• 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.
Wild speculation that I had before but might not be as wild as you might think.


PS5 GPU has the same number of CU's as PS4 Pro (36 CU's) but it's clocked to 1.85Ghz & it's able to perform 3 operations per clock instead of 2



So someone going by old standards would see 36CU's clocked at 1.85Ghz & think 36 x 64 x 2 x 1850 = 8.52 TFLOPS but because it can perform 3 operations per clock it's 36 x 64 x 3 x 1850 = 12.78 TFLOPS


MIMD better for raytracing & all that plus PS4 BC
So basically, if they released the consoles that we have today with better efficiencies and upgraded hardware, we would be where we want to be.
 

ethomaz

Banned
Wild speculation that I had before but might not be as wild as you might think.


PS5 GPU has the same number of CU's as PS4 Pro (36 CU's) but it's clocked to 1.85Ghz & it's able to perform 3 operations per clock instead of 2



So someone going by old standards would see 36CU's clocked at 1.85Ghz & think 36 x 64 x 2 x 1850 = 8.52 TFLOPS but because it can perform 3 operations per clock it's 36 x 64 x 3 x 1850 = 12.78 TFLOPS


MIMD better for raytracing & all that plus PS4 BC

Do you want to change the base of processors of the last 30 years??? It is 2 instructions per clock and there is a reason for that.
 

HeisenbergFX4

Gold Member
AMD To Host “Next Horizon Gaming” Event & Product Unveil at E3 2019: June 10th, 3pm PT

“Instead this is going to be an AMD-exclusive event, so hopefully it will be a more focused affair than AMD’s past efforts. Though make no mistake, this is an E3 keynote rather than a technical presentation – complete with host Geoff Keighley – so while Dr. Lisa Su is presenting, expect to see AMD playing to the E3 crowd who are just getting out of the Ubisoft keynote.”
 
Last edited:
AMD To Host “Next Horizon Gaming” Event & Product Unveil at E3 2019: June 10th, 3pm PT

“Instead this is going to be an AMD-exclusive event, so hopefully it will be a more focused affair than AMD’s past efforts. Though make no mistake, this is an E3 keynote rather than a technical presentation – complete with host Geoff Keighley – so while Dr. Lisa Su is presenting, expect to see AMD playing to the E3 crowd who are just getting out of the Ubisoft keynote.”
It would be interesting to see if Sony skipped E3 to be on stage with AMD. AMD hasn't really had a presence at E3, so this is new. It would be a great way to talk specs.
 
Last edited:

FranXico

Member
AMD To Host “Next Horizon Gaming” Event & Product Unveil at E3 2019: June 10th, 3pm PT

“Instead this is going to be an AMD-exclusive event, so hopefully it will be a more focused affair than AMD’s past efforts. Though make no mistake, this is an E3 keynote rather than a technical presentation – complete with host Geoff Keighley – so while Dr. Lisa Su is presenting, expect to see AMD playing to the E3 crowd who are just getting out of the Ubisoft keynote.”
I think that's for the next gen Xbox. I could just as easily be wrong.
 
Do you want to change the base of processors of the last 30 years??? It is 2 instructions per clock and there is a reason for that.

Yes, dammit! I want a quantum processor that is performing all the instructions at the same time*!

(* - until I choose to measure the result, that is)
 
Last edited:

Evilms

Banned
AMD-E3-Next-Horizon-Hero-1000x273.jpg

AMD-E3-2019-1000x217.png
 

SonGoku

Member
there is a lot of unused space on vega20:

next_horizon_david_wang_presentation-06.png


im not sure why the are doing this. maybe thermal reasons.
  • 28-nm Fiji: 8,900 M transistors, 596 mm2, 14.9 M transistor/mm2
  • 14-nm Vega 10: 12,500 M, 486 mm2, 25.7 M / mm2, 1.72x density
  • 7-nm Vega 20: 13,230 M, 331 mm2, 40.0 M / mm2, 1.55x density

Density increase is unexpectedly low compared to Vega, so you might be onto something with the unusused spaced, maybe to get better yields on a new process and its not like they had to worry about silicon size. Or maybe Pro cards have more padding hence the 2x density increase over M125 claim.
Also doesn't the increased FP64 perf add more space along with other machine learning stuff?
 
