upnorthsox
Veteran
Clearly.
Baseless Next Generation Rumors with no Technical Merits [post E3 2019, pre GDC 2020] [XBSX, PS5]
- Thread starter BRiT
- Start date
- Status
upnorthsox
Veteran
In your request to be reasonable, the quote you are referencing:
I don't see under any reasonably reading of this you can come up with him calling it a definitive version number or that he is or represents himself as part of any marketing statement.
A reasonable reading would be that he saying it's close but not quite RDNA2 but if you want to call it RDNA2 then that's fine.
And exactly how is he qualifying this statement if no RDNA2.0 architecture is out there right now?
He's never seen or had access to the PS5 kit as per his own words. Everything is 3rd party for him. The amount of knowledge required to make this call is massive. This RDNA 2.0 is a micro architecture change. This requires a good look at what they did.
Can’t wait to read the next few hundred pages of discussions over what exactly 1.9 means versus 2, based on what some guy somewhere wrote. Riveting stuff.
On first glance one would assume it means it's 0.1 less. However, the guy who wrote it never established which numeral system he was using and were there any conversions to other systems on the way, so the real difference is anyone's guess
I'm done with it so don't worry, there will be no debate, you'll be fast forwarded to previews about XSX using Zen 1.5, we'll circle back to 12 TF GCN 4.7 flops.
all reasonable logic is not prevailing here:
a) marketing materials from MS and their partners at AMD has been officially announced as RDNA 2.0 micro-architecture
b) AMD is the one responsible for determining what is and isn't RDNA2.0
c) we have a person who is not directly involved with the hardware claiming it's not quite RNDA2.0 but has no method of proving this
d) we have a whole bunch of people jumping onto this bandwagon.
It's no different than allowing MisterXMedia claims to fun aloof here and re-open the discussion around a 3D stacked GPU and whatever else he raved about.
I aligned with the notion by removing all talk of github, it's had its time in the spotlight and should no longer be discussed, but the turn this thread took is significantly worse imo.
A GPU can self identify as it sees fit! We’re in 2020 guys, let GPUs be their true self. Even if it feels like a CPU, then who are we to judge?
The gpu isn't yet self-aware, don't tell the enclosure what it can do with it's body!
D
Deleted member 13524
Guest
AMD will even let them call it RDNA 99 SUPER-ADVANCED XT, as long as Microsoft pays for it.
Why else would they let Intel call "Vega M" to what is essentially a Polaris chip?
Who wrote RDNA 1.9 when mentioning the SeriesX GPU?
I confess I can't remember the person who wrote it and about what, but I had the impression it was referring to the PS5 and not the SeriesX.
Like you said, it's AMD who defines what RDNA2 and no one else. AMD also defines what "Vega" is, and as soon as the KBL-G solutions emerged then "Vega" became a mess of a name.
If you are a multiplatform developer who's seen Big Navi's RDNA2 ISA, and you have had access to the consoles' devkit documentation, then you might notice how the later are missing some instructions that define the first (or should define in your opinion). In this situation you'd be telling your friends that you don't really think the console is using RDNA2, it's more like RDNA1.9 because it's missing this and that.
Similarly, many people have claimed that KBL-G's iGPU is actually a Polaris chip with a tiny bit of Vega dust.
Yet AMD called it Vega (which should include RPM for example), and AMD is the one who owns the rights to call Vega to whatever they see fit.
Last edited by a moderator:
@ToTTenTranz can't speak for @iroboto but I think the 1.9 discussion is done here. Stick a fork in it.
Tommy McClain
Tommy McClain
Per the amd slide, rdna2=7nm+, right? For some reason some xbox guys are saying xsx is not on 7nm+. (the ~400mm2 die size would agree with them). Would that lack of 7nm+ make it "<rdna2"?
Guess I'm asking if the density increase associated with 7nm+ could be responsible for some of the optimizations rdna2 has over rdna1, besides removing the remaining traces of gcn?
Which xbox guys? Or do you mean supposed insiders? Cuz there's an easy answer to that conundrum?
And without knowing the exact contents of that die... how many mm2 in a teraflop?
D
Deleted member 13524
Guest
400mm^2 seems to line up nicely with 3 shader engines that are 10 WGP big each, plus 8* Zen2 cores with 8MB L3, plus leeway for additional RT units (which are tiny on Turing, so people are assuming they should also be tiny on RDNA2), memory controller, glue, etc.
The chip being 400mm^2 suggests it's not taking advantage of the 17% area reduction that TSMC's N7+ offers over N7, at which the chip should come at closer to 350mm^2.
I don't remember any insider ever mentioning fab process. What they tell is coming from game developers which rarely follow that stuff AFAIK.
There was that (foxconn?) PCB leak on reddit, but I don't remember if it mentioned process either.
It's for their own GPUs, not for the architecture, AMDs architectures aren't tied to specific process or even foundry.
So you've made a bunch of assumptions and 7nm+ isn't lining up with your assumptions therefore 7nm+ must be wrong?
Just to be clear I have my doubts it's 7nm+. But that's not really evidence.
D
Deleted member 13524
Guest
My assumptions?
Where did I ever assume N7+ must be wrong?
As for the other numbers they didn't even come from me, I'm just paraphrasing what other posters wrote here when the SoC picture came out.
I don't mean insiders. I mean people who have no problem telling all of twitter and gaf/era that they prefer xbox.
There were a couple write ups showing what can fit in a ~400mm2 die. The limit on 7nm was 56/60 cu accounting for lowered l3 cache.
7nm+ is supposed to bring a 20% density increase.
https://www.resetera.com/threads/ne...arks-for-chance-at-glory.165711/post-28427929
Thanks.
- Status
Similar threads
- Locked
