• Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
  • We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.

Overall maximum teraflops for next-gen launch consoles?

  • 8 teraflops

    Votes: 43 1.9%
  • 9 teraflops

    Votes: 56 2.4%
  • 12 teraflops

    Votes: 978 42.5%
  • 14 teraflops

    Votes: 525 22.8%
  • Team ALL THE WAY UP +14 teraflops

    Votes: 491 21.3%
  • 10 teraflops (because for some reason I put 9 instead of 10)

    Votes: 208 9.0%

  • Total voters
    2,301
Status
Not open for further replies.

Deleted member 38397

User requested account closure
Banned
Jan 15, 2018
838
I have a quick computation and the probability of getting >10Tflops is now

KecK7cL.png
 

zedox

Member
Oct 28, 2017
5,215
I wouldn't put my hopes on ML-based Super Resolution solutions. They're really computationally expensive to run with meaningful quality and temporally stable in real time.

I mean, DLSS uses models trained for individual games and runs on hardware with quite a bit of inference grunt and only delivers, eh, questionable results.
You know what else is computationally expensive? Raytracing...but we're getting that too.
DirectML and Super Resolution solution will be used for Lockhart . :)
 

modiz

Member
Oct 8, 2018
17,844
You know what else is computationally expensive? Raytracing...but we're getting that too.
DirectML and Super Resolution solution will be used for Lockhart . :)
The difference is reconstruction techniques are meant to reach higher resolution with inferior hardware. Ray tracing is aimed at having much more accurately looking games.
 

Theorry

Member
Oct 27, 2017
61,028
My prediction on all this with my limited knowledge about all of these numbers etc. :P

Lockhart - 8TF - 299
PS5 - 11TF - 399
Anaconda - 14TF - 499

Take it as you will.

:D
 

VX1

Member
Oct 28, 2017
7,000
Europe
My prediction on all this with my limited knowledge about all of these numbers etc. :P

Lockhart - 8TF - 299
PS5 - 11TF - 399
Anaconda - 14TF - 499

Take it as you will.

:D

The thing with Lockhart is that DF boss Richard Leadbetter explicitly said recently that he thinks it will be ~ 4TF and if that turns out to be true we have to lower PS5 and Anaconda expectations comparing to what you suggested.

Btw,i agree with your pricing range: 299-399-499.
 

Pheonix

Banned
Dec 14, 2018
5,990
St Kitts
Obviously a lot of disappointing claims in this video. First, I agree with the heuristics he applies to come up with performance predictions for Navi. Additionally, one only need look at AMD's claim of the same performance for 50% the power on 7nm to see the numbers aren't far-fetched.

Regarding the concerns Jim's video lays out:
  • Clocks are below expectation and thermals are bad- Honestly this one wouldn't be a surprise at all. AMD has struggled with this for the past several generations, and this was feeding much of the skepticism regarding clocks originally. It's why many of us were predicting clocks around 1300-1400 to begin with.
  • Engineers are fed up and can't wait to be done with it- This one is very multi-faceted and has a ton of interpretations to it. Is it simply because it's been a project they knew would be troubled from early on? Is it because they know there are better projects out there on-going? The optics are bad no matter what, but having a fence with greener grass behind it could have huge implications for off-shoot spins of the design for consoles or just for RTG and future architectures.
  • Navi can't beat Vega performance- This is the most troubling one, in my opinion. All along, many of us had been preaching that Navi TF =/= Vega TF =/= Nvidia TF because of architecture differences. This has a lot of potential explanations I'll try to split off.
    • Navi fails to fix many of the architectural deficiencies of Vega, such as primitive shaders, NGG, etc.
    • One or more of Navi's key architectural features was broken, and it was the reason for the re-spin. In this scenario, it's possible the early returns we have for performance targets are based on these initial chips that were not fully functional. One would presume performance from this part is how Jim was able to slightly revise his performance, TDP, and pricing claims to begin with.
    • Navi is a better architecture, but failed to do enough to address bandwidth shortfalls compared to HBM2-equipped Vega, and it simply sits starved for bandwidth as a result.
    • Navi does have a superior architecture, but the early drivers do nothing to take advantage of these improvements, leaving a lot of performance on the table that is yet to be unlocked. This would not be a surprise either, as AMD has often been maligned for their drivers compared to Nvidia.
In all, the news regarding clocks and thermal doesn't surprise me one bit, and it alone would not make me budge my own personal performance predictions, particularly given the Gonzalo leak which must be reckoned with in my opinion. As such, I still expect both PS5 and Anaconda to come in above 10TF, just as Justin suggested they would.