Last edited:

NickFire

Member
Random thoughts about next generation.
1) I am very happy that both MS and Sony seem to be including BC with current generation.
2) I think both MS and Sony are teasing us consumers with leaks and tidbits to try getting the other to give more details about target specs.
3) For me personally, the talk about play anywhere and streaming offerings are snake oil / bullshit tonics. Streaming is just a way for platform holders to make more money without giving a better experience by and large. And play anywhere as a console feature is utter bullshit. Cross platform makes perfect sense for specific games. But there is no way that the average game pushing 13 TF to the limit will run very well on my cell phone.
 

ethomaz

Banned
Why not both? 56 CU at 1.8Ghz gives the magic 12.9TF number. (or 60CU at 1700)
Well because it is easier to reach high clocks with lower SPs :D

For example the same Arch can reach 2Ghz in clock with 40CUs but it won't pass 1.7Ghz with 64CUs (numbers from my ass just to exemplifique).

BTW 50CUs at 2000Mhz give you 12.8TFs :p
 
Last edited:

SonGoku

Member
Well because it is easier to reach high clocks with lower SPs :D

For example the same Arch can reach 2Ghz in clock with 40CUs but it won't pass 1.7Ghz with 64CUs (numbers from my ass just to exemplifique).
That's a good point (56 CU at 1.8Ghz>60CU at 1.6Ghz)
However lower CU parts tend to hit diminishing returns when pushed too far for performance.
Hopefully we'll get some saucy details at E3. I expect a little tease here and there and sharing of corporate vision, etc. Would have thought Computex would be where they unveil tech specs for Zen 2 and Navi.
I wish they finally revealed Navi, that way we can know if it is massively improved over vega or just barely.
 
Last edited:

ethomaz

Banned
Is being again GCN help Sony with BC or just AMD beinh AMD?
Being GCN so the same drivers/API helps a lot.
Most issues they will have will be with the CPU... maybe they create a closed sandbox like they did with PS4 Pro to have the GPU and CPU running at similar performance than PS4.
 

SonGoku

Member
Is being again GCN help Sony with BC or just AMD beinh AMD?
Sony/MS wouldn't go with an old arch just for the sake of BC, IF GCN replacement was ready in time they would have used it.
maybe they create a closed sandbox like they did with PS4 Pro to have the GPU and CPU running at similar performance than PS4.
So no enhanced rez then? (for dynamic resolution titles) or performance? (for fluctuating 30 fps games or unlocked fps in general)
 

ethomaz

Banned
So no enhanced rez then? (for dynamic resolution titles) or performance? (for fluctuating 30 fps games or unlocked fps in general)
That is a different option in Pro, no? It allow to use all the GPU/CPU but it can give issues in some games.

Games that have locked fps will never has unlocked fps unless the dev do a new patch.
 
Last edited:

xool

Member
  • 28-nm Fiji: 8,900 M transistors, 596 mm2, 14.9 M transistor/mm2
  • 14-nm Vega 10: 12,500 M, 486 mm2, 25.7 M / mm2, 1.72x density
  • 7-nm Vega 20: 13,230 M, 331 mm2, 40.0 M / mm2, 1.55x density

Density increase is unexpectedly low compared to Vega, so you might be onto something with the unusused spaced, maybe to get better yields on a new process and its not like they had to worry about silicon size. Or maybe Pro cards have more padding hence the 2x density increase over M125 claim.

According to https://www.semiwiki.com/forum/content/6713-14nm-16nm-10nm-7nm-what-we-know-now.html GF 14nm has SRAM cell size of 0.065 micrometers square, while TSMC (proposed) 7nm would have cell size of 0.027 - so that should be 2.4x density over GF 14/16nm (TSMC's website claims even higher figure compared to their own 16nm)

Amongst other things we might be seeing early 7nm, just like we had 16nm, 16nm FINFET, 16nmFF+, 16nmFF++ etc as they got better at their jobs over the years.

