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

[dobwal]

I've been searching for a good 20-30 minutes now - long enough for my cup of tea to go cold - and I'm losing the will to live, especially when I know I'm just procrastinating and should be looking for a job.

If I think on, I'll carry on having a look tomorrow, but in the meantime, I know there's a fair bit of discussion around GPU chiplets in this thread.

There is a research paper that Nvidia produced regarding a MCM based GPU. And the work necessary to produce a performant MCM design isn’t trivial.

https://research.nvidia.com/sites/default/files/publications/ISCA_2017_MCMGPU.pdf

I’m not sure if this is the exact paper but a similar one was part of a previous B3D discussion

 

[Megadrive1988]

There is a research paper that Nvidia produced regarding a MCM based GPU. And the work necessary to produce a performant MCM design isn’t trivial.

https://research.nvidia.com/sites/default/files/publications/ISCA_2017_MCMGPU.pdf

I’m not sure if this is the exact paper but a similar one was part of a previous B3D discussion

Supposedly, MCM will be implemented with Nvidia's Hopper GPU architecture, which will come after Ampere.

https://techreport.com/news/3467028/nvidia-mcm-hopper-ampere-successor/
https://videocardz.com/newz/nvidia-registers-hopper-and-aerial-trademarks
https://www.3dcenter.org/news/hardware-und-nachrichten-links-des-1112-november-2019
https://www.guru3d.com/news-story/r...named-hopper-gets-a-registered-trademark.html
https://wccftech.com/nvidia-hopper-gpu-mcm-leaked/

Again, if any of this is true, one might expect Hopper to occupy the 2022 time frame, assuming Ampere occupies 2020 and 2021.

Then again, I wouldn't be surprised if Nvidia will have another architecture, mainly for consumer cards, like how Turing came after Volta, or how Pascal got inserted on the roadmap between Maxwell and Volta.

 

[chris1515]

https://www.resetera.com/threads/ne...8-the-dark-tower.159131/page-89#post-27506956

I was listening to the Giant Bombcast and Jeff Gerstmann said that he had heard from developers that both of these consoles are super powerful and actually on par with PCs and not behind like most generations

 

[Globalisateur]

18gbps chips might be really expensive. Perhaps 17gbps in retail.

How many CUs can fit in a 316mm² APU + Zen 2 8GB L3 at 7nm EUV ? Have you already try that combination ?

 

[PSman1700]

Vague again, on par with what kind of pc’s? RTX titan/2080Ti with a 12core ryzen at over 4ghz or a rtx 2070/r7 combo.

Current or late 2020 pc’s with rtx3000/navi2 parts.

 

[Proelite]

How many CUs can fit in a 316mm² APU + Zen 2 8GB L3 at 7nm EUV ? Have you already try that combination ?

My guess is 44-48 total (-4 disabled for yield). Obviously a lot can change depending on size of RT HW and secret sauces.

 

[Shifty Geezer]

Some of my baseless theorizing as to why Sony/MS may be aiming for 12+ TF: Perhaps the chances of a Pro version of either console in the next 3-4 years is rather slim simply due to the slow down in process tech.

Or even because, after trying it out, maybe they found the market ain't really interested and it cost more money than it was worth? There's zero precedent for improved power mid-gen consoles. This gen could just be a blip rather than a new standard, with normal service resumed next gen.

 

[PSman1700]

PS5 is rumored to be having a 300mm2apu.

 

[Shifty Geezer]

Rumoured where by whom? We need sources to decide whether we believe in the rumour or not.

 

[Pinstripe]

Or even because, after trying it out, maybe they found the market ain't really interested and it cost more money than it was worth? There's zero precedent for improved power mid-gen consoles. This gen could just be a blip rather than a new standard, with normal service resumed next gen.

Yup. Perhaps 5nm/3nm Slim-versions down the road, but no Pro versions. That made sense this gen due to 4K gaming, but I don't see the same holding true for next-gen.

 

[chris1515]

Vague again, on par with what kind of pc’s? RTX titan/2080Ti with a 12core ryzen at over 4ghz or a rtx 2070/r7 combo.

Current or late 2020 pc’s with rtx3000/navi2 parts.

With current PC I suppose and Jason Schreier told more powerful than a RTX 2080. At least on Microsoft side it seems possible.

 

[PSman1700]

Thought i was in the baseless section, it was ’komachi’, according to ppl on resetera.

 

[PSman1700]

With current PC I suppose and Jason Schreier told more powerful than a RTX 2080. At least on Microsoft side it seems possible.

Could be, could also not be. I suppose a 2080 isn’t that common dominator (yet). For 2020 it might be true then. I just mean ’on par with pc’s’ can mean alot of things, it is as vague as anything else so far.

 

[Shifty Geezer]

Thought i was in the baseless section, it was ’komachi’, according to ppl on resetera.

This is where baseless rumours are collated and discussed, but they still need to be explained for the purposes of that discussion. The Tech Thread discusses possible designs based on tech analysis, whereas this one is based on 'insider info', but we still kneed to know whether a rumour is from a notable insider or some random pastebin.

 

[Proelite]

Rumoured where by whom? We need sources to decide whether we believe in the rumour or not.

I posted it in the technology forum since it has technical merits.

https://forum.beyond3d.com/posts/2092462/

 

[PSman1700]

Well it was supposed to be someone named komachi, it’s all over the last page over at resetera next gen speculation topic.

No idea if that komachi is reliable or if it is another ’klee’ with a friend that has a source.

 

[Proelite]

Komachi has being the single most reliable leaker of this generation.

He and to a lesser extent, Tom_Apisack, leaked Gonzalo, Oberon, Ariel, Sparkman, etc.

He had access to AMD roadmaps and/or chip tests results.

 

[Deleted member 13524]

there's nothing else MS can measure to derive a 2x figure, unless they're running something like 3DMark.

If Sony uses 3DMark, Microsoft probably does it too.

https://twitter.com/x/status/1118106567579062272

I think there's 3 possible reasons why Phil Spencer didn't claim any TFLOPs numbers:

1 - Clocks are still not final;
2 - Clocks are final but they don't want to open their game for the PS5, whose clocks may not be final;
3 - Like Digital Foundry suggested, he wants to avoid comparing GCN TFLOPs to RDNA TFLOPs which would lead to disappointment for people who don't know that RDNA TFLOPs > 1.2x GCN TFLOPs.

 

[Proelite]

If Sony uses 3DMark, Microsoft probably does it too.

https://twitter.com/x/status/1118106567579062272

I think there's 3 possible reasons why Phil Spencer didn't claim any TFLOPs numbers:

1 - Clocks are still not final;
2 - Clocks are final but they don't want to open their game for the PS5, whose clocks may not be final;
3 - Like Digital Foundry suggested, he wants to avoid comparing GCN TFLOPs to RDNA TFLOPs which would lead to disappointment for people who don't know that RDNA TFLOPs > 1.2x GCN TFLOPs.

Phil is doing what he did for the 1X. PR 4-4.5X when the real world increase was 5-6X (900p to 4k). Straight up math.

 

[goonergaz]

Phil is doing what he did for the 1X. PR 4-4.5X when the real world increase was 5-6X (900p to 4k). Straight up math.

So undersell and overdeliver? Seems an odd PR.