I'm still just waiting for that moment where folk realize they should just wait for the specs and leave Komachi and Github out of it.
You're not going to fully understand that stuff.
-
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.
PS5 and Xbox Series speculation launch thread |OT9| - For flops sake!
- Thread starter Mecha Meister
- Start date
- Status
You're not going to fully understand that stuff.
correct
once upon a time 13F8 was Ariel.
I'm trying to keep track. I think it goes something like this:
Gonzalo = Platform/APU name (for dev kit)
Ariel = iGPU name for dev kit APU
Prospero = Dev kit name
Flute = Platform/APU name (for retail)
Oberon = iGPU name for retail APU (for retail)
That is how I'm organising it in my head even if it is completely wrong....
I am linking the code names completely different (from my understanding). I love it that we all have different opinions LOL.
Lets see:
Retail
Gonzalo platform
Ariel chip
Dev kit
Prospero platform
Oberon chip
Last edited:
Do you know how oberon (36cu 2ghz no hw rt) is connected to final ps5 hardware?
Last edited:
Between 11 and 12 TFLOPs? hehe, I kid.
Perhaps Oberon with its apparent 36 CUs / 9.2 TF is for a PS5 SKU that will face Lockhart and there's a high-end PS5 that will face XSX. I hope.
Some reflections are just not subtle though!
And I would not count on what you are saying exactly - for one, the rendering of transparency is in and of itself much more expensive and complex to Model than opaque hard surface. You end up having to spend a lot more rendering Power on Such surface for subtle distibctions, diminish Returns are strong there.
And if ray tracing is involved, subtle effects requires vastly higher ray counts. For example, quake 2 rtx just introduced double sided glass, and at Standard real time ray counts it barely Shows up for its added cost. It either requires many more rays per Pixel to Show up at all or not be undersampled, and/or a specialised denoising Pass to bias the results more to make it more visible (and denoisers are expensive).
In reality, i would expect next gen console RT solutions to dramatically increase ray/Pixel over what we currently see.
This makes 0 sense in the context of a console launch month. MS's production isn't going to be gated on what they can sell, it will be gated on what it is possible to produce within the pre-launch window the same as Sony's will be. The capacity to produce the components that make up the console, the consoles themselves, the storage of same and the logistical capacity for moving all of these pieces around between these places and out to retail is all finite. There is an absolute upper limit on the amount of PS5's that Sony can possibly get onto retail shelves within it's launch month no matter how many they can sell.
Now if you want to talk about the first year, then sure, I could see a massive sales difference being a real possibility, but the first month? That would genuinely surprise me.
maybe they're not actually console related but a separate product
Welcome Osiris. If you do ever choose to throw your hat into the ring and share any info you might have please remember the whole mod verification process :)
I look forward to any insights you might choose to share with us.
Who can really say? The only somewhat solid piece of info that has three sources is the Prospero dev kit name (Jason, DF and Gizmodo leaker).
Valid point, but remember that the PS3 wasn't merely $200 more, it was a year late and actually less performant in third-party games. That's a bit different to Lockhart launching alongside or after the PS5 with a tremendous drop in power due to the lower price. This is another example of why I'm iffy with using the PS3 as a precedent even if, as you correctly state, it's kind of all we've got.
Makes me wonder how they'll deal with the different files for each Xbox. Which version will be on the disc? Who will have to download a big patch?
You're making a slightly different, but equally interesting point, IMO.
What you are cluing into is that we are reaching a point of diminishing returns on Moore's Law, whereby to extract a significant (or worthy) advance in performance, you're not going to rely on silicon improvements alone anymore. From the "modern age" of gaming between PS1 Gen and PS4 Gen, the silicon advancements alone were astronomical. The PS1 CPU was on a 500nm process. The PS4 Pro is on a 16nm process. Next gen looks to be targeting 7nm. Combine that with an amazing run on improvements in rendering techniques, plus all the integrated "secret sauce" rendering that was being built natively into the silicon. There were also other market factors (desktop rendering, growth of CD/DVD/BD, massive advancements in storage density), all combined to create a golden age of graphics computing.
