• 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.

Next-Gen PS5 & XSX |OT| Console tEch threaD

Status
Not open for further replies.

Loxus

Member
I saw someone post this on one of Rogame Tweets.


"XSX = 360mm2. So XSX is 360/308=~17% bigger. Has 56/40=40% more CUs. Hmmm "
"XSX=15B transistors, PS5 =15*308/360=12.8B, but should be ~=40/56*15=10.7B (has 2 less MC too). "
"So it has 2B more than it should. 32MB of 6T SRAM = 32Mbit*6*8bits=1.5B transistors."

To summaries:
-XBSX has 40% more CUs, but is only 17% bigger.
-PS5 has 12.8B transistors, but should have 10.7B based on 40% less CUs than XBSX.
-Probably has 32MB+ cache more than XBSX.

Speculation Time
What do you guys think?
 
Last edited:

FunkMiller

Gold Member
4+ weeks away to the slaughter of new innocents as they start up Demon souls for the first time thinking, wow this is going to be so Fun lol. I love killing monsters :messenger_heart:

The #DemonsSouls hashtag on Twitter for the whole of that first week is going to be fucking glorious. The amount of piss boiling is going to be stratospheric 😋👹 I essentially just want to rush through the whole thing fast enough to be levelled for some lovely, lovely invasions...
 
Last edited:

sircaw

Banned
I started my Soulsborne adventures with Bloodborne, and worked backwards.. never getting to Demon's Souls.. so this is gonna be goooooooooood

Same with me brother, when i do eventually get to play Demons souls, i hope in some ways it can beat Bloodborne in how it felt to me as a game, but i doubt it.

I liked the other souls games but nothing has come close to BB. "lollipop_disappointed:
 
Last edited:

sircaw

Banned
I saw someone post this on one of Rogame Tweets.


"XSX = 360mm2. So XSX is 360/308=~17% bigger. Has 56/40=40% more CUs. Hmmm "
"XSX=15B transistors, PS5 =15*308/360=12.8B, but should be ~=40/56*15=10.7B (has 2 less MC too). "
"So it has 2B more than it should. 32MB of 6T SRAM = 32Mbit*6*8bits=1.5B transistors."

To summaries:
-XBSX has 40% more CUs, but is only 17% bigger.
-PS5 has 12.8B transistors, but should have 10.7B based on 40% less CUs than XBSX.
-Probably has 32MB+ cache more than XBSX.

Speculation Time
What do you guys think?


I am afraid to say anything in case i get banned. "lollipop_disappointed:

Wheres my infinity cache, geordiemp geordiemp help.
 

SlimySnake

Flashless at the Golden Globes
I saw someone post this on one of Rogame Tweets.


"XSX = 360mm2. So XSX is 360/308=~17% bigger. Has 56/40=40% more CUs. Hmmm "
"XSX=15B transistors, PS5 =15*308/360=12.8B, but should be ~=40/56*15=10.7B (has 2 less MC too). "
"So it has 2B more than it should. 32MB of 6T SRAM = 32Mbit*6*8bits=1.5B transistors."

To summaries:
-XBSX has 40% more CUs, but is only 17% bigger.
-PS5 has 12.8B transistors, but should have 10.7B based on 40% less CUs than XBSX.
-Probably has 32MB+ cache more than XBSX.

Speculation Time
What do you guys think?

You are forgetting that there is a cpu and io block in there. Ps5 is based on the 10.7 billion 5700 xt. The rest is the Zen 2 and io block.
 

geordiemp

Member
I saw someone post this on one of Rogame Tweets.


"XSX = 360mm2. So XSX is 360/308=~17% bigger. Has 56/40=40% more CUs. Hmmm "
"XSX=15B transistors, PS5 =15*308/360=12.8B, but should be ~=40/56*15=10.7B (has 2 less MC too). "
"So it has 2B more than it should. 32MB of 6T SRAM = 32Mbit*6*8bits=1.5B transistors."

