If true and RT has dedicated chip this would free up some GPU resources, and 10TF PS could still compete 12TF XSX.
Baseless Next Generation Rumors with no Technical Merits [post E3 2019, pre GDC 2020] [XBSX, PS5]
- Thread starter BRiT
- Start date
- Status
If true and RT has dedicated chip this would free up some GPU resources, and 10TF PS could still compete 12TF XSX.
AbsoluteBeginner
Regular
Gonzalo ES2 1.0GHZ January
Gonzalo QS 1.8GHZ April (20K FireStrike)
Oberon A0 2.0GHZ August
Oberon B0 2.0GHZ October
There is no hidden chip here or 2nd SOC, this is 99% PS5 chip.
They may have planned to release it in 2019 @ 1.8GHZ (I doubt it), but I am 100% you do not create QS and OPN of your chip and cancel it for new one.
Gonzalo QS 1.8GHZ April (20K FireStrike)
Oberon A0 2.0GHZ August
Oberon B0 2.0GHZ October
There is no hidden chip here or 2nd SOC, this is 99% PS5 chip.
They may have planned to release it in 2019 @ 1.8GHZ (I doubt it), but I am 100% you do not create QS and OPN of your chip and cancel it for new one.
That can't be the case because Nvidia CEO has said with enough confidence from his espionage network that it is slower than his mobile products.
He has spoken!
Fits perfectly with the rumors a while ago regarding this, and the fact PS5 was Navi1/2 hybrid. I guess this is why NV said their 2080 was faster, meaning the PS5 perhaps, as PS is the most popular console.
AbsoluteBeginner
Regular
Most telling part IMO is TF per GB/s of BW.
Oberon shows 448GB/s 36CUs and 2.0GHZ clocks.
This brings us to :
448/9.2TF - 48GB/s per TF
For Scarlett, as per E3 video and rumored 12TF on 320bit bus it brings us to :
560/12TF - 46GB/s per TF
So it seems per TF, Navi will require less BW then GCN (which was well known) and 448/560GB/s might just be enough for 9.2TF and 12TF.
Oberon shows 448GB/s 36CUs and 2.0GHZ clocks.
This brings us to :
448/9.2TF - 48GB/s per TF
For Scarlett, as per E3 video and rumored 12TF on 320bit bus it brings us to :
560/12TF - 46GB/s per TF
So it seems per TF, Navi will require less BW then GCN (which was well known) and 448/560GB/s might just be enough for 9.2TF and 12TF.
I think they have enough in support for their first party games to compete against a small power differential.
It’s a higher priority that they nail backwards compatibility, CPU, Hard drive before nailing add-ons like RT.
Failing to hit those will be much more detrimental to their success than not having the power crown.
1. The 36 CUs APU fits the rumor very well (PS5 was originally planned for 2019).
Imo they reuse the previous version for devkits.
2. After changing the plan to release PS5 a year later, they have to add ray tracing in GPU, and change the whole APU from layout. They may even add some CUs and move to 7nm+, so SONY’s target performance increase to surpass xex. This result fits the information from the insiders Kleegamefan and others.
3. Leaks from Asia BBS said that PS5 APU was in early stage of verification in Oct. And the APU was under revision and re-testing even in November. It means PS5 APU has changed very much from the original 2019 design.
If Sony has decided to push PS5 a year later, they must know it costs hundreds of millions and IMO Sony has already invested to ensure PS5 leads.
Imo they reuse the previous version for devkits.
2. After changing the plan to release PS5 a year later, they have to add ray tracing in GPU, and change the whole APU from layout. They may even add some CUs and move to 7nm+, so SONY’s target performance increase to surpass xex. This result fits the information from the insiders Kleegamefan and others.
3. Leaks from Asia BBS said that PS5 APU was in early stage of verification in Oct. And the APU was under revision and re-testing even in November. It means PS5 APU has changed very much from the original 2019 design.
If Sony has decided to push PS5 a year later, they must know it costs hundreds of millions and IMO Sony has already invested to ensure PS5 leads.
Even though this was a jab at me, it was really funny.
AbsoluteBeginner
Regular
Absolutely nothing points in that direction.
You do realize current leaked Oberon A0 and B0 leaks point to recent revisions of chip? Both pretty much the same. Same amount of CUs, same amount of BW, same bus, same frequency.
