If raytracing is working bad on consoles this is because they will not have raytraced hardware. If they have it, it work correctly... There is no miracle...
-
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
If raytracing is working bad on consoles this is because they will not have raytraced hardware. If they have it, it work correctly... There is no miracle...
Read my post above. Nothing related to hardware or software, it's just pure marketing and commercial reasons. How hard it is to explain? As if this is not happening rigth now.
That is the point. You also need a memory controller if you want to go HBM + DDR4 instead of just HBM.
And if you want to use a decent interface with 256 bit that is at last somewhat useful DDR4 bandwidth, you increase the power consumption by a lot.
It's not like they run a desktop PC setting with just a 64bit interface.
It's the total bandwidth, that determines the total power draw of your memory solution. And for that DDR4 and GDDR6 are not very different. They're the same technology anyways. Anything you would accomplish in power saving by going with HBM, would be negated by DDR4 to a great degree.
so the RX 5700 series was the navi 10 after all? so these GPUs and the consoles are indeed made of the same base.
In that case i think its safe to say that in terms of raw teraflops number we wont surpass the 9.75TF mark.
As long as the games are using DXR, AMD hardware that supports ray tracing will be fine.
You basically have three layers. The API, which is the language/instruction set the game uses to talk to the hardware. Then the drivers, which take those API instructions and interprets them for the hardware, and finally the hardware itself. DXR is the API, RTX is the driver and hardware (RT cores).
DXR is platform agnostic. If AMD wanted to, they could put out drivers today that take those DXR calls/instructions, and have them run on compute cores. It would be horribly slow, like ray tracing on 10xx Nvidia cards, but there's technically nothing to stop AMD from supporting it today.
This is different from something like PhysX, which was a proprietary Nvidia physics API (but is now open-source, so AMD could even support PhysX now).
Totally! all the behind the scenes twists and turns are what I really find fascinating.
Also that almost nothing is ever certain. Except death ofc (for now).
So I like to believe that PS5 was 2019 and then change to 2020 for whatever reason (maybe MS forced them), and that MS had 2 SKU's and now they're not that sure anymore, etc
This was ZhugeEX on an institutional investor he happened to talk to.
This coupled with some of the interviews that Matt Booty gave show that they are looking for AAA games, but they have brought in studios that can give Game Pass content when AAA games are not showing up.
It is not hard to imagine that they launch within the first year we will see Forza Motorsport, Halo Infinite, a third party game (because the have always had one/two at launch year). And not to mention that Microsoft studios have generally not taken that time (4 or 5 years) to get AAA games out.
Developers are already using different setups for PC, they are already making more and more games for base console and mid gen consoles. I honestly do not know why people think that it is a big thing, not when both Microsoft and Sony have stated that the porting process between the consoles they have something that can be done in minimal time with minimal staff.
We have no idea what version of Navi MS is using. Sony could be using Navi 10 as early silicon to get dev kits out sooner. They could even do a chiplet Zen 2 Navi 10 APU out of off-the-shelf parts before their custom APU die is sampling. It's good to be skeptical about a 1.8GHz GPU in consoles, but you have to peg Navi 10 to the max to approximate your console if you plan to have more CUs.
That doesn't make any technical sense. Have you ever designed a memory subsystem or worked with someone who has?
delaying lockhart is even weirder, you force developers on launch to make 2 versions of a game, with the second one not even being used until a year later.
good post.
i'd add 5. if lockhart were canceled, MS would be crystal clear that scarlett was one console. there would be no reason to be vague.
lockhart+anaconda is the plan. bet me.
It doesn't change my point that you do benefit from going wider and reducing clock speeds. Spreading heat across more units at lower clock speeds should also help improve the ability to dissipate heat, but I don't have a reasonable way to estimate the effect. Nor, as I pointed out, can I guess how a second iteration of RDNA will improve matters. Adopting v1.0 of anything generally means you wind up with some sub-optimal aspects that another small revision can iron out. I suspect the 2020 launch reflects dissatisfaction with where Navi was going to be this year, suggesting that there's something worth waiting for coming in the next iteration.
We shall see, of course.
It would definitely be expensive, no argument there. I would be shocked if the 40CU designs didn't have any redundancy built in or the same applies, so I assumed the 50% increase in size accounted for increased redundancy as well. Hoping for zero defects at ~250mm2 is likely to have serious yield implications.
Last edited:
A hypothetical system with a total memory bandwidth of 700GB/s uses around the same power with either a DRR4 or GDDR6 solution.
So there is no significant power saving at all using DRR4 on a per throughput basis.
That is the point.
HBM on the other hand is a lot better when it comes to power consumption per GB/s
There's going to be a PS5pro later down the road. Assuming those numbers hold on.
