Support NeoGAF

[DeepEnigma]

Not mutually exclusive.. you can cover both ends of the spectrum with a 2 or 3 SKUs

Not if the sunk cost doesn't justify the means. It's all a balance, you see.

 

[PeteBull]

Ask Sony who almost crashed and burned by pricing their premium console at $600.

The R&D and associated costs with developing a machine you'll sell for $800 don't make sense in the console space.

Obviously, but the thing is, ps3 launched in 2006, 600$ back then is equivalent to over 900$ now, and we talking pr0 console, aka sony could be ok with it selling only 15-20% of total instal base of ps5, ofc its super unrealistic/wishful thinking/dream, but i would love at least 800$ ps5pr0 with specs that back up such price point, not standard 400-500$ pricepoint when machine is at most midrange pc components specs wise

 

[twilo99]

Obviously, but the thing is, ps3 launched in 2006, 600$ back then is equivalent to over 900$ now, and we talking pr0 console, aka sony could be ok with it selling only 15-20% of total instal base of ps5, ofc its super unrealistic/wishful thinking/dream, but i would love at least 800$ ps5pr0 with specs that back up such price point, not standard 400-500$ pricepoint when machine is at most midrange pc components specs wise

There are many like you out there...

 

[Meicyn]

Selling a $800+ console in this economy is straight up comedy. Some of you need to interact with the outside world.
PS4 Pro accounted for only 20% of units sold in previous gen, and that debuted at $400. Everyone else bought the cheaper option.

 

[Mr.Phoenix]

I understand all of that, my question is why use a CPU architecture from 2019 in 2025? It made sense to use Zen2 in 2020, but it makes very little sense to use the same thing 6 years later, and since you know a lot about tech, you would also know that 6 years in CPU development is a VERY long time.

Of course the devs will make the games work on it, thay can technically make those game run on Zen2 in 2030, but why bottleneck the whole system like that.. it makes very little sense to me.

You are in the wrong market.

And in CPU terms, 6 years isn't that bad. GPU on the other hand... yes. But more importantly, and what you keep ignoring. The base console is the PS5. Its not like the PS5pro is a completely new platform, or that the games running on it are different. Its running the exact same games that the PS5 runs, why would you need a much better CPU? And another thing, look at its target market. Its a console, that is primarily plugged into a TV. All it needs to do is run games at 60hz for most people.

That's exactly it.. the PC market has shown that gamers have a very real thirst for performance, and I don't see why that can't be replicated on the console side.



Nah, I don't buy it.. until someone tests the market for a high end console and fails I don't think you can make such conclusions.

You cant be serious... don't get it twisted. 100M people are not buying $2000 GPUs. Why don't you look at the most common GPU on Steam? I think you will be surprised. And until someone tests the market? Did you sleep throughout the PS3 gen?

 

[sankt-Antonio]

Selling a $800+ console in this economy is straight up comedy. Some of you need to interact with the outside world.
PS4 Pro accounted for only 20% of units sold in previous gen, and that debuted at $400. Everyone else bought the cheaper option.

You can get a Apple phone for 399 all the way to 1299, these 20% of pro consoles are a better ROI for Sony so its worth having that SKU.
I can even imagine them having a PS5 slim, PS5, PS5pro and PS5pro max. from 299 all the way up to 800+. In almost every electronics niche you'll find these spreads of options. TV's, Cameras, HiFi equipment, etc...

 

[twilo99]

You can get a Apple phone for 399 all the way to 1299, these 20% of pro consoles are a better ROI for Sony so its worth having that SKU.
I can even imagine them having a PS5 slim, PS5, PS5pro and PS5pro max. from 299 all the way up to 800+. In almost every electronics niche you'll find these spreads of options. TV's, Cameras, HiFi equipment, etc...

 

[rofif]

You can get a Apple phone for 399 all the way to 1299, these 20% of pro consoles are a better ROI for Sony so its worth having that SKU.
I can even imagine them having a PS5 slim, PS5, PS5pro and PS5pro max. from 299 all the way up to 800+. In almost every electronics niche you'll find these spreads of options. TV's, Cameras, HiFi equipment, etc...

That's not exactly what we want for consoles though right?
I mean, 1 pro ever gen is ALREADY PUSHING IT, Especailly this gen with barely anything using ps5 properly since lazy devs like remedy are just using ps5 as bad pc and slap low pc settings on their ps5 ports..... which is already stupid.

