• 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.

MastaKiiLA

Member
HLFU4fH.png


DEzNB8a.png


Did AMD add an RA to free up the 4 TMUs? Does that give RDNA2 CUs 4 texture + 1 ray op per clock?

Likewise, is XSX's RDNA implementation limited to either texture or ray ops, or can it be a mix of 4/0, 3/1, or 2/2 op splits per clock?

I'm wondering if the PS5 CUs have separated the ray accelerator from the TMUs, or if it repurposes TMUs as RAs.

Does this mean that the XSX can process more rays per clock than the PC part, but at the loss of texturing capacity? I didn't see anything about how many rays ops an be processed per clock for the PC cards, and just assumed it was 1, though it might well be 4 per CU.
 

LucidFlux

Member
So it isn't possible to have lower amounts of Infinity Cache? Like 32MB, 16 or even 8MB?

Also people are sleeping on "Cache Scrubbers" that Cerny specifically mentioned as exclusive to PS5 in his presentation.

Based on the APU size estimates there just doesn't seem to be room for anything more than maybe 8MB if the estimates were low.

Regardless, I do however think that the cache scrubbers were Cerny's answer to not being able to fit the huge 128MB cache AMD had planned for RDNA 2.

So whether or not PS5 has that additional cache, their L0/1/2 should all be much more efficient with the addition of those scrubbers.



Cache scrubber in action.
 

jamwest24

Member
I felt a bit stupid after noticing this (mainly because I didn't read the entire article from the start).

It’s hilarious to me that people are falling for all this stuff yet again.. Sony has never used DX12 or any DX... They have their own APIs that are actually much more efficient and get closer to their specific hardware than the much broader/generic APIs of DX12 that are also meant for PC.
 

FunkMiller

Gold Member
This.. ^ Anyone who has worked in PR knows how important those specific words are....And just like literally everything else we’ve seen so far from Microsoft, they’ve been all talk and PR spin and no substance.. (besides backwards compatibility and faux HDR that my 3 year old TV does).

Er, excuse me. Are you fucking suggesting that people in PR spend their entire time thinking of ways to make things sound far better than they actually are? I've never been so offended!







you're absolutely right
 

onesvenus

Member
LOL., you think XSX has frequency adjustment down to each CU to acheive high clocks ? What ?

And then you think MS just turned it off because sustained sounds better, and turned down the frequency by 20-30 % lol

Really ? Thats being silly.

What about the CU looking like a 5700 and nothing like a 6800 ? Maybe we all cannot see ?

XSX has more CU but they are not same as PC parts or ps5.
Ok, so they are lying when they say they are using the full rdna2 architecture and they waited for rdna2 to be developed to choose RDNA1 CUs, that really makes sense, you are right
 
Oh shit! I'm late to the party! Hey guys....
tumblr_inline_n00bb6Rg7y1rgzq2t.gif



Well, well, well...ain't this place a bitch.

I guess PS5 doesn't have infinity cache? Or maybe we still don't know yet? I dunno. Whatever. Seems like the usual assholes on Twitter are arguing over which console is "true" RDNA2. At some point they (certain Xbox fanboys on Twitter in particular) will realize RDNA2 is a foundation and how it's built upon matters most.
 

Zathalus

Member
HLFU4fH.png


DEzNB8a.png



Did AMD add an RA to free up the 4 TMUs? Does that give RDNA2 CUs 4 texture + 1 ray op per clock?

Likewise, is XSX's RDNA implementation limited to either texture or ray ops, or can it be a mix of 4/0, 3/1, or 2/2 op splits per clock?

I'm wondering if the PS5 CUs have separated the ray accelerator from the TMUs, or if it repurposes TMUs as RAs.

Does this mean that the XSX can process more rays per clock than the PC part, but at the loss of texturing capacity? I didn't see anything about how many rays ops an be processed per clock for the PC cards, and just assumed it was 1, though it might well be 4 per CU.
Both are the exact same thing. The Amd diagram shows 1 ray accelerator per compute unit, and the Xbox diagram shows 1 ray accelerator per compute unit.
 

