Next Generation Hardware Speculation with a Technical Spin [pre E3 2019]

To be fair to the argument, there are ways to do RT light as well.
There's fair argument on whether or not there is hardware fixed function support or without. I generally think there will be, because I'm not sure yet if we're at the point in time in which we can go without it just yet.
I have a 'strong' hunch that AMD is working on an accelerated solution.

 

No, it illustrates what could have happened and how people are reading between the lines given a lack of explicit info.

hmm, why? It's a buzz word and a clear next-gen tech. I would expect tech articles to ask about it.

the only effort we have is in talking about using raytracing for non-graphics tasks, which is a reasonable fit for compute.

Personally I think there will be some form of hardware design to accelerate raytracing, but his remarks aren't evidence for that by any stretch. Without explicit "there is RT hardware in PS5", the current "the GPU will support raytracing" is suitably ambiguous that if PS5 doesn't come with any RT hardware, Wired's remark won't be a lie.

 

About ps5 pricepoint

 

would be even more impressive if they aimed for less than 4K and maybe Checkerboard 4K at the maximum. Ridiculous resolutions will hold back next-gen. Not liking that fact even if I'll have to upgrade to a modern screen at some point. Even if I owned a 4K screen I'd take innovative worlds over native resolutions all day because I don't think it's important.

 

Remembering these were stated as preliminary specs, and subject to change:
Zen - check
8 cores - check
GPU Navi Based - check
14.2 Tflops - yet to confirm
32 GB Gddr6 - yet to confirm
1 TB SSD - check
All SEW games started after April 1, 2018, exclusive to project epsylon - yet to confirm, but possibly true
PS4/PS4 Pro games compatible - check

It can be pure coincidence, since in my oppinion, with the exception of the 3D sound card, and improvements on the SSD performance, there was really nothing new in Cerny’s statement, so these were predictable statements for a informed person.

 

Meaning, the magical $399 leprechaun is going away.

 

There are steps between having full "RT cores" as defined by Nvidia and being no better suited for RT than PS4/Pro. Turing without RT cores is still significantly faster than Pascal for ray tracing.

Navi will no doubt be better than Vega or Polaris regardless of the degree to which there is dedicated hardware.

 

I'm sure it will support RT at least the same way it supports 8k.

 

My main argument is about substance, really. Regardless of how it’s done, to me the fact that he mentioned it together with the main big features of the system is that it will be ‘usable’ and he would expect it to be used.

It sounds different from throwing 8K out there, which we all know is just not going to happen in reality, expect maybe some special cases.

Also, I reiterate my utter disdain for the concept of pushing absolutely ridiculous resolutions such as 8K. What a total waste of resources.

 

In AMDs favour, I think a pure compute based RT challenge, with the emphasis on async compute to do the BVH update and the intersection calculations, AMD would
Be fairly competitive against Turing non RT Core.

 

I hope that the CPU will be a big leap (3+x) even with the same amount of cores

 

I was referring to the rumor of next Xbox having 12 cores with most likely same CPU architecture as PS5's. Should have mentioned it, sorry.

 

This seems to be a popular prediction. I think the opposite is true. If the NAND Flash price reduction curve stays on track, by late 2020 the cheapest way to provide a storage device with "faster than anything for PC" speeds is just a simple PCIe 4.0 NVMe controller attached to at least a terabyte of cheapest available flash. Right now, the spot price of a terabyte of flash is ~$100, and it's estimated to be cut in half by this time next year. Large, preplanned purchases pay less than spot price. It would be very hard to fit any kind of tiered storage at a price point below this, and this would also provide the best cost reductions over time.

I predict it will only be superfast at reads, btw. Writing NAND flash takes a lot more power than reading it, and cheapest grades of flash are naturally much slower at it. If they just arbitrarily limit their write speeds, they can reduce the TDP of the drive and save that power for something more useful. It's not like it's needed -- with just a single pool of storage, the maximum write speed they need is the speed of their network interface/optical drive, whichever is faster.

An interesting corollary is that a very fast mass storage pool reduces the size pressure on the ram. If I had to pick between having much faster ram, but say only 12-16GB of it, vs having a main ram pool of a more pedestrian speed but 24-32GB, if my storage device was HDD I'd go for for more ram every time, whereas with a fast enough storage that I can stream >4 gigabytes into ram every second, I'd probably prefer the faster ram. That would also be very good for RT, btw...

It absolutely will. Jaguar -> Zen2 will be 3-4x improvement, even with relatively low clocks on the Zen.

 

That fake confidential material is so terribly made. But it does sounds similar to the supposed Pastebin/Reddit PS5 leaked specs from December 2018.

 

I think fast writes to SSD would enable stuff too though. A system that could support that would be better than one without, allowing for more dynamic, persistent worlds and working data. eg. procedural content creation could be operated in a lower priority thread and the results accumulated in storage for use without impacting read speeds and general game streaming.

 

Or the speed also comes from having 20 GB of ram for the game instead of 5GB.

 

Well that’s not really a ‘super cool new technology’ as described in the interview. It’s just more RAM.

 

It is for Console-land...

 

I'd say 16GB GDDR6 is likely since they bet the farm on fast NAND. Feeding 16GB or more with a 5400rpm spindle wasn't going to cut it. Hoping for cheap DDR4 side pool...

 

The SSD part might be a custom implementation of what AMD did with the Radeon Pro SSG. I can't post links but the 2016 article from Toms "Examining AMD Radeon Pro SSG: How NAND Changes The GPU Game" claimed a 4.59 GBps throughput on that card and they could probably squeeze out a bit more performance without doing anything more exotic than was done for the SSG.