But what if we had ps5, ps5 light, ps5 pro, ps5 ultra, ps5pro ultra fold, ps5+ etc... every year.
If phones can do it, why can't consoles? They are cheaper than phones.
Don't give them ANY IDEAS

 

[Meicyn]

these 20% of pro consoles are a better ROI for Sony so its worth having that SKU

citation needed

 

[HeisenbergFX4]

Selling a $800+ console in this economy is straight up comedy. Some of you need to interact with the outside world.
PS4 Pro accounted for only 20% of units sold in previous gen, and that debuted at $400. Everyone else bought the cheaper option.

$800 really beefy console?

How many would I buy?

 

[Insane Metal]

$800 really beefy console?

How many would I buy?

Another one to gift me please

 

[Salty Popsicle]

You can get a Apple phone for 399 all the way to 1299, these 20% of pro consoles are a better ROI for Sony so its worth having that SKU.
I can even imagine them having a PS5 slim, PS5, PS5pro and PS5pro max. from 299 all the way up to 800+. In almost every electronics niche you'll find these spreads of options. TV's, Cameras, HiFi equipment, etc...

Consoles will never compete with high-end PC's, not even these "Pro" consoles. Anyone who wants the latest and greatest in graphics and performance in games will need to look elsewhere.

 

[leizzra]

Nah, I don't buy it.. until someone tests the market for a high end console and fails I don't think you can make such conclusions.

Don’t buy what? That you don’t have millions of expensive cards (and part of that sales are not purely for gaming) on the marked vs tens or hundreds millions of consoles? Or that PS4Pro and XOX weren’t used to their full potential (especially Xbox) and the same will be with next console upgrades? That dev’s have a headache with more then one SKU? That QA with each SKU has additional game to test?

Also the talk about Apple and many devices - I started my career in the industry from mobile games in well respected company. It was a nightmare to work with and the games were very small in comparison to what we have at consoles now. It’s not a good way to go, especially when AAA games have already a huge issues with overblown budgets. In the end the quality of games would take a hit.

 

[HeWhoWalks]

Consoles will never compete with high-end PC's, not even these "Pro" consoles. Anyone who wants the latest and greatest in graphics and performance in games will need to look elsewhere.

 

[MikeM]

Consoles will never compete with high-end PC's, not even these "Pro" consoles. Anyone who wants the latest and greatest in graphics and performance in games will need to look elsewhere.

If they can offer mid range gaming PC equivalents for less than an equivalent PC than thats all they need to do.
The high end PC gamers is a fraction of PCMR.

 

[twilo99]

Don’t buy what? That you don’t have millions of expensive cards (and part of that sales are not purely for gaming) on the marked vs tens or hundreds millions of consoles? Or that PS4Pro and XOX weren’t used to their full potential (especially Xbox) and the same will be with next console upgrades? That dev’s have a headache with more then one SKU? That QA with each SKU has additional game to test?

Also the talk about Apple and many devices - I started my career in the industry from mobile games in well respected company. It was a nightmare to work with and the games were very small in comparison to what we have at consoles now. It’s not a good way to go, especially when AAA games have already a huge issues with overblown budgets. In the end the quality of games would take a hit.

I don't buy the one size fits all thing that the console world has been stuck in for decades now is the best way to do it and I think that there is a whole untapped market for a high end console that no one has exploited yet.

Big studios can figure out how to make games for multiple SKUs.. you will be paying $100 a game soon enough anyway

 

[mckmas8808]

That would be a significantly jump over the actual ps5.

How many TF is the 4070?

 

[rsouzadk]

How many TF is the 4070?

About 29 TF.

 

[mckmas8808]

About 29 TF.

Jesus man wow! That's insane!

 

[OverHeat]

If they can offer mid range gaming PC equivalents for less than an equivalent PC than thats all they need to do.
The high end PC gamers is a fraction of PCMR.

Love being on top

 

[Mr Moose]

Consoles will never compete with high-end PC's, not even these "Pro" consoles. Anyone who wants the latest and greatest in graphics and performance in games will need to look elsewhere.

Yup, they will be

And that's fine. For the price it will be fantastic. They will hit a bit above their weight and be worth the price.

 

[leizzra]

Big studios can figure out how to make games for multiple SKUs.. you will be paying $100 a game soon enough anyway

There is, it’s called PC.

