Yes! Bet is on :)
I say One of the consoles will be at least 12.9tf or more, you say 10.7tf or less. Week long avatar bet.
If it happens to be in between, like 11.somthing we'll call it a stalemate. Sound good?
-
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.
Next-gen PS5 and next Xbox speculation launch thread |OT5| - It's in RDNA
- Thread starter Mecha Meister
- Start date
- Status
I say One of the consoles will be at least 12.9tf or more, you say 10.7tf or less. Week long avatar bet.
If it happens to be in between, like 11.somthing we'll call it a stalemate. Sound good?
I've wondered if motion matching will be improved beyond TLOU2 in the next gen, or if it's more that we'll just see it in every game rather than a single really advanced one.
Maybe. But we can also remember than the 2 previous Sony consoles where more advanced than Xbox consoles. And Sony last one was released one year before the Xbox and was still more advanced.
People are getting really hung up on the whole "custom designed" and "collaborated with" when it comes to Sony and Microsoft working with AMD. I remember these arguments back in the early parts of this generation with Xbox using custom parts and PS4 using "off the shelf" components and it turned out that both of them were pretty much identical with customizations here and there. I feel this is what's going to happen again. It's kind of like Microsoft not wanting to say Jaguar CPU in the One X and choosing the say custom designed CPU, and people fell for the PR, and it's really just jaguar. Also how people took the FP16 features of the PS4 Pro and overblew them. With it having that feature of Vega, I'm sure Sony and Microsoft get to choose which customizations they want to add to their next gen consoles but it's not like they're going to be gamechanging customizations.
Parasite76 is a mrxmedia acolyte lol
I need to get some nasty bass over here, because I'm feeling all Alexander O'Neal
A 2060/2070 level GPU, proper PC level 8 core CPU, and PCIe4 level SSD will be a fantastic baseline for the next gen
Yes it can be used and they will even go further with motion synthesis using AI to improve motion matching
No, he didn't confirm it. You would have to watch the Giant Bomb interview for the exact quote, but I believe he stated his expectations are that the process would be similar to 360 BC
Thanks, Colbert for the update. Goes in line with what I am personally expecting. To the letter. My range or Navi was 10TF -10.5TF. Which falls within my GCN range of 10-12TF lol.
I am curious as to the Mem bandwidth thing though. That a lot of potential bandwidth for sony to give u just to use a cheaper PCB.... especially if also using GDDR6.
I am curious as to the Mem bandwidth thing though. That a lot of potential bandwidth for sony to give u just to use a cheaper PCB.... especially if also using GDDR6.
Indeed, I'm very happy with how these consoles are shaping up. Not to mention that with CBR and VRS they can punch above their weight performance wise with minimal impact on the graphics quality.
I hope so to.
That said, imagine what Anaconda and PS5 could do graphically, if an AAA developer targeted 1080p, choosing to push the most "CGI-like" visuals as they can.
i.e. Toy Story in KH3, but going much further. Nobody should be barred from doing that, and yet another reason Lockhart is a bad idea. If Anaconda was used to do something utterly spectacular like a realtime version of Toy Story 2 or 3 (or insert your favorite CG movie/cutscene) in 1080p, where who that leave Lockhart, sub HD?
Obviously things like that would be the exception, not the rule.
Navi's larger caches will help reduce memory bandwidth dependency.
After what we know about Navi this seems very unrealistic to me even for $499 console.
Y
yeah it could basically deliver that ray traced troll demo in-game easily!
Last edited:
It seems to be still 50GB/s for every teraflop.
I.e. the ~9 teraflops in RX 5700 XT for 448GB/s
Hey now, Lisa did say it very clearly.
PS5 = "Next generation Radeon archictureS"
XBox="RDNA"
It's pretty clear to me this confirms PS5 is the amalgation of all the future GPU architectures AMD plans to build, while new XBox is just a Navi.
That's a weird assumption. Wasn't Sony who wanted "desperately" Navi for their hardware and now MS uses exclusively it meanwhile sony adds next gen stuff to their Navi? Doesn't make sense to me.
Last edited:
Sony:
"Working collaboratively with Sony over the last few years to build a custom chip based on the Zen 2 and next gen Radeon Architectures."
Microsoft:
"Custom Processor co-engineered between our team and the Microsoft team, with one goal in mind, to deliver the ultimate gaming experience. It uses our Zen 2 CPU Core and next generation Radeon RDNA gaming graphics architecture."
They both said Navi. The rest is PR fluff until we get architecture details. We don't even know if their HW RT implementations are the same or derived from RDNA 2.0, as likely as that may be.