Honestly really sure not what the cause is ..

Also doesn't the increased FP64 perf add more space along with other machine learning stuff?

Yep better fp64 can use a lot of transistors. Does Navi have machine learning?
 
Last edited:

Mega Man

Member
BC compatibility will be one of the biggest selling points for me as a PS4 gen skipper. I have had two Xbox Ones, but now with a very capable PC I will be making the switch permanently. That said, the OneX will house all of my still console exclusives from generations prior.

PS5... here I come!
 

SonGoku

Member
That is a different option in Pro, no? It allow to use all the GPU/CPU but it can give issues in some games.

Games that have locked fps will never has unlocked fps unless the dev do a new patch.
I'd rather they copied MS BC method
 

ethomaz

Banned
I'd rather they copied MS BC method
It is emulation and that generates a series of new issues that should be worked.
MS choose this method because X360v and Xbox hardware are incompatible with XB1.

MS could use all the money they put to emulate 360 and Xbox games on XB1 to make new games for XB1 imo.

PS5 BC for PS4 games will be probably native (same for XB1 games on Nextxbox)... it is cheaper and you have all games compatible without any change in the game.
 
Last edited:

LordOfChaos

Member
Wild speculation that I had before but might not be as wild as you might think.


PS5 GPU has the same number of CU's as PS4 Pro (36 CU's) but it's clocked to 1.85Ghz & it's able to perform 3 operations per clock instead of 2



So someone going by old standards would see 36CU's clocked at 1.85Ghz & think 36 x 64 x 2 x 1850 = 8.52 TFLOPS but because it can perform 3 operations per clock it's 36 x 64 x 3 x 1850 = 12.78 TFLOPS


MIMD better for raytracing & all that plus PS4 BC


Very, very unlikely, this would not only be a break from GCN but from decades of convergent evolution in GPUs. To even call it Navi, a GCN derivative, would be a massive disservice for one.

Two, each core is capable of dispatching two operations per cycle, but that's not to say they get there every cycle, real IPC is lower than the paper calculation of cores * clock * 2 ops per cycle (part of why Gflops can't be compared accross architectures), so to add another instruction displach per core may not add as much as you might think it would.

Three, GCN ALUs get to those two ops per cycle in the form of a fused multiply-add in the first place, where does this third come from.

Edit: Bonus Four, IPC in GPUs is inversely correlated to clock speed capability.
 
Last edited:

Fake

Member
Software emulation its not a easy way to current gen and less perfomance can be extract.
Maybe software emulation for ps2/ps3 games, but ps4 are unlike.
 

SonGoku

Member
It is emulation and that generates a series of new issues that should be worked.
MS choose this method because X360v and Xbox hardware are incompatible with XB1.

MS could use all the money they put to emulate 360 and Xbox games on XB1 to make new games for XB1 imo.

PS5 BC for PS4 games will be probably native (same for XB1 games on Nextxbox)... it is cheaper and you have all games compatible without any change in the game.
nvm I should correct myself, i would like Sony to handle BC like MS does boost mode for xbone games on X (so enhanced resolution and fps for dynamic titles) and use Pro games as the base.
Would that be too hard to achieve?
 

ethomaz

Banned
Software emulation its not a easy way to current gen and less perfomance can be extract.
Maybe software emulation for ps2/ps3 games, but ps4 are unlike.
Exactly.

I believe they will use the actual PS2 emulation for PS4 on PS5 too.
 

ethomaz

Banned
nvm I should correct myself, i would like Sony to handle BC like MS does boost mode for xbone games on X (so enhanced resolution and fps for dynamic titles) and use Pro games as the base.
Would that be too hard to achieve?
MS handle emulation for each title adding new options and features... that is exactly the same if a Dev choose to do a patch to the game but instead MS do that patch for it emulation.
The emulation team needs to work in these games to make them run enhanced.