Of course I’m exaggerating because there is a difference in a way. But as I said you are watering the advantages of the console by multiplying the numbers of game versions. Especially if you want to take the advantages of more powerful platforms.

Also I like that you believe in magic. Big studios are figuring out how to cut costs. We are thinking a ways to make the process easier not harder. And there are other studios that also needs to be taken into consideration.

In the end I see this is a pointless discussion. You don’t want to listen and offer nothing in return. At end of the day console manufacturers want to earn money. They do what they do to achieve that and have years of experience in doing so. They already were thinking about hardware pricing and there is a reason why it’s not this expensive. People here need to realize that they are only a small percentage in global market.

 

[DeepEnigma]

There is, it’s called PC.

Of course I’m exaggerating because there is a difference in a way. But as I said you are watering the advantages of the console by multiplying the numbers of game versions. Especially if you want to take the advantages of more powerful platforms.

Also I like that you believe in magic. Big studios are figuring out how to cut costs. We are thinking a ways to make the process easier not harder. And there are other studios that also needs to be taken into consideration.

In the end I see this is a pointless discussion. You don’t want to listen and offer nothing in return. At end of the day console manufacturers want to earn money. They do what they do to achieve that and have years of experience in doing so. They already were thinking about hardware pricing and there is a reason why it’s not this expensive. People here need to realize that they are only a small percentage in global market.

Exactly. It's not an honest discussion.

 

[sankt-Antonio]

Consoles will never compete with high-end PC's, not even these "Pro" consoles. Anyone who wants the latest and greatest in graphics and performance in games will need to look elsewhere.

Sony’s mirrorless cameras aren’t competing with cine cameras but are still available from 1000 all the way up to 6000 $

Sometimes having something that’s rather simple in operation but still relatively high power is what people want.

I would rather spend 800€ on a PS5pro max and have a seamless experience than buying a gaming PC and fiddle around with it.
People buy Sony’s 199€ controllers… they will buy a 1000€ PS5 SKU.

 

[twilo99]

I'm trying to offer you an alternative outlook on this but y'all stuck in a rut..

I can't really make it intellectual, its a hypothetical scenario where you have an offering at different price points so that the consumer has more options instead of just one, which I genuinely believe to be beneficial.

 

[twilo99]

Sony’s mirrorless cameras aren’t competing with cine cameras but are still available from 1000 all the way up to 6000 $

Sometimes having something that’s rather simple in operation but still relatively high power is what people want.

I would rather spend 800€ on a PS5pro max and have a seamless experience than buying a gaming PC and fiddle around with it.
People buy Sony’s 199€ controllers… they will buy a 1000€ PS5 SKU.

100%

A premium PlayStation will do extremely well, but instead, you will be getting the same old outdated hardware because that's your only choice.

 

[sankt-Antonio]

Gaming as an Industry evolved. It’s not the kid’s toy it was back in the 80/90s.
People who grew up with gaming are financially able and imo willing to spend more now. For them gaming is a very cheap hobby, compared to most other hobbies.

If there was a PS5pro at launch I feel the take rate would exceed 50%.

 

[Go_Ly_Dow]

I'm sticking with $549-$649.

 

[ShakenG]

Random nobody here.. just throwing in a guess of 15 - 16 TF if we are using that general metric. For reasons i cant explain, maybe a hunch. Im not expecting anything around 20 TF.
And Sony need it to be worthy of a PRO title so it cant be to small of a jump.
(Not expecting 12 -13.)
Im sure ive made guesses in the past but this is where i sit as of now.

 

[Brigandier]

Random nobody here.. just throwing in a guess of 15 - 16 TF if we are using that general metric. For reasons i cant explain, maybe a hunch. Im not expecting anything around 20 TF.
And Sony need it to be worthy of a PRO title so it cant be to small of a jump.
(Not expecting 12 -13.)
Im sure ive made guesses in the past but this is where i sit as of now.

If the GPU in the PS5 Pro is based on a 7800XT then it will be much more capable than 15tf.

The desktop 7800XT is supposed to be around 37tf so a modified version for a console should be hitting at least 25-30tf or I should imagine so.

 

[Mr.Phoenix]

If the GPU in the PS5 Pro is based on a 7800XT then it will be much more capable than 15tf.

The desktop 7800XT is supposed to be around 37tf so a modified version for a console should be hitting at least 25-30tf or I should imagine so.

