• 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.

R600

Banned
IO and memory bus is already accounted on the GPU die size
Its calculated for 64mm² (256 bit bus). For 384 its 32mm² more.

So system with 40CU, Zen2, RT and 256bit bus would be ~340mm².

For system with 384bit bus we are looking at ~370mm² without adding any new CUs and front end, so ~400mm2+ is safe bet.

Is this achievable or is PS4 die size (bigger then Pro) much more likely to be used on more expensive process and while packing SSD?

Remember, with 40CU Navi XT and 8GB of RAM we are already well over console TDP (225W, historical at that).

So with 8GB more you can add another 16W more. Zen2 is another ~30-35W. Which brings us to total of 225 + 16W (additional 8GB) + 30W Zen 2 = 271W at LEAST.

I think your math is of tbh.
 

R600

Banned
I'm expecting at least 15-20% difference with HBM2 space savings as the main factor. PS5 could fit at least 10 more CU because of it.
HBM2 for Vega was 185$ IIRC. That is 4x per GB vs GDDR6 (and less perf to boot).

They are not using it.
 

bitbydeath

Member
All can be fabricated, but Gonzalo type of leaks we get by no nonsense guys who dig through benchmark/codename databases.

Rumors like "OMG 450mm² die size with 14TF Navi and GTAVI launch title!!! Btw my dad works at Nintendo" are terrible. They are not even rumors. They fly in the face of everything we know about 7nm node, Zen2 and Navi die sizes and TDP.

The accounts only been around for a year and the reason it’s suggested as being PS5 related is at best grasping at straws.

Not to mention it’s way too early for anything like this to be benchmarked.

It just doesn’t make sense. Throw in the tests being run on Windows and it’s pretty much debunked.
 

R600

Banned
The accounts only been around for a year and the reason it’s suggested as being PS5 related is at best grasping at straws.

Not to mention it’s way too early for anything like this to be benchmarked.

It just doesn’t make sense. Throw in the tests being run on Windows and it’s pretty much debunked.
Mate, 99% of his tweets are one liners consisting of codenames for AMD/Intel/Nvidia chips. He is not insider for PS5 nor he claims to be.

Gonzalo is 8core 1.8GHZ Navi APU designated for console. Which one? As I already explained, codename says clearly PS5.
 

Aceofspades

Banned
Songoku, please re read my post and calculate everything again, you are missing the point. And point is :

PS4 die size = 348mm²
PS4 CPU + GPU = 70 + 212 (282mm²)
66mm² remaining is IO / 256 bit bus

PS5 die size = ???
PS5 CPU + GPU(40CU) = 75 + 251 + ~10RT (336mm²)
IO + 256 bit bus = 38mm² + 64mm² (102mm²)

So by my calculations ~ 438mm² for 40CU part if we follow PS4 die size dimensions.

Boy oh boy will some people turn out disappointed.

Ok wait, your prediction stated that Scarlet will have a 44Cu on a 380mm die size.
What changed now? Now you are saying 40Cus won't fit on 430mm die. Or is it some MS engineering wizardry BS?
 

luca_29_bg

Member
Not with all.

1) 13TF Navi would be gigantic
2) It would suck ~300W
3) It would be faster then 2080, perhaps matching 2080TI
4) No one would be able to know target specs and die sizes. This is not something that is provided to devs, you can be sure of it.

Not one thing about it makes any sense whtsoever.

So, someone from Era or Gaf went there, posted it and now enjoys people gushing over it.

Actually Reiner talked about target specs, known to developers :/
 

ethomaz

Banned
Updated die size estimates (for real this time):
75mm for CPU
81.6mm2 for 320 bit bus (20GB)
97.92mm2 for 384 bit bus (24GB)
46.66 mm2 for 4SEs (64ROPs, cache, etc.)
IO 20.72mm2
10.85mm2 GPU CC (CP, ACEs etc.)
156.48mm2 32DCUs with RT (64CUs)
Total:
391.31mm2 - 320 bit bus
407.63mm2 - 384 bit bus
4d900c.png
crisis-averted.jpg
I don’t know but it feel that here are something off with these estimates... in what they are basing to make the estimate?

Like in the size of what?
 

R600

Banned
Ok wait, your prediction stated that Scarlet will have a 44Cu on a 380mm die size.
What changed now? Now you are saying 40Cus won't fit on 430mm die. Or is it some MS engineering wizardry BS?
Sorry, was using IO/bus for 384 in addition to already used 256 bit one.

