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

Status
Not open for further replies.
My opinion? Cerny is as far from a PR guru as you can find. He really is a matter of fact kind of guy, as demonstrated at the PS4 reveal where he simply explained what the system had inside, showed off some rather underwhelming demos (compared to what we actually got from some games) and off he went. As opposed to Sony’s older attitudes, which need no further mention.

I’d be very surprised if Wired explicitly asked him whether PS5 supports RT, and at the same time I don’t think Cerny would have mentioned the feature - and with this much effort trying to explain it - if he thinks he can just get away with, ultimately, it not being used in reality.

It follows that he would expect RT being accelerated somehow in the platform and that devs will be using it in some shape or form.
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.
 
Are these true dialogues??
No, it illustrates what could have happened and how people are reading between the lines given a lack of explicit info.

I’d be very surprised if Wired explicitly asked him whether PS5 supports RT...
hmm, why? It's a buzz word and a clear next-gen tech. I would expect tech articles to ask about it.

...and at the same time I don’t think Cerny would have mentioned the feature - and with this much effort trying to explain 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.
 
The CPU and SSD are going to bring next-gen world possibilities.

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.

daefuievwaaeq8t-jpg-large-jpg.14183
 
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.
 
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.

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.
 
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.
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.
 
Not all CPU cores are created equal and the Zen 2 cores going into PS5 have SMT, so can run up to 16 threads. There is no comparison between the two, the CPU portion of the PS5 is going to be a huge leap over the one in the PS4.
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.
 
I'm expecting a hybrid system, perhaps something like Apple's 'Fusion' drive approach. If they're doing something ridiculously fast as claimed ("Cerny claims that it has a raw bandwidth higher than any SSD available for PCs") then it's going to be small in size. Or expect this box to be a lot more expensive.

Of course it is possible that Sony are using some genuinely new tech with the promise of speed, reliability and size. Actually, you can probably scratch reliability or rather the number of write-cycles. There are lots of promising pre-production solid state storage technologies whose only drawback is that it's limited to < 1,000 write-cycles which probably wouldn't be an issue in a console.

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...

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

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

daefuievwaaeq8t-jpg-large-jpg.14183


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.
 
It's difficult to achieve speeds claimed by cerny with small conventional ssd. Has to be some tricks, big ssd or intel optane like ssd.

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

AMD's solution is simple from a hardware standpoint. The company connected two M.2 slots (PCIe 3.0 x4) directly to the GPU with a PEX8747 PCIe bridge chip, which creates a "private" PCIe bus and gives the GPU complete p2p control of the SSDs (and not the CPU/OS). AMD claims this reduces latency up to 10x and reduces PCIe bus contention.
Either the GPU can have its own private storage volume that is hidden from the OS, or it can share the SSD with the operating system, which adds several interesting possibilities for sharing the same data set with the CPU for additional processing. The current developer kit allows the operating system to see the storage pool, but AMD indicated that developers could use it as a private or shared pool in the future. AMD uses two 512GB Samsung 950 PRO SSDs in a RAID 0 configuration, which provides the bandwidth needed to use the storage as a 1TB extended frame buffer.
 
Status
Not open for further replies.
Back
Top