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

[thuway]

I think we might be a little optimistic on the Microsoft front if we are predicting a 50 plus cu spec.

If any thing the latest devkit chatter indicates they are going for a lower power cieling but who knows what's true any more.

Since Lockhart died I can't tell if that benefited Scarlett in raw power or not


There were older targets that claimed targets of 14 tflops but those could be Vega teraflops and once again just targets.

 

[iroboto]

Andrew Reiner and Colin are correct with PS5 being more powerful.

Did they indicate it was more powerful than Lockhart or Anaconda?

 

[thuway]

Did they indicate it was more powerful than Lockhart or Anaconda?

Reiner just said he's hearing PS5 is more powerful. Which would make sense since Lockhart is dead.

He didn't specify devkits or anything else. Just that he heard it from devs.

Colin Moriarty echoed those sentiments.

Than a shit storm happened where people cried foul that they were referring to devkits and that specs are moving targets. It wasn't pretty

 

[Proelite]

I personally think 40CUs at 1800MHZ is too much in terms of TDP. I think 36CUs at 1800MHZ boost would be ~8.3TF and given that Navi XT allegedly beats 2070 by few %, end 2070 outperforms GTX1080 by 15-20%, that would be better fit for rumor.

Any way I think 40CUs at those clocks is too much. 36CUs would also tie up better with ~320mm² die size, especially considering RT hardware will have to be included.

If 20 CUs a limit for a Navi shader engine, then I think 36 total make sense. Maybe 44 with 4 disabled would work too.

 

[Deleted member 13524]

I wonder if Andrew Reiner and Colin are correct with PS5 being more powerful.

If anaconda is 380mm roughly with 560 GB /s memory that should atleast get to 10.5 Navi tflops or nearly 14 Vega Tflops.

Something must be happening on Sony's ssd front because even if the rumors are true of Microsoft sending out documentation of conservative Xbox Scarlett specs, there has to be something to trigger developers chattering that PS5 is more powerful.

Sony going with chiplets for dGPU, raytracing coprocessor, CPU and IO all interconnected through IF in a single substrate could mean getting higher performance (at the cost of considerably higher sum of die area) than a monolithic large SoC and without breaking the bank on low yields.

I don't think Sony would order monolithic SoCs much larger than 400mm^2 because then they'd have the problem of not supplying enough chips on time to meet the demand.
It could be that Microsoft's approach for RT (which according to rumors is CPU-driven) takes considerably more transistors and die area than Sony's, and/or that Scarlett's demanded flexibility for non-gaming workloads in Azure servers also takes a toll on gaming performance-per-transistor or performance-per-area.
E.g. Microsoft wants Scarlett to have 32MB L3 and 256bit AVX2 units on the CPU essentially for non-gaming server workloads, whereas Sony is fine with 16MB L3 and smaller 128bit FPU units because it doesn't make that much difference for games written in their proprietary API and compiler. Sony then takes those extra mm^2 to put 4 or 8 more CUs in the iGPU instead. Going with HBM instead of GDDR6 would also save some die area that could be spent on a couple of CUs too, and two "Flashbolt" HBM2 stacks can provide up to 820GB/s to provide enough bandwidth for the larger GPU.

Did they indicate it was more powerful than Lockhart or Anaconda?

Their statements are the devs during E3 had been saying the PS5 is the most powerful console.
Besides, all proof points to Lockhart being canceled prior to E3, not during. E3 reports said "no one is hearing about Lockhart anymore", not that "Lockhart is being cancelled right now".

 

[iroboto]

Reiner just said he's hearing PS5 is more powerful. Which would make sense since Lockhart is dead.

He didn't specify devkits or anything else. Just that he heard it from devs.

Colin Moriarty echoed those sentiments.

Than a shit storm happened where people cried foul that they were referring to devkits and that specs are moving targets. It wasn't pretty

Fair enough. I’ve no reason to believe they would lie about it. Information can change over the course of time, however so theoretically they can be both right and wrong depending on when the statement is made.

But we’ve no intel on whether MS has sent 1 or 2 devkits? And we have no intel on when MS sent their kits right ?

 

[iroboto]

Their statements are the devs during E3 had been saying the PS5 is the most powerful console.