Technically, the 7800XT is actually around 18.5TF. Dual issue compute, is what gets it up to 37TF.

In each compute unit, there are 32 ALUs (stream processors) and as such a 32-wide wavefront (instruction set). What AMD (and Nvidia) is doing (or claiming) is allowing the ALU get two instructions instead of one. Hence the dual issue thing. But its extremely dependent on parallelism, which is a problem the GCN architecture had and then AMD shifted to RDNA. So basically, unless code is extremely optimized for a 32-wide wavefront, dual-issue computing will not work.

Evidence of this...

On a game like Avatar, let's look at the performance in 1440p on 3 GPUs

• 6900XT - 23TF, 16GB, No DIC = 48fps
• 6800XT - 20.7TF, 16GB, No DIC = 45fps
• 7800XT - 18.66TF, 16GB, 37.3TF DIC = 44fps

As you can see, the 7800XT is not showing anywhere near double the performance that it should be showing considering the DIC number but is instead more accurately represented by its standard compute number. Ideally, it should be significantly outperforming not just the 6800xt but also the 6900xt, but that would require devs to write very very very specific and optimized code for it.

The PS5pro should have support for DIC, since its based on RDNA3, and maybe Sony first-party devs would take advantage of it, as for third parties, considering that we have yet to see a single game on either AMD or Nvidia GPUs take advantage of it, I won't hold my breath.

I believe the PS5pro GPU will be between 16-18TF. Then you can double that if you want to use the DIC number.

 

[ShakenG]

If the GPU in the PS5 Pro is based on a 7800XT then it will be much more capable than 15tf.

The desktop 7800XT is supposed to be around 37tf so a modified version for a console should be hitting at least 25-30tf or I should imagine so.

I cant help but be highly conservative on my expectations. I want it up there around 20 but 25-30!!! dont see it happening.
Too many people in here expecting PS6 specs and are forgetting what a pro console most likely will bring.
Don't mean to be a buzzkill..

 

[Bernoulli]

I cant help but be highly conservative on my expectations. I want it up there around 20 but 25-30!!! dont see it happening.
Too many people in here expecting PS6 specs and are forgetting what a pro console most likely will bring.
Don't mean to be a buzzkill..

It would be surprising if even the next gen xbox in 2025 or 2026 could reach 30 TF

 

[Mr.Phoenix]

I cant help but be highly conservative on my expectations. I want it up there around 20 but 25-30!!! dont see it happening.
Too many people in here expecting PS6 specs and are forgetting what a pro console most likely will bring.
Don't mean to be a buzzkill..

Its not even you being conservative. Its you being practical. I don't know where people are getting the 20, 25 or 30TF+ numbers from. There is nothing, if looking at what Sony has done thus far, or what AMD has done, or just using a base knowledge of chip/console design, to suggest what some people are hoping or expecting.

First things first, its a console, so it has to fit within a certain chassis and be built to a very specific thermal design. Then there is the cost, and size of the chip, which will also determine its thermal ceiling, which also determines its clocks. And we pretty much have all the info to make a very accurate educated guess as to what it would be.

Its supposed to be based on RDNA3 + Zen2. So that's the 7xxx series GPUs. The 7800XT is the best 7xxx GPU out there, and it has 60CU. It also is 200mm2, not including infinity cache and mem controllers. So this means that at best, the PS5 is some variant of 60CU. Be that 60 with 54/56 active, or 64 with 60 active. Can't be more than that. Then the 7800xt has a boost clock of 2400Mhz.

From that, we can say the PS5pro will have an APU of around 300-320mm2. I am pegging it at around 2300 MHz. Regardless of what compute unit config we then choose, that puts the PS5pro at around 16-17TF. I once thought, the PS5pro could be clocked as high as 2500Mhz, until I saw the 7800xt. If on 5nm that thing cant be clocked higher than 2400Mhz, then no way a PS5pro (even on 4nm) would hit 2400Mhz.

I don't get where people are getting 20TF or 25TF from. It's just not practical.

 

[RagnarokIV]

Hilariously shit ideas here.

A $1000 PS5 SKU? What, and still no backwards compatibility? LMAO

 

[Brigandier]

Technically, the 7800XT is actually around 18.5TF. Dual issue compute, is what gets it up to 37TF.