Aceofspades

Banned
Oh shit! I'm late to the party! Hey guys....
tumblr_inline_n00bb6Rg7y1rgzq2t.gif



Well, well, well...ain't this place a bitch.

I guess PS5 doesn't have infinity cache? Or maybe we still don't know yet? I dunno. Whatever. Seems like the usual assholes on Twitter are arguing over which console is "true" RDNA2. At some point they (certain Xbox fanboys on Twitter in particular) will realize RDNA2 is a foundation and how it's built upon matters most.

Those fanbois are hanging on that AMD slide where they say they support DirextXU . Followed by a clever tweet by Xbox....rational people would know that PS5 thankfully won't be supporting DirectX api..

Its just Xbox exec playing on emotions and promises hoping to slow down PS5 massive train lol
 

LucidFlux

Member
What does clock gating mean? I know it helps them achieve over 2 GHz sustained clk speeds even on an 80 CU GPU. But I don't get the 'gating' part.

Essentially It refers to the logic gates of which all processors are built on. The clock controls the speed of all transistor gates to keep the chip in sync and all gates at the same speed.

The pervasive-fine grain clock gating to me could mean selective groups of gates receiving different clock rates depending on workload. Or in a broader sense the GPU could select specific CUs to be clocked higher/lower for a specific task to more efficiently control power usage.
 

timmyp53

Member
Not sure if this was pointed out yet but there looks to be a potential patent with PS5 related implications for Auto-HDR:
New Sony Patent Hints At PS5 Auto HDR Feature - PlayStation Universe (psu.com)
Related patents appear to pertain a display controller, display device, content playback device, and various devices such as a game devices, and other devices containing it.

This would probably need some further digging to understand the application:

PS5-Auto-HDR-2.png

xfl1Doy.jpg

stw8elgFSoSedVnT-R4hSauWx1NyBGJRapvH3auSGoItaRAY-TS7vskmcgt35Hwm0DM-FBLTZ8dBk8iU92SD3m2uDY1O7eT-LVvDQhBE8oY1NCTU0AMwitWx21ScrsBrLoFgqJumcRmobXerMZQjmxFjNdEZglPA-oUNaLJ8la4
 

THE:MILKMAN

Member
Essentially It refers to the logic gates of which all processors are built on. The clock controls the speed of all transistor gates to keep the chip in sync and all gates at the same speed.

The pervasive-fine grain clock gating to me could mean selective groups of gates receiving different clock rates depending on workload. Or in a broader sense the GPU could select specific CUs to be clocked higher/lower for a specific task to more efficiently control power usage.

Basically variable clocks....
 

Zathalus

Member
Nop Xbox diagram show ray accelerator use TMU for calculate. AMD diagram Show you can use RT accelerator & 4TMU at the same time.
202008180220211.jpg
The TMU is part of the compute unit, the AMD diagram makes no claims if a RA and TMUs can be used at the same time. Please point out where this is claimed.
 

MastaKiiLA

Member
Both are the exact same thing. The Amd diagram shows 1 ray accelerator per compute unit, and the Xbox diagram shows 1 ray accelerator per compute unit.
I don't think that's the case. XSX slide shows 4 texture or 4 ray ops per clock, which aligns with the 4 TMUs per CU. I'm just not sure if it's an all or nothing thing per clock. Maybe each CU can handle 1 ray op + 3 texture op per clock, or any other permutation of 4 ops. I'm not really clear on that. The slide from today appears to clearly separate the ray accelerator from the TMUs, as the TMUs are still visible as 4 texture filter units, and 4 texture mapping units, to the right of the ray accelerator.
 

geordiemp

Member
Ok, so they are lying when they say they are using the full rdna2 architecture and they waited for rdna2 to be developed to choose RDNA1 CUs, that really makes sense, you are right

Yes they are using full RDNA2 Architecture and features,.

But the XSX cUs are not RDNA2 from specs, die shots or hot chips block diagram. Is CU a feaure or a hardware item ?