You're concerns are unfounded. These games being marketed as RTX titles will have no effect on whether the raytracing functionality runs on non-Nvidia hardware anymore than a title being marketed as a "The Way It's Meant To Be Played" prevents it from running on non-Nvidia hardware.
Here's my Anaconda prediction now:
3 shader engines, 6 workgroup processors, 1CU pair per SE disabled - 54 total CUs. 1500MHz. 10.4TF.
3 shader engines, 6 workgroup processors, 1CU pair per SE disabled - 54 total CUs. 1500MHz. 10.4TF.
I really hope to be wrong and to see the same games that are having a partnership with NVidia for RTX RT to have RT on consoles and working on ATI GPUs. If it doesn't happen, then don't get surprised. Only the future will tell us.
Looking back after watching all the conferences, I think it was a brilliant marketing move for Sony to announce the PS5 a few weeks ago, and other hardware makers will surely take notice for the future.
The news would have been buried underneath all the other gaming news around E3. I feel only the techies are even talking about Scarlett right now while the PS5 dominated mainstream press cycles for a good while. To be honest, it didn't help that their reveal was basically a complete re-tread of what Sony mentioned a few weeks ago.
The news would have been buried underneath all the other gaming news around E3. I feel only the techies are even talking about Scarlett right now while the PS5 dominated mainstream press cycles for a good while. To be honest, it didn't help that their reveal was basically a complete re-tread of what Sony mentioned a few weeks ago.
You think that both MS and Sony will further customize the architecture of current NAVI to include more CUs to have more TFlops or do you think that they will wait a little more for AMD to finalize their
Last edited:
Microsoft has never been shy about getting people into their ecosystems with free trials. 6 months is way too aggressive. but even 1 month trial to gamepass (two weeks feels short) could be a big selling point. It feels a lot easier to swallow an expensive console when you can point to the fact that upgraders wouldnt have to buy any accessories or games at launch. No "hidden cost" of a second controller. Play our biggest hit for a few weeks....gut sucked into our eco system...... It certainly feels like they could persuade people to buy in early by alliviating day one cost (even if they have to pay it later, little by little, to keep those games). Thats just how subscriptions work. Get people sucked in the short term and make it just easy enough not to cancel.
Good, you are confirming that a HBM2 + DDR4 solution has a better pj/bit metric than an equivalent GDDR6 solution (say, 8GB HBM2 + 16GB DDR4 vs. 24GB GDDR6. Note that total bandwidth is similar, but not identical and that is intentional on part of the electrical engineer). Given that the memory access patterns for a 8-core/16 thread Zen 2 are widely different than those of a Navi GPU, I'm sure you can clearly see that the efficiency gains from separating the memory traffic into two memory pools is high. Fortunately, the two memory pools appear as one at the software API level, meaning there is no complication factor for the game programmer.
Are they suddenly cancelling Halo Infinity, Gears 5, Forza 8, Forza Horizon, and a possible Fable? I get people may not like the Gears/Forza/Halo constant hammering, but if these are their major AAA franchises, and they work in a bunch of AA stuff on a steady basis for cheap with gamepass, it seems a valid stategy. It may not be for everyone but it could be for them. And with Spencer saying he doesnt care how many consoles they sell, only who is using their services, then the Gamepass way may be good for them.
i dont know about that. i have no idea what happened here overnight but when i went to sleep we were discussing 11-12 tflops, 56-60 CUs because the Scarlett APU is huge, way bigger than the base PS4, Pro and the X1X. Some estimates are 400 mm2.
The Navi GPU revealed yesterday is only 250mm2 and 40CUs. Even if you add 70mm2 for the Zen 2 CPU, you still have a lot of space to add more CUs.
I am back on the 12.9 Tflops train after that.
Scarlet - 64 CU at 1.5 Ghz - 11.5 tflops
PS5 - 56 CU at 1.8 Ghz - 12. tflops (HBM2 + DDR4 lets them push clocks higher)
I have no idea why we all thought they would settle for a 350mm2 die when they are no longer going for $399 standard cooling consoles. These consoles are going to be $499 which means they can create a bigger die than a PS4. With Vapor chamber cooling they can push clocks up to high levels as well.
project awakening is hinted at being a next gen title in the interview. based off the TGS trailer this game looks very impressive.
They both have hardware raytracing.
It's part of the Navi roadmap, so could ne custom hardware for these Navi chips but based on the upcoming standard RT hardware for the Navi line of GPU.
RDNA 2 is confirmed to have hardware ray tracing, its likely MS and Sony will modify their chip with that feature (like ps4 pro including vega features)
Welp, it's good to be back.
I've missed you fine upstanding ladies and gentlemen.
What are the currentastroturfing objectives topics of conversation currently being kicked around?
I'm sure there's lots to dig into.
TF navel gazing is always a favourite to fall back on. Although I still don't think either Anaconda or PS5 should be below 12.0 TF in platforms that are expected to launch next year and last 7-8 years.
