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

[Betanumerical]

Yes, it says Native for Oberon in 36CU - 2.0GHz.

Or Navi is TDP/perf underperformer (as it was rumored late last year), but 2-3 years ago when Sony and AMD started designing the chip, 36CU and high clocks were a very logical target considering chip size and 7nm node.

This is just a theory, but could easily explain this situation. Think about it, 36CUs is clearly something Sony would aim to put in console for following reasons :

So, if Sony started early with designing its Navi with AMD as it was rumored, I can definitely see their targets for Navi being slightly higher then 1.8GHz sweetspot. When this turned out to be false hope, they had nothing left but continue to work on chip and hope for 2.0GHz.

If the assumption was that Navi would clock high then wouldn't we be seeing a Arden that is smaller and more highly clocked?, Arden went the complete other direction with more CUs and a seemingly lower clock (no clocks in the GitHub leak for Arden).

 

[AbsoluteBeginner]

If the assumption was that Navi would clock high then wouldn't we be seeing a Arden that is smaller and more highly clocked?, Arden went the complete other direction with more CUs and a seemingly lower clock (no clocks in the GitHub leak for Arden).

They would be either way, no matter how highly clocked Navi can be, as MS apparently always shot for entry and premium SKU, giving them headroom for bigger chip as they would not have to find sweetspot like Sony would with single one.

I also dont know how Navi scales, since rumored Arden is 3x10WGPs, while PS5 rumored is 2x10WGPs, so maybe something between, while possible, would not be ideal (say 44 or 48CUs).

This is why rumored big Navi is 80CUs, there is probably sweetspot for one shader engine being 10WGPs in Navi.

 

[Betanumerical]

They would be either way, no matter how highly clocked Navi can be, as MS apparently always shot for entry and premium SKU, giving them headroom for bigger chip as they would not have to find sweetspot like Sony would with single one.

But the rumours would have been that the XBSEX was more powerful then the PS5 then, because the target specs (which it wouldnt meet) would also be higher then the PS5.

 

[Deleted member 13524]

Yes, it says Native for Oberon in 36CU - 2.0GHz.

Or Navi is TDP/perf underperformer (as it was rumored late last year), but 2-3 years ago when Sony and AMD started designing the chip, 36CU and high clocks were a very logical target considering chip size and 7nm node.

This is just a theory, but could easily explain this situation. Think about it, 36CUs is clearly something Sony would aim to put in console for following reasons :

• Navi CUs are considerably larger then GCN, and entire arch was made to be clocked much higher
• 40CU Navi in desktop is 251mm²
  • This is 18% more then counterpart used in PS4 chip (HD7870)
  • This is 9% more then counterpart used in Pro chip (RX480)
• It matches perfectly to 2x18CU and 36CU from last gen consoles they need for BC
• Zen2 will take more space then Jaguar cores
• RT hardware will take additional
• 7nm design and process is more expensive then 28nm and 16nm
• There is clear reduction in die size for Sony's consoles since PS2 to rumored PS5, and clear increase of clock speeds

So, if Sony started early with designing its Navi with AMD as it was rumored, I can definitely see their targets for Navi being slightly higher then 1.8GHz sweetspot. When this turned out to be false hope, they had nothing left but continue to work on chip and hope for 2.0GHz.

If Navi sweetspot was ~2.0GHz, everyone would be talking about design being smart (which it would be, very much, but we are looking from outside in after this has been designed for 3yrs).

None of this explains how all actually verified industry insiders are claiming both consoles are within splitting distance of each other and both at double digit TFLOPs, where the 9.7 TFLOPs GPU that is at least 24% slower than the competition clearly doesn't fall into.

 

[Betanumerical]

None of this explains how all actually verified industry insiders are claiming both consoles are within splitting distance of each other and both at double digit TFLOPs, where the 9.7 TFLOPs GPU that is at least 24% slower than the competition clearly doesn't fall into.

It also doesn't explain why the bandwidth in the leaks is nearly identical for both consoles (530GB/s for Oberon and 560GB/s for Arden) but the TFLOP difference is ~24%.

 

[AbsoluteBeginner]

None of this explains how all actually verified industry insiders are claiming both consoles are within splitting distance of each other and both at double digit TFLOPs, where the 9.7 TFLOPs GPU that is at least 24% slower than the competition clearly doesn't fall into.

Would not take industry insiders as gospel, especially 1.5yr before release considering how it it went last time around...Anyway, if you slash Pro vs XBX difference by half, would you say they are close?

It also doesn't explain why the bandwidth in the leaks is nearly identical for both consoles (530GB/s for Oberon and 560GB/s for Arden) but the TFLOP difference is ~24%.

Would not take it at face value. Arden/Sparkman only had theoretical values - therefore 560GB/s. Oberon had theoretical value of 448GB/s, but Ariel A0 tested 447GB/s, while Oberon tested 512GB/s with A0 and 530GB/s with B0, meaning AMD likely used 14Gbps as default value for theoretical numbers.

 

[Betanumerical]

Would not take it at face value. Arden/Sparkman only had theoretical values - therefore 560GB/s. Oberon had theoretical value of 448GB/s, but Ariel A0 tested 447GB/s, while Oberon tested 512GB/s with A0 and 530GB/s with B0, meaning AMD likely used 14Gbps as default value for theoretical numbers.

If we are not taking Arden/Sparkman at face value why assume any of the values are correct at all?.

 

[N2O]

But most people won't slash PS5 vs XSX as half of PS4Pro vs XB1X if the numbers end up real.