I said few pages ago, Sony going for narrow and fast console and pricing it at 399$ seems TOO SMART to miss. Glad it looks like Cerny thought thats the best way forward.
Mind you, 36CU Navi at 2.0GHZ is very very strong and with Zen2, propiratery SSD and 7nm node it wont come cheap.
With 36 CU's, I certainly hope they are going for a $399 price point. I hope they are going for something like an October release as well.
The only thing I find missing in these leaks is a good explanation for the difference between Ariel and Oberon. Both have B0 stepping versions. They both seem to be separate APU's, but they seem to have the exact same specs. Perhaps they are testing different process technologies (or fabs if dual sourcing) for the same design?
The only thing I find missing in these leaks is a good explanation for the difference between Ariel and Oberon. Both have B0 stepping versions. They both seem to be separate APU's, but they seem to have the exact same specs. Perhaps they are testing different process technologies (or fabs if dual sourcing) for the same design?
forum meta is finally happening. I was there!
AbsoluteBeginner
Regular
I think Ariel was dev kit SOC codename for Sony, Oberon was retail unit.
Ariel has been found since 12.12.2018, Oberon has Ariel referrences but first appeared in August (A0) and October (B0)
I totally get the naming confused - but wouldn't one be owned by MS and the other by Sony? So by default they would be named differently?
edit: I see I was wrong.
Which are the 2 APUs that MS owns then?
IMO all of them are credible: Kleegamefan, Taiwan BBS, and Japan Komachi. A lot of things point to the direction in previous posts.
Just asking is the B0 chip the same as the APU tested in Taiwan in Oct. 2019?
AbsoluteBeginner
Regular
Could be. B0 is after A0, and A0 is from August (at least we know it since then).
9.2TF Navi ~ 13TF GCN. If Sony sent dev kits in late 2018, and this is when 13TF leaked, they must have been 13TF GCN.
But Klee clearly mentioned double digit, over 10 RDNA TFs. And he also confirms 12 TF RDNA for XsX, so if nothing changes with PS5 still 7-10% stronger it'd be around 13 TF RDNA. This is what's so contradicting with all the new info today, nothing matches up any more. You either believe one "credible" source or another. Or all of them are wrong.
And the BS continues.
How the hell would a 4 TF Lockhart "Sparkman" have higher memory bandwidth than PS5?
Last edited:
AbsoluteBeginner
Regular
Klee...I wont comment on that guy. He gave 0 specs. That might work for Resetera, but I doubt anyone will take his leaks here hs genuine.
Oh look
I think regardless if ps5 is $400 9.2TF or $500 11.5TF people will come to terms with it and say they played it smart.
Not because people are being fanboys but because Sony has much more leeway to play with.
So with that being said, if 9.2 is the most financially better approach i.e. less loss on hardware that's what they will go for.
I'm not sure if that's the case yet though, I don't know if there's less or more of a loss at 4 or 500 price points due to cost of chips, cooling, power delivery.
Not because people are being fanboys but because Sony has much more leeway to play with.
So with that being said, if 9.2 is the most financially better approach i.e. less loss on hardware that's what they will go for.
I'm not sure if that's the case yet though, I don't know if there's less or more of a loss at 4 or 500 price points due to cost of chips, cooling, power delivery.
I'll take it as seriously as any other leak, worthy of discussion.
When you don't have all the information, you don't have all the information.
it's that plain and simple.
Finding evidence that corroborates other evidence is stronger than singular data points.
The issue in this type of scenario is identifying unique data points. Does everyone at Resetera who claims 10TF+ are all unique data points or are they the same data point just told by different people.
These SOC leaks are actual different and unique data points which is why you're hitting conflict.
Wait for more data. Jan 6 isn't far away anyway.
It doesn't work that way. Every computer is made up of processing and storage. If you have more storage and faster storage than processing, you use algorithms weighted in that direction. If you have more processing power than storage potential, you use methods that favour computation.
By and large, memory bandwidth has not been improving as fast as computational power, and so we have less gbps per flop than we used to. First Googled result.
Code:
| G80 | GP100 | GV100
GFLOPS (SP) | 384 | 10600 | 15000
GPU↔GPU memory | 84 GB/s | 720 GB/s | 900 GB/s
FLOP/Byte | 4.5 | 14.7 | 16.67Any 'sweet spot' will be economic in nature, not architectural.
- Status
Similar threads
- Locked
