Baseless Next Generation Rumors with no Technical Merits [post E3 2019, pre GDC 2020] [XBSX, PS5]

[BRiT]

I don't see how any third party dev would be privy to any changes to the hardware regarding cooling. Yes they will obviously have performance targets and so forth but I get very sceptical when they start talking about design decision that only the hardware guys at AMD and Sony would know.

Even more sketch when third party software developers talk about matters that only the hardware architects at all three would know (AMD, Microsoft, and Sony). That would require such exact and detailed information from not just 1, not just 2, but all 3. Especially in a time when no one else has even the slightest hint of details.

[obligatory clip of Inconceivable here]

 

[ultragpu]

His first party dev friend told him that if that means anything.

 

[BRiT]

Dynamic stuff. Fluid/matter simulation could be a game changer in some titles, creating new game mechanics.

Whenever anyone speaks to new physics simulations creating new game mechanics I always have flashbacks to the PC video game Trespasser.

 

[Mskx]

Credible sources, like Jason Schreier of Kotaku, has been saying for a while that devs familiar with both nextgen consoles have said they're both in the same ballpark, with one having an advantage on GPU and the other I/O, but fundamentally nothing earth shattering separating them.

The problem is that without numbers or other specifics this kind of stuff does not mean much.
What kind of TF difference is allowed in the same ballpark? What kind of advantage in GPU or I/O? What is "earth shattering" in terms of hardware difference?
Answers might change greatly from person to person or even dev to dev.

 

[disco_]

Where's this cooling revamp info coming from?

Bloomberg and now o'dium seems to be piggybacking off that.

 

[AbsoluteBeginner]

I collected few pictures from 3 independent sources in different time spans that all seem to match up pretty damn great. I sometimes talked about it separately, but decided to make a case for it because some have been calling me out to be "Github cult" lead, but I personally think I am just following well known and verified leaks that all seem to match (and have been deleted from the Internet at AMDs request).

Here is how it goes :

1. Flute Benchmark leak (source - Userbenchmark 23rd of July)



In blue I highlighted following :

• AMD Eng Sample
• OPN 100-000000004-15_32/12/18_13F9
• Platform : AMD Flute
• Socket : BL5
• Memory amount - 16GB
• Memory speed - 528GB/s

2. Github leak (source - Github December 24th, dates from June 2019)



In blue and yellow I highlighted

• File containing OPN number called Oberon_100-000000004-20_4_GPUID_ThermTripLimit_Feb-10
• kysy_if files which contain following...

• Chip name - Oberon
• Proj - OBR
• Stepping - A0
• Socket - BL5

3. Oberon GPU leak (without a score) - (source Userbenchmark date ??)



In blue I highlighted

• Chip PCI ID - 13F9:OBR
• Stepping - A0

What to make of this?

• Flute Benchmark contains OPN number that matches Oberon OPN number found in one of Github docs (100-000000004)
• Flute Benchmark contains socket type BL5 that also matches Oberon socket type found in one of Github docs
• Userbenchmark Oberon leak contains PCI ID - 13F9, which matches PCI ID used in Flute Benchmark AND also matches to OBR proj A0 stepping chip found in Github
• Flute Benchmark contains 16GB of what appears to be GDDR6 RAM on 256 bit bus (check latency), and modules higher then 14Gbps (448GB/s) or 16Gbps (512GB/s), which coincidentally, also matches to Github Oberon leak

Conclusion?

• Flute benchmark showed custom SOC, judging by socket type almost certainly not for PC or laptop
• Flute matches Github data perfectly
  • same OPN number
  • same socket
  • same GPU - 13F9
  • same stepping - A0
• We therefore have two independent sources confirming legitimacy of chip named Oberon, which was tested at the time Sony released "V" style PS5 dev kits to developers

In addition to this, we have a well known AMD leaker from Taiwanese message board (guy got Vega, Polaris and Navi 10 sizes right months before they were announced).