I've missed you fine upstanding ladies and gentlemen.
What are the current
I'm sure there's lots to dig into.
TF navel gazing is always a favourite to fall back on. Although I still don't think either Anaconda or PS5 should be below 12.0 TF in platforms that are expected to launch next year and last 7-8 years.
It isn't customization to add CUs. The architecture is inherently scalable. I think we'll see 80 CU flagship GPUs next year.
RDNA 2 is probably not going to be named Navi 2. For now, I'd call consoles RDNA 1.5.
Didn't even realize you were banned for console warring. Not surprised though.
My bad I meant RDNA 2. :p
What is better? A GPU with more CUs and lower clock speed or a GPU with less CUs and higher clock speed? I know the latter brought you higher flops but on the terrain which will have better results in games, streaming textures and shaders etc..?
Last edited:
I think they are still undecided. Maybe they are talking to devs and getting new feedback. They have no reason to commit and annonce two skus right now anyway.
I guess people have different senses of humour and some are more touchy about certain subjects than others.
I must and will try better to be more accommodating to all the different perspectives going forward.
Hehehe....firestarter.
They have announced RT support on the one GPU series that has native hardware support for RT. How about you wait till AMD has GPUs with native hardware support for RT before you start peddling exclusive partnerships.
There is RT, then there is what Nvidia calls it (RTX) and there is what AMD will call it. A dev saying they support RTX doesn't mean they wouldn't support RT in whatever form AMD and hence consoles will have it.
I've asked this several times myself. i dont think it makes a difference tbh.
That said, i was looking at vega 64 and vega 7 (60 CU) benchmarks and vega 7 had drastically better performance despite having fewer CUs and only around 7% more tflops. But that was likely due to the 1 Terabyte memory bandwidth.
🤔 from May 25th
im joking of course, but would be funny if both of you are right on the money, and if this leak is real. I also imagine that even if this pb is real, final clocks may still change
There is no limitation on what platforms a dev may implement raytracing on in a deal with NV. DXR/VK raytracing is working on Pascal GPUs right now, without any h/w acceleration, it's pure GPU compute essentially, RT h/w just makes parts of said compute run faster.
Now, it is certainly possible that future DXR-compatible RT h/w won't be well suited to running current DXR implementations which are made and tuned for Turing's RT cores. In some cases some patching may be required. But that's hardly NV's or developers fault that nobody but NV has RT h/w available right now.
I think we're at Xbox is more powerful, or was that yesterday? Maybe it was PS5 and Xbox is today?
You need to keep up it changes by the hour..
I've been thinking about this, because of that Gran Turismo SXSW ray tracing demo
but it's bloody 8k and 120fps.
For 4k rendering wouldn't we start to see diminishing returns on how much these GPUs can achieve after a certain point? Back when the Pro was announced a lot of people thought 8TF was a good place to be for this objective.
I think the CPUs will be far more relevant this next generation.
I think the CPUs will be far more relevant this next generation.
Quick and dirty:
No need for Lockhart since XBOX One S (and maybe X, but I don't think so!) can take care of the price sensitive customers. MS announcement not to deliver any next-gen exclusive 1st party titles in the forseeable future strongly indicates that XBOX One will still be around once Anaconda arrives.
Looking forward to anyone's feedback to those maps btw.
I like a good picture! They really are worth a thousand words (or wrong assumptions).
I don't believe the Post E3 image is quite right. If we disregard performance for now and concentrate solely on price.
Sony is targeting the mass market consumer, not the performance enthusiasts. So while Lockhart is still a thing (and it is) the PS5 price will always be skewed towards that rather than Anaconda. Even though the PS5 and Anaconda have similar specs.
Lockhart wasn't mentioned at E3 because it confuses the simple soundbite type of messaging that is required at these events. It needs more explaining to customers so they don't get the wrong idea or bad perception about it. So better to avoid that at this stage.
I'm sure that MS would be delighted if Sony bought the 'Lockhart is dead' rumour to tempt Sony to play at the same price point as Anaconda but that is never going to happen.
edit ugh. Auto correct
I think it's pretty obvious now that Navi 1x are RDNA while Navi 20 (and other 2x) is RDNA2.
Navi 12 and 14 should be below 10 so we're probably looking at a 20 and 10 CU parts.
I don't know if it was in SXSW, but googling GT Sports 8K demo returns some CES 2019 results
And here's the thing... GTPlanet.net says it's upscaled to 8k. It's not like it was shown in a manner where pixel counting would be possible to make sure. So that's one.
Two, it would not be the first time for PD to show off a GT demo running on multiple consoles to hit insane resolution and frame rates that will never make it to the retail GT versions.
Yeah I know. There's a few, just that I wonder was it using the devkit or not.
Didn't know it was upscaled though!
- Status