In each compute unit, there are 32 ALUs (stream processors) and as such a 32-wide wavefront (instruction set). What AMD (and Nvidia) is doing (or claiming) is allowing the ALU get two instructions instead of one. Hence the dual issue thing. But its extremely dependent on parallelism, which is a problem the GCN architecture had and then AMD shifted to RDNA. So basically, unless code is extremely optimized for a 32-wide wavefront, dual-issue computing will not work.

Evidence of this...

On a game like Avatar, let's look at the performance in 1440p on 3 GPUs

• 6900XT - 23TF, 16GB, No DIC = 48fps
• 6800XT - 20.7TF, 16GB, No DIC = 45fps
• 7800XT - 18.66TF, 16GB, 37.3TF DIC = 44fps

As you can see, the 7800XT is not showing anywhere near double the performance that it should be showing considering the DIC number but is instead more accurately represented by its standard compute number. Ideally, it should be significantly outperforming not just the 6800xt but also the 6900xt, but that would require devs to write very very very specific and optimized code for it.

The PS5pro should have support for DIC, since its based on RDNA3, and maybe Sony first-party devs would take advantage of it, as for third parties, considering that we have yet to see a single game on either AMD or Nvidia GPUs take advantage of it, I won't hold my breath.

I believe the PS5pro GPU will be between 16-18TF. Then you can double that if you want to use the DIC number.

Gotcha clearly explained in that case yeah realistically the Pro will be 15-18tf.

Hopefully Sony's 1st party studios take full advantage of DIC though I'd like to see some real generational leaps!!

I'm still hoping that the Pro will at least have some RT secret sauce

 

[HeisenbergFX4]

Hilariously shit ideas here.

A $1000 PS5 SKU? What, and still no backwards compatibility? LMAO

 

[HeisenbergFX4]

Its not even you being conservative. Its you being practical. I don't know where people are getting the 20, 25 or 30TF+ numbers from. There is nothing, if looking at what Sony has done thus far, or what AMD has done, or just using a base knowledge of chip/console design, to suggest what some people are hoping or expecting.

First things first, its a console, so it has to fit within a certain chassis and be built to a very specific thermal design. Then there is the cost, and size of the chip, which will also determine its thermal ceiling, which also determines its clocks. And we pretty much have all the info to make a very accurate educated guess as to what it would be.

Its supposed to be based on RDNA3 + Zen2. So that's the 7xxx series GPUs. The 7800XT is the best 7xxx GPU out there, and it has 60CU. It also is 200mm2, not including infinity cache and mem controllers. So this means that at best, the PS5 is some variant of 60CU. Be that 60 with 54/56 active, or 64 with 60 active. Can't be more than that. Then the 7800xt has a boost clock of 2400Mhz.

From that, we can say the PS5pro will have an APU of around 300-320mm2. I am pegging it at around 2300 MHz. Regardless of what compute unit config we then choose, that puts the PS5pro at around 16-17TF. I once thought, the PS5pro could be clocked as high as 2500Mhz, until I saw the 7800xt. If on 5nm that thing cant be clocked higher than 2400Mhz, then no way a PS5pro (even on 4nm) would hit 2400Mhz.

I don't get where people are getting 20TF or 25TF from. It's just not practical.

 

[Gaiff]

Technically, the 7800XT is actually around 18.5TF. Dual issue compute, is what gets it up to 37TF.

In each compute unit, there are 32 ALUs (stream processors) and as such a 32-wide wavefront (instruction set). What AMD (and Nvidia) is doing (or claiming) is allowing the ALU get two instructions instead of one. Hence the dual issue thing. But its extremely dependent on parallelism, which is a problem the GCN architecture had and then AMD shifted to RDNA. So basically, unless code is extremely optimized for a 32-wide wavefront, dual-issue computing will not work.

Evidence of this...

On a game like Avatar, let's look at the performance in 1440p on 3 GPUs

• 6900XT - 23TF, 16GB, No DIC = 48fps
• 6800XT - 20.7TF, 16GB, No DIC = 45fps
• 7800XT - 18.66TF, 16GB, 37.3TF DIC = 44fps

As you can see, the 7800XT is not showing anywhere near double the performance that it should be showing considering the DIC number but is instead more accurately represented by its standard compute number. Ideally, it should be significantly outperforming not just the 6800xt but also the 6900xt, but that would require devs to write very very very specific and optimized code for it.

