Eurogamer Spec Analysis: PlayStation 5 Pro - the most powerful console yet |UP| 4/1 How Did Sony Beat Microsoft In Machine Learning For Consoles?

rinzler

Banned
12 Mar 2024
352
244
Question for people that understand tech.

These upscale techniques that they use, if an image is upscaled from, say, 1080 to 4k, if the original image was 30fps, does that stay when converted or can that be changed to 60fps?
That stays unless you have the CPU overhead to handle it.
 
  • Informative
  • haha
Reactions: Ezekiel and Sircaw

ethomaz

Rebolation!
21 Jun 2022
9,383
7,831
Brasil 🇧🇷
PSN ID
ethomaz
Question for people that understand tech.

These upscale techniques that they use, if an image is upscaled from, say, 1080 to 4k, if the original image was 30fps, does that stay when converted or can that be changed to 60fps?
30fps means 30 images (frames) per second… the upscale techs do it to each image/frame so it won’t increase the images/frames to 60 per second.

You will still have 30 images/frames per second but now they frames/images will be uoscalled to 4k.
 
  • Informative
Reactions: Sircaw

ethomaz

Rebolation!
21 Jun 2022
9,383
7,831
Brasil 🇧🇷
PSN ID
ethomaz
On PS systems it's been that if upscaled, fps could be raised as well as long as the game was "uncapped" in framerate
There is no upscale fps.

When you cap the framerate to 30fps it means the GPU is rendering over the target like 35, 37, 40, 48, etc and so it is caped to only output 30 and discard the rest.


When you uncap it the discard is cancelled and the output start to be what is being rendered: 35, 37, 40, 48, etc.

Upscaling refer to each frame… and not how many frames is being rendered… in fact upscaling has a overhead so doing upscaling will probably make the GPU render a few lower frames per second than if not upscaling… depend of course how costly is the upscaling algorithm to the GPU.

Plus reconstruction techs like DLSS are way more costly to GPU than upscaling… needs around 30% more GPU power… for example a 1080p target running at average 60fps will probably drop to around 40fps when using DSLL 1080p to 4k (Performance profile).

Of course the cost for DLSS 1080p to 4k is still lower than native 4k… if we take the 30% then it will be around 65% less expensive to the GPU than native 4k.
 
Last edited by a moderator:

Alabtrosmyster

Veteran
26 Jun 2022
3,230
2,840
Let me guess, "Its a good upgrade but wholly unnecessary. We look forward to what xbox is doing next though."
Pretty much, they avoid comparisons, except to imply that Sony somehow went in a more like series X path (even tho it has less CUs per TF)... but they are very much downplaying it with their analysis.

Also, they are still very "concerned" about the power envelope and those varying clock speeds.

These people are repeating the 2020 talking points.
 

anonpuffs

Veteran
Icon Extra
29 Nov 2022
9,006
10,278
We are all very aware of this as well, but what nobody else seems to be aware of is that a 35 teraflop RDNA 3 GPU performs to the same degree that a 16 teraflop RDNA 2 GPU does.
That's not quite true. For example, the 7800xt has 18.5 "rdna2" tflops and yet performs almost the same as the 6800xt which has over 20. There were architectural improvements so I expect performance on the level of an 18-19tf rdna2 card in raw performance.
 
OP
OP
Gamernyc78

Gamernyc78

MuscleMod
Moderating
28 Jun 2022
20,160
16,444
There is no upscale fps.

When you cap the framerate to 30fps it means the GPU is rendering over the target like 35, 37, 40, 48, etc and so it is caped to only output 30 and discard the rest.


When you uncap it the discard is cancelled and the output start to be what is being rendered: 35, 37, 40, 48, etc.

Upscaling refer to each frame… and not how many frames is being rendered… in fact upscaling has a overhead so doing upscaling will probably make the GPU render a few lower frames per second than if not upscaling… depend of course how costly is the upscaling algorithm to the GPU.

Plus reconstruction techs like DLSS are way more costly to GPU than upscaling… needs around 30% more GPU power… for example a 1080p target running at average 60fps will probably drop to around 40fps when using DSLL 1080p to 4k (Performance profile).