To summaries:
-XBSX has 40% more CUs, but is only 17% bigger.
-PS5 has 12.8B transistors, but should have 10.7B based on 40% less CUs than XBSX.
-Probably has 32MB+ cache more than XBSX.

Speculation Time
What do you guys think?


Ps5 is not going to have 128 MB of L2 like an 80 CU PC part, so max half at 64 MB, but its a console so 32 MB would be a good bet....

You are forgetting that there is a cpu and io block in there. Ps5 is based on the 10.7 billion 5700 xt. The rest is the Zen 2 and io block.
Thats RDNA1 and GPU, Rogame is doing a fair esstimate. Both XSX and ps5 have Audio silicon so...Ps5 has more IO but XSX has more server stuff in there so......its fair guess IMO
 
Last edited:
I personally did not want another tear-down from them as i have watched about 10 already from other people already.

But saying that, this breakdown from Sony has a gazillion views already, i mean i find it extremely interesting that one of the premier video gaming channels would not want a slice of that action.

Are they in such a good financial position now that they, digital Foundry can really turn their nose up at potentially the most views video on their site ever, i am sorry that sounds Strange, please tell me that sounds strange to you?>

so what if its been covered 100 times, 100% guaranteed it would of brought more views than that assassin creed video, did we really need, what was it 14 mins of talking about assassins creed hitting 60 fps.

And i get Richard was working on a video or what ever, but surely this is one you drop all tools for and run too, just look at the Views.

very very strange, at least i think so.

They have multiple people, right? Is the Alex dude in charge? If it's lucrative, maybe one of the others is shilling on tape as we type.
 

geordiemp

Member
I am afraid to say anything in case i get banned. "lollipop_disappointed:

Wheres my infinity cache, geordiemp geordiemp help.

Rogame is estimating, based on size which is hard to do from pics and big estimates, there could be infinity cache, 32 MB would be a good estimate.

Dont know why your afraid, its a speculation thread and others have leaked this (RGT) and Rogame now saying similar....

Dont know how people can stare at pixels and measure up so accurately lol
 
Last edited:
I saw someone post this on one of Rogame Tweets.


"XSX = 360mm2. So XSX is 360/308=~17% bigger. Has 56/40=40% more CUs. Hmmm "
"XSX=15B transistors, PS5 =15*308/360=12.8B, but should be ~=40/56*15=10.7B (has 2 less MC too). "
"So it has 2B more than it should. 32MB of 6T SRAM = 32Mbit*6*8bits=1.5B transistors."

To summaries:
-XBSX has 40% more CUs, but is only 17% bigger.
-PS5 has 12.8B transistors, but should have 10.7B based on 40% less CUs than XBSX.
-Probably has 32MB+ cache more than XBSX.

Speculation Time
What do you guys think?


It's hard to work the math like that. 40% of the chips are probably near identical. If you want to scale transistors based on CUs, you'd have to know what percentage of the XSX transistors are for CUs and then compare that to the percentage of the Sony chip used for the same.

If this math held up, the XSS chip would only be a little more than a third the size of the XSX chip, where in reality it is more than half the size. This is because a lot of the components remain the same (the CPU and the parts of the GPU that are not scaled up for XSX).
 
Last edited:

sircaw

Banned
They have multiple people, right? Is the Alex dude in charge? If it's lucrative, maybe one of the others is shilling on tape as we type.

i am sure they will probably be putting in a late video in the next couple of days, it will be interesting to see how many views it gets vs the assassins creed one.
 

sircaw

Banned
Rogame is estimating, based on size which is hard to do from pics and big estimates, there could be infinity cache, 32 MB would be a good estimate.

Dont know why your afraid, its a speculation thread and others have leaked this (RGT) and Rogame now saying similar....