Also both consoles do not have infinity cache.

Careful wording goes allong way.

if you telling me that XSX now has finesse freqency control and gating, then thats opposite of what MS ahve said for past year.

You cant have it both ways, is XSX variable high freqhency clocks or fixed ?


.
 

MarkMe2525

Member
I don't think that's the case. XSX slide shows 4 texture or 4 ray ops per clock, which aligns with the 4 TMUs per CU. I'm just not sure if it's an all or nothing thing per clock. Maybe each CU can handle 1 ray op + 3 texture op per clock, or any other permutation of 4 ops. I'm not really clear on that. The slide from today appears to clearly separate the ray accelerator from the TMUs, as the TMUs are still visible as 4 texture filter units, and 4 texture mapping units, to the right of the ray accelerator.
You're debating with someone who has a brain as their avatar, how can you compete? :)
 

ROPs - 96!!!!

Perhaps the 80 ROPs for Xbox Series X was right all along.

I debated this a couple of month ago and got called a fanboy and stupid for even suggesting the idea that Xbox Series X could have more than 64 ROPs.

Well now the RX 6800 has 60 CUs and 96 ROPs (cut down from the bigger cards - 80 and 72 CUs with 128 ROPs).
That's because it removes two arrays ( each 10 CUs) my friend. So 128 - 32=96.
XsX is 14 CU array. Thus still 64 ROPs.
 

LucidFlux

Member
Basically variable clocks....

Variable clocks but instead of the whole GPU shifting clocks up and down, it's different parts of the chip simultaneously running at different speeds.

Rudimentary example:
Say the frame time is 16.67 (60fps)

GPU task 1 could complete in only 4 ms with lots of idle time after, so the portion of the GPU tasked with those operations could down clock to say 1500mhz and still finish within the time it takes to render the frame.

GPU task 2 needs 16ms to complete work, these cu's stay at 2230mhz and finish work in time for the frame target.
 

THE:MILKMAN

Member
Variable clocks but instead of the whole GPU shifting clocks up and down, it's different parts of the chip simultaneously running at different speeds.

Rudimentary example:
Say the frame time is 16.67 (60fps)

GPU task 1 could complete in only 4 ms with lots of idle time after, so the portion of the GPU tasked with those operations could down clock to say 1500mhz and still finish within the time it takes to render the frame.

GPU task 2 needs 16ms to complete work, these cu's stay at 2230mhz and finish work in time for the frame target.

All very interesting possibilities...If only Sony would open up on the their implementations and customisiations. I guess they are too busy eating their popcorn right now though.
 

LucidFlux

Member
All very interesting possibilities...If only Sony would open up on the their implementations and customisiations. I guess they are too busy eating their popcorn right now though.

Well this is part of RDNA 2 so even if Sony doesn't mention anything it should be detailed in the RDNA 2 whitepaper from AMD.

It's certainly fascinating stuff. We've seen it in CPUs but afaik this is the first time a GPU will have parts of the chip running at different clock speeds simultaneously. At least that's how I'm interpreting it.
 
Last edited:

NEbeast

Member
Custom RDNA2 > RDNA2. MS clearly wanted to keep unity with PC,they are releasing games day on on there. PS5 clearly has the upper hand at this point,the games and RT at launch clearly show that.

I'm not sure why people are so excited, we have known xbox has VRS for months, we saw it in Halo. We just need to know Sonys custom solutions.
 

MarkMe2525

Member
Custom RDNA2 > RDNA2. MS clearly wanted to keep unity with PC,they are releasing games day on on there. PS5 clearly has the upper hand at this point,the games and RT at launch clearly show that.

I'm not sure why people are so excited, we have known xbox has VRS for months, we saw it in Halo. We just need to know Sonys custom solutions.
You can make that assumption, but that is all that it is. The only conclusion that can be drawn at this point is that Sony "clearly" has games further along than MS. Judging future capability based solely on a launch window lineup is futile.
 
Last edited:
Status
Not open for further replies.
Top Bottom