• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Discussion Playstation 6 speculation

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
MrMPFR said:
Yeah well MS clearly doesn't care + XSS GPU is useless and weaker than a 6500XT, RX 470/1650 territory. Would've never worked when RX 6600 struggles with FSR4 INT8.
Click to expand...

AFAICT the only RDNA2 GPU that may be able to use FSR4 INT8 without killing performance would be N21 because it's a pretty wide chip with plenty of compute. RDNA3 with WMMA and/or dual-issue should be able to do better.


marees said:
For ex: FSR4/PSSR has a magical effect in RE Requiem

similar thing could have been attempted with int8 version if it had true RDNA 2
Click to expand...
Not if 10-12 TFLOPs in a RDNA2 GPU were never going to have enough INT8 matrix throughput for an upscaler that targets 4K presentation.
It's a bit like people asking Valve to enable the RT option in windows games for the Steam Deck.
 
ToTTenTranz said:
AFAICT the only RDNA2 GPU that may be able to use FSR4 INT8 without killing performance would be N21 because it's a pretty wide chip with plenty of compute. RDNA3 with WMMA and/or dual-issue should be able to do better.
Click to expand...
It's still 1/3 more compute vs RX 6600, but yeah big compromises need to be made to hit frame rate target.
RDNA 3 only has dual issue for floating point not integers. Not sure about the WMMA instructions though.
 
MrMPFR said:
RDNA 3 only has dual issue for floating point not integers. Not sure about the WMMA instructions though.
Click to expand...
Yes, WMMA + dual-issue should allow RDNA3 to run FSR4 FP8 (promoted to FP16) at higher performance than INT8 because throughput would be twice as large. Assuming the hit on memory bandwidth isn't significant.
 
ToTTenTranz said:
Yes, WMMA + dual-issue should allow RDNA3 to run FSR4 FP8 (promoted to FP16) at higher performance than INT8 because throughput would be twice as large. Assuming the hit on memory bandwidth isn't significant.
Click to expand...
AMD's RDNA 4 slides show INT8 ops = FP16 ops. FP8 gotta run as full FP16. No tricks unfortunately + size doubled due to higher precision.

Momoka_ said:
SIE announced the news on the PS Blog, and DF also shared their take on PSSR2.0.
Click to expand...
I don't think this is the right thread to post it. Do we have a PSSR or PS5 thread?
 
MrMPFR said:
AMD's RDNA 4 slides show INT8 ops = FP16 ops. FP8 gotta run as full FP16. No tricks unfortunately + size doubled due to higher precision.


I don't think this is the right thread to post it. Do we have a PSSR or PS5 thread?
Click to expand...
I have a ps5 pro thread (sort of)

Discussion - Will the PS5 pro have the performance of 4070 or 4070 super ?

Usecase for the pro: the PS5 is like pairing a zen 2 CPU with 4070 / 4070 super in gpu bound gaming scenarios Large monitor 40 fps 4K with DLSS RT enabled Which GPU can do this ? My understanding, in raster: 7700xt = 4070 Ps5 pro = ??? 7800xt = 4070 super In RT N48xtx = 4070 ti N48 xt =...
forums.anandtech.com forums.anandtech.com
 
Last edited:
marees said:
I have a ps5 pro thread (sort of)

Discussion - Will the PS5 pro have the performance of 4070 or 4070 super ?

Usecase for the pro: the PS5 is like pairing a zen 2 CPU with 4070 / 4070 super in gpu bound gaming scenarios Large monitor 40 fps 4K with DLSS RT enabled Which GPU can do this ? My understanding, in raster: 7700xt = 4070 Ps5 pro = ??? 7800xt = 4070 super In RT N48xtx = 4070 ti N48 xt =...
forums.anandtech.com forums.anandtech.com
Click to expand...

@marees can you please remove the lengthy quote at the bottom?

It seems like the only Anandthread about PS5 Pro. But maybe the poll needs to be closed + header changed? ¯\_(ツ)_/¯
 
Last edited:
MrMPFR said:
AMD's RDNA 4 slides show INT8 ops = FP16 ops. FP8 gotta run as full FP16. No tricks unfortunately + size doubled due to higher precision.
Click to expand...

Correct, I calculated it wrong. In any case, if FSR4 FP8-promoted-to-FP16 fits into the VOPD instructions that allow for dual-issue, RDNA3 should be able to run FP8 FSR4 with similar performance to INT8 FSR4.


marees said:
the ps5 pro era is just getting started with pssr2. so will keep it open for some more time
Click to expand...
Hold your horses, now... PSSR2 is in one (1, uno) game so far.
 