It was enough to rely on process improvements between generational shifts that you could mostly count on those advancements to not only bring down costs (reducing price over time) but have a significant performance bump without the need to really increase die area. The run on console improvements (as a percentage of growth) on $200-$300 boxes from 1995 - 2013 may never be seen again.
Going forward, we have a slowing of Moore's First Law, the application of Moore's Second Law (or Rock's Law) becomes more apparent. It's going to get more costly to produce these upcoming nodes. And the market dynamics around storage and media are no longer in play. So not only are the things we come to associate with consoles fixed in cost, in some cases they are getting either stagnant or more expensive.
So you are completely right - one of 3 things has to happen.
1. You're going to see smaller and smaller % leaps in performance between generations if you want low cost and price drops.
2. The price of consoles are going to have to go up in order to provide meaningful differentiation. You will see less price drops over time. This is your "advancement tax"
3. Console manufacturers will have to eat more loss on HW to offset #1 and #2.
So your point is super valid. Inflation has nothing to do with it. If you're paying attention to technology, the console business is going to come under tremendous pressure as it becomes harder and harder to deliver huge leaps in performance every few years at price points that were seen has historically successful.
I don't paint this as doom-and-gloom for consoles. Only that the traditional boundaries are going to have to be broken and accepted at some point.
Fun times ahead!
But if you take ??? and add in !!!!, divide the result by @@@ you'll ended up with ?!?!@ worth of power.
Or he could just put what his sources told him in other forums to circumvent verification at Era.
We know that this works.
Thought that most measurements put the Xsx at a little over 400mm2.
Most of that stuff was just a confluence of bad luck, had the same lead-free solder/board seperation problem that gen, just not as high as the 360.
This is a good breakdown of how I think about it as well. Thanks for that.
This one?
https://pastebin.com/UAbUBpMk
Any reliable rumors about early MS devkits using nvidia? By June a lot of devs should've been getting the updated PS5 devkits, but this post doesn't specify which one was still on Vega. I have no idea what is known about Navi 21 LITE these days.
395 is probably within the margin of error for those measurements. Probably doesn't mean anything though.
If Osiris wants to have a good time here, it is better he says nothing. Every "insider" that gave even a hint of what they know behind the scenes regretted it. Reiner hounded by fanboys on Twitter. Jason got sick of sites that lurk this thread for easy news hits taking his every word and using it as a big, attention grabbing headline. Klee suffering for weeks of passive-aggressive sniping until he requested a ban.
Yeah but it was not Final
395mm2 was not final?
Well that does bring up the usual pastebin question, how do they know how big the APU is? Is it the one in the devkit, and if so, how did they find out? I promise they're not taking it apart, and if they did it's a Vega/Ryzen part or discrete GPU. If it's the one in the spec sheet, why is that in the spec sheet in the first place?
Sounds about right.
With scarlett not so far AFAIK. But I wouldn't be surprised, since they also used nvidia cards on xbox one devkits at E3 2013.
You would expect next gen console RT solutions to dramatically increase ray/Pixel over nvidia rtx2xxx ?
Someone on the old site gave a good point on the github leak
"
I don't take these Github tests as reliable because it lacks context.
But if you look into them you will see they only tested 18CUs and 36CUs that match the PS4 and PS4 Pro number of CUs.
So they only tested two modes: 18CUs and 36CUs with both original clock and up to 2Ghz.
The Github testes shows to you that the PS5 in PS4 Pro mode can reach 9.2TFs in a stress test.
The clock up to 2Ghz is probably a stress test... no Navi based chip will run at 2Ghz in a comercial product... you can forget about that.
Now we need to know how many CUs the PS5 chip will have? That the Github tests didn't give you."