It is still of some interest that Microsoft specifically said "real-time hardware accelerated ray tracing" explicitly as compared to what Sony said about ray tracing, and it will remain that way until Sony provides further detail on their ray tracing implementation. I know that one of the mods on here Matt said that he had heard they will both have some type of hardware ray tracing, but there is a whole spectrum of what that could mean of course in practice until both sides reveal the architectures fully.
Sorry there's so much FUD in here all I see are bogies 😬
I'm just a layman not a dev.
But I think every answer from today's standpoint will be vague even if you are asking a dev.
I expect rather limited integration of ray tracing and see the current Geforce RTX hardware as a usable reference point.
There is no AVX512 support in Zen 2.
AMD kept it simple and made an upgrade from 128-Bit pipes to 256-Bit.
They use relatively lightweight cores and more of them.
For multithreaded applications you still get good vector performance.
RDNA has multiple working modes on different levels.
Wave32 or Wave64 is used depending on the workload.
Depending on the circumstances Wave32 or Wave64 is more efficient.
On a higher level a RDNA Compute Unit can still be compared to one GCN Compute Unit because under RDNA a Compute Unit can still work indepentendly.
Every Compute Unit can work on a Workgroup, which is also called CU mode, where the registers and caches are all exclusive per Compute Unit.
But it's also possible to have two Compute Units working on a Workgroup together and sharing ressources, this is called WGP (Work Group Processor) mode.
Last edited:
Yh I get that much from that nexus guy youtube breakdown. But if there is a choice to go with a 384bit bus ad still ony have 16GB'ish of RAM... then why not do it? What the benefit of gong with a smaller 256bit bus? Less space is taken up in the chip for mem controllers? Less power draw? Cheaper PCB
Smaller PCB, less die space usage, less power used by less amount of chips and less buses.
Hahaha this will be yoursDEAL!
If you lose, you have to wear this for a week:
Oh I know, seems illogical today.
But wait till you see the leaks coming out over the next year and a half. 10tf will look like a wet sock.
Well, a wider wave will have a bigger impact when it stalls due to the nature of a wider wave - it has more data in it. It will also have a bigger waste of resources if only part of said wave goes into execution. That's expected. The question is if there are any benefits at all to running 64 wide waves over 32 ones.
All current DXR titles denoising were pure compute added to TAA routines. Denoising is kinda trivial.
Them highlighting BVH intersection tests point to this being their primary acceleration point. Why would they highlight this portion specifically otherwise? I agree that there are many options in how they may approach the acceleration though.
The way I understand it the primitive shader runs on incoming geometry stream and culls invisible triangles. This workload is handled by general ALUs as it would be with any shader. Vega isn't really a great reference as neither work as intended on it so we can't really know what they intended for either.
NV has tensor denoiser but no one uses it. One reason is likely because it would tie your code to NVAPI as said denoiser isn't accessible through DX (or wasn't until DXML release with Win10 1903 at least). Another is that tensors seemingly can't run alongside main SIMDs due to internal b/w constraints which means that you can use either/or anyway.
I don't know if this answers any of my questions really. My point still stands: instead of calling half of WGP a "CU" they would be better off calling the WGP a new CU, much like NV did back with Maxwell for example, calling the new SM "SMX".
You could argue either way because it's kinda in the middle of it.
For example in WGP mode you have 4x SIMD32 working on the same Workgroup, sharing the registers (2x256KB) and LDS but the TMUs and the L0$ is still per CU (2xSIMD32).
That's not the case with an Nvidia SM or GCN Compute Unit, where everything is shared.
In CU mode a Workgroup is executed by one Compute Unit (2xSIMD32) using 256KB of registers, half of the LDS and its own TMUs + L0$.
You could of course take the freedom and call 4xSIMD32 (WGP mode) a Compute Unit but then you would need a name for CU mode and what half of the WGP (or half of the Compute Unit then) should be called.
lol haha.Hahaha this will be yours
i will wear the 12 tflops shame badge too. YOLO!
Are they? This wasn't really clear in any of the docs released so far.
The bigger issue though is that they haven't really provided any reason for such design. Why would anyone want to run a workload on WGP instead of a CU or vice versa?
Technically should we? In all fairness, most of us made our 12TF prediction at a time when we were calculating based on GCN flops. Now It's clear that 9.7TF'ish Navi flops is about equivalent to a 12TF GCN card. And if those consoles comen at anything over 10TF then they are about 12TF+ compared to a GCN card.
Basically, we have an out.....
This is what I want. I'm not hung on game being native 4K. Give me 60fps and betters assets. Native 4K should be an option, but I don't make it the only option.
Question is are you talking about RDNA 10.7TF? Or 8.56TF?
One matches the stadia numerically, but the Stadia would need around 13.3TF to match it, the other matches the Stadia in performance but will have a smaller TF number.
- Status