Dont know how people can stare at pixels and measure up so accurately lol

Sorry for the stupid question is that large for a cache size, 32 Mb, the only cache sizes i know are for pcs and they are tiny if i recall.
 

geordiemp

Member
Sorry for the stupid question is that large for a cache size, 32 Mb, the only cache sizes i know are for pcs and they are tiny if i recall.

Yeah 4 MB or 5 MB for XSX, so 32 MB would be massive, thats almost 800 %.

NXGamer NXGamer has teased this as well. I have no idea.

Looking at the numbers again, we wouldbe lucky to see 8 MB -12 MB as cache is just too big area.
 
Last edited:

Aceofspades

Banned
It's hard to work the math like that. 40% of the chips are probably near identical. If you want to scale transistors based on CUs, you'd have to know what percentage of the XSX transistors are for CUs and then compare that to the percentage of the Sony chip used for the same.

If this math held up, the XSS chip would only be a little more than a third the size of the XSX chip, where in reality it is more than half the size. This is because a lot of the components remain the same (the CPU and the parts of the GPU that are not scaled up for XSX).

I think XSX GPU portion occupy 47% of the SoC. I thought I read that somewhere after Hot chip presentation

Edit: the GPU portion of XSX is 47% of the 360mm APU, so ~170mm for GPU, that leaves 190mm for everything else. Using same structure PS5 GPU would be 40CUs or around 120mm, assuming everything else is similar to XSX we end up with 311mm SoC so pretty close to this estimate.

In conclusion, I don't expect increased caches on PS5 vs XSX
 
Last edited:

Alex Scott

Member
Like Alex explained, Richard is only human and has x-amount of time per day. He was working on another video.
I believe DF should release a video on PS5 teardown when they have time to do so , so they can do it's justice. I don't think people care if DF didn't release a video on PS5 teardown on time. It was that DF member said PS5 teardown was unimpressive. This made some people mad and rightly so. If they get any backlash from this, it is deserving.
 

JTCx

Member
1280x720.jpg
LMAO with the XSX on its side you can see it just fucking looks off.
 

Andodalf

Banned
I believe DF should release a video on PS5 teardown when they have time to do so , so they can do it's justice. I don't think people care if DF didn't release a video on PS5 teardown on time. It was that DF member said PS5 teardown was unimpressive. This made some people mad and rightly so. If they get any backlash from this, it is deserving.

Wait, so they deserve backlash for not praising Sony? Literally just saying a video was dull is bad? Man, It really is give Sony a blowjob or bust for you guys, huh? There can be no middle ground.
 
I saw someone post this on one of Rogame Tweets.


"XSX = 360mm2. So XSX is 360/308=~17% bigger. Has 56/40=40% more CUs. Hmmm "
"XSX=15B transistors, PS5 =15*308/360=12.8B, but should be ~=40/56*15=10.7B (has 2 less MC too). "
"So it has 2B more than it should. 32MB of 6T SRAM = 32Mbit*6*8bits=1.5B transistors."

To summaries:
-XBSX has 40% more CUs, but is only 17% bigger.
-PS5 has 12.8B transistors, but should have 10.7B based on 40% less CUs than XBSX.
-Probably has 32MB+ cache more than XBSX.

Speculation Time
What do you guys think?

Are we speculating that the PS5 has a version of AMD rumored "Infinity cache"?
 

TLZ

Banned
Typical DF BS, now one of the most infamous games in history, AC Unity, running at 60fps is somehow interesting. Seems too dry and they're trying to make shit smell good. Glad more people are getting it and calling them out.

No freebies from Sony. :lollipop_tears_of_joy:

tenor.gif


EDIT: Wow, all that praise about AC Unity and it was 900p@60fps!

98E47928967144C577388AAB74A1117837CD1374
A whole video about how the XSX can run a 900p game with uncapped fps at max 60fps. Waaaow. So interesting.......







:messenger_neutral:
 
I saw someone post this on one of Rogame Tweets.