The PS5pro should have support for DIC, since its based on RDNA3, and maybe Sony first-party devs would take advantage of it, as for third parties, considering that we have yet to see a single game on either AMD or Nvidia GPUs take advantage of it, I won't hold my breath.

I believe the PS5pro GPU will be between 16-18TF. Then you can double that if you want to use the DIC number.

Well thought out and I really wish people stopped focusing on TFlops so much. Even after 3 years of the current generation and whacky TFlops numbers that scale all over the place, people still use it as THE reference for performance. They should take a look at how Lovelace's compute scales in gaming vs workstation applications.

Things like ROPs, pixel/texel fillrate, memory bandwidth, memory type, bus width, die size/area, cache, etc, all get ignore to focus on compute.

 

[HeisenbergFX4]

Well thought out and I really wish people stopped focusing on TFlops so much. Even after 3 years of the current generation and whacky TFlops numbers that scale all over the place, people still use it as THE reference for performance. They should take a look at how Lovelace's compute scales in gaming vs workstation applications.

Things like ROPs, pixel/texel fillrate, memory bandwidth, memory type, bus width, die size/area, cache, etc, all get ignore to focus on compute.

Its going to be hard for the majority of people to not use the TF metric to base performance on since (especially Xbox) was screaming it every chance they got

Have been saying it for maybe close to a year here TFs wont be the buzzword but RT will be for the Pro

Plus again people are going to see the actual TF number for the Pro and think its weak but wait for performance numbers to roll in

Another side note once the TF numbers are out there I am quite certain the "armies" have their marching orders to get out there the supposed TF numbers of the next xbox the closer the Pro gets to launching

 

[leizzra]

I'm trying to offer you an alternative outlook on this but y'all stuck in a rut..

Yes, you are like Prometheus who brings people fire, but people already have an electricity.

I can't really make it intellectual, its a hypothetical scenario where you have an offering at different price points so that the consumer has more options instead of just one, which I genuinely believe to be beneficial.

Exactly, hypothetical which means your thesis is right or wrong. Now it needs to be tested. Non of us can do it as it should be, we can only speculate using data we have. And that’s why some people here are showing you that it’s not a good and viable option. I haven’t seen here any good reasons why it is doable better then ‘trust me, I would buy it’.

Gaming as an Industry evolved. It’s not the kid’s toy it was back in the 80/90s.
People who grew up with gaming are financially able and imo willing to spend more now. For them gaming is a very cheap hobby, compared to most other hobbies.

If there was a PS5pro at launch I feel the take rate would exceed 50%.

If we look at the numbers of consoles and what (if I recall correctly) Microsoft has shown during ActiBlizz case, number of potential customers isn’t growing much for consoles. Moreover you don’t see even numbers as high as PS2. Where are those gamers, playing on PC? If they are willing to spend so much why those graphic cards aren’t selling in tens or hundreds of millions? Or why consoles aren’t selling 200 millions or more.

There are limits. Hardcore gamers can invest more in their hobby, but they are like 1:5 with casuals. For those, gaming isn’t important the most, especially now after pandemic has ended.

For me video games are very important. Heck, they are my love and life. I also work in the industry because of that. If the consoles would be for 1k USD/euros I’m not sure if I would buy them right away. I don’t feel comfortable with that price even though I can afford it.

Expensive consoles have sense only if they are the main platform for dev’s. If not it’s only a waste of money. PC hardware is good example of it - those GPUs are so much underused and rapidly replaced with newer models.

 

[yurinka]

A $1000 PS5 SKU? What, and still no backwards compatibility? LMAO

PS5 is BC

 

[sankt-Antonio]

Yes, you are like Prometheus who brings people fire, but people already have an electricity.


Exactly, hypothetical which means your thesis is right or wrong. Now it needs to be tested. Non of us can do it as it should be, we can only speculate using data we have. And that’s why some people here are showing you that it’s not a good and viable option. I haven’t seen here any good reasons why it is doable better then ‘trust me, I would buy it’.



If we look at the numbers of consoles and what (if I recall correctly) Microsoft has shown during ActiBlizz case, number of potential customers isn’t growing much for consoles. Moreover you don’t see even numbers as high as PS2. Where are those gamers, playing on PC? If they are willing to spend so much why those graphic cards aren’t selling in tens or hundreds of millions? Or why consoles aren’t selling 200 millions or more.

There are limits. Hardcore gamers can invest more in their hobby, but they are like 1:5 with casuals. For those, gaming isn’t important the most, especially now after pandemic has ended.