He posted few tidbits :

• Oberon (he refers to it as OBR) is PS5 chip (October)
• Arden is Xbox chip (October)
• Oberon is 15% smaller then Arden (~300mm2 / 350mm2) (October)
• Sparkman is new SOC and is not for laptop, but most likely semi custom as it doesn't have FPx pin but BLx (October)
• 50% smaller then Arden (December)
  
• Oberon is going through constant revisions and he wonders if they will make it for Holiday 2020 (November)

Now, beside him being very much correct for years through leaking AMD stuff made at TSMC (where he either has a good source or actually works), his post about Oberon's revisions match perfectly with Komachi's post about Oberon now on stepping E0.

This would not be that interesting had it not been for AquariusZi saying this 2 months before Komachi found it in data he mined.

His posts can be found :

https://www.ptt.cc/bbs/PC_Shopping/search?q=author:AquariusZi

 

[JoeJ]

Well done, priest

Question: Do we have anything like this about Lockhart? I do not remember why some months ago there was new confidence of 2 tiers from MS.
There is no official confirmation and no leak about chip either. Shouldn't we start to doubt it again?
Or would it make sense to disable more CUs for that?

 

[AbsoluteBeginner]

Well done, priest

Question: Do we have anything like this about Lockhart? I do not remember why some months ago there was new confidence of 2 tiers from MS.
There is no official confirmation and no leak about chip either. Shouldn't we start to doubt it again?
Or would it make sense to disable more CUs for that?

Well, apart from Github's Sparkman and AquariusZi saying there is a chip at TSMC foundries for consoles half the size of Arden, only this :

I see people say this a lot -- Lockhart existing isn't a rumor. If you put a retail Xbox One into dev mode there are several references to it

eg

// TODO: Final branding for Lockhart/Anaconda (TFS 23410665)
"lockhart": "Constrains CPU, GPU and memory resources to resemble a Lockhart retail console, but should not be considered an exact match for performance nor a replacement for testing on a Lockhart.",
"anaconda": "Constrains CPU, GPU and memory resources to resemble an Anaconda retail console",
"game streaming": ""

https://www.resetera.com/threads/ne...-contest-inside.168732/page-322#post-29495051

 

[JoeJ]

Thanks, forgot about 'sparkman' already and looked some old posts up

 

[mpg1]

Well, apart from Github's Sparkman and AquariusZi saying there is a chip at TSMC foundries for consoles half the size of Arden, only this :

Didn't the testing docs show Sparkman as a 56CU chip?

 

[AbsoluteBeginner]

Didn't the testing docs show Sparkman as a 56CU chip?

It had testing template from Arden copied in Sparkman spreadsheet, there were no tests done on either, just template with theoretical values from what I assume is Arden.

 

[AbsoluteBeginner]

https://medium.com/auedbaki/amd-ray...en-raytracing-addressed-rx-5600xt-8704575a704

How did we miss this?

AMD as a company…strongly believes in the value and capability of raytracing. RDNA 2, the next-gen, will support raytracing. Both the next-gen Xbox and PlayStation will support hardware raytracing with Radeon natively. We will be sure to have the content that gamers can actually use to run on those GPUs

We believe in our raytracing, and we will have it when the time is right.

- Mithun Chandrashekhar, Product Management Lead, AMD

So both consoles will have RDNA2 and both will have AMD's ray tracing solution.

Ariel never really fit in with Oberon IMO, and I speculated that Ariel was first Navi (its codename was GFX1000) that Sony needed early to get their BC correctly before AMD had RT and VRS ready (because we have Ariel tests from December 2018, and it was not actual SOC but GPU).

Entire Github test was regression test done on Oberon chip, and Ariel Spec was used for Native/BC1/BC2 tests hence no RT/VRS.

 

[Shifty Geezer]

Oh. How boring.