40CU/Zen²/RTHW/256bit ~ 340mm²
44CU/Zen²/RTHW/320bit ~ 380mm²

If MS goes with 14Gbps memory speed while Sony goes with 16Gbps, difference between BW will be 48GB/s, while 4CUs-8CUs more at lower clocks would result in ~0.6TF adv.
 

vpance

Member
HBM2 for Vega was 185$ IIRC. That is 4x per GB vs GDDR6 (and less perf to boot).

They are not using it.

Got to factor in PC card fat profit margins.

Going all in at Samsung fab 7nm+ with HBM Aquabolt. 16GB will give them 614GB/s only on 2 stacks with a 2048-bit bus.
 
AMD are often negatively compared to Nvidia.

It's not too much of a stretch to consider that a non AMD HW RT solution would have a smaller die footprint freeing up the space for more CUs.
 
Last edited:

ANIMAL1975

Member
GPU:

1.8Ghz
64CU

14TF Navi... ~ 18TF GCN, 10x PS4.
Are you predicting or you have Intel 😳

Join #Team16TF

Win hard or fall hard!
It's 16.8 team, stay focus!

How trustworthy are those Navi 5700 measurements?

Hype train de-railing. Only coding to the metal can save us now.
Ms already coded direct X to the metal on Scorpio, so i think Scarlett is a given.
Our hype train has bs stabilizers so is very improbable to derail... but lets say that it does (very unlikely), we got you covered _ it features a very advanced magnetic propulsion system (sideways and on top too!) so that we can still get to our destination in safety at a cruise controlled 14.2 miles per hour speed.

Anywhere between 11-13TF
For the 80CU Kaiju 14TF might be doable
But expect around ~12TF to not be disappointed
Goku, always expect the best for you and the Universe will provide
 

bitbydeath

Member
Mate, 99% of his tweets are one liners consisting of codenames for AMD/Intel/Nvidia chips. He is not insider for PS5 nor he claims to be.

Gonzalo is 8core 1.8GHZ Navi APU designated for console. Which one? As I already explained, codename says clearly PS5.

It doesn’t say clearly PS5.
It states Gonzalo and Ariel which could mean anything as they are not confirmed code names for PS5.
 

R600

Banned
RX 5700XT(9.75) < PS5 < next gen RDNA cards, that's easily pass 10Tflops we shouldn't compare 5700's TBP to PS5, because PS5 custom made silicon most likely draws less W at same performance than these cards.
Wasnt the case with Pitcairn (PS4) nor with Polaris (PS4 Pro).
 

R600

Banned
It doesn’t say clearly PS5.
It states Gonzalo and Ariel which could mean anything as they are not confirmed code names for PS5.
Check numbering behind ZG for Sony chips (PS4, Pro and now Gonzalo) and compare it to MS console nomenclature. Its very clearly different, while Ariel falls in line with other custom Sony PCI IDs in database.
 

bitbydeath

Member
Check numbering behind ZG for Sony chips (PS4, Pro and now Gonzalo) and compare it to MS console nomenclature. Its very clearly different, while Ariel falls in line with other custom Sony PCI IDs in database.

Why just compare it to MS, they’re not the only manufacturers out there?

Can you explain why this would be benchmarked so early and why it was run in Windows? Because they’re the two biggest things that debunk your theory.
 

Thedtrain

Member
Update: We might not be fucked boys
I am a roller coaster of emotions
 

R600

Banned
Why just compare it to MS, they’re not the only manufacturers out there?

Can you explain why this would be benchmarked so early and why it was run in Windows? Because they’re the two biggest things that debunk your theory.
Because there are non remaining bar these two? What where other high performance APUs AMD made that got into console and wasnt Sony, MS or Subor. Ok, Subor was meh considering it was 2018 console.

Mind you, we would probably know of console if its chip was already in QS status.
 

Darius87

Member
Wasnt the case with Pitcairn (PS4) nor with Polaris (PS4 Pro).
it was PS4 APU draw is rated at 75W vs equalent 7790HD 85 W it's not much around 12% less.
PRO is 150W at 16nm vs RX470 120W at 14nm so prety much same results i don't know if even that includes jaguars W in apu. this time i think it could be around 20% less W with ps5.
 