"XSX = 360mm2. So XSX is 360/308=~17% bigger. Has 56/40=40% more CUs. Hmmm "
"XSX=15B transistors, PS5 =15*308/360=12.8B, but should be ~=40/56*15=10.7B (has 2 less MC too). "
"So it has 2B more than it should. 32MB of 6T SRAM = 32Mbit*6*8bits=1.5B transistors."

To summaries:
-XBSX has 40% more CUs, but is only 17% bigger.
-PS5 has 12.8B transistors, but should have 10.7B based on 40% less CUs than XBSX.
-Probably has 32MB+ cache more than XBSX.

Speculation Time
What do you guys think?


I'm curious as to how much dosce the I/0 complex takes up. It has alot of stuff in it like the co processors and the SRAM.
 

jonnyp

Member
I saw someone post this on one of Rogame Tweets.


"XSX = 360mm2. So XSX is 360/308=~17% bigger. Has 56/40=40% more CUs. Hmmm "
"XSX=15B transistors, PS5 =15*308/360=12.8B, but should be ~=40/56*15=10.7B (has 2 less MC too). "
"So it has 2B more than it should. 32MB of 6T SRAM = 32Mbit*6*8bits=1.5B transistors."

To summaries:
-XBSX has 40% more CUs, but is only 17% bigger.
-PS5 has 12.8B transistors, but should have 10.7B based on 40% less CUs than XBSX.
-Probably has 32MB+ cache more than XBSX.

Speculation Time
What do you guys think?


He doesn't seem to account for the IO complex and cache scrubbers that will occupy additional die space.
 

Gamerguy84

Member
Eh whatever. DF was working on something else at the time. That's what was going on. They were never given a console to teardown. Sony does their own teardown.

As for DF yes they are bias. I don't consider this a thing though. Alex is Alex.
 
I saw someone post this on one of Rogame Tweets.


"XSX = 360mm2. So XSX is 360/308=~17% bigger. Has 56/40=40% more CUs. Hmmm "
"XSX=15B transistors, PS5 =15*308/360=12.8B, but should be ~=40/56*15=10.7B (has 2 less MC too). "
"So it has 2B more than it should. 32MB of 6T SRAM = 32Mbit*6*8bits=1.5B transistors."

To summaries:
-XBSX has 40% more CUs, but is only 17% bigger.
-PS5 has 12.8B transistors, but should have 10.7B based on 40% less CUs than XBSX.
-Probably has 32MB+ cache more than XBSX.

Speculation Time
What do you guys think?


The relative size difference is because stuff like the GDDR6 memory PHYs are quite large. The G6 PHYs in say for example Navi 10 (5700/XT) take up almost 100mm2 of the 250mm2 total die area. The actual CU's are quite small:
EPJuORtUUAAhx-j
Doing some napkin math on the die shot of Navi 14 gets me about 7mm2 for a DCU, which are 2 CU's conjoined together into a workgroup. Feel free to correct me on those numbers, its just an estimate. Navi 14 is 158mm2 for reference. We know that the RT on RDNA2 has a negligible area cost, so the CU's/DCUs won't be much bigger than Navi 10/14's.

Oberon has 20 DCU's - 7*20mm2 = 140 mm2
Arden has 28 DCU's - 7*28 = 196 mm2.

There's the 40% difference.

Let us not forget the other aspects of a GPU outside of the compute units. There is the render backends, i.e the ROPs. The ROPs we know are configured based on the number of Shader Engines (each contain 2 Shader Arrays) there are. Both Arden and Oberon have 2 Shader Engines (4 Shader Arrays), so both contain 64 ROPs. These will be pretty much exactly the same size for both consoles. Then there is the central GPU command processor, which will likely be pretty much exactly the same for both consoles. Given the slightly different memory config, Arden likely has 25% more L2 cache than Oberon 5 MB vs 4MB. This is an assumption based on the fact that Navi 10 has 4MB L2 cache and Microsoft has told us Arden has 5MB. The L1 Caches are likely to be pretty much the same for Arden and Oberon and thus take up the same amount of space.