Like I said I think MS spend a lot of money in that BC Emulation.

But forget that about PS5 BC.

It will be native... PS4 games will run native on PS5 just like PS4 games runs on Pro... it will be a sandbox to match the PS4 (or Pro) level of performance which won't have any incompatibility issue and probably a Boost Mode where the game runs with the full PS5 hardware.

But the Boost Mode is just like the Pro... the enhancements depends of the game (if it has unlocked fps, dynamic resolution, etc) and in some cases generates incompatibilidades which Sony recommends to disable Boost Mode.

BC native with Boost Mode is easy, very easy, cheaper, no hard work is done at all.
BC emulation like MS did is hard, very hard, time consuming and expensive.

People forget but PS2 emulation for PS4 is basically the same MS did with BC on XB1... the difference is that MS focused in BC spending a lot of money to put a lot of titles there and Sony choose to spend little with a small team so the releases are really sporadically and it is probably dead today.
 

xool

Member
Also doesn't the increased FP64 perf add more space along with other machine learning stuff?

[more]

ok all the evidence points to VegaVII being a Radeon Instinct MI50 with the juicy stuff disabled/fused out, and sold at a lower price - Radeon Instinct die size 331 mm2 , VegaVII die size 331 mm2

But the Instinct part has 1:2 FP32:FP64 rate (reduced to1:4 in VegaVII) compared to 1:16 in Vega64 .. that's a huge increase in the SIMD unit transistor count to accomplish that.

So - as far as console parts the comparison VegaVII can be thankfully ignored (they won't have fp64 levels like that) .. and hopefully we'll see scaling more in line to the 2-3x smaller from 14/16nm expected - with Vega64 being a better starting point.

..so 64(60) CU next gen GPU should be around 486mm2 divided by 2.4 - means next gen GPU die size =~ 203mm2 - more reasonable . thermals should be down too - up to about the same amount as size. (120W?)

I'd take reasonable bets on this (or better)
 
Last edited:

ethomaz

Banned
[more]

ok all the evidence points to VegaVII being a Radeon Instinct MI50 with the juicy stuff disabled/fused out, and sold at a lower price - Radeon Instinct die size 331 mm2 , VegaVII die size 331 mm2

But the Instinct part has 1:2 FP32:FP64 rate (reduced to1:4 in VegaVII) compared to 1:16 in Vega64 .. that's a huge increase in the SIMD unit transistor count to accomplish that.

So - as far as console parts the comparison VegaVII can be thankfully ignored (they won't have fp64 levels like that) .. and hopefully we'll see scaling more in line to the 2-3x smaller from 14/16nm expected - with Vega64 being a better starting point.

..so 64(60) CU next gen GPU should be around 486mm2 divided by 2.4 = 203mm2 - more reasonable . thermals should be down too - up to about the same amount as size.
Are you sure it is not the same chip? Because you can reduce FP64 performance via GPU ROM (software) like nVidia did for years.
 
Last edited:

xool

Member
Are you sure it is not the same chip? Because you can reduce FP64 performance via GPU ROM (software) like nVidia did for years.

Yep. I think it's the same chip/design - and they're doing just that - nerfing it in GPU ROM/bios (whatever it's called)
 

ethomaz

Banned
Yep. I think it's the same chip/design - and they're doing just that - nerfing it in GPU ROM/bios (whatever it's called)
I just checked.

Vega 20 chip is used in 4 cards (it is the same chip):

AMD Radeon Instinct MI50
AMD Radeon Instinct MI60
AMD Radeon Pro Vega
AMD Radeon VII

These FP64 difference are done via GPU BIOS (it is called BIOS you are right... you can do BIOS Mod in some cards).
 
Last edited:

Ellery

Member
So we are still at GCN and supposedly Navi (6th gen) is making a bigger jump from Vega (5th gen) than Vega (5th gen) did from GCN 1st generation from way back 2011?