Reality Check Gentlemen!
Which Console is Much powerfull That's the only talk here?? Let us say even if MS Scarlett is More and More powerfull then PS5 if wont matter a Shiiiiiiiiit, Why?
Microsoft has only 343 Studio's which uses ingame engine The rest use Unreal Engine which Every 3 rate developer uses. So the games Will never show the Scarlett true Power Maybe 70 percent. Look at Epic games also using Unreal Engine. PS5 on the otherhand if even inferior I doubt thAt iT Will be equal in power. Will totally destroy MS Scarlett because PS5 games, the best games use Ingame engine look at Naughty dog Guerilla Studio's Sony own studios and more developers. Not thAt universally bland Unreal engine. So Scartlett may be this time more and More powerfull but we will never SEe then hardware actual Potential. Only 343 Studios. The Games are the ultimate Factor when showing the power of a Hardware and honestly Said ingame engines are better then MS 1000 developers using Unreal Engine While Sony first and second Party uses ingame engines. Remember The order 18.... Still better looking then Xbox Scarlett Halo
It is called in-house not ingame. Get it right before you start spewing out random nonsense.
 

CrustyBritches

Gold Member
wft? I thought the bet was a stock 5700XT (1950MHz)/RTX2070
4
They stated "opportunistic boost clock", so I'm guessing that's just a function of the silicon lottery and good cooling.

P.S.- If it ends up somewhere in between our predictions then we will call a truce.
 

R600

Banned
it was PS4 APU draw is rated at 75W vs equalent 7790HD 85 W it's not much around 12% less.
PRO is 150W at 16nm vs RX470 120W at 14nm so prety much same results i don't know if even that includes jaguars W in apu. this time i think it could be around 20% less W with ps5.
Nope, no it didnt. For start, PS4 would go up To 140W in gaming. 7850 was 175W from the wall, ~100W in games with 2CUs less and 8% higher clocks. Add Jaguar and 8GB of GDDR5 and surprise surprise, you get exactly ~140W.

RX470 on the other hand can be found in PS4Pro, but its heavily downclocked and undervolted, therefore slower, which brings its perf/watt up (RX470 comes overclocked OOB).

This is why I dont understand people talking about 2x5700, let alone XT. Because there is no magic decrease of TDP under load in consoles compared to PC.

This is TDP of last Sony/MS/Nintendo consoles.
History.png
 
Last edited:

Darius87

Member
Nope, no it didnt. For start, PS4 would go up To 140W in gaming. 7850 was 175W from the wall, ~100W in games with 2CUs less and 8% higher clocks. Add Jaguar and 8GB of GDDR5 and surprise surprise, you get exactly ~140W.

so to you it's ok to compare total ps4 W vs just 7850 gpu W + 2gb of ram W?

RX470 on the other hand can be found in PS4Pro, but its heavily downclocked and undervolted, therefore slower, which brings its perf/watt up (RX470 comes overclocked OOB).
again total pspro W vs rx470 gpu W + 4gb W and yea it's better then pro gpu but that's closest to it that i found, and 470 is 14nm vs pro 16nm.
 
Last edited:

Mass Shift

Member
5700XT is 251mm2 Zen2 75 mm2 add an extra 10 mm2 for RT
Total: 336mm2 on plain 7nm

7nm EUV its a 20% reduction->336*0.8 = 268.8 mm2
That's a very rough estimate, console chips are more optimized and waste less space

Your "prediction" doesnt fit 7nm euv

A 7nm 400mm2 die would be 320mm2 on 7nm EUV

Okay, while this sounds good but I think a couple things should be considered first.

#1. I believe that MS and Sony both have final silicon. So even if the 7nm EUV process is being used, it's not going to change what is already on the die. The process is worth waiting for as it's both cost effective and increases performance.

#2 MS has already shown an approximation of their APU in their Project Scarlett announcement, and it has been estimated to be 380mm2 - 400mm2.

I don't believe MS' APU is getting any smaller. I believe that the Scarlett APU we saw in that announcement is POST 7nm EUV, not a chip waiting for any further reductions. Otherwise MS would not have revealed it that way to us. So MS' APU is going to be large regardless, as it makes sense as they are claiming accelerated RT right out of the gate.

So I would exercise a bit of caution before being convinced that the 7nm EUV process will bring additions to what may already be final silicon.

Don't mean to sound like a killjoy but tempering our expectations is still probably the best way forward. At least until MS and Sony reveal otherwise.
 

LordOfChaos

Member
Ian Cutress, AnandTech: Given that Ray Tracing has been announced in the PS5 specs, can you say if that’s a specific Sony optimization or is that inherent in RDNA?

Lisa Su:
We certainly have done very specific optimizations for Sony. They are a very deep partner with us on semi-custom, and there are optimizations there. However we view ray tracing as a very important element across the portfolio. We will have ray tracing in a number of places. Will you look at that, you got me to say more about ray tracing!


