Next Generation Hardware Speculation with a Technical Spin [post E3 2019, pre GDC 2020] [XBSX, PS5]

[Malo]

I did some basic math using the perspective lines. I reached at ~373mm for the 1X SOC. The Anconda SOC is nearly the same length as the 1X's, but it's a tiny bit wider.

I believe he's asking what component you're using as a basis for scale.

 

[anexanhume]

I did some basic math using the perspective lines. I reached at ~373mm for the 1X SOC. The Anconda SOC is nearly the same length as the 1X's, but it's a tiny bit wider.

How are translating between the two though? What known value do you have in the Anaconda image to base off of?

 

[Proelite]

How are translating between the two though? What known value do you have in the Anaconda image to base off of?

GDDR5 package size is 14x12mm. I extrapolated the 1X SOC size from the lines to the GDDR5 packages. I arrived at 373mm2. The actual SOC is 24mmx15mm.

GDDR6 package size is 14x12mm. I used the same methodology and arrived at 24mmx16/17mm.

 

[Nisaaru]

I think that unlikely. I doubt they’d have development silicon to show off yet.

I don't really wanna speculate about the nature of the render and how it relates to the current DevKit but

TSMC 7nm+(7nm with EUV) is in production.

https://www.techspot.com/news/80237-tsmc-7nm-production-improves-performance-10.html

the comments are also interesting there. Apple's A13 is 7nm+.

 

[anexanhume]

GDDR5 package size is 14x12mm. I extrapolated the 1X SOC size from the lines to the GDDR5 packages. I arrived at 373mm2. The actual SOC is 24mmx15mm.

GDDR6 package size is 14x12mm. I used the same methodology and arrived at 24mmx16/17mm.

Thanks. This assumes the top of all parts measured are in the same plane, so the difference in perspective will yield some error since they’re not in the same plane.

 

[Proelite]

Thanks. This assumes the top of all parts measured are in the same plane, so the difference in perspective will yield some error since they’re not in the same plane.

I think my errors come from using the corners of the metal covers of the SOC. The actual die is inside the metal cover.

 

[anexanhume]

Intel hired chief Xbox SoC architect John Sell. Confirmed via LinkedIn.

https://www.fudzilla.com/news/pc-hardware/48873-intel-hires-chief-architect-of-xbox-one

 

[Ike Turner]

Intel hired chief Xbox SoC architect John Sell. Confirmed via LinkedIn.

https://www.fudzilla.com/news/pc-hardware/48873-intel-hires-chief-architect-of-xbox-one

Fudzilla's definition of "Our knowledgeable sources" is literally LinkedIn. At what point in time are internet "journalist" (idiots) going to be factual (ie: "while doing some goggling I found that" & "one of our readers just pointed to us that xxxx updated his LinkedIn profile") instead making up fake knowledge & secret sources?

 

[anexanhume]

Fudzilla's definition of "Our knowledgeable sources" is literally LinkedIn. At what point in time are internet "journalist" (idiots) going to be factual (ie: "while doing some goggling I found that" & "one of our readers just pointed to us that xxxx updated his LinkedIn profile) instead making up fake knowledge & secret sources?

The article makes the claim he’s gone to work on security applications at Intel, which is not captured on LinkedIn.

 

[borntosoul]

I don't really wanna speculate about the nature of the render and how it relates to the current DevKit but

TSMC 7nm+(7nm with EUV) is in production.

https://www.techspot.com/news/80237-tsmc-7nm-production-improves-performance-10.html

the comments are also interesting there. Apple's A13 is 7nm+.

This is really good news, and something l hope either Sony or ms are able to take advantage of next year.
If they do it’s going to make these gen consoles truely cutting edge!

 

[Ike Turner]

The article makes the claim he’s gone to work on security applications at Intel, which is not captured on LinkedIn.

It is (when accessed on mobile at least) :neutral:

 

[anexanhume]

It is (when acceded on mobile at least) :neutral:

Thanks. It wasn’t showing for me.

