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

Sure, I meant before we got the sdk docs. After that it made little sense to speculate (some did, balanced at 14, clocks not final, etc..). We had a pretty wild range even for scorpio, like a triple sku, a streaming stick, $99 odd-less xb1s, and last minute upgrade to 10TF. That was all crazy and it was the official technical thread, and it continued after MS literally announced it was a single sku at 6TF. We had excursions into crazy distribution media before ps4/xb1 launch, but it wasn't anything close to a concensus.

Ram amount always remain a reasonable question, the target can be fluid, since the ram market is unstable and they can increase it as late as 9 months before launch.

Personally I really didn't think we'd go back to >200W and the unveiling of the first navi cards TDP made me go back to 10TF absolute max. But from that point, it took two seconds looking at the SeriesX case that everything falls into place. If they have a high wattage compromise, we still predicted it was possible but less likely for marketing and BOM reasons. The concensus on technical limits still apply.

Once we established rational limits based on the general technical knowledge of posters here, it really narrow down the possibilities. There can still be surprises!

 

That is interesting at the least.

 

Impressive how some users seems unable to generate an unbiased response to everything shown.
The Teraflop Metric even compared between two different archs from AMD, is not really a great measure even more comparing to Turing.
Each arch has it's strengths and weakness on each type of task so they are barely comparable, Turing is probably much better at tessellation than even RDNA, but maybe on other tasks it loses the TF metric will not show this, so software optimizations is far more relevant than any type of "raw" power because it will balance out the hardware differences.

 

I lack the tools to make a measurement from the initial reveal trailer (on mobile) but I think it's fair to say it more like 30cm by 25cm

Unless this is an elf's workshop that hand means it's larger than 16cm deep which is the estimated case outer size.

Up till now all devkits of actual hardware have been pretty close or identical to final hardware cases as far as I know.


The small circular cut out from the motherboard is quite interesting as it's too small for disk drive and that is the usual suspect as they tend to use full pc parts.

Also the APU is very much top left and not Central to allow airflow round. Perhaps if this is a devkit the width and depth differ, and having the chip there meant it's low and a large cooler get floor support and loads of surface area going across the board (upwards) for venting?

It must be related as they showed it and were keen to hide most of it. How does this relate to what we are going to get?

If it was a test rig and never for external use why have a circular cut out? That seems a conscious design element here.

 

Test rigs are not that elaborate. Could be a slim revision of the X?

It's blue. My biggest problem is that it's blue.

 

Cutout for Optical Drive Motor Mechanism to be set in it so it sits at the edge corner of the case, assuming motherboard is a bit offset like PC motherboards are mounted? Or are those skim units usually set with same height all the way across?

 

Isn't that board basically just an engineering board? It's obviously far too large to fit into the case that was shown.

What I'm more interested in is how they are going to arrange things internally.

• Multiple PCBs?
• Single PCB on one side with heatsink fins extending across the rest of the space?
  
• Multiple PCBs mounted at the edges with the central area filled with heatsink fins?
• PCB (components on both sides) mounted centrally with heatsink extending from the center to the edges cooling both sides?
• Heatsink fins running the full height of the enclosure?

There's so many possibilities beyond what Silverstone did with their similarly sized case as the Xbox Series X doesn't need to be assembled by the person buying it.

Regards,
SB

 

It is still on or before the knee.



edit: there are now some claiming that the retail GPU matches the SDK. This was the Dante SDK rumor from a ways back:

DANTE ( XDK )

CPU: Custom AMD Zen 2 8C/16T @ 3.2GHz
GPU: Custom AMD Navi @ 1475MHz
MEMORY: 48GB GDDR6 @ 672GB/s
STORAGE: SSD 4TB NVMe @ 3GB/s

64 CUs is 12.083TF.

double edit: another source claiming the above CU count and TF count correct.

 

Pretty hard to believe that retail XSX would have clocks that low and 64 CU's. That would not make much sense since the sweet spot for current Navi seems to be around 1800Mhz. 64 CU part with 56 active CU's around 1700MHz - 1800MHz could be possible though, so I'm not totally dismissing this but it seems unnecessary expensive setup.

 

VIrtual Texturing has other extreme problems with pop-in, so it uses almost no game except maybe for terrain. Many GamecCompanys threw it out completely.

So Textures are not the Problem , Problem are things i wrote above, so things there must be permanently in the Memory and cannot be streamed or you build and SSD that is in the same way fast then Ram Memory so a SSD with 560 gbyte/s direct connected to the GPU via Interposer.

 

Not if done right, and especially with fast storage. I don't know that any game companies 'threw it out'. Rather, they didn't swap to VT and stuck with the system they knew notably on engines that already worked a certain way and which could work well on consoles. Next gen establishes a new baseline that'll better support VTs, although VT wasn't developed with RT in mind so we'll have to see how it fairs.

 

Or perhaps it is possible that MS is going for low clocks and wide chip if they are really committed to provide as silent console as possible. But that would also provide the opportunity to up clock if necessary depending on what Sony is doing.

 

I get it, but that's the thing. If textures don't need any more memory, than that means every extra Gigabyte will be left for everything else. And with better VT wich SSDs definetly will make possible, devs can free up even more memory for everything else while keeping graphical quality still good.
Keep in mind, currently, textures are the thing that takes up the majority of memory in a typical game. In KZSF, it took about 70% if my memory serves me right. If that can be optimised down to 256Mb with VT like sebbbi sujested, that would be a HUGE win. And that's huge with a soft "h" like "Yoooge" and orange face and all.

 

Credit to DemonCleaner at NeoGAF for his table.

A theoretical 64 CU GPU at 1500 MHz with 10 GDDR6 chips and 3.5GHz Zen 2 APU might draw:

140W GPU
35W/2 memory
50W CPU
TOTAL: 208W
TOTAL AT THE WALL:~260W (18W memory power, 10W SSD, OD, misc, 90% efficient supply)

 

64CU -> 140W ?

 

That’s die only, no memory, assuming an aggressive voltage tailoring to frequency.

 

Did current gen also have the lower tier of VT?

 

PS3 and 360 did with Rage, unless you guys are talking about a different technology than space virtual texturing (megatexture). So any of the id tech 5 engine games would have used virtual texturing. Brink, Wolfenstein: New Order, etc.

 

I'm talking about hardware implementation.

 

Damn, does that even exist?