 

pawel86ck

Banned
Guys if Gonzalo leak is legit, then PS5 results looks exactly like RX5700 XT performance estimations, and that's 9.75TF GPU already, 8TF Navi GPU wouldnt beat 12.5TF Vega 64. Also people have to keep in mind windows drivers for PS5 are probably not optimized to run 3dmark unlike for example vega64 drivers so that PS5 GPU should be even faster than results suggest (maybe even 11-12TF like SonGoku estimate).
 
Last edited:

R600

Banned
so to you it's ok to compare total ps4 W vs just 7850 gpu W + 2gb of ram W?


again total pspro W vs rx470 gpu W + 4gb W and yea it's better then pro gpu but that's closest to it that i found, and 470 is 14nm vs pro 16nm.
What do you mean? I am comparing PS4 vs PC GPU equivalent (7850) and adding additional RAM and CPU to prove that no, we wont be able to fit in 225W GPU in console because MS and Sony have magic talent for semicodnuctor design that AMD does not have.

Lets wait and see what Navi bring in weeks time and then see where it exactly is in terms of tdp.
 
Guys if Gonzalo leak is legit, then PS5 results looks exactly like RX5700 XT performance estimations, and that's 9.75TF GPU already, 8TF Navi GPU wouldnt beat 12.5TF Vega 64. Also people have to keep in mind windows drivers for PS5 are probably not optimized to run 3dmark unlike example vega64 drivers so that PS5 GPU should be even faster (maybe even 11-12TF like SonGoku estimate).
Exactly, not to mention, both Sony and Microsoft's designs need to be complete at this stage.
 
Ian Cutress, AnandTech: Given that Ray Tracing has been announced in the PS5 specs, can you say if that’s a specific Sony optimization or is that inherent in RDNA?

Lisa Su:
We certainly have done very specific optimizations for Sony. They are a very deep partner with us on semi-custom, and there are optimizations there. However we view ray tracing as a very important element across the portfolio. We will have ray tracing in a number of places. Will you look at that, you got me to say more about ray tracing!



The article puts the Renoir (Zen + Navi APU) speculation to bed.

It seems Renoir was for mobile.

John Taylor, AMD: I don’t think we’ve said anything about Renoir. I think Ian’s going into uncharted territory here. What’s Renoir? [laughs] Future Ryzen Mobile.
 
Last edited:

Darius87

Member
What do you mean? I am comparing PS4 vs PC GPU equivalent (7850) and adding additional RAM and CPU to prove that no, we wont be able to fit in 225W GPU in console because MS and Sony have magic talent for semicodnuctor design that AMD does not have.
there's something doesn't add up PS4 gpu 75W < HD7850 gpu 130W
how is that ps4 in games go up to 150W but 7850 just 100W?
 

ethomaz

Banned
there's something doesn't add up PS4 gpu 75W < HD7850 gpu 130W
how is that ps4 in games go up to 150W but 7850 just 100W?
HD7850 is only the board, GPU, VRAM.
PS4 is GPU, CPU, VRAM, SystemRAM, board, etc... the fully system.
 
Last edited:

R600

Banned
there's something doesn't add up PS4 gpu 75W < HD7850 gpu 130W
how is that ps4 in games go up to 150W but 7850 just 100W?
HD7850 max 100W in game. GPU only.

PS4 is 140W in games with CPU, IO and memory. I think PS4s GPU uses ~90W in this case.
 

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
Ian Cutress, AnandTech: Given that Ray Tracing has been announced in the PS5 specs, can you say if that’s a specific Sony optimization or is that inherent in RDNA?

Lisa Su:
We certainly have done very specific optimizations for Sony. They are a very deep partner with us on semi-custom, and there are optimizations there. However we view ray tracing as a very important element across the portfolio. We will have ray tracing in a number of places. Will you look at that, you got me to say more about ray tracing!



I think this answer from Lisa is underrated. Has she said AMD is a "deep partner" with MS? And Sony has done "very specific optimizations" to their Navi GPU? Can it really be THAT different from Xbox Scarlett's?
 