 

[Tkumpathenurple]

Intel hired chief Xbox SoC architect John Sell. Confirmed via LinkedIn.

https://www.fudzilla.com/news/pc-hardware/48873-intel-hires-chief-architect-of-xbox-one

So Intel are looking to get into the TV, TV, TV game?

 

[rokkerkory]

>24mm * >16mm = >384mm2.

Nice job

 

[PSman1700]

What advantages this virtual ram will give us? Is it a way to have less GDDR6 ram, and have the SSD/Optane like SSD as a ram instead?

 

[Panino Manino]

Watching Kanter talking about Wave32 and Wave64 and how it works for backwards compatibility, does this mean that is this first version of RDNA that the consoles will and MUST use?

 

[Pinstripe]

Watching Kanter talking about Wave32 and Wave64 and how it works for backwards compatibility, does this mean that is this first version of RDNA that the consoles will and MUST use?

I'd assume at least the PS5 does for PS4 backward compatibility. XBox Scarlett might end up using RDNA2 with AMD's own RTRT, since backward compatibility there works mostly on a DX12/OS level, like a PC.

 

[3dilettante]

What advantages this virtual ram will give us? Is it a way to have less GDDR6 ram, and have the SSD/Optane like SSD as a ram instead?

Solid state storage is a poor substitute for DRAM. It's faster than a spinning disk, but still orders of magnitude too slow or lacking endurance. Having fast storage can indirectly reduce the need for DRAM capacity by reducing the amount of time it takes to load sufficient data for use in a game's simulation or rendering for a given frame. Streaming data on the fly often relies on loading data sufficiently far in advance of when it is needed. The slower the storage relative to the demand, the further the system must load things in advance. A downside of this is that the more data that is loaded, and the earlier it is loaded, the less certain it is what must be loaded. This means a certain amount of extra data can be pulled in just in case, since the information needed to narrow down the range may not be available in time.

If the storage system is faster, then a more current list of assets can be submitted, and buffered data can be more readily dropped from RAM if it can be safely brought back in time.
Reducing the need for having a lot of data just sitting around "just in case" can reduce the pressure for more DRAM, not that devs couldn't find a use for it if available.

Watching Kanter talking about Wave32 and Wave64 and how it works for backwards compatibility, does this mean that is this first version of RDNA that the consoles will and MUST use?

The presence of the wave64 mode is up to the designers. If they want another version to support it they can decide to have it.
It's possible the first generation RDNA GPUs and the next-gen consoles are in a similar situation to the first generation of GCN GPUs and the current consoles.
Southern Islands was the first GCN, but there were elements to the architecture and surrounding hardware that hadn't moved as far from the prior architecture as they could have. The consoles had Sea Islands, which updated a number of IP levels the first GCN products weren't able to modernize sufficiently.
Since the consoles were in development at the same time as much of the GPU hardware that came out just before or around their launch, they had features nobody new about prior to launch.
The advance disclosure of some features is different this time, but semicustom clients have access to IP from future products and are often developed in parallel with products that won't launch until much later.

 

[bgroovy]

What advantages this virtual ram will give us? Is it a way to have less GDDR6 ram, and have the SSD/Optane like SSD as a ram instead?

Traditionally, it's mostly useful for keeping a program from crashing because it ran out of memory... Using the SSD as virtual memory was a really weird thing to boast about for Scarlett. They might be a bit self conscious about their planned RAM capacity.

 

[turkey]

Solid state storage is a poor substitute for DRAM. It's faster than a spinning disk, but still orders of magnitude too slow or lacking endurance.

This is very true on PC and even more so on current generation consoles.

How if at all do you think it will compare with a good controller and very tight IO path, allowing dependable access to the drive at say 50 to 75% of its theoretical maximum?

It sounds like not only will the drive itself be much faster than current gen HDDs but the whole access path is refined to eliminate bottlenecks that exist on pc say.