What is more troubling is the lack of efficiency gains and the apparent mood within RTG. Many of us had been taking for granted that Navi would be a big boost in throughput simply because it fixed Vega's flaws and added its own advantages on top. Now we're dealing with the possibility that Navi FLOPs equal Vega FLOPs which paints you into a smaller performance window. If RTG is genuinely upset about how it has turned out, this points to some combination of insufficient resources as rumored for Vega, or inept management that is incapable of driving their teams toward design goals that the Zen team has clearly been able to hit.


Why? What we have now are competing rumors. Gonzalo is very likely a real chip, and also very likely a gaming console chip. It could be that we are misinterpreting the code to mean 1.8GHz when it doesn't, or that that's only a boost frequency, but it's still a rumor to contrast with this new information. In Jim's own list, he even has claims that are at odds with each other - Navi can't match Vega 20 clocks vs. Navi can reach those clocks, but over the originally allotted power and thermal budgets.



There is no alternative. They're already committed to the path.


They were not predictions. They were based on information directly leaked to him.

Keep in mind that the positive spin on Navi was coming directly from Lisa Su. Of course she'd go out of her way to say something positive about the architecture.


Perhaps, but buildzoid also pointed out that reference/evaluation PCBs can often be overbuilt. Indeed, one wouldn't expect to cool the amount of power that PCB can supply with a simple blower cooler.


It's not so much that the TF numbers have been reduced, but that the possibility that Navi has no per clock advantage over Vega would be devastating.
Well said.

Maybe the only "significant' gain with Navi just comes down to its process node. Basically making 10-12Tf available at a much lower TDP by AMD standards. I mean from the chart they seem to be able to make 11TF+ available at as little as 175W. Three years ago they needed a little over 200W to hit about half that. That's gotta count for something right?
 

zedox

Member
Oct 28, 2017
5,215
The difference is reconstruction techniques are meant to reach higher resolution with inferior hardware. Ray tracing is aimed at having much more accurately looking games.
Please reread what I wrote in which why I compared the two.
*hint* I know the difference between the two
 

bcatwilly

Member
Oct 27, 2017
2,483
While I don't expect any Xbox specific news out of it, the info starved may find it of minimal interest that the Microsoft Build conference in Seattle for developers starts tomorrow and runs through Wednesday. Given that some of the technologies such as DirectML being talked about in this thread are applicable to the PC platform or even Azure cloud related technology, there may be a couple of interesting things come out of this.
 

zedox

Member
Oct 28, 2017
5,215
While I don't expect any Xbox specific news out of it, the info starved may find it of minimal interest that the Microsoft Build conference in Seattle for developers starts tomorrow and runs through Wednesday. Given that some of the technologies such as DirectML being talked about in this thread are applicable to the PC platform or even Azure cloud related technology, there may be a couple of interesting things come out of this.
Your mind is in the right place. You'll get more out of the sessions in relation to DirectML (I haven't seen any so far...but I haven't looked today) and DXR during the conference. Plus, I watch build each year. :D
 

Theorry

Member
Oct 27, 2017
61,028
You have no idea of how much meltdowns would generate an output like this ahhah
Like i said its all based on my limited knowledge if its all possible. Just thinking about more logically about it. Still think Sony wants that 399 pricepoint. It did wonders for them and they said many times i believe its the sweetspot. MS isnt "scared" of putting down a 499 price and Spencer said and showed they want to be the brand of performance leader. Then you have lockhart wich needs to be a step away from Anaconda price wise and it needs to be a step up from X also TF wise imo. Even if the CPU and ram is already a step up. I simply cant believe MS putting a 4TF machine out.

This is all more gut feeling then knowing the ins and outs of how the specs work etc.
 
Nov 12, 2017
2,877
Like i said its all based on my limited knowledge if its all possible. Just thinking about more logically about it. Still think Sony wants that 399 pricepoint. It did wonders for them and they said many times i believe its the sweetspot. MS isnt "scared" of putting down a 499 price and Spencer said and showed they want to be the brand of performance leader. Then you have lockhart wich needs to be a step away from Anaconda price wise and it needs to be a step up from X also TF wise imo. Even if the CPU and ram is already a step up. I simply cant believe MS putting a 4TF machine out.

