• 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.
  • We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.

When do you think the PS5 reveal will take place?

  • January

    Votes: 6 0.3%
  • February

    Votes: 1,172 65.7%
  • March

    Votes: 273 15.3%
  • April

    Votes: 81 4.5%
  • May

    Votes: 116 6.5%
  • June

    Votes: 48 2.7%
  • Later

    Votes: 89 5.0%

  • Total voters
    1,785
  • Poll closed .
Status
Not open for further replies.
Aug 26, 2019
6,342
Speaking of The Witcher 3, I recall that it was announced a couple of weeks ahead of the PS Meeting in February 2013. It was announced via Gameinformer for "next-gen consoles and PC", so hopefully we start seeing next-gen games get announced if the console reveals aren't happening yet.
 

MykhellMikado

Alt account
Banned
Jan 13, 2020
823
Cost per wafer. Size of chip. Yields. All of this is important when considering price. Have you seen how big NVIDIAs big GPUs are? The 2080 is 545mm2 on 12nm. The 5700xt by comparison is 251mm2 on 7nm.

Scorpio APU is 359mm2 on 16nm. We estimate the Series X APU to be around 380mm2. Very similar. Wafers are more expensive at 7nm but yields are good so the price won't much different. Consoles get orders in the millions. We're going to be ok.

I'm not worried at all, personally my expectations for the PS5 is somewhere between a 5700 and 5700XT with a bunch of added features bolted on.
With that in mind I can see between mid 8Tflops and low 9Tflops for PS5.

I still think the XSX is sub 10 RDNA Tflops but we will see for sure soon.
 
Oct 27, 2017
4,018
Florida
I'm not worried at all, personally my expectations for the PS5 is somewhere between a 5700 and 5700XT with a bunch of added features bolted on.
With that in mind I can see between mid 8Tflops and low 9Tflops for PS5.

I still think the XSX is sub 10 RDNA Tflops but we will see for sure soon.

Well, we estimated the die size of the X to be 380-400nm. If the 5700xt is 250mm2 and the zen 2 CPU is around 50mm2 (maybe smaller depending on the cuts to cache) so we still have 80mm2 -100mm2 to work with. Memory controller isn't that big so I'm going to assume many more CUs.
 

DammitLloyd

Member
Oct 25, 2017
779
I'll take my stab at it. I envision a chiplet configuration with a small bank of HBM directly on the interposer.

The specific configuration can be found here:


index.php
index.php

Nvidia's next GPU after Ampere, Codenamed Hopper is a MCM GPU too.
 

Deleted member 12635

User requested account closure
Banned
Oct 27, 2017
6,198
Germany
Having memory stacked on top of the GPU die is definielty future stuff.
But just having a simple MCM APU with a CPU-Chiplet connected to the GPU which has the Memory-Controllers and then talking to GDDR6 would be relatively easy to do.

The Xbox 360 followed that concept and with newer process nodes they put the CPU and GPU on the same chip later on.
They did "crazy" stuff back then.
Xbox_Architecture_large.gif
Indeed they did crazy stuff. Wasn't it several steps until it was one chip. First 2 chips, then MCM and then single chip? I do not remember.
 

anexanhume

Member
Oct 25, 2017
12,913
Maryland
This is not linear, the gross margin of lower end GPU is probably lower than 64% and the gross margin of Turing GPU is probably higher. On the table I gave the lowest Nvidia GPU is only price 50% above manufacturing cost the highest Nvidia GPU price is 143% above the manufacturing cost.
Precisely. And low margin GPUs make up a majority of the sales. A lot more people buy 1650, 1660, and 2060s than they do 2080 and 2080 TIs.
 

MykhellMikado

Alt account
Banned
Jan 13, 2020
823
This is not linear, the gross margin of lower end GPU is probably lower than 64% and the gross margin of Turing GPU is probably higher. On the table I gave the lowest Nvidia GPU is only price 50% above manufacturing cost the highest Nvidia GPU price is 143% above the manufacturing cost.

I'm not claiming it is, the two I referenced earlier had manufacturing costs in the $150-$200 range. If we have the manufacturing costs of a product
we don't have to use MSRP to guess anymore.

The fact is we don't know the manufacturing costs of each of the current cards but we do have MSRPs which on average and in the price tier we are talking about have manufacturing costs of 1/3.

Using the MSRP is not a hard and fast rule, it's just a guideline in the absence of actual manufacturing costs. It's an estimation tool, I'm not sure why we are trying to turn it into a hard equation.
 