Of course both APU's also contain 8 Renoir Zen 2 cores and their associated caches. As both are likely to be virtually identical, they're both the same size, at around 30-40mm2. I must reiterate, this will take up the same die area for both chips.

Then there are the massive GDDR6 PHYs, of which Oberon has 8*32-bit PHYs and Arden has 10*32-bit Phys, and their associated memory controllers. 10 PHYS will only be 25% larger than 8 PHYs. Memory controllers will be marginally larger for Arden vs Oberon.

Then there will be other miscellaneous stuff on each die - the display engine, encoders etc. Each of Oberon and Arden may also contain additional fixed function/custom units which will take up an unspecified amount of die area.

Here's a die diagram of Arden graciously provided by Microsoft:

Series_X_die_breakdown.png


Each black box I've drawn here is a DCU (someone smarter than I can probably make much more accurate annotations). You have 28 boxes, which amounts to 56CU.
We know Oberon (probably) has 40CUs. We know for certainty that the main differences between Arden and Oberon are the number of CU's and the number of Memory PHYs. So in essence, you just need to remove what I've highlighted in red, and boom, you have an approximation of the PS5's APU.

I shouldn't need to say it, but this is a gross oversimplification. In reality there will be additional differences, as I've briefly touched upon above. This is just intended to be a visual aid.
This image should be a good enough representation of what a 56CU vs 40CU actually looks like on a die. The differences is ultimately quite small, because there is a hell of a lot of uncore stuff that is present on the die.
40% more shaders does not at all imply 40% larger. There areas of the APU that will be pretty similar between each console.
 
Last edited:
I saw someone post this on one of Rogame Tweets.


"XSX = 360mm2. So XSX is 360/308=~17% bigger. Has 56/40=40% more CUs. Hmmm "
"XSX=15B transistors, PS5 =15*308/360=12.8B, but should be ~=40/56*15=10.7B (has 2 less MC too). "
"So it has 2B more than it should. 32MB of 6T SRAM = 32Mbit*6*8bits=1.5B transistors."

To summaries:
-XBSX has 40% more CUs, but is only 17% bigger.
-PS5 has 12.8B transistors, but should have 10.7B based on 40% less CUs than XBSX.
-Probably has 32MB+ cache more than XBSX.

Speculation Time
What do you guys think?

So that's why John from DF is saying people will get impressed by PS5 performance. SECRET. MOTHAFUCKING. SAUCE.
 

MHubert

Member
What a fucking take, Alex. Very nice.

First of all, I'll be the first to say that Unity running at 60fps is amazing news, as I thoroughly enjoyed the setting of the game, but pretty much stopped playing because of the bugs and janky frames. So, Richard releasing that video? Amazing, kudos to him and I'm happy with that.

Now, the "rather uninteresting" PS5 video has more than 7 million views already, so, if a good reason to prioritize content is saying that that very same content is something "people actually care about", do the fucking math, Alex, mate. I don't keep count on how many Xbox videos they do versus PS videos, and I think looking at it like that is not necessarily the right way, but, the disdain with which he dismissed the idea of providing content to people that are so obviously, with statistics in hand, eager to consume it and know just a bit more about a platform, goes to show his equity and how profesional he is.

Seriously, is not even the fact that they're not releasing a video about it, I think that's perfectly fine, it's the fucking disrespect towards people excited for PS5. Some fans are more worthy of content than others, apparently. His attitude and words are saying exactly that.

Disgraceful.

And I stated many times, I have no issue with Digital Foundry as a whole, I watch their videos and Richard, John and Thomas, whatever their biases, do good work and are often informative and nice to hear. But this guy, seriously, I don't understand his fucking problem.