Of course the cost for DLSS 1080p to 4k is still lower than native 4k… if we take the 30% then it will be around 65% less expensive to the GPU than native 4k.
You took my comment out of context. There should have been a comma after upscaled seperating the two. Wth says upscaled fps? 😂😂
 
  • noneofmybizz
Reactions: ethomaz

ethomaz

Rebolation!
21 Jun 2022
9,383
7,831
Brasil 🇧🇷
PSN ID
ethomaz
That's not quite true. For example, the 7800xt has 18.5 "rdna2" tflops and yet performs almost the same as the 6800xt which has over 20. There were architectural improvements so I expect performance on the level of an 18-19tf rdna2 card in raw performance.
That is one of the biggest issue with AMD marketing…

7800XT should perform on pair with 6900XT not 6800XT.
6800XT performance level should be 7700.

When you release a new generation of cards you have to increase the performance in the same range level… not continue the same.

See nVidia.

2080… 1080TI
2070… 1080
3080… 2080TT
3070… 2080

7800XT performing similar to 6800XT is basically killing your own product… because 7800XT is more expensive and don’t delivery more.
 
  • Like
Reactions: Alabtrosmyster

Alabtrosmyster

Veteran
26 Jun 2022
3,230
2,840
That is one of the biggest issue with AMD marketing…

7800XT should perform on pair with 6900XT not 6800XT.
6800XT performance level should be 7700.

When you release a new generation of cards you have to increase the performance in the same range level… not continue the same.

See nVidia.

2080… 1080TI
2070… 1080
3080… 2080TT
3070… 2080

7800XT performing similar to 6800XT is basically killing your own product… because 7800XT is more expensive and don’t delivery more.
AMD has done fake new products for a very long time now.

The Radeon 480 and 580 were pretty much on par with each others and even back then it was talked about as if this was the usual way.
 

rinzler

Banned
12 Mar 2024
352
244
That's not quite true. For example, the 7800xt has 18.5 "rdna2" tflops and yet performs almost the same as the 6800xt which has over 20. There were architectural improvements so I expect performance on the level of an 18-19tf rdna2 card in raw performance.
While it's not an exact 2:1 science it very closely aligns with what can be expected.
 

ToTTenTranz

Veteran
Icon Extra
4 Aug 2023
733
791
We all know this, but we also know it will perform to the sum of the rough equivalent of a 16.7 teraflop RDNA 2 GPU.

Not really. We know a X TFLOP RDNA3 GPU used to perform similar to X/2 RDNA2 GPU on the PC, but even that is changing. RDNA3 cards have been slowly but steadily increasing performance with new driver releases.

For example, when the 7900 XT released in late 2022 it was ~22% on average faster than the 6900XT at 1440p:

Relative Performance 2560x1440


But in last month's batches of reviews, the 7900XT now appears as 26% faster than the same 6900XT:

relative-performance-2560-1440.png




On top of that, developers on the consoles can make much better use of AMD-specific features. Not only because they're certain that a larger share of their users will be making use of those, but also because Sony can automate many of these optimizations in their compiler. Such as making sure those dual-FP32 operations are used as much as possible.

You can't deduce the projected use and gain of vendor-specific optimizations between a console solution and the PC market (much less from a competitor that only has like 12% of marketshare).
 

ethomaz

Rebolation!
21 Jun 2022
9,383
7,831
Brasil 🇧🇷
PSN ID
ethomaz
For these interested in the RDNA3 vs RDNA2 at same computer power… that means same numbers of CUs at 2Ghz with the same memory bandwidth.


Rasterization: ~4% better
RT balanced: ~7% better
RT maximum: ~12% better
 
  • brain
Reactions: rinzler

Airbus

Veteran
30 Jun 2022
1,888
1,695
Bro ign and digital foundry already launch smear campaign against ps5pro for gta6

Lets see how gta6 perform on ps5pro next year

If theyre wrong it will prove theyre official fanbot shill