And I still haven't seen any evidence that singular CUs can be gated off at run time. No, factory disabled ones are a very, very different thing.
-
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.
Next-gen PS5 and next Xbox speculation |OT10| - We aim to transition those from OT9 at a pace never seen before. [SEE THREADMARKS FOR CHANCE AT GLORY]
- Thread starter Mecha Meister
- Start date
- Status
And I still haven't seen any evidence that singular CUs can be gated off at run time. No, factory disabled ones are a very, very different thing.
DVFS can get very fine-grain, but it can complicate the overall design and make it less efficient in normal usage.
The answer may be beyond my depth, but why is it possible to transiently shutdown groups of CUs live, but not individual CUs?
So think about this, OG PS4 was 18 CU, but that's 9 RDNA work groups. 2x9 work groups both feeding into a pair of shader engines would replicate the Pro's butterfly. What we know is that a SE can handle more than 10 WG's feeding into it, so I suppose the only question that remains is whether individual WG can be deactivated for BC modes, or does it need to be an entire path, I.e. SE et al.
That's a very good question. It's simpler, cleaner and more transparent to software to just shut down entire data paths. The way these GPUs are organized they have clear and even work division, be it as a bunch of CUs under a SE or a bunch of WGPs under a SA (and a bunch of SAs under a SE).
One thing that kinda points towards this "evenness" being a somewhat desirable is that CUs/WGPs are always disabled as pairs.
I still subscribe to my theory that they'll do 56 active CUs as two 14WGPs SEs, 5-5-4 + 5-5-4 WGPs, use only 5-4 for PS4, 5-4 + 5-4 for PS4Pro and the whole thing for PS5.
Where the hell are all of these assumptions coming from? Sony weakens the main money maker to focus of VR? Do you realise how stupid that would be?
Idk if you guys are joking, but either way, quite silly to entertain the thought of sony pulling an xbox with kinect instead of putting their main focus on the PS5 itself 🤨
I take it you're assuming 15WGP SE in silicon, with two disabled overall?
Last edited:
Checks out with what Klee has always said, that it was not small Navi and not quite big Navi, either.
\
Going to lose our god damn minds.......... probably nothing, we'll just make another thread and life will....find a way....
Since this thread seems to be an [annoyingly] hotbed of rumors and the like, I will just issue a blanket statement here.
Official Staff Communication
Unless specified, assume all information, verified rumor or not, is just that - a rumor. Legit insiders may get things wrong, and fake insiders may slip through the cracks. There is only so much I can verify. While I will do my utmost best to provide verification for credible info, I cannot promise 100% success. Please keep any expectations in check.
Thanks. Please feel free to PM me if you have questions.
Thanks. Please feel free to PM me if you have questions.
I don't know, that doesn't sound convincing to me.
That statement was made in november, there would be no reason to point out that there will not be any next-gen exlusive in the next year, because there is no new console to begin with
Please ask Sega to make Phantasy Star 5 and connect it strongly to the original RPG series, thanks Phil!
Does that work with the rops?
32 rops for PS4,
64 rops for PS4 pro + PS5.
Since the RBs are part of the SAs, it would be 96 ROPs for PS5.
EDIT: damn, just remembered there was that one rumor that big Navi would be 80CUs with 96 ROPs. One way to do that would be with six 7 WGP SAs, two WGPs disabled for yields.
techpowerup has placed Xbox Series X with 96 rops.
www.techpowerup.com
Of course they are only guessing, it seems to be placeholder.
AMD Xbox Series X GPU Specs
AMD Scarlett, 1825 MHz, 3328 Cores, 208 TMUs, 64 ROPs, 10240 MB GDDR6, 1750 MHz, 320 bit
Of course they are only guessing, it seems to be placeholder.
That's 3 shader engines, since each shader engine so far looks to be 32 rops each.
techpowerup has placed Xbox Series X with 96 rops.
AMD Xbox Series X GPU Specs
AMD Scarlett, 1825 MHz, 3328 Cores, 208 TMUs, 64 ROPs, 10240 MB GDDR6, 1750 MHz, 320 bit
Of course they are only guessing, it seems to be placeholder.
They're guessing.
Github leak already shows 64 ROPS for all next gen consoles. 320 bit bus for Xsx, like in the teaser from e3 2019.
Nope, that's 2 Shader Engines with 3 Shader Arrays each, 4 RBs per Array, 16 ROPs. Navi rearranged a bunch of stuff, RBs are coupled to a SA now.
I am pretty sure each SE in Navi 10 has two shader arrays. Are you talking about a version with 3 shader arrays?
You're right about SA / RBE coupling.
Technically they can do 52CU, 5,4,4 and 5,4,4.
Last edited:
Could do a premium SKU.
Yup, 5 WGP per SA.
EDIT: quick block diagram to illustrate:
Dat Geometry Processor thicc, brah...
Gotta lift some heavy workloads, crunch dem triangles, dang.
Just scaled it like whatever to fill up the blank space, lmao. Probably would need extra ACEs, a bigger GCP, etc.
If the latter scenario is possible, then why not mine? View it as two 13 WGP SEs, 5-4-3 + 5-4-3. Use only 5-4 for PS4, 5-4 + 5-4 for PS4Pro, and the whole thing for PS5. There's a slight hit to yields since the lead array can't have any WGP disabled. But on the other hand, the chip as a whole is smaller, pushing yields back up (I would think more than the losses).
Really? That would put XSX with the worst ROP/CU ratio of any Navi product, by far. Is there a technical reason why this wouldn't matter? From my admittedly very limited understanding, modern rendering techniques are trending toward more and more MRTs. Doesn't that require greater ROP resources?
On the other hand, GCN cards had lower ratios across the board. So maybe it's no big deal.
Last edited:
X1X had 32 ROPS only and it was fine.
I don't think ROPs are that important. More than 64 and you're likely to be memory bandwidth bound anyways.
I imagine that having SAs that are almost half as capable as others might end up creating work scheduling issues, especially since the SAs are much more of "standalone" entities than CUs were in GCN (I mean, they have their own cache level, WGPs don't talk to L2 at all). I mean, having any kind of unevenness at all is probably less than ideal, but you likely wouldn't want to take it very far.
Don't know if it will be used but Sony patent a light transport accelerated method Photon mapping linked to raytracing most of the time or surflet if without raytracing phase.
Important to remember what Reiner said, things change anyway. We'll see
Yeah rumors....
Important to remember what Reiner said, things change anyway. We'll see
I mean, can we really trust Reiner, Klee, and the others? 🤔
No. It would have to be 1800 or less. Kaijus theory is sound.
This one? https://respawnfirst.com/details-what-changes-with-ps5-game-invites-system/
Don't think it was
You are telling me that there is not much money to make on PC........or trying to convince me what exactly? All of this coming at a time when Halo did 2 million in Steam sales in 2 days at the back end of last year. Free money.
There have been a few. Just earlier today there was a neat one about dynamic VRS.
What do you guys think do these two relate? Has Microsoft mentioned Navi anywhere or just RDNA? Do we know what's different between NAVI and RDNA2 other than RT and process node?
- Status