This is all more gut feeling then knowing the ins and outs of how the specs work etc.
Yes i understand your points ) well all in all if it end like that......Ms would be very...very happy
 

zedox

Member
Oct 28, 2017
5,215
I am saying that comparing a reconstruction technique to ray tracing is wrong.
I wholeheartedly disagree...they are both computationally intensive tasks...which was the reasoning for the "downplay" of using DirectML in what I quoted. The same could be said about raytracing which is computationally intensive and people thought it wasn't going to be in the next gen consoles (I thought it was) and here we are. Same type of scenario...hence the comparison. Like I said, reread what I wrote...
 

GhostTrick

Member
Oct 25, 2017
11,315
I wholeheartedly disagree...they are both computationally intensive tasks...which was the reasoning for the "downplay" of using DirectML in what I quoted. The same could be said about raytracing which is computationally intensive and people thought it wasn't going to be in the next gen consoles (I thought it was) and here we are. Same type of scenario...hence the comparison. Like I said, reread what I wrote...


Wait for more before thinking Ray tracing is fully in.
 

zedox

Member
Oct 28, 2017
5,215
Yes you do. Considering the huge impact of ray tracing in performance, people should temper their expectations here.

That next console will support Ray Tracing is a given. To what extent is a different matter here.
I know that DirectX Raytracing will be supported on the next xbox console. I don't need temper my expectation of that because that's been my expectation for the past year.

If I'm wrong, I'm wrong...I don't get "let down" if something doesn't happen.
 

anexanhume

Member
Oct 25, 2017
12,913
Maryland
I totally agree with you both.

Then there is Graphene.

Graphene breakthrough could push your gaming PC's clock speeds into new territory

https://www.pcgamesn.com/terehertz-cpu-clock-speeds-graphene
Don't believe anything about graphene without a plan to scale to production.

E.g. TSMC will actually use graphene in small amounts on 5nm.

https://www.semiwiki.com/forum/content/8149-2019-tsmc-technology-symposium-review-part-i.html

Advanced Materials Engineering
In addition to the N5 introduction of a high mobility channel, TSMC highlighted additional materials and device engineering updates:

  • super high-density MIM offering (N5), with 2X ff/um**2 and 2X insertion density
  • new low-K dielectric materials
  • metal Reactive Ion Etching (RIE), replacing Cu damascene for metal pitch < 30um
  • a graphene "cap" to reduce Cu interconnect resistivity
 

More Butter

Banned
Jun 12, 2018
1,890
Here is a question to all you native 4K true believers who firmly believe it is better than reconstruction.

How many of you would have preferred native 4K versions of Uncharted, Horizon or God of War be released on the PS4Pro?

Ones that look significantly worse, and I mean significantly worse, but are native 4K?

I don't think anyone would prefer that, even the people who claim native 4K is the holy grail. Yet that is what they are asking for in this thread.

Because that is the trade off here, we aren't comparing the Xbox One X to the Pro. We're talking about the games running on the same hardware. Which means the native ones will look worse compared to the same game made using reconstruction. A lot worse.

The only reason the X gets native 4K games is because the games are designed to run on much worse hardware than the X and it's far easier to bump the resolution than actually improve the effects, assets etc. But again, if you made a game with the X in mind, then a CB game would look far better than a native game.

So I guess the question is, why focus on one aspect of visual quality to the detriment to the overall visual quality of the game? Why want worse looking games, overall, just so you can say you have native 4K?

Nobody is saying Horizon or GoW look bad because they aren't native 4K. But if you released native 4K versions with far worse assets, draw distance and effects then they would certainly look worse than the versions we got.
If the desired graphical features were possible in naive 4K you would be crazy to want cb. Given the hardware limitations it made sense for the games listed above to employ checherboarding. The problem with checkboard begging about next gen is that we don't know what the hardware capabilities or what efficiencies will come about. It may not be as necessary to cb while still getting great results. We just don't know yet.
 

anexanhume

Member
Oct 25, 2017
12,913
Maryland
One more thing. Excessive thermals, you say?

So this seems pretty significant. Sony has applied for a patent on a heatsink that sits on the opposite side of the PWB of the APU has thru-hole thermal conduits that touch the bottom side of the package (the area is voided by virtue of a package with fan-out that avoids the center of the device). It also appears to depict (amongst other things), stacked die within the package (5c, 5d) and another heatsink or other assembly on the top of the package (9).

This is pretty exciting.

 

Gamer17

Banned
Oct 30, 2017
9,399
There's no way the gap will be that large between Snek and PS5. This place would be a war one if that happened.



Bwahaha!
thats around 25% difference . many fans mention 45% difference between PS5 and anaconda.I said 15 to 20%
so 25% isnt that much more(comapred to 20% that i expect) vs the 45 % that some fans suggest
 
Status
Not open for further replies.