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

I’ll add a little tidbit found in an article on digitimes on a new Realtek video chip. Since Digitimes articles are locked behind a paywall after a few days, I’ll cut and paste. The last paragraph is probably the most interesting here.

“Realtek 's RTD2893 TV chip can upgrade a 4K TV platform to a fully functional 8K one, as the chip supports 8K AV1/HEVC/VP9 video decoding, and is able to convert low refresh rate 8K video into 8K 60Hz content using FRC technology, or upscale front-end 4K TV SoC output into 8K 60Hz images.

The RTD2893 chip is also able to secure HDMI 2.1 8K input and output and video bit stream transmission over USB/PCIe, and supports all 8K video processing including HDR formats, making it a very cost-competitive 8K TV solution.

Project Scarlett comes with product specifications that are comparable to Sony's PS5 with regard to CPUs, SSD storage capacity, CD-tracking and support of related 8K technology.”

Digitimes don’t care about the consoles per se, they care about the Taiwanese supply chain businesses. So “comparable” probably shouldn’t be interpreted as “identical”. But still.

 

Do you have a link?

EDIT: find it

https://www.digitimes.com/news/a20190719PD201.html?mod=3&q=Realtek+s+RTD2893&chid=9

 

Would that enable HDMI IN functionality on Next-Box similar to Microsoft Xbox One family? Or do they mean this does the HDCP for HDMI?

 

Or a "free" hardware upscale from 4k to 8k output

Sounds a bit like the ana and then Dana (with hdmi so digital) chips in the 360?

 

Why bother? Why not let the 8K TVs included such chips to upscale any content? Seems inefficient to add cost to every box for a feature a small percentage will use, versus adding that cost only to those who need it (in the TV).

 

Without looking at the concrete specification this looks like a chip for inside the TV. Not for the console.

 

The advantage of having upscaling on the console is that you control quality, and get more predictable latency. A TV can take whatever time it wants for upscaling as long as it keeps the sound in reasonable sync. That’s less than ideal when you want a good control experience.

My point with the post though was that it provided a neutral confirmation of basic similarities in terms of for instance SSD presence and optical drives.

 

https://www.3dmark.com/fs/19933757
19289 points 8.5tf navi (similar as 1850mhz 36cu), gonzalo is probably 40cu 1800mhz

 

But will a Gonzalo running in a console APU configuration and console API fair better than a bork standard desktop config running the same hardware?

 

Another "framework for discussion" kind of post. I came across an interesting graph of Zen 2 frequency v. Power draw. While I think you should dismiss the average system power under load number, the frequency at which you hit the (too pronounced?) knee is probably pretty valid assuming the same process being used. Which in turn would imply at least a 3 GHz CPU clock speed in an upcoming console. Even a Zen 2 core, and the consoles may use an improved version, is pretty damn performant at those frequencies. If you were a PC gamer that wanted twice the frame rate of a 30Hz console game that is CPU limited, you might have to fiddle around a bit with settings since just buying a twice as performant CPU might be impossible for a good long time to come (as such a limitation is unlikely to be when running all cores full out, so throwing more cores at the problem won’t help).

 

Looks like the Zen 2 cores in Rome/Matisse chiplets were mostly designed to operate at up to ~3.2GHz.
Which is probably what we should expect for the new consoles. Approximately twice the 1.6GHz clocks of the original consoles.

 

This is weird. Why would there be a knee? I would expect a standard exponential growth curve

 

There is no physics giving a specifically exponential rise, the mechanisms are different. Yes, the knee looks a bit sharp, but we are not dealing with a continuous function but discrete measurements and rounded numbers. The curve looks OKish with a largely linear function progressively taking off due to a number of reasons. I think it tells a largely correct story.

 

All finfet processes seem to behave in this manner, planar processes used to be more linear in clock increase to volt increase, thats why sandy bridge (32nm planar) was such a loved CPU once Ivy bridge (22nm finfet) came out.

 

So, Flute.
https://www.tomshardware.com/news/amd-flute-soc-xbox-scarlet-benchmark-leak,40000.html

Rename of Gonzallo, or "identical" chip coming to both consoles?

 

Can't be Scarlet as Scarlet has 320bit bus with 10 chips instead of 256bit bus with 16 chips shown in the now deleted benchmark.

I think it's a new version of the Gonzalo devkit. 18gbps chips matches up with the OQA leak.
Though the presence of a HDD is puzzling. If it was a devkit containing only the SOC, then there shouldn't be GDDR6. If it was a devkit like the OQA leak, then we should be seeing SSD in the bench.
OQA didn't mention SATA for HDD.

Puzzling.

 

My opinion - Gonzalo revision. Scarlett video clearly showed 320bit bus and 10 GDDR6 chips, while this one features 16.

With reduction on L3 cache, I am pretty sure we have clear picture of PS5 and design philosophy.

IMO we can expect 16GB of very fast RAM (downclocked 18Gbps chips as PCB leak suggested) on 256bit bus. Cut down desktop Zen2 and Navi XT with RT hardware.

 

I wonder if actual SSD for PS5 is custom made, therefore not exactly easy to pick up in Userbenchmark test. Adding to that, could new consoles also have SATA port for HDD in case users would like to save alot of games outside of SSD from where they are actually played?

Seems so weird these where removed so swiftly. Could someone from QA gotten PCB and tested it, so there is where the leaks came from? I think PCB ~ Flute 18Gbps RAM connection is just too specific to be fake.

 

Or neither.

 

That could be simply an external or secondary HDD somehow plugged to the board. SSD and GPU were hidden, probably on purpose.

Another thing was interesting in the Flute bench: background CPU was at only 1%. That's really odd compared to all others tests I have seen until now, all of them have it above 4% or much more (because they usually run windows 10), particularly odd considering the modest CPU.

From fuzzy memory there was like only 3, 4 or 5 GB available from the 16GB. So the background CPU was super low, but most of the ram was used by some OS.

Also this was called "AMD Flute" so this was not a new model of "AMD Gonzalo": this was not Gonzalo.