I wish it is true, but I will keep my expectations low. I don't expect big architectural gains
 

LordOfChaos

Member
Radeon's Maxwell sauce
That part went over my head earlier, if Navi is as big a jump as Maxwell was over Kepler it will be huge.


This?

 

ethomaz

Banned
This?

But that is Turing feature not Maxwell... Maxwell was a big change in how nVidia handled perf/watt and what put nVidia one or two generations ahead AMD.
 

xool

Member
I just checked.

Vega 20 chip is used in 4 cards (it is the same chip):

AMD Radeon Instinct MI50
AMD Radeon Instinct MI60
AMD Radeon Pro Vega
AMD Radeon VII

These FP64 difference are done via GPU BIOS (it is called BIOS you are right... you can do BIOS Mod in some cards).

Thanks - this makes more sense now. Somehow I missed all this - looks like it is mainstream knowledge
 

HeisenbergFX4

Gold Member
Just adding this here though in no way do I believe it.

PS5 over 12tf
Xbox premium over 11tf

PS5 blowing us away. Seems way more advanced than Xbox even the upper end model. Reveal June 3 for PS5.


 
Last edited:
Just adding this here though in no way do I believe it.

PS5 over 12tf
Xbox premium over 11tf

PS5 blowing us away. Seems way more advanced than Xbox even the upper end model. Reveal June 3 for PS5.


Microsoft would have a hell of a time if they are not more powerful than the PS4 PS5.

EDIT: I noticed my typo. Whoops :messenger_beaming:
 
Last edited:

FranXico

Member
Just adding this here though in no way do I believe it.

PS5 over 12tf
Xbox premium over 11tf

PS5 blowing us away. Seems way more advanced than Xbox even the upper end model. Reveal June 3 for PS5.


In the same thread you can read this reply, though:
Work at studio...in the US. OP is technically correct. We have had ps5 devkit for three weeks and scarlett dev kit for one week.

Right now, ps5 has more tf than anaconda. From our limited testing, anaconda will will actually push better visuals though. Anaconda has a pretty advanced (maybe not advanced I'm not a chip nerd) custom version of navi. Basically, MS did a lot of custom work on navi that helps push better visuals than ps5. OP nailed the tf number but it was either a lucky guess or he is lying because ps5 devkit isn't more powerful or advanced than Anaconda.

The conclusion to take is the same as always, 4chan isn't a reliable source for leaks, because anybody can say whatever they like, and in this subject in particular, there are already a bunch of contradictory leaks.
 
Last edited:

CyberPanda

Banned
Just adding this here though in no way do I believe it.

PS5 over 12tf
Xbox premium over 11tf

PS5 blowing us away. Seems way more advanced than Xbox even the upper end model. Reveal June 3 for PS5.


This looks similar to another leaker in ERA.
Lol it’s 4chan though, I guess we’ll see. But this made me laugh:
My dad works at Nintendo and he told me that Bowser told him that I should tell you guys that they will come with optical drive
 

thelastword

Banned


Radeon's Maxwell sauce
That part went over my head earlier, if Navi is as big a jump as Maxwell was over Kepler it will be huge.

I believe this rumor, yet, it's nothing new..... It's what I've been saying all this time. When folk were comparing Vega CU's to Navi CU's, I shook my head, I constantly told them they should not do that. Navi has many more instructions per CU, the whole CU structure was redesigned and beefed up considerably for Navi.

It will take you much less CU's on Navi to get similar performance or more against Vega, and by extension, Pascal and Turing.....It's the same way people are thinking 7nm Vega can be compared to Navi built on 7nm, it's just not how you go about analyzing future technology.
 

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
Just adding this here though in no way do I believe it.

PS5 over 12tf
Xbox premium over 11tf

PS5 blowing us away. Seems way more advanced than Xbox even the upper end model. Reveal June 3 for PS5.



No way is June 3rd the reveal date for the PS5. How much sense does that make?
 
Status
Not open for further replies.
Top Bottom