Locuza

Member
Mar 6, 2018
380
Indeed they did crazy stuff. Wasn't it several steps until it was one chip. First 2 chips, then MCM and then single chip? I do not remember.
First it were two chip packages.
One package had the GPU die + the eDRAM die and the other the CPU (everything was build under 90nm for the Xenon/Zephyr versions).
mb.jpg


Jasper was the final revision before the fusion, where the CPU and GPU were manufactured under 65nm and the eDRAM die on a 80nm process but the packaging was still the same:
motherboard.jpg


https://www.anandtech.com/show/2682
https://www.anandtech.com/show/3774/welcome-to-valhalla-inside-the-new-250gb-xbox-360-slim

After that Vejle (Valhalla was not the codename) under the 45nm process node was the turning point, where everything was on a single package.
CPU and GPU got unified but the eDRAM was still separate.
adf59e96c3096e79915d1dd3fe8a56b7-970-80.jpg

dims
 

MykhellMikado

Alt account
Banned
Jan 13, 2020
823
Well, we estimated the die size of the X to be 380-400nm. If the 5700xt is 250mm2 and the zen 2 CPU is around 50mm2 (maybe smaller depending on the cuts to cache) so we still have 80mm2 -100mm2 to work with. Memory controller isn't that big so I'm going to assume many more CUs.

We have to make that assumption but does that mean bigger and slower as well, how many CUs enabled/disabled, clocks etc. If they went for more CUs, disabled a few, kept clocks moderate, the we would still need to see the hardware solutions AMD has such as RT.

And no I'm not saying they are going to dedicate 100mm2 to just raytracing, just that as it stands now we have no idea if is a wide and slow, wide and fast, total CUs and CUs disabled for yield, or what else is on die.

actually do we have confirmation that Xbox Series X is indeed monolithic and not chiplet based?
 

MykhellMikado

Alt account
Banned
Jan 13, 2020
823
Ok so I'm going to put forth and clarify my thinking:

the XSX is the equivalent of 12 GCN Tflops or 2x X1X

XSX is actually 9.6 RDNA Tflops
PS5 is actually 9.2 RDNA Tflops
This makes:
Phil's statements true
rumors that both are close true
That XSX slightly out performs PS5 true
9 Tflop PS5 leaks true

further we already have prospective AMD GPUs which would fall into the needed performance/price range for this to be true

ok I'm adding to this theory and likely going to catch a lot of hate for this, but...maybe the Xbox series X actually uses a chiplet design.

All the above stays true with two important differences:

Lockhart is a 5500/XT at 4.8 Tflops
Xbox Series X is 2 5500 XTs duck taped together putting out 9.6Tflops.

this keeps the costs down for the XSX and takes the 158mm2 and doubles it in the XSX to 316mm2 which would also help explain the larger die size of the XSX.

Yes I know it's a little wild but this is a theory I came up with to make the pieces fit.
 

III-V

Member
Oct 25, 2017
18,827
You intend to tell me there is a significant markup on the premium GPUs?

pretendtobeshocked.gif
 

MykhellMikado

Alt account
Banned
Jan 13, 2020
823
Xsx die officially revealed to be monolithic APU.
You know there is a photo of the Xbox Series X chip in the public already?
(no hate 🤓)
Nope.

We've already seen the die and it's monolithic.

And I doubt they're clocking their 52 CUs at 1450 MHz. That would make the X1X a bigger jump in clock to the S than the SX is to the 1X.

Man that fell apart quick.. oh well.
 
Oct 26, 2017
20,440
Ok so I'm going to put forth and clarify my thinking:

the XSX is the equivalent of 12 GCN Tflops or 2x X1X

XSX is actually 9.6 RDNA Tflops
PS5 is actually 9.2 RDNA Tflops
This makes:
Phil's statements true
rumors that both are close true
That XSX slightly out performs PS5 true
9 Tflop PS5 leaks true

further we already have prospective AMD GPUs which would fall into the needed performance/price range for this to be true

This seems all plausible to me. Where did the 12 teraflop report come from btw?
 
Aug 26, 2019
6,342
Oct 26, 2017
20,440
Tom Warren teased 12TF since X019, and this WC article posted before TGAs reveal:

www.windowscentral.com

Inside the target specs of the next Xbox 'Project Scarlett,' 'Anaconda', and 'Lockhart'

Microsoft is gearing up to reveal Xbox "Scarlett" in full in the near future, and we've finally got some credible information on the target specs.

Is this site usually accurate?

Series X won't be 9.6TF and Ps5 won't be 9.2TF

Why not?
 
Oct 25, 2017
2,884
I suspect Microsoft are confident Series X is more powerful than PS5 given they've revealed the box already. They won't say it's the most powerful as they shouldn't know that.
 

Gamer17

Banned
Oct 30, 2017
9,399
I suspect Microsoft are confident Series X is more powerful than PS5 given they've revealed the box already. They won't say it's the most powerful as they shouldn't know that.
I m not so sure about that . Every time they refer to xsx they mention "our most powerful console "(even 2 days ago in investor call)which is a far from what they did with x1x when they were confident about power.
If anything I feel they are both close otherwise ms language would be different .
 
Status
Not open for further replies.