-
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.
- Status
Entroyp
Member
hmm you are right, the 1080ti and 2080 ti have 88 ROPS. Interesting.
Darius87
Member
i can assure you 100% rdna 2.0 will have dual compute units which means it also will have L0 cache per dual CU's unless Sony/Ms want it downgraded arch to their new consoles, gcn has nothing to do it here
you also admited to look at RDNA white papers(even page 17 was right)and still got it wrong
it's like dumb and dumber movie all over again 
DiddyInAHellcat
Member
The fake insiders should be banned at that time
Last edited:
DJ12
Member
No one's talking about l0 cache. When calling names probably best to make sure your post is correct. Loses its affect.i can assure you 100% rdna 2.0 will have dual compute units which means it also will have L0 cache per dual CU's unless Sony/Ms want it downgraded arch to their new consoles, gcn has nothing to do it here
you also admited to look at RDNA white papers(even page 17 was right)and still got it wrong
gaming_noob
Member
Ban dat bumbaclut.
Darius87
Member
he can mention whatever he want's but if he says L1 is per cu which actually L0 are per dual cu's so that's his mistake.
L0 cache is 32KB and 4-way set-associative
Darius87
Member
doesn't matter mentions or not he said: L1 Cache: 32 KB (Per CU) which is wrong!!! stop defending frauds.
megreotsugua
Banned
What's happening in January?
THE:MILKMAN
Member
I am curious about this line from the Wired article regarding BC:
In what way could PS4's [GCN] architecture be incorporated into the Navi architecture for BC? Butterfly wing design or something else?
In what way could PS4's [GCN] architecture be incorporated into the Navi architecture for BC? Butterfly wing design or something else?
DJ12
Member
Lol. You don't know that at all. Maybe the IPC gains for Rdna2 mean l1 cache is per cu.
Maybe this helps with ray tracing.
I'm not saying Tommy is right or wrong what I am saying is you and misterxmedias cronnies don't know either.
I hope RDNA 2 does have l1 cache per cu, not because it will prove Tommy right, but to see the mental gymnastics from you lot.
Last edited:
The PlayStation meeting most likely will happen in February so January is the last time they will be able to give out these fake specs for attention.
Nickolaidas
Member
Hell, no. Quiet those bitches down - and give us the power of Grayskull as well.
Disco_
Member
What’s Next For High Bandwidth Memory
Different approaches for breaking down the memory wall.
semiengineering.com
Not info yet whether a recent huge order was put in to drop these or whether a sony sized order would lower them dramatically.
FranXico
Member
The new hardware architecture can be designed to also be backwards compatible, no?
DeepEnigma
Gold Member
How do PCs do it? (rhetorical question)
THE:MILKMAN
Member
That is sort of why I'm asking. If BC is basically "automatic" like PC then why does it even say it is based in part on PS4 architecture? RDNA is quite different to GCN and Mark Cerny detailed how difficult BC was even with Pro.
Not that this matters much. I'm just trying to make the time pass quicker to the reveal......
SmokSmog
Member
Sony can use 2 stacks HBM2E, 12GB per stack total 24GB 820GB/sWhat’s Next For High Bandwidth Memory
Different approaches for breaking down the memory wall.
Not info yet whether a recent huge order was put in to drop these or whether a sony sized order would lower them dramatically.
Thats mean smaller memory controller combined with 7nm EUV ( 20% area reduction compared to 7nm) you can fit 40cus RDNA under 200mm2
Samsung Develops 12-Layer 3D TSV DRAM: Up to 24 GB HBM2 per stack
Last edited:
DeepEnigma
Gold Member
That is sort of why I'm asking. If BC is basically "automatic" like PC then why does it even say it is based in part on PS4 architecture? RDNA is quite different to GCN and Mark Cerny detailed how difficult BC was even with Pro.
Not that this matters much. I'm just trying to make the time pass quicker to the reveal......
They probably mean x86-64 and even though the architecture is new with RDNA (instruction set), it is still based on GCN as well. Just like Zen is a new architecture (instruction set) for them based off of x86-64 foundation.
RDNA (microarchitecture) - Wikipedia
Darius87
Member
fixed Tommy boy
saintjules
Member
Looks like he's being discredited from the other side.
The ninja farmer
Member
and you are now taking the word of ‘mistercteam’ !!!
the lapdog of another mister I’m afraid and famed on the internet for his cobbling tech papers together and baffling people with ‘science’ !!!!
sorry, but even my cat doesn’t believe in dark silcone hiding inactivated cu units, Gpu’s hidden under power bricks and other countless madness statements. that have almost virtually never been close to correct. Or made a lot of sense
not saying that cache isn’t per shader array, but guess where the Cus live ?
Insane Metal
Member
TF is for Tommy Fraud
Darius87
Member
there's no lapdog or anything it's all clear writen in RDNA whitepaper.
mckmas8808
Mckmaster uses MasterCard to buy Slave drives
I never trust Avatars. Dudes on here be having Avis of women knowing darn well they aren't women themselves.
