• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Next-Gen PS5 & XSX |OT| Console tEch threaD

Status
Not open for further replies.

SonGoku

Member
They are. Navi with 8-9TF is definitely match for Vega ~12TF. Its that some people cling on TF number and wont accept we are getting less then 10TF, likely even 9.
The TF doesnt even bother me, what doesnt make sense its a small power hungry chip
I just know it will be 56CUs that's the best option for perf/watt while still keeping die size reasonable
 

Ar¢tos

Member
I would actually care more for the specs if we had any idea of what res psvr2 is targeting, so we could better judge the improvement over ps4+psvr1. As far as non-vr games go, I couldn't care any less if it's 9tf or 11tfs.
 

DeepEnigma

Gold Member
I would actually care more for the specs if we had any idea of what res psvr2 is targeting, so we could better judge the improvement over ps4+psvr1. As far as non-vr games go, I couldn't care any less if it's 9tf or 11tfs.

I think there is a good reason they are keeping mum on that until on or after the PS5 reveal.
 
Last edited:

SonGoku

Member
I would actually care more for the specs if we had any idea of what res psvr2 is targeting, so we could better judge the improvement over ps4+psvr1. As far as non-vr games go, I couldn't care any less if it's 9tf or 11tfs.
I think vr games will look a gen behind to mantain high fps
Mid gen refreshes might have enough juice for decent vr

As far as non vr games go every tf counts for 4k and rt
 
Last edited:

Ar¢tos

Member
I think vr games will look a gen behind to mantain high fps
Mid gen refreshes might have enough juice for decent vr

As far as non vr games go every tf counts for 4k and rt
I really don't need realistic graphics in VR games, cartoon graphics can create as much immersion if done properly. What I want is no Vaseline smeared image and no/or little screen door effect.
 

R600

Banned
The TF doesnt even bother me, what doesnt make sense its a small power hungry chip
I just know it will be 56CUs that's the best option for perf/watt while still keeping die size reasonable
This is where we will have to agree to disagree. We dont know if 320mm² chip with less CUs and higher clocks is less economical then 360-370mm² one.

What we do know for sure is that bigger die is most definitely more expensive, especially at 7nm node, and that Navi does not have low MHZ sweet spot such as GCN (which MS also clocked MUCH higher then Sony). We also know that space on die will be taken by RT hardware and that every Sony console released gets smaller and smaller die as node gets lower and more expensive.

What we also know is that 5700 Navi has 180W TDP. So even that in itself is too much for console, so I really dont see where you are getting 56CU part. Maybe 56CU at 1200/1300mhz, which would still be in 8-9TF ballpark but only with much bigger die.
 
Last edited:

SonGoku

Member
I really don't need realistic graphics in VR games, cartoon graphics can create as much immersion if done properly. What I want is no Vaseline smeared image and no/or little screen door effect.
I remember reading 8k was needed to completely remove the screen door effect
 

FrostyJ93

Member
I hope this is true. It was such a silly next-gen baseline to have for 3rd parties (based on rumored specs).

The way I see it One X will become the new baseline and get a $300 7nm slim revision next year. One X slim and Xcloud will be for those who cannot afford Scarlett.
 

THE:MILKMAN

Member
RIP Lockhart most likely.


That is the video I'm commenting on, Frosty!

For one, given we know Scarlett is a (big) APU, then how would a cheaper 'Lockhart' console be done within a reasonable budget? Two separate APUs and therefore two different consoles seems very expensive.

I'm thinking the two SKU model was an idea proposed in the very early stages and ripped to shreds as soon as Microsoft first-party teams heard about it!

I'm just confused how the likes of Brad had it down as something all but confirmed to be happening.....?

OR...tin foil hat. "Lockhart" was a Microsoft ruse to try and fool Sony!!
 

SonGoku

Member
. We dont know if 320mm² chip with less CUs and higher clocks is less economical then 360-370mm² one.
But we do know long term (and short term 6nm) a big lower clocked chip will go through more cost reductions thus being cheaper to produce long term, which is more important for a console imo
and that Navi does not have low MHZ sweet spot such as GCN
and you know this how? This isn't even up for debate
Lower clocks -> lower voltage
We also know that space on die will be taken by RT hardware
We know nvidias takes 22% on 12nm, so AMDs will take even less than that
Even accounting RT logic 64CU would fit on a 380-390 mm2 die
every Sony console released gets smaller and smaller die as node gets lower and more expensive.
That's my point, the ps5 chip will get smaller and cheaper to produce with subsequent node transitions.
 
