Middle Generation Console Upgrade Discussion [Scorpio, 4Pro]

If they can figure out a way to avoid memory contention while still using a unified pool of memory, that would render the ESRAM almost redundant. Yes, it'd still have some benefits but it'd lose some of the main benefits that it currently enjoys. No memory contention when using it and faster speed than the main pool of DDR3. Then again if the memory is fast enough (320 GB/s is a massive step up from the XBOs combined bandwidth) the memory contention may not be as much of an issue.

The ESRAM takes up a relatively massive amount of die space. I'm sure Microsoft would prefer to use that for additional CUs. It solved a problem they had during the design of the XBO, but I don't feel it's necessary for the XBO Scorpio.

Regards,
SB

 

I'm not suggesting esram for the scorpio, I'm saying could use hbm. Was just saying why I think a simular set up (smaller fast bit of memory) isn't necessarily evil.
Like the esram you solve contention, have fast speed, large enough size, and don't need to take up die space.
Do we know how they got the 320 GB/s?

IF it's cheaper than 12+GB of gddr5/x, it could be a good solution.

 

I could not believe it. I went and checked and checked again pictures... to finally notice the one memory controller that run all along the short side of the PS4. So you measurement might have been done right.

They had different design goal, MSFT wanted lots of RAM. I would not vouch MSFT approach more efficient without proper data, the best metric should be average gigapixels per seconds by the power consumption.
Now my argument is that discrete parts and NUMA were fitting the requirement of both Sony and MSFT.

I don't think so but I'm pretty sure we won't agree at this point . Dealing with NUMA is pretty much the basis, VRAM are huge enough nowadays and PCI -express offer plenty of bandwidth to move some assets when needed.

It seems MSFT settled on a SOC again. As things work till somebody DOES something differently and it proves more efficient nobody will even reconsider. I think it is worse considering looking at the constrains the consoles manufacturers set on themselves through the use of SOC. I acknowledge that it is a difficult matter to discuss as we can get enough data (it goes for both side of the argument though).
I quit the conversation, as MSFT confirmed it is a SOC, so my speculation and the related argument wrt SOC vs discrete and NUMA vs UMA are getting OT

 

Do you have a link for this?

 

From the Eurogamer interview:

 

If we needed more bandwidth, wouldn't more affordable alternative would be to use GDDR5X?

 

Unfortunately, not off hand. It's in one of the numerous interviews he did. I don't even remember if it was a video interview or a transcribed to web page interview. Had you asked the day I posted it, there's a decent chance I may have remembered which one it was in since I only had time to look at maybe 3-4 things a day related to XBO-T on the day of the conference and for a few days after. And I unfortunately don't current have the time to spend a few hours trying to track it down again. My apologies.

He does state in multiple interviews that the 6 TFLOPs was a number they settled on after getting developer feedback. But I think the blurb about being able to launch a more powerful console this year was one of their options, but for 6 TFLOPs they'd have to wait until 2017 was only in 1 or 2 interviews.

[edit]Thanks Mrcorbo for digging that up.[/edit]

Regards,
SB

 

Now see, I see the ESRAM as being a necessary evil that then turned out to not be necessary. So it was just evil. It's pretty much *the* reason the XBOne was so much less powerful than the PS4.

 

Given the design goals I disagree, I also think that if the ps4 ended up with 4gb which was very possible, the ps4 would have been a very compromised machine overall imo.
As I said I'm in the minority and recognise that.

edit: Better way to put it, if xo ended up with 4gb gddr5, there's no way it could do all the things that it can and is planned to do

 

That is the headline design goal, but he has also said it's up to the devs how they use the power.
They just wanted to make sure they had a machine that could play all xo games at 4k without loss of framerate and graphic quality.

 

agreed, that's the benefit of making it powerful enough to do 4k without optimization. If dev has time then can do other stuff, it almost becomes the minimum really. Not a bad thing
with the way engines are designed though, optimise for xo, you'll reap the benefit on Scorpio also more than likely

 

I just don't see all the fuss with 4K at the moment. We are a long time away from mass 4K adoption and considering the extra processing power required to up Xbox One games from 900p (which they are normally) to 4K would diminish any other extra graphical extras they could add to the games.

I remember everyone jumping on the bandwagon and assuming we would be getting 1080p@60 for every single game on the Xbox One and PS4, but developers decided to up the look of games and use the processing power advantage for other features rather than locking in 1080p and 60fps as standard. This is the same issue here. If 4K is a lock in, then forget anything else.

 

adoption of 4k isn't that far away, plus there's vr. Making a machine that can run current games at 4k, then leaving it up to devs to use it as they see fit where 4k is the quick win isn't a bad approach imo.
Especially when planning ahead a couple years after it's release.
This will become the low end/mass market device when 'next gen' starts

 

A fewthings for me; 1) people are saying 4.5x the power' - it's not a true 4.5x, only the TF - not the CPU, not the bandwidth and not the RAM. 2) XBO already struggles to hit 1080p so yeah, how realistic is it to expect this 'not really 4.5x the power' going to produce 4x the pixels by default as many seem to be suggesting? 3) it seems far too soon for 4K, so few gamers care and those that do are PC gamers sitting right in front of their screens...4k is more next gen, hell this gen can't even get 1080p right.

 

MOst liekly yes, but are AMD memory controller ready? Nvidia seems pretty proud of their GDDR5X memory controller and they dedicated a couple marketing slides on the matter when they launched the 1080.

 

they need a console that can do vr, want one that can do 4k.
its not their mainstream offering, so it only needs to sell within their forecasts.
both vr and 4k may currently be niche but that doesn't mean you don't want to be in that market.

 

In my opinion though, decent 4K adoption is some way off. Look at HD and BR for one. We are just about getting to a level where there are multiple HD tv's around peoples houses and even BR isn't properly adopted. Yeah some houses will have 1 or more BR drives while others may just have one and they never buy BR discs for it and still buy DVDs.

It's good going forward, but going by the specs, Scorpio will just about manage 4K based on the medium settings of current XBox One visuals. I personally think we should get to a point where home consoles are up there with the very high or ultra settings at 1080p before we move up to displaying 4K. There is planning in advance, but if the device is out for 3, 4 or 5 years before we see proper 4K adoption, then it will be ready to be replaced.