44alltheway
Member
in Klee we trust...
48086
Member
Do we? Why isn't he posting here?
Free will knight
Neo Member
Except for the L1 cache which is 512KB on the rx5700 with 40 cus but here it's 1664KB over 3 times that.
The ninja farmer
Member
wasn’t talking about you my friend
There maybe little nuggets of sense in there
but if your that keen on that persons posts, do a little research, especially on info updates that were made since just before the Xbox 1 was released and since (before backing that horse
)
chemical boy
Member
Everything related to MisterXmedia is an insult to intelligence.
Proelite
Member
Each shader engine in Navi has exactly 32 ROPS. They have always been powers of 2. i.e. 2, 4, 8, 16, 32, 64, 128 in AMD GPUs.
88 ROPS is nonsense.
Last edited:
chemical boy
Member
Tommy could lie (I don't think so) but for the moment he has the benefit of the doubt. We are not talking about a hidden dGPU with alien technology.
Tommy's data can be perfectly real. There is one sure thing, Sony right now wants to sell all the PS4 it can, it is not the time to talk about PS5. So it seems that until February we will not go out of doubt.
Tommy's data can be perfectly real. There is one sure thing, Sony right now wants to sell all the PS4 it can, it is not the time to talk about PS5. So it seems that until February we will not go out of doubt.
Darius87
Member
so you're telling me that rdna whitepaper is wrong where it says in page 9:wasn’t talking about you my friend ❤
There maybe little nuggets of sense in there
but if your that keen on that persons posts, do a little research, especially on info updates that were made since just before the Xbox 1 was released and since (before backing that horse
and page 17:
because that what blue guy on twitter said, so if he's wrong meaning amd is also wrong because they wrote rdna white paper.
Darius87
Member
blue guy from twiter and i proved that he's wrong and a fraud which i hope gone for good from this thread.
Sosokrates
Report me if I continue to console war
Meltdowns in reaction to sony's ssd tech?
iamvin22
Industry Verified
February will be the reveal
TeamGhobad
Banned
RDNA 2.0 cards will be revealed jan 6. so we will know the GPUs then.
True lol
Lort
Banned
It be here...
L0 cache 2 per dual cu each 16kb 32 way
L1 cache for each 5 dual cus 128kb 16 way
L2 4mB 16 way
This has 40 cus ( 4x10) ... most likely upgrade is to add one dual cu to each group = (4x12) 48 cus .. or add a whole extra group = (5x10) 50 cus ... or do both = (5x12) = 60
56 active seems likely
If you add a whole new group you will add more latency and wont be able to clock as high.
The ninja farmer
Member
I’m not saying the white paper, amd or you are wrong. Especially if it means that much to you
. Think a little chilling and Xmas cheer would do you good there 
I’m more warning
about that messenger and that even a stopped clock can be right, twice a dayand not to get so hung up on a tiny detail, or showing how clever you can sound to a bunch of strangers on the internet
there is far too much of that sparky, Feisty behaviour kickin around in all this next gen excitement and speculation. (And there is still almost a year to go, till hands on time
)
Insane Metal
Member
88ROPS is a monster. If something like that is confirmed it'll run 4K with 4x MSAA.
THE:MILKMAN
Member
Did you see Proelite's post above? 88ROPS is apparently nonsense for AMD GPUs.
TLZ
Banned
I had a dream..
The truth is most likely 96 or 128ROPS for VR & TV at the same time
Insane Metal
Member
Where is it? Can 't find
THE:MILKMAN
Member
post 19578 just a few posts back!
Dee Dah Dave
Member
Is misterxmedia still on the prowl these days?!
Insane Metal
Member
thicc_girls_are_teh_best
Member
What’s Next For High Bandwidth Memory
Different approaches for breaking down the memory wall.
Not info yet whether a recent huge order was put in to drop these or whether a sony sized order would lower them dramatically.
The order isn't the only factor here; it's a question of HBM3 manufacturers have the capacity to produce in the quantities Sony would need. If fabs and production lines are allocating tons of space to, say, DDR5 and GDDR6 (keep in mind Apple, IIRC, is looking to use GDDR6 in newer phones), that leaves less capacity for HBM3 production. Which can also affect pricing of components.
Maybe for some. The only real options here are ReRAM (which has a lot of ground to gain, but they could maybe pull it off) or 3D Xpoint persistent RAM. Either of them would be between 64GB-128GB in size, realistically. Probably soldered to the board, to a DRAM memory controller.
That would be the best surprise in terms of storage-based solutions for either system so I'm hoping at least one of them (if not both) implements that route because I do feel it'd cause some meltdowns indeed (especially if ONLY one implements it). That'd be a hefty, very high-speed cache sitting between main RAM and whatever more standard SSD the systems will come with.
Last edited:
- Status