Last edited:

FrostyJ93

Member
That is the video I'm commenting on, Frosty!

For one, given we know Scarlett is a (big) APU, then how would a cheaper 'Lockhart' console be done within a reasonable budget? Two separate APUs and therefore two different consoles seems very expensive.

I'm thinking the two SKU model was an idea proposed in the very early stages and ripped to shreds as soon as Microsoft first-party teams heard about it!

I'm just confused how the likes of Brad had it down as something all but confirmed to be happening.....?

OR...tin foil hat. "Lockhart" was a Microsoft ruse to try and fool Sony!!

True may they just couldnt get the hardware at the price they wanted. Oh well. There is One X.

And even if Scarlett is $500 USD...remember the Xbox All Access monthly financing program?
 

R600

Banned
But we do know long term (and short term 6nm) a big lower clocked chip will go through more cost reductions thus being cheaper to produce long term, which is more important for a console imo

and you know this how? This isn't even up for debate
Lower clocks -> lower voltage

We know nvidias takes 22% on 12nm, so AMDs will take even less than that
Even accounting RT logic 64CU would fit on a 380-390 mm2 die

That's my point, the ps5 chip will get smaller and cheaper to produce with subsequent node transitions.
7nm to 6nm is NOTHING in terms of cost saving. During PS3/360 gen we had 90nm > 60nm > 45nm. During PS4 we had 2years of 28nm > 16nm and now we have 7nm (and they wont be shrinking 8th gen consoles on 7nm).

All of these where much bigger then 7nm > 6nm. Shrinking dies becomes harder and more expensive as we are getting closer to physical limits. This is actually why you would want to stay at smaller SOC side more then ever, because shrinking is getting harder and harder.

NaviXT with 40CUs is exactly 251mm². Pitcairn with 20CUs was 212mm² and exact same GPU ended up in Sonys 348mm² PS4. Where do you find place for 70mm² Zen2, 56CUs and RT hardware as well and go for bigger and more expensive die then what was 28nm PS4? It makes 0 sense IMO.
 

SonGoku

Member
7nm to 6nm is NOTHING in terms of cost saving.
You are out of the loop on this one, let me fill you in
7nm designs can be shrunk to 6nm with minimal retooling/investment
Launch die size at 7nm"6nm" die size (15% reduction)
400 mm2​
340 mm2​
390 mm2
331.5 mm2
380 mm2​
323 mm2​

Another factor in favor of a bigger chip is that the next node transitions favor density increase over power reductions

Where do you find place for 70mm² Zen2, 56CUs and RT hardware as well and go for bigger and more expensive die then what was 28nm PS4? It makes 0 sense IMO.
The calculations were posted earlier on this thread
 
Last edited:

FrostyJ93

Member
I hope the PS5 Casing is more flat. The rhombus shape of the PS4 family makes plugging things in the back of the consoles a pain lol.
 

R600

Banned
You are out of the loop on this one, let me fill you in
7nm designs can be shrunk to 6nm with minimal retooling/investment
Launch die size at 7nm"6nm" die size (15% reduction)
400 mm2​
340 mm2​
390 mm2
331.5 mm2
380 mm2​
323 mm2​

The calculations were posted earlier on this thread
So why would Sony shrink 400mm² SOC (already very expensive) to 340mm² (PS4s original die size @28nm) when they can pump up the clocks with better cooling (like MS with X) and shrink 320mm² to 280-290mm² and be in profit sooner? That 400mm² die will never be small enough tbh, there is a reason why MS bumped up clocks so much instead of remained in PS4Pro teritory with 16CUs more and bigger die size.

PS2 die size (520mm²) > PS3 die size (470 mm²) > PS4 die size (348 mm²) > PS4Pro die size (320 mm²). Notice pattern as we go towards lower node?
 

ethomaz

Banned
Navi flops are exactly the same as Vega flops.

