-
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
Nah we are more even now .
It's anyone game so it both of them at the same time lol.
I've posted a few times in the subject. The quick summary is more CUs are "better" in the sense that they'll yield better performance within the same thermal constraints, presuming tasks can be efficiently scheduled. Higher clock speeds are "better" in that smaller designs are easier and cheaper to manufacture, and they scale predictably (but one of the those predictable factors is that clock scaling results in exponential increases in heat which will ultimately limit how high you can go.)
Last edited:
Oh you mean that game coming to PS4? Yeah I've heard of it.
How does it even relate to Scarlett at all?
My predict is based on precedent. The Xbox One X had clocks that roughly matched the RX570, which was around 100 MHz lower than RX580. RX 580 is the 180W card to compare to 5700 (non XT), so I subtracted a little more than 100 MHz from the game clock to get 1500MHz. Only a 300 MHz bump from X1X.
RX 570 is a full 30W below RX 580 with those clocks, despite only being 4 less CUs. 5W is probably due to slower RAM, but you're probably saving 15W normalized for CU count just by lowering clock ~100 MHz.
....wait shit he actually said this 2 years ago?
edit: nvm, found the article and it seems he's saying the assets for gran turismo can run at 8k
edit: nvm, found the article and it seems he's saying the assets for gran turismo can run at 8k
User Banned (2 Weeks): System wars, trolling, history of similar infractions
I think the main thing when astroturfing is to ensure your chosen product is always in the conversation, no matter what the topic. At least that way there is a chance that it can be relevant once again rather than completely forgotten by potential customers.
Maybe some of the astroturfers can correct this assumption if I've got things wrong? I'm happy to take all feedback on board ;)
Definitely referring to Polyphony there. Whether or not we're seeing PS5 Devkit footage in those videos is another story.
Quoting this from the Radeon RX 5000 family GPU thread :
Pretty much de-mystifies how next-gen consoles will implement hardware ray-tracing. This is not some in-house secret-sauce from Sony or Microsoft camps specifically but rather incoming features on the Radeon roadmap (2020-2021 on PC). We will see if next-gen consoles will be based on RDNA2 or if they will be based on RDNA with the hardware ray-tracing aspect brought over early (I would bet on the latter). Just like the PS4 Pro Polaris GPU had rapid-packed-math implemented early despite it being a Vega GPU feature on the roadmap.
Pretty much de-mystifies how next-gen consoles will implement hardware ray-tracing. This is not some in-house secret-sauce from Sony or Microsoft camps specifically but rather incoming features on the Radeon roadmap (2020-2021 on PC). We will see if next-gen consoles will be based on RDNA2 or if they will be based on RDNA with the hardware ray-tracing aspect brought over early (I would bet on the latter). Just like the PS4 Pro Polaris GPU had rapid-packed-math implemented early despite it being a Vega GPU feature on the roadmap.
Oh. Just saw that this is from may. Thought it was a recent, post ms e3, "leak"
In short:
RDNA Tearflops ≠VEGA Teraflops
5700XT = 11 VEGA Teraflops
You're failing to take into account the additional die area taken up by the RT hardware. In Turing, this is not insignificant.
a) consoles will have a bigger userbase than entirety of RT supported cards on PC out there in a week from launching. You can expect that's when RT adoption really starts, and CDPR has always been very port-happy with their updates.
b) in all probability, the nextbox SDK will include DXR implementation (it doesn't matter if software or hardware based), so at least support for MS console will be relatively simple to adopt (barring performance problems of course). PS5 will be the odd man out in terms of API support, but then that's been the case for the past 25 years on all non MS consoles, so nothing new there either.
The problem is, that HBM2 + DDR4 has not the same bandwidith.
24GB GDDR6 RAM 14Gbps on a 386bit bus has a total bandwith of 672GB/s
HBM2 in a 2 stack (2*4GB) configuration at 1000MHz on a 2048 bit bus has a total bandwidth of 512GB/s
So you still need 160GB/s bandwidth with your DDR4 RAM solution to match GDDR6
And that would be a 256 bit interface with 5000MHz DDR4 RAM (8* 2GB)
HBM makes sense. HBM in combination with DDR4 does not make sense, unless you're okay with less total bandwidth.
And only in that case would you save any significant power savings over GDDR6 only.
Yeah,i guess we can expect something like that from Anaconda.
Any predictions for PS5 Anex? It's a bit tricky to extrapolate Pro unlike 1X i suppose?
I just caught up with this thread in time to see SBK was freed and re-incarcerated within a few hours...
Don't do this to ya self man.
Don't do this to ya self man.
I don't think its even near official that Lockhart is out of the game yet. MS just stated they have their overall concept of scarlett in the works but didn't mentioned anything about how many consoles will be out. Its still up in the air and probably being decided in the upcoming year.
TSMC 2nm node:
TSMC To Commence 2nm Research In Hsinchu, Taiwan Claims Report
TSMC To Commence 2nm Research In Hsinchu, Taiwan Claims Report
It's more a matter of "peak teraflops is a terrible measure of gaming performance." There is no simple multiplier at work. The 1.25x factor people have adopted here is likely not a reliable indicator (it was a description of how many instructions can be issued per clock cycle, which will already be factored into the published peak TF number.) So no, the 5700XT is not theoretically capable of 11 billion floating point operations per second. What it should be capable of is spending more time closer to its actual theoretical 9.75TF peak than an equivalent Vega GPU could with the same workload.