But at the same time, no one know are these "insiders" even heard from right developers when the XSX dev kits are being later than PS5's.

 

[AbsoluteBeginner]

If we are not taking Arden/Sparkman at face value why assume any of the values are correct at all?.

Have you seen repo? There is column with theoretical value and one that was achieved.

For Ariel/Oberon theoretical is 448GB/s, therefore 14Gbps on 256bit bus.

Achieved were the following :

Ariel A0 - 447GB/s (14Gbps)
Oberon A0 - 512GB/s (16Gbps)
Oberon B0 - 530GB/s (18Gbps - underclocked)

Entire time theoretical value remained 448GB/s because template used for theoretical assumes 14Gbps chips for Ariel/Oberon and since Arden had no test results, only theoretical it remained 560GB/s (14Gbps on 320bit bus). Thats all.

 

[Globalisateur]

Arden with 560GB/s theoretical bandwidth (so maybe like max 540GB/s achieved) is going to be bandwidth starved with 12 tflops GPU + 8 cores Zen 2 maybe clocked at 3.4ghz. We know this from 5700XT benchs when it's overclocked. We also had the confirmation from a known MS insider that Arden (well XSX) uses RDNA1.

At 530GB/s achieved bandwidth, PS5 with 9.2 Tf should have enough available bandwidth for both CPU and GPU.

 

[Deleted member 13524]

Would not take industry insiders as gospel,

Not gospel, just a lot more trustworthy than some log dumps someone found without relevant context.

especially 1.5yr before release considering how it it went last time around...

Last time around where they confirmed the XBone would be considerably slower than the PS4?
Or are you talking about the mid-gens where the specs dump of the Pro didn't mention it had dual-rate FP16?

Anyway, if you slash Pro vs XBX difference by half, would you say they are close?

No, I wouldn't. That's a difference large enough to create two distinct performance tiers in the OC market. It's the difference between a 5700XT and a RTX2070 Super which are on completely different price brackets.

 

[Proelite]

Arden with 560GB/s theoretical bandwidth (so maybe like max 540GB/s achieved) is going to be bandwidth starved with 12 tflops GPU + 8 cores Zen 2 maybe clocked at 3.4ghz. We know this from 5700XT benchs when it's overclocked. We also had the confirmation from a known MS insider that Arden uses RDNA1.

At 530GB/s achieved bandwidth, PS5 with 9.2 Tf should have enough available bandwidth for both CPU and GPU.

Did you not read the previous post lol.

If Arden uses 16gbps and 18gbps chips, the bandwidth would be 640 and 720.

 

[AbsoluteBeginner]

Not gospel, just a lot more trustworthy than some log dumps someone found without relevant context.

We will agree to disagree.

Last time around where they confirmed the XBone would be considerably slower than the PS4?
Or are you talking about the mid-gens where the specs dump of the Pro didn't mention it had dual-rate FP16?

I am talking about Durango and Orbis before specs leaked. And that difference is easily 2x rumored Arden vs Oberon.

No I wouldn't. That's a difference large enough to create two distinct performance tiers in the OC market. It's the difference between a 5700XT and a RTX2070 Super which are on completely different price brackets.

You maybe wouldnt, but 2070 v XT difference is less then 10%. How is that not close, especially for non hardcore PC audience is beyond me.

 

[Deleted member 7537]

https://twitter.com/i/web/status/1212177557736411136

Edit: Probably not related.

 

[Globalisateur]

Did you not read the previous post lol.

If Arden uses 16gbps and 18gbps chips, the bandwidth would be 640 and 720.

I thought we only talked about facts written in the github and not imagining stuff ? It's clearly written 560GB/s for Arden which was a double confirmation about the Scarlett reveal render analysis. We have no reason to think they are going to use faster chips.

There could be new version of Oberon with 56CUs and 13tf then. Why not compare Arden with this imagined PS5 ?

 

[Betanumerical]

https://twitter.com/i/web/status/1212177557736411136

Im not sure what this means?

 

[Proelite]

The first pastebin last gen that leaked Durango vs Orbis was very unkind to Orbis.

Sweetvar26, an AMD insider, leaked the Jaguars cores for Durango. Also said Durango was a supercomputer and more powerful than Orbis. He and his AMD friend were two of four people sued by AMD.

I was at MS at the time and was aware that earliest XB1 dev kits were Intel Xeons with 7900 GPUs. I mistakenly thought the 1.2 teraflops processor was related to the CPU.

The narrative prior to complete Durango picture was that it was definitely the more powerful machine. People thought PS4's 1.84TF GPU and 2GB GDDR5 sucked.

 

[Proelite]

I thought we only talked about facts written in the github and not imagining stuff ? It's clearly written 560GB/s for Arden which was a double confirmation about the Scarlett reveal render analysis. We have no reason to think they are going to use faster chips.

There could be new version of Oberon with 56CUs and 13tf then. Why not compare Arden with this imagined PS5 ?

448GB/s was theoretical for Ariel / Oberon. We see tests showing upgrade to 512GB/s and 530GB/s when they used faster GDDR6 for the real product.

560GB/s was theoretical for Arden. If it follows the same upgrade path as Ariel / Oberon, 640GB/s and 663GB/s are achieved.

It's a simple change to upgrade to faster GDDR6, they can do it right before mass manufacture.

 

[N2O]

https://twitter.com/i/web/status/1212177557736411136

I doubt it's even related to next gen console at all.

 

[Deleted member 7537]

I doubt it's even related to next gen console at all.

One does wonder if other Device IDs were fake.