higher FPS than you showed, but I chalk that up to silicone lottery. (Posted chart for point of reference.) I can match everything with my 7950x tests, but I can't reach the same temps you show in your game testing.

Ryzen has quite the wide spread of temp, core quality, core V-F and IHS contact quality variation ever since they moved to chiplets. If you do that much hardware testing, you should already know that:

- if you are only monitoring Tctl/Tdie, then it is functionally a hotspot sensor that is highly dependent on your sample

- if you are monitoring per-core temps, then CPPC variation means always unique distribution of preferred cores, and the aforementioned temp variations between samples of the same CPU

 

[Miss_Cherry_Bomb]

What temps do you see with 7950X + RTX 4090 @ 1080p ? What CPU power usage?

At 1080p on a normal day temps are in the mid to high 70c area, on a hot day in the low to mid 80c area but never reaches high 80c. Power usage is about 85-110w depending on the game and intensity of what's happening and what's being rendered/loaded. The highest power draw I've ever seen on my CPU when gaming is around 120-150w at 1080p, and the temps never go above 82.7c in that game.

 

[dgianstefani]

Getting back on topic, seems others are starting to take issue with the significant IQ difference between DLSS and FSR.

"It's very disappointing to me that FSR hasn't received any improvements to the upscaling component since FSR 2.2, released in Nov 2022, with the exception of frame generation"

"...but it feels like we've been saying that a lot about AMD GPU features lately, so lets hope it doesn't launch late and half baked (AI FSR)"