... and we're seeing that borne out in benchmarks. The next generations consoles will be a significant leap forward in terms of what developers can actually get on the screen, which is what matters in the first place. The proof is in the pudding, as they say, and it's likely to be at least the end of this year before we start to see any actual games. Next E3 will be the blowout once all the remaining big current generation titles have already shipped.
No PS5 prediction because we have no die to make guesses off, but I would guess less CUs based off Gonzalo clock.
The 1.25X is not included in TF numbers because it doesn't affect them. The IPC number is a utilization or efficiency metric. TF is a theoretical capability measurement. The amount of theoretical compute per CU has not changed, but the way those ALUs are organized is vastly different than before.
Gonzalo was said to have Navi 10 lite.
So 36-40 CU with ~1800 mhz.
That being said the addition of RT and the rumors of high tfs in PS5 makes me doubt Gonzalo was PS5.
Gonzalo might have been Lockhart(half ROPs), or some other announced Chinese console.
Yeah that response didn't even make sense. Not to forget Age of Empires 4, Microsoft Flight simulator, etc
Yeah, was def not saying you made your predictions off of the pastebin or anything, just a funny coincidence. Your posts in this thread have been very level headed and reasonable. I'm just glad more people are coming around to the idea of next gen consoles having higher clocks than previously expected now.
I actually think both consoles will have the same number of active CUs (56), any differences in clock speeds will come from cuts or innovations in other areas of the console, such as different memory and cooling setups..
IMO Navi is capable of doing 4 SEs of 10 dual CUs each. I think next year we will see 80 CU cards.
Consoles with have 3 SEs and 60 CUs. 54 active like Anex said.
Consoles with have 3 SEs and 60 CUs. 54 active like Anex said.
Maybe the gaming industry as a whole is also done with AAA gaming soon?
If you wouldn't have posted the "status update" I wouldn't have noticed as he is on my ignore list since ages.
What did this say again, I had the page open and it was a chart. I refreshed it and the tweet is dead now.
It depends what they do. One of their research paper talks about only a 5% die cost for speeding up kd tree traversal.
AMD has already stated they're going to target a limited number of RT effects for RDNA 2. I would expect their solution to be more compact.
He keeps deleting the tweet to update the table. I edited the post to include the latest.
Devs get to choose what effects RT is used for. Heck, there is even a project where it's used for super sampling along aliasing edges.
Like NVidia's RTX cards, AMD are suggesting that there isn't the power to fully ray trace all lighting, shadows & reflections, so developers will choose where to apply it.
NVidia had a good talk on Ray Traced Irradiance Fields at GDC this year, hinting at a good use for rays that won't be as accurate as per pixel, but doesn't have the issue of scenes taking multiple seconds before their lighting data is complete.
Don't forget how that bandwidth is actually being used.
One pool of GDDR6 @672GB/s is sharing that bandwidth between the GPU and CPU. And we have all heard the horror stories of memory contention in a shared pool of RAM. So even if the CPU only needs like 50GB/s of bandwidth it ends up taking anywhere from 75GB/s and up.
With that split pool, all of that 512GB/s of HBM bandwidth is going to the GPU exclusively.
I can see a set up where sony puts 16GB of DDR4 RAM, 8GB reserved for OS and SSD and 8GB for games coupled with 8GB of HBM2 RAM. If so doing would be cheaper (and less power hungry) than going with one pool of 24GB GDDR6? I can see it happening.
What! can't take a light hearted comment or what.. I have know idea what you're even implying LOL.
I thought forums were supposed to be light hearted fun?
Now thats one game I would be there day one for on new Xbox, Microsoft's conference didn't really interest me with what they showed but that looked incredible, especially if they can pull off what it looks like their trying to do.
Can only imagine how good that will look and feel on a 60" Tv next gen
EDIT:' Sorry for double post my internets jumping all over the place
So basically the latest legitimate info (by the GI editor) suggests PS5 may be more powerful...but there still seems to be a lot of speculation that the next xbox will be more powerful based on what? Neither situation is confirmed in anyway, but it doesn't seem like there is any confirmed insider info to indicate the latter, except for speculation based on uncertainty around the two sku model...
Apart from the most recent tweet by the GI editor, is any think with decent credence that indicates either way? Or is it largely down to speculation orientated to a degree by the preferences of the speculator?
I am sure this is just a diversion. Most of the interesting detective work is simply around working out specs based on info, rather than the comparative stuff obviously.
Apart from the most recent tweet by the GI editor, is any think with decent credence that indicates either way? Or is it largely down to speculation orientated to a degree by the preferences of the speculator?
I am sure this is just a diversion. Most of the interesting detective work is simply around working out specs based on info, rather than the comparative stuff obviously.
Yeah, this is what I was going to say. The rumour described this exact thing. The stated bandwidth of GDDR6 is not what you actually get in reality, unlike with the other method.
We have that tweet, and we have other insiders saying that nothing is final, so speculation is wide open at the moment.
Man, it's lile SpinningBirdKick is playing a game of whack-a-mole with the mods -- and he's the mole... loooool.
He just can't help himself.
He just can't help himself.
Wouldn't that mean no unified memory for the CPU and GPU? A feature Mark Cerny advertised as a highly requested dev feature and a big advantage of the PS4 ?
No. The High Bandwidth Cache Controller causes the RAM to appear as a single pool to developers and manages it all in the background, so the issues with split RAM pretty much go away.
- Status