"
Continuing...
Taking the Github tests in mind I can guess Sony will have 2 options for each BC mode.
PS4 BC Normal Mode: 18CUs @ 800Mhz
PS4 BC Boost Mode: 18CUs @ Up to 2000Mhz (they will choose a clock they thing is comfortable)
PS4 Pro BC Normal Mode: 36CUs @ 911Mhz
PS4 Pro BC Boost Mode: 36CUs @ Up to 2000Mhz (they will choose a clock they thing is comfortable)
2000Mhz looks to me stress test but I can see something between 1600 and 1900Mhz.
We need a test with all PS5 CUs active to start to guess how much power the PS5 will have.
That the Github doesn't give you."
"
The tests are about PS4 / PS4 BC modes so Ray-tracking is disabled.
There is no tests from what I know in PS5 Mode with full CUs and ray-tracking enabled."
"
I don't take these Github tests as reliable because it lacks context.
But if you look into them you will see they only tested 18CUs and 36CUs that match the PS4 and PS4 Pro number of CUs.
So they only tested two modes: 18CUs and 36CUs with both original clock and up to 2Ghz.
The Github testes shows to you that the PS5 in PS4 Pro mode can reach 9.2TFs in a stress test.
The clock up to 2Ghz is probably a stress test... no Navi based chip will run at 2Ghz in a comercial product... you can forget about that.
Now we need to know how many CUs the PS5 chip will have? That the Github tests didn't give you."
"
Continuing...
Taking the Github tests in mind I can guess Sony will have 2 options for each BC mode.
PS4 BC Normal Mode: 18CUs @ 800Mhz
PS4 BC Boost Mode: 18CUs @ Up to 2000Mhz (they will choose a clock they thing is comfortable)
PS4 Pro BC Normal Mode: 36CUs @ 911Mhz
PS4 Pro BC Boost Mode: 36CUs @ Up to 2000Mhz (they will choose a clock they thing is comfortable)
2000Mhz looks to me stress test but I can see something between 1600 and 1900Mhz.
We need a test with all PS5 CUs active to start to guess how much power the PS5 will have.
That the Github doesn't give you."
"
The tests are about PS4 / PS4 BC modes so Ray-tracking is disabled.
There is no tests from what I know in PS5 Mode with full CUs and ray-tracking enabled."
I think this is because people can't just wrap their heads around how different things are now.
I tend to look at the APU as its own component. With a number of fixed agreed-upon parameters. Like PCIe lanes, PCIe gen, memory bus width, CPU cache..etc. But the Actualcompute components will be in lux and have multiple revisions. They do not need an 18month lead time anymore to get hardware finalized, no more than laptop/phone makers need any me to get the latest ad best new chip into their systems.
Unless people here really think that it took around 18 months to finalize the PS4pro APU which was made on a fab process that just hit the market a few months before it did. The only reason these hips would be finalized before June this year is because you have to get final dev kits out to devs.
I do, as well.
Remember the Xbox Scarlett clip at E3? They say "nextgen raytracing", which to me implies that Turing raytracing is current gen.
So I believe it will be much better than current nvidia's. But then nvidia will highly increase their "GigaRays" with Ampere.
IMHO, this first raytracing generation is completely experimental. And the moore law might not work anymore with current processor cores, but raytracing will most likely enjoy it again.
Last edited:
I have no special knowledge here and would love to be wrong, but id be shocked if nextgen consoles have anywhere near the RT of the 2080. its highly expensive new technology, and I'd expect the few glimpses of nextgen games we've seen would be highlighting rt a lot more if it was this amazing killer app.
Silicon has a few more node rolls, and a major topology change with GAAFET, but the future beyond that is more than silicon.
If i had to make an educated guess I doubt any console would be $600. Even if they clocked in at around that price you have to eat $100 and make it up on the back end. Live and Plus subs and software.
- Status