For me video games are very important. Heck, they are my love and life. I also work in the industry because of that. If the consoles would be for 1k USD/euros I’m not sure if I would buy them right away. I don’t feel comfortable with that price even though I can afford it.

Expensive consoles have sense only if they are the main platform for dev’s. If not it’s only a waste of money. PC hardware is good example of it - those GPUs are so much underused and rapidly replaced with newer models.

I’ll never buy a 1000€ gpu but would buy a 1000€ premium console.

Can’t really compare the two. It’s DIY vs out of the box.

If only 15% of the potential 100.000.000 userbase thinks like that it’s a viable business decision.

Just look at the yearly released TVs from Sony. They go up to 3500€ and only sell low volume, but it still makes sense for Sony to sell them.

 

[ChiefDada]

Technically, the 7800XT is actually around 18.5TF. Dual issue compute, is what gets it up to 37TF.

In each compute unit, there are 32 ALUs (stream processors) and as such a 32-wide wavefront (instruction set). What AMD (and Nvidia) is doing (or claiming) is allowing the ALU get two instructions instead of one. Hence the dual issue thing. But its extremely dependent on parallelism, which is a problem the GCN architecture had and then AMD shifted to RDNA. So basically, unless code is extremely optimized for a 32-wide wavefront, dual-issue computing will not work.

Evidence of this...

On a game like Avatar, let's look at the performance in 1440p on 3 GPUs

• 6900XT - 23TF, 16GB, No DIC = 48fps
• 6800XT - 20.7TF, 16GB, No DIC = 45fps
• 7800XT - 18.66TF, 16GB, 37.3TF DIC = 44fps

As you can see, the 7800XT is not showing anywhere near double the performance that it should be showing considering the DIC number but is instead more accurately represented by its standard compute number. Ideally, it should be significantly outperforming not just the 6800xt but also the 6900xt, but that would require devs to write very very very specific and optimized code for it.

The PS5pro should have support for DIC, since its based on RDNA3, and maybe Sony first-party devs would take advantage of it, as for third parties, considering that we have yet to see a single game on either AMD or Nvidia GPUs take advantage of it, I won't hold my breath.

I believe the PS5pro GPU will be between 16-18TF. Then you can double that if you want to use the DIC number.

Excellent write-up. My question: Isn't difficult optimization the reason why dual issue performance didn't pan out for RDNA 3? Wouldn't the console space have a much easier time executing? So 2x tflop number should at least be taken somewhat more seriously.

 

[Panajev2001a]

Another side note once the TF numbers are out there I am quite certain the "armies" have their marching orders to get out there the supposed TF numbers of the next xbox the closer the Pro gets to launching

I can believe that .

 

[TheRedRiders]

Excellent write-up. My question: Isn't difficult optimization the reason why dual issue performance didn't pan out for RDNA 3? Wouldn't the console space have a much easier time executing? So 2x tflop number should at least be taken somewhat more seriously.

You have two options when it comes to dual issue compute, the first is you can leave it to the driver to auto-compile your code to take advantage, but AMD's compiler for dual issue compute is kind of crap, and the ability to even compile to dual issue would be relative to the kind of workloads being performed.

The second option is written assembly by developers, in which developers can write their own code to take advantage of the feature, not sure what the gains would be through this method but likely higher than auto compilation. However developers may be reluctant to adopt this feature because of lack of experience, and more importantly developer time and effort.

This will be complicated further by memory bandwidth constraints, RDNA 3 already has a weak memory hierarchy, so running wave32 dual issue and wave 64 code can easily choke out the memory bandwidth if there's poor optimisation. RDNA 3 cards are already suffering from this problem.

 

[Mr.Phoenix]

Gotcha clearly explained in that case yeah realistically the Pro will be 15-18tf.

Hopefully Sony's 1st party studios take full advantage of DIC though I'd like to see some real generational leaps!!

I'm still hoping that the Pro will at least have some RT secret sauce

The RT "secret sauce" is the only hunch I am most confident in. Having a slightly bigger RTU in each CU that accelerates the BVH tree in an RT pipeline (which its inability to do so is the primary reason RDNA2/RDNA3 GPUs suck at RT and are so far behind Nvidia cards) is a far smarter silicon spend than throwing more CUs at the problem. And there have been too many patent leaks and noise about RT from Sony and talk about 3rd gen RT in RDNA4 for there to not be something happening on that front.