 

[Proelite]

https://medium.com/auedbaki/amd-ray...en-raytracing-addressed-rx-5600xt-8704575a704

How did we miss this?



So both consoles will have RDNA2 and both will have AMD's ray tracing solution.

Ariel never really fit in with Oberon IMO, and I speculated that Ariel was first Navi (its codename was GFX1000) that Sony needed early to get their BC correctly before AMD had RT and VRS ready (because we have Ariel tests from December 2018, and it was not actual SOC but GPU).

Entire Github test was regression test done on Oberon chip, and Ariel Spec was used for Native/BC1/BC2 tests hence no RT/VRS.

View attachment 3621

We found data that suggests Ariel was pre-silicon in another repo.

 

[AbsoluteBeginner]

We found data that suggests Ariel was pre-silicon in another repo.

And then there is Shader testbench available that specifies Ariel as GPU (also in testlog its referred to as adapter 0x13E0 and runs on Vulkan) so I think that chip certainly existed, I just think they used early Navi as testbed for BC until RDNA2 is ready with RT and VRS, hence Oberon.

I also think Oberon is 40CU chip, as is Ariel. This is why BW results show ~440GB/s for Ariel and 530GB/s for Oberon, because Oberon is actual APU accessing full API BW, while Ariel was GPU with 256 bit bus and 14Gbps chips.

 

[mpg1]

And then there is Shader testbench available that specifies Ariel as GPU (also in testlog its referred to as adapter 0x13E0 and runs on Vulkan) so I think that chip certainly existed, I just think they used early Navi as testbed for BC until RDNA2 is ready with RT and VRS, hence Oberon.

I also think Oberon is 40CU chip, as is Ariel. This is why BW results show ~440GB/s for Ariel and 530GB/s for Oberon, because Oberon is actual APU accessing full API BW, while Ariel was GPU with 256 bit bus and 14Gbps chips.

But this theory doesn't totally square though. RT was missing from Oberon tests was it not?..

 

[Mskx]

https://medium.com/auedbaki/amd-ray...en-raytracing-addressed-rx-5600xt-8704575a704

How did we miss this?



So both consoles will have RDNA2 and both will have AMD's ray tracing solution.

View attachment 3621

Well, technically the quote does not actually say that both will have RDNA2 but rather that RDNA2 and both next gen consoles will have Radeon RT.

 

[Scott_Arm]

https://medium.com/auedbaki/amd-ray...en-raytracing-addressed-rx-5600xt-8704575a704

How did we miss this?



So both consoles will have RDNA2 and both will have AMD's ray tracing solution.
...

Sure sounds like it. I expect that Series X and PS5 are going to be incredibly similar at this point. May as well be the same console with different brand stickers on the cases.

 

[AbsoluteBeginner]

But this theory doesn't totally square though. RT was missing from Oberon tests was it not?..

That is true, but Oberon Native regression test is running Ariel-SPEC-iGPU tests as noted below

If Ariel is GFX1000 (Navi10_lite), and it is, then there won't be RT/VRS test when running regression test with Ariel specs. Therefore, this does not prove Oberon has no RT/VRS, it proves that Oberon Native regression test was running Ariel SPEC iGPU (as was the case with BC1 and BC2), thus I speculate Ariel was early Navi used by Sony as testbed for PS5 BC, until RT/VRS become available.

 

[anexanhume]

That is true, but Oberon Native regression test is running Ariel-SPEC-iGPU tests as noted below

If Ariel is GFX1000 (Navi10_lite), and it is, then there won't be RT/VRS test when running regression test with Ariel specs. Therefore, this does not prove Oberon has no RT/VRS, it proves that Oberon Native regression test was running Ariel SPEC iGPU (as was the case with BC1 and BC2), thus I speculate Ariel was early Navi used by Sony as testbed for PS5 BC, until RT/VRS become available.

I’ve had the same theory for a while. Just remains to be seen if the latest stepping of Oberon is indeed 40 native CUs.