Now the efficiency in games needs to be benchmarked yet but 8.3TFs is ridiculous low for a next-gen console... barely better than actual X.
 
Last edited:

demigod

Member
.
Hallah at ya boi lmao


What a fucking tool. Mr 40% should stop talking.
 

R600

Banned
.


What a fucking tool. Mr 40% should stop talking.
Well, one of his predictions regarding ~8TF is getting closer and closer.

This 2012 again. While GPUs where approaching 4TF and Epic was screaming "we need at least 2TF" we ended happy with console that had most, PS4 with 1.8TF.
 

SonGoku

Member
So why would Sony shrink 400mm² SOC (already very expensive) to 340mm² (PS4s original die size @28nm)
400mm2 would be 72CUs the only way i see them going there is if they target 12-13TF
380-390 mm2 for a 11TF target would net a proper next gen leap. Sony wants a clear jump Cerny made this clear and MS is using 385mm2 for scarlet
when they can pump up the clocks with better cooling (like MS with X) and shrink 320mm² to 280-290mm² and be in profit sooner?
A 64CU chip will be able to be shrunk to 330 mm2 in 2021 (6nm) better than launch PS4s 350mm2 and XBONEs 360mm2
As we transition to 5nm and beyond the size difference between 330mm2 and 390mm2 chips on 7nm will get smaller

There will be more density increases than power reductions with subsequent node transitions so as i said a bigger chip will go through more cost reductions than a power hungry small chip
That 400mm² die will never be small enough tbh, there is a reason why MS bumped up clocks so much instead of remained in PS4Pro teritory with 16CUs more and bigger die size.
They also had a 360 mm2 die (bigger than Pro) and profit from day one
The x and pro are unlikely to get node reductions so that skews comparisons with base models.
 
Last edited:

R600

Banned
SonGoku I just dont agree with you. I dont see where you read 12-13TF target, there was never any official target. If anything, Cerny and Phil have been cought talking SSD, RT, immersion and Zen2, but not TF. This is very telling, as your prediction of 13TF would result in absolute best GPU to date, Nvidia 2080TI included. Do you find this reasonable?

Scarlett seems big, but MS perhaps went with bigger console duo to Lockhart/Anaconda split that is now out of the picture. They will also surely have lower clockspeeds if Sony is 1.8GHZ and 36CUs. I dont expect more then 48CUs (with 4 disabled), with 1.5GHZ that would be ~8.5TF, give or take few GFLOPS and Sony with smaller die would likely be winner in this case because difference between the two would be fee % while die size difference would be 20-25%. Lets see what future holds :)

And of course Xbox One X had bigger die and profit from day one, but that was on less expensive node and at 100$ premium.
 
Last edited:

demigod

Member
.
Well, one of his predictions regarding ~8TF is getting closer and closer.

This 2012 again. While GPUs where approaching 4TF and Epic was screaming "we need at least 2TF" we ended happy with console that had most, PS4 with 1.8TF.

No it is not. He was talking about Polaris/Vega TF and not Navi.
 
RT cores, as in Nvidia style, are a complete waste of space that could be used for better things on consoles. I hope sony/MS find a better solution (modified hybrid CUs, whatever).
No, they're not.

That's just like saying that fixed-function units (like TMUs/ROPs) should be replaced by full programmable shader ALUs.

Does that make sense? Fixed-function ASIC is always more efficient.



Because Pro's 36CUs are on 16nm node, not 28nm
Pro doubled the amount of CUs, so who's to say that PS5 won't do the same?

28nm (18 CUs) -> 16nm (36 CUs) -> 7nm (72 CUs)

Because NAVI CUs are quite a bit larger then GCN CUs
Larger? Why?

It's Vega CUs that are larger due to FP64 support, IIRC.

The way I see it One X will become the new baseline and get a $300 7nm slim revision next year. One X slim and Xcloud will be for those who cannot afford Scarlett.
A Jaguar-based console will be the next-gen baseline?

Trust me, Lockhart would be a much better baseline for next-gen games (CPU IPC > GPU flops).