And on the secret sauce, you can also look out for the AI stuff. ML accelerators are present in RDNA3 (two per CU to be exact). And the way it's integrated would seem like a stretch for Sony to cut out. But you never know. If taken as far as it can be taken, it could mean all PS5pro native apps (basically games designed with native support for the PS5pro) could have a specific AI-based reconstruction preset. Or at the very least it would further help RT with AI denoising.

Well thought out and I really wish people stopped focusing on TFlops so much. Even after 3 years of the current generation and whacky TFlops numbers that scale all over the place, people still use it as THE reference for performance. They should take a look at how Lovelace's compute scales in gaming vs workstation applications.

Things like ROPs, pixel/texel fillrate, memory bandwidth, memory type, bus width, die size/area, cache, etc, all get ignore to focus on compute.

I blame Nvidia and AMD. They had used ONLY TF for ages to describe the performance of GPUs, so now it's going to be really hard getting people to start taking note of all the other stuff. When in truth, TFs are just like polygons and sprites before it. At some point, they are enough, and then it's something else that becomes the problem. Right now that something else is RT (and everything that ties to it) as it has become obvious now that lighting is even more important than polygons.

Excellent write-up. My question: Isn't difficult optimization the reason why dual issue performance didn't pan out for RDNA 3? Wouldn't the console space have a much easier time executing? So 2x tflop number should at least be taken somewhat more seriously.

Not just for RDNA3, but for GCN. GCN had a 64-wide wavefront in a world where code was mostly written for 32 wavefront. That was why GCN "TF" were never as good as Nvidia TFs. We can even say its the whole reason AMD ditched GCN and switched to RDNA. The difference now, however, is that the ALUs are now 32 wide as a standard, but can now (with RDNA3 and Nvidia stuff from I think the 30xx series) handle two instructions per cycle.

You are spot on though, most FP32 instructions fit in 32 wide wavefront, however, a lot of them don't, this is an arbitrary explanation, but think of it as some instructions are 32 wide, 16 wide...etc. Everything that is not 32 wide, becomes the problem because that thing would still take up ONE ALU.

With RDNA3+, its easier to get DIC to work as its something that the GPU does for the devs and that scheduling can improve (drivers), basically the GPU should be able to group all instructions and see to it that as many 32-wide instructions are occupying ALUs in pairs. And yes, Sony's first party can utilize that, at the end of the day it's an optimization thing, if done, GPUs that support that kinda code would perform significantly faster, and those that don't would work just the way they always did.

I am just not that optimistic, because DIC has been around for almost 2 years now, and we are yet to see one game using it.

 

[playsaves3]

The cpu may be new in the sense if they take advantage of a smaller node and clock it up. That's not much of an upgrade but if they decide it doesn't need anymore it'll need to get proven when the games come out.

If I had to guess rn, I'd say games still manage 60fps but not increases world density settings, which is most of what a much better CPU would allow

It's got to do stuff that doesn't effect the design of the regular game, so even turning up world density (more cars, pedestrians, etc) might not even be on the table since you could argue it affects the gameplay.

What I want to see if is how much weight these supposed improved features can carry. It can have twice the raster power but in a lot of cases that's not going to get you much as far as on-screen impact. If the upscaling is real and the rt hardware is real, future games that make better use of rt, being able to run multiple effects could really make a difference in a game's look.

Avatar is promising because they took the little rt performance they had to work with and made a difference you can appreciate. More games like that could make a pro that focuses on rt more useful. But if games keep using it for shadows and ao and stuff, it will still be unimpressive.

unless they are adding 3d cache sticking with zen 2 is not going to give us 60fps on a demanding game in like 2026. Also rt eats more cpu resources not just gpu

 

[Mr.Phoenix]

unless they are adding 3d cache sticking with zen 2 is not going to give us 60fps on a demanding game in like 2026. Also rt eats more cpu resources not just gpu

I am curious, how many games on the current gen consoles currently do not have a 60fps mode period. So basically, comes with only a 30fps mode...even after optimization and patches.

If the answer is NONE. Or something that translates to maybe under 3-5% of all games on the platforms.

Then why are you talking like this is Jaguar CPU all over again and the CPU has been shown to be an obvious bottleneck and everything released in 2 years time would be crippled by it?

 

[Kings Field]

Come on where’s the fucking leaks