Do you have a link to said shit storm? I didn’t follow this story fully through.

 

[thuway]

Do you have a link to said shit storm? I didn’t follow this story fully through.

It's on twitter but the cliff notes is he said he heard ps5 is more powerful around E3.

The question than started was Reiner referring to devkit power or actual system target spec from documentation provided with said devkit.


It blew up after that for fanboy reasons and no further clarification was made

 

[Deleted member 13524]

Do you have a link to said shit storm? I didn’t follow this story fully through.

I couldn't care less about the shitstorm. Only the statements from journalists with a decent track record that talked with devs during E3.

https://forum.beyond3d.com/posts/2074903/

 

[iroboto]

I couldn't care less about the shitstorm. Only the statements from journalists with a decent track record that talked with devs during E3.

https://forum.beyond3d.com/posts/2074903/

You are referring to this tweet?

https://twitter.com/x/status/1137833936682274816

He has successfully inserted doubt into his own tweet. I think this is a deviation from “definitely more powerful” if we are looking at running his tweet through a sentiment analysis.

 

[Deleted member 13524]

There are tweets that go up to June 17, after which he said he wouldn't share any more information because he was trolled to death.

 

[iroboto]

There are tweets that go up to June 17, after which he said he wouldn't share any more information because he was trolled to death by microsoft fanboys.

I think it’s clear he had information from developers. What is not clear is what the developers received.

What if they purposefully and only sent a devkit to represent Lockhart because Anaconda isn’t ready? We don’t know enough, and he may not know enough. It’s critical for Andrew to not divulge too much information because it could expose his sources.

Like what if some developers received Lockhart only and others got anaconda. And some got both like MGS teams. It’s a pretty straightforward blue dye test to see who is talking.

If you want to assign risk levels to different development groups and you have two consoles that have the same features but different power characteristics how would you handle it. You have two possible configurations that you haven’t decided on yet, maybe one or maybe the other, or maybe both, why not A|B test who you send them out to and not let them know there are 2 consoles. Or if they received the more powerful variant
Or the weaker variant.

 

[thuway]

I think it’s clear he had information from developers. What is not clear is what the developers received.

What if they purposefully and only sent a devkit to represent Lockhart because Anaconda isn’t ready? We don’t know enough, and he may not know enough. It’s critical for Andrew to not divulge too much information because it could expose his sources.

Like what if some developers received Lockhart only and others got anaconda. And some got both like MGS teams. It’s a pretty straightforward blue dye test to see who is talking.

If you want to assign risk levels to different development groups and you have two consoles that have the same features but different power characteristics how would you handle it. You have two possible configurations that you haven’t decided on yet, maybe one or maybe the other, or maybe both, why not A|B test who you send them out to and not let them know there are 2 consoles. Or if they received the more powerful variant
Or the weaker variant.

Lockhart has been dead for greater than 6 months according to a few insiders. This is around the time final specs get nailed which is why you hear the whispers of who is more powerful

 

[AlNom]

E.g. Microsoft wants Scarlett to have 32MB L3 and 256bit AVX2 units on the CPU essentially for non-gaming server workloads, whereas Sony is fine with 16MB L3 and smaller 128bit FPU units because it doesn't make that much difference for games written in their proprietary API and compiler.

A larger LLC should help mitigate requests on a shared external bus.

 

[AlNom]

If they're wiring the pipeline for distributing workloads to 10 DCUs, the current max might be 16DCUs per Navi Shader Engine.

 

[iroboto]

Lockhart has been dead for greater than 6 months according to a few insiders. This is around the time final specs get nailed which is why you hear the whispers of who is more powerful

Links? This is new for me. 6 months dead is pretty significant.

 

[Ike Turner]

Links? This is new for me. 6 months dead is pretty significant.

 

[Proelite]

If they're wiring the pipeline for distributing workloads to 10 DCUs, the current max might be 16DCUs per Navi Shader Engine.

Any chance of >2 Navi Shader Engines?

 

[MrFox]

I see no credible source lockhart ever existed.

 

[Globalisateur]

It doesn't matter how we spin this, GCN flops < RDNA flops and all, but 36 CUs for not even 9Tflops would be pretty disappointing in 2020. I think 36CUs could have being a 2019 plan, but that plan would have changed.