7nm to 6nm is NOTHING in terms of cost saving. During PS3/360 gen we had 90nm > 60nm > 45nm. During PS4 we had 2years of 28nm > 16nm and now we have 7nm (and they wont be shrinking 8th gen consoles on 7nm).

All of these where much bigger then 7nm > 6nm. Shrinking dies becomes harder and more expensive as we are getting closer to physical limits. This is actually why you would want to stay at smaller SOC side more then ever, because shrinking is getting harder and harder.
7nm yields will improve and the wafer costs will drop. It's destined to be a long-term node (just like 28nm).

Who's to say that Sony won't die shrink PS4 APU to 7nm? It's one of their most successful consoles and extremely appealing to casuals (FIFA/Fortnite MTX -> higher PSN profits -> possibly more subsidies for PS5 hardware R&D).
 

R600

Banned
.


No it is not. He was talking about Polaris/Vega TF and not Navi.
No he wasnt. Xbox One X is GCN chip with 6TFs on 16nm. They could have easily gotten 10TF GCN chip there. Smaller CUs therefore more of then on die where you can fit 50% more. So pretty much add 24CUs on Xbox One X and with 1200MHZ you have 10TF chip.

He obviously wasnt talking about GCN.
 

demigod

Member
No he wasnt. Xbox One X is GCN chip with 6TFs on 16nm. They could have easily gotten 10TF GCN chip there. Smaller CUs therefore more of then on die where you can fit 50% more. So pretty much add 24CUs on Xbox One X and with 1200MHZ you have 10TF chip.

He obviously wasnt talking about GCN.

Yes he was, lol. Why do you think he's a joke around here?
 

R600

Banned
I am out of here guys. Please read some AMD papers, dont hang onto CUs and TFs and compare apples to oranges. Navi10 is 251mm² chip on 7nm node, RX590 is 230mm² chip on 16nm node.

Quite clearly architecture from GCN to RDNA was changed. More die is wasted per CU, but perf per TF have gone 30% up. Saying "well they fit 36CUs in Pro, why wouldnt they be able to fit 2x as many in PS5" is completely nonsensical. They cant because they fit 40 on 320mm² die when same part in PC space (RX580) was 230mm². Now take 251mm² Navi10 and fit double of CUs in same die space.

Makes 0 sense.
 

R600

Banned
The doof predicted that PS5 is 8TF while Anaconda is 12TF ... Lets see how that pan out
I saw him saying something akin to "Eh...8TF is not so much out of realm of possibility"

This happened because people where laughing off 8TF as FUD but he must have known from MS that even at 7nm and X being 6TF, RDNA TF are not GCN TF and we could easily get sub 10TF consoles. Which we are getting btw...This was at time everyone was going "Oh yea 13TF!!!".
 

demigod

Member
I saw him saying something akin to "Eh...8TF is not so much out of realm of possibility"

This happened because people where laughing off 8TF as FUD but he must have known from MS that even at 7nm and X being 6TF, RDNA TF are not GCN TF and we could easily get sub 10TF consoles. Which we are getting btw...This was at time everyone was going "Oh yea 13TF!!!".

Clearly he's not talking about RDNA.

I wonder if the smart move for Sony is to ditch the PS4 base and make the Pro the new low-end. 4tflop Pro for $299, 8tflop PS5 for $399 would be super compelling, if BC works as well as everyone hopes and even moreso if they sort out cross-buy.
 

THE:MILKMAN

Member
Or they threw it out there to see if gamers would like the idea.

Of course I wasn't really being serious but then again seeing some of the posts over the way and on Twitter are very interesting now. One post in particular over there has me scratching my head as to why the idea of Lockhart wasn't immediately quashed?
 

R600

Banned
Sub 10TF is incredibly unlikely for either.
Just how? Here :

Pitcairn - GCN 20CUs - 212mm² - 170W
PS4 - 20CUs(2disabled) - 348mm² APU

R580 - GCN 36CUs - 231mm² - 185W
PS4Pro - 40CUs (4 disabled) - 321mm²

Navi10 - RDNA 40CUs - 251mm² - 225W
PS5 - 40CUs + Zen2 +RT hardware - ?
 
Status
Not open for further replies.
Top Bottom