ToTTenTranz said:
Correct, I calculated it wrong. In any case, if FSR4 FP8-promoted-to-FP16 fits into the VOPD instructions that allow for dual-issue, RDNA3 should be able to run FP8 FSR4 with similar performance to INT8 FSR4.
Click to expand...

Do you know if people have tested both models side by side on linux? There has to be more changes to the INT8 model than just INT8.
 
MrMPFR said:
Do you know if people have tested both models side by side on linux?
Click to expand...

AFAIK, no one has been able to run FSR4 with WMMA on RDNA3, only through regular compute shaders. Which means the FP8 version runs with a large performance deficit since it's running at twice the compute cost of the INT8 version.
 
ToTTenTranz said:
AFAIK, no one has been able to run FSR4 with WMMA on RDNA3, only through regular compute shaders. Which means the FP8 version runs with a large performance deficit since it's running at twice the compute cost of the INT8 version.
Click to expand...
Crap I forgot.
Well if it's >2X ms cost then INT8 is def a lighter model. Haven't checked the numbers though.
 
NTMBK said:
At best, I'm hopeful the games industry stabilises again in 5 years or so.
Click to expand...
~2031+
With this timeline the PS6 is DOA, could suggest such a thing will never even happen.

I'll leave you all to speculate as to when PS6 launches, if it ever will.
 
MrMPFR said:
~2031+
With this timeline the PS6 is DOA, could suggest such a thing will never even happen.

I'll leave you all to speculate as to when PS6 launches, if it ever will.
Click to expand...

People are assuming the culprit for all this damage to consumer electronics, all these gigawatt AI factories being announced, are going to be built.


I don't think they are. I think eventually everyone will acknowledge that foundation and frontier models hit a performance wall because they have no more high quality data to feed on, and at some point Sam Altman and his empty AGI promises will be ignored by regulators and investors. The sole reason the major governments haven't started to step in with regulations over blasting senseless power in training to increase model performance by 0.1% is because they still think AI is a race they can't lose or they risk national security, but that too will fade as soon as VCs start to drop out of this.

If anything, we're seeing the smaller distilled models that run on consumer GPUs get major leaps in performance (just look at Qwen3.5), and the near/mid-term future of actually useful AI will be in agents with increasingly advanced orchestrators using local inference of smaller models wherever possible for data privacy compliance. Physical AI will mostly use local inference because factory managers aren't going to put production lines dependent on internet connection and/or the mood swings of mega corporations. This is a lot more boring than claiming humanity developed AGI but it's real, tangible, actually useful and it's a good business proposition.



And what this actually means is most of those AI factories won't really get built, eventually there'll be a ton of memory chips free to put in consumer devices and we will eventually see a PS6, a Xbox Next, etc.
 
MrMPFR said:
From the recent Loose Ends livestream:
wccftech.com

PlayStation 6 Reportedly Won't Be Delayed Due to RAM Costs: "It Would Cost More to Delay Than to Pay Extra"

Sony is reportedly unlikely to delay the PlayStation 6, as it would cost more to delay the system than to pay the extra RAM costs
wccftech.com wccftech.com

Argument is if they delay they risk their contracts pushing the launch back by multiple years.
Uses Pandemic pricing disaster as an example that they can take a loss.

@Kepler_L2 thoughts?
Click to expand...
contracts can be renegotiated.

AMD's server cpus are selling like gang busters.

so if the 3 parties come to an understanding then it can be done
 
MrMPFR said:
From the recent Loose Ends livestream:
wccftech.com

PlayStation 6 Reportedly Won't Be Delayed Due to RAM Costs: "It Would Cost More to Delay Than to Pay Extra"

Sony is reportedly unlikely to delay the PlayStation 6, as it would cost more to delay the system than to pay the extra RAM costs
wccftech.com wccftech.com

Argument is if they delay they risk their contracts pushing the launch back by multiple years.
Uses Pandemic pricing disaster as an example that they can take a loss.

@Kepler_L2 thoughts?
Click to expand...
I don't think PS6 will be delayed.
 
marees said:
contracts can be renegotiated.

AMD's server cpus are selling like gang busters.

so if the 3 parties come to an understanding then it can be done
Click to expand...
on second thoughts while this is in theory possible, what will happen is

  1. TSMC has announced consecutive hikes for 3nm and advanced nodes for next 4 years
  2. sony probably already locked in supply at a 'cheap' price
  3. relinquishing their quota now, will only benefit AMD but for Sony they might lose access to cheap & pre-agreed 3nm prices
  4. so even if sony delays to 2029, still they pay MORE for the same 3nm but by that time RDNA 6 would be on the horizon & selling an RDNA 4.8 based console when RDNA 6 is about to be released (& all this at a still inflated price) will not be a good look
  5. what Sony can do is release on schedule (2028) but cut down on memory — rely on AMD's software tricks to optimize memory usage
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top