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

You said size - and it's around the same size. If you said shape...

Either way let's see how it compares regarding previous console comparisons...there have been some juicy ones before!

 

I'm on board with the 12TF now. I think Phil's statement is stated in such a way as to give him plenty of wiggle room if for some unforeseen issue they can't hit 12 TF in retail units. Even if they have to downclock it, they can still say it's over 2X powerful (which it would be).

Although given that design, I doubt they'll have issues. With that volume, they could fit a very nice cooler into that design. One of the things that has been very disappointing for me with my PS4s is the noise, Xbox has this nailed, and I have no reason to doubt them that this box will be very quiet.

 

Break your back lifting it up.
Huuuge heat sink running the length of it.
One way to cool that APU.

I doubt power will go down from what has been stated. They have no reason to give a figure that they wasn't 100% confident they could hit. Will only stay the same or go up.
But it's not like they've given a definitive figure.

 

it clearly has more volume if you want to be precise.

 

Well we don't know the exact figures yet do we?

 

How do you think someone come up with that mockup?

https://www.resetera.com/threads/xb...project-scarlett.158955/page-46#post-27356948

 

Which says it's an assumption that the figures in OP are correct...so again, do we know the exact figures? That'll be a no.

 

Watching people overheating their brains trying to make this new information fit with their preconceptions is hysterical. Some of these posts will not age well.

 

They have both the controller and ODD to scale from. It will be correct to within +/- 5%. In fact, Microsoft noted the controller is slightly smaller, so any error will be on the negative side.

If you want to be a pedant, go right ahead.

 

On pictures it looks symmetrical long cube, and on those mockups it doesnt. So i would not trust any of it, if I am right and it is "A x A x B" not "A x B x C" sized. That would make him really bad at estimating sizes if he is so blind that he doesnt see it is " long cube"

 

Well as you said, the controller is smaller and with minimal photoshop work I figure it's around 5.5 x 5.5 x 10.5 = 317 compared to 263 of the X.

So yes, it's potentially bigger...by around 20%, but again I reiterate that size doesn't mean anything...the Xbox One was massive at 13.1 x 10.8 x 3.1 = 438...and that didn't have a PSU!

 

it will roll either way. But I appreciate the effort into comprehension of the statements made on XSX. In the end it’s not going to be known until DF gives the full run down of CU and clock speed. Once a that is known this debate is cleared; but until then I think this debate needs a hard pause since it’s going to be about word/article interpretation.

 

12 TFLOP Navi GPU would be quite impressive! Way above what i thought was realistic.

 

That's very unrepresentative. It's twice the width yet half the depth. Go by volume as others have calculated - it's basically 'normal' console size. It looks pretty cool standing upright, but I don't think it's a great fit for AV cabinets. The advantage is a central vortex-type cooling I suppose which should provide better cooling for the size. It worked well with the Mac Pros (which seemingly ditched it to help with serviceability, going back to huge boxes where professionals don't care about size versus functionality and raw perofrmance).

 

Could you please link this quote?

 

Now I understand why phil said its form follows its function. You can use the console as a stool to sit on as you play!
I hope that doesn't affect the cooling though.
Trully revolutionary.

 

For me it will fit nicely into the void behind my TV!

 

I don't think expecting 10% more efficiency in RDNA2 is realistic. It has never been the case is the recent few years of microarchitecture changes for AMD GPUs. Someone once posted here a great link to an analysis of the efficiency changes in the GCN generations.

The RDNA did bring much higher performance per TF than GCN, but RDNA delivers less TFs per transistor. Since we're talking 7nm to 7nm+ differences here, we should be comparing performance per number of transistors.

https://images.anandtech.com/doci/15206/Radeon RX 5500 XT Press Deck_03_575px.jpg

This image shows the RX 5500 has 1.12x the performance of RX 480. The RX 480 has 5.7 billion transistors and the RX 5500 has 6.4 billion, or 1.12x more.

So I don't think RDNA is amazingly more efficient than GCN, as many people seem to think. People seem to reach that conclusion because they want to keep looking at the meaningless TF numbers

 

Since the Series X is shown to be cooling focused, and the rumored high clock speeds are looking to be true, I've been thinking:

Could it be that building high clock speeds console is not the most cost effective way to achieve the desired performance?

I haven't done any math, this is just speculation. Could it be that these high clockspeed consoles could result in, for instance, an average of +$70 in power consumption over the console lifetime compared to going low clockspeeds and bigger die, but the bigger die would result in a $50 more expensive console?

So the console makers are choosing the less cost effective way because it's not them who will pay for the energy consumption, and they will benefit a lot by being able to sell a cheaper console or take less loss per console.

Is that plausible or are additional energy costs negligible compared to bigger die size costs?

 

No, people reach the conclusion that RDNA is more efficient than GCN by using the more (as you point out) relevant metric performance/transistor.
Radeon Vega 64: 12.5billion
Radeon 5700 XT: 10.3billion
At similar bandwidth, Navi still outperforms Vega 64 by (ballpark) 20%. Even if we normalize clocks, Navi still has a substantial performance/transistor advantage.

It is a game of definitions to try to pin an exact number on the difference. RDNA-console can probably tweak this a bit further, if nothing else because they can drop everything in the architecture that is motivated by computational loads or anything else that might be largely irrelevant in the consoles. (This is not likely to bring huge advances though, the biggest benefit probably lies in having the target for the architectural decisions being exactly defined.)