I think this answer from Lisa is underrated. Has she said AMD is a "deep partner" with MS? And Sony has done "very specific optimizations" to their Navi GPU? Can it really be THAT different from Xbox Scarlett's?
I mean AMD says the same thing about Microsoft:

"Our relationship with Microsoft is based on silicon design deeply rooted in tight collaboration. Our engineering organizations work together as one design team to continually innovate and significantly advance the overall gaming experience. "

 
Last edited:

Aceofspades

Banned
Im still confused, in the quote below:

Songoku, please re read my post and calculate everything again, you are missing the point. And point is :

PS4 die size = 348mm²
PS4 CPU + GPU = 70 + 212 (282mm²)
66mm² remaining is IO / 256 bit bus

PS5 die size = ???
PS5 CPU + GPU(40CU) = 75 + 251 + ~10RT (336mm²)
IO + 256 bit bus = 38mm² + 64mm² (102mm²)

So by my calculations ~ 438mm² for 40CU part if we follow PS4 die size dimensions.

Boy oh boy will some people turn out disappointed.

You calculated PS4 die size based based on 40CUs and ended up with 438mm² and stated that people will be disappointed while at the same time, you say this:

Sorry, was using IO/bus for 384 in addition to already used 256 bit one.

40CU/Zen²/RTHW/256bit ~ 340mm²
44CU/Zen²/RTHW/320bit ~ 380mm²

If MS goes with 14Gbps memory speed while Sony goes with 16Gbps, difference between BW will be 48GB/s, while 4CUs-8CUs more at lower clocks would result in ~0.6TF adv.

So according to you, Sony CANNOT put 40CUs on the die because it will be too big 438mm² while MS will fit 44CUs+ bigger memory bus on 380mm² die?!

Sorry man, the bias is strong with this one.
 

R600

Banned
Im still confused, in the quote below:



You calculated PS4 die size based based on 40CUs and ended up with 438mm² and stated that people will be disappointed while at the same time, you say this:



So according to you, Sony CANNOT put 40CUs on the die because it will be too big 438mm² while MS will fit 44CUs+ bigger memory bus on 380mm² die?!

Sorry man, the bias is strong with this one.
Mate, it was mistake, calculated double buses (256+384).

If Sony revealed video showing 340-380mm² die I would change my tune. Since Im working with 1.8GHZ Gonzalo and PCB rumor (as well as Sonys ever smaller die sizes in following gens), Ill go with what I wrote.

I have no favorite tbh, if anything PS5 being faster would suit me better duo to BC and me having only PS4pro.
 

Evilms

Banned
We now know almost all the ranges of Navi GPUs. It is a Russian retailer, as for the old releases, that reveals the existence of all the other GPUs to come. In particular, models from Sapphire, AMD's spearhead.

RX 5950XT - RX 5950 - RX 5900XT - RX 5900
RX 5850XT - RX 5850 - RX 5800XT - RX 5800
RX 5750XT - RX 5750 - RX 5700XT - RX 5700
RX 5650XT - RX 5650 - RX 5600XT - RX 5600
RX 5550XT - RX 5550 - RX 5500XT - RX 5500

RX 590XT - RX 590XT

HbnavMo.png



 
Last edited:

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
I mean AMD says the same thing about Microsoft:

"Our relationship with Microsoft is based on silicon design deeply rooted in tight collaboration. Our engineering organizations work together as one design team to continually innovate and significantly advance the overall gaming experience. "


Welp! Nothing to see here then. lol
 

pawel86ck

Banned
Mate, it was mistake, calculated double buses (256+384).

If Sony revealed video showing 340-380mm² die I would change my tune. Since Im working with 1.8GHZ Gonzalo and PCB rumor (as well as Sonys ever smaller die sizes in following gens), Ill go with what I wrote.

I have no favorite tbh, if anything PS5 being faster would suit me better duo to BC and me having only PS4pro.
According to you Gonzalo leak is legit, but at the same time your calculations suggest PS5 will be 8.3TF at best. Let me quote you:
315 - 320mm² die
Zen2 3.2GHZ 8c/16t
Navi GPU HW RT - 36CU@1.8GHZ(8.3TF)
16GB of GDDR6 256 bit bus - 512GB/s
6GB of DDR4
1TB SSD

The thing is, Gonzalo with just 8.3TF GPU would never beat vega64 score, and especially considering PS5 was using unoptimized drivers (unlike vega 64) for windows and 3dmark. Gonzalo 3Dmark score suggest at least 10TF GPU in PS5, therefore your 8.3TF estimations must be wrong.
 
Last edited:
I am confused wat does it mean?

Renoir was something on the AMD roadmap that had very few details officially.

Pieces here and there suggested it was an APU, first Zen/Vega then Zen/Navi.

I thought it looked like a good fit for console. It's not.

The interview clarifies things that it's Ryzen mobile. Probably for the Samsung/AMD collaboration.
 
Status
Not open for further replies.
Top Bottom