Edit: And besides what I already stated, somebody correct me if I'm wrong, but I believe having liquid metal as a stock cooling solution on a piece of hardware mass produced on the volume PS5 is going to be produced is a first, so, as tech guys, are they not interested in something like that? I find it odd. But that's beside the point, as I said, my main issue is not with DF not releasing a video, is the disdain on Alex's words. Everyone, Xbox fans or PS fans, deserve respect, and this guy clearly seems uninterested in such a thing.
Like many of you, I too, am possessed by the Sony boogaloo - but I am not so blinded as to not see that what Alex said makes perfect sense. Right now DF has an XBsX and I would honestly be disappointed in them not focusing 100% on the next gen machine they actually have in their hands. What is it, that you would like them to tell you, exactly?
 
Last edited:

geordiemp

Member
The relative size difference is because stuff like the GDDR6 memory PHYs are quite large. The G6 PHYs in say for example Navi 10 (5700/XT) take up almost 100mm2 of the 250mm2 total die area. The actual CU's are quite small:
EPJuORtUUAAhx-j
Doing some napkin math on the die shot of Navi 14 gets me about 7mm2 for a DCU, which are 2 CU's conjoined together into a workgroup. Feel free to correct me on those numbers, its just an estimate. Navi 14 is 158mm2 for reference. We know that the RT on RDNA2 has a negligible area cost, so the CU's/DCUs won't be much bigger than Navi 10/14's.

Oberon has 20 DCU's - 7*20mm2 = 140 mm2
Arden has 28 DCU's - 7*28 = 196 mm2.

There's the 40% difference.

Let us not forget the other aspects of a GPU outside of the compute units. There is the render backends, i.e the ROPs. The ROPs we know are configured based on the number of Shader Engines (each contain 2 Shader Arrays) there are. Both Arden and Oberon have 2 Shader Engines (4 Shader Arrays), so both contain 64 ROPs. These will be pretty much exactly the same size for both consoles. Then there is the central GPU command processor, which will likely be pretty much exactly the same for both consoles. Given the slightly different memory config, Arden likely has 25% more L2 cache than Oberon 5 MB vs 4MB. This is an assumption based on the fact that Navi 10 has 4MB L2 cache and Microsoft has told us Arden has 5MB. The L1 Caches are likely to be pretty much the same for Arden and Oberon and thus take up the same amount of space.

Of course both APU's also contain 8 Renoir Zen 2 cores and their associated caches. As both are likely to be virtually identical, they're both the same size, at around 30-40mm2. I must reiterate, this will take up the same die area for both chips.

Then there are the massive GDDR6 PHYs, of which Oberon has 8*32-bit PHYs and Arden has 10*32-bit Phys, and their associated memory controllers. 10 PHYS will only be 25% larger than 8 PHYs. Memory controllers will be marginally larger for Arden vs Oberon.

Then there will be other miscellaneous stuff on each die - the display engine, encoders etc. Each of Oberon and Arden may also contain additional fixed function/custom units which will take up an unspecified amount of die area.

Here's a die diagram of Arden graciously provided by Microsoft:

Series_X_die_breakdown.png


Each black box I've drawn here is a DCU (someone smarter than I can probably make much more accurate annotations). You have 28 boxes, which amounts to 56CU.
We know Oberon (probably) has 40CUs. We know for certainty that the main differences between Arden and Oberon are the number of CU's and the number of Memory PHYs. So in essence, you just need to remove what I've highlighted in red, and boom, you have an approximation of the PS5's APU.

I shouldn't need to say it, but this is a gross oversimplification. In reality there will be additional differences, as I've briefly touched upon above. This is just intended to be a visual aid.
This image should be a good enough representation of what a 56CU vs 40CU actually looks like on a die. The differences is ultimately quite small, because there is a hell of a lot of uncore stuff that is present on the die.
40% more shaders does not at all imply 40% larger. There areas of the APU that will be pretty similar between each console.

Thanks for drawing it out - we could just look at the 4 MB L3 cache on CPU cluster, there is no way there is an additional 8 of them, 32 MB would be too much.
 
Status
Not open for further replies.
Top Bottom