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

Status
Not open for further replies.
That's the thing. We can't really use Nvidia's new made up definition (which applies to its own HW & is here to differentiate between their GPUarchs) with hardware from other vendors which most probably won't support the same features in the same way.
When the time comes, with other vendors arriving, I know we will all do our best for labelling conventions.
 
I still think it might be a small ssd (like 256GB) with a 2TB hdd for main storage. This continues to allow both internal and external expansion.

The main storage is normally compressed in LZ which is not exactly easy to decompress at many GB/s. So they could have an uncompressed ssd storage, with some algo for automatic space management, or an sdk for devs to manage what they need to be copied there. Or otherwise go wild on a very fast fixed function LZ decoder.

Pro or high end models can still have a big ssd and no hdd. And more importantly devs get a guaranteed performance from the ssd, regardless of which drive is replaced or added externally. It allows users to change the internal drive for a 5TB drive inexpensively. (if they have a slot with 15mm clearance this time).
 
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.
 
If they weren't headed this way already, I hope MS are also going down the fast storage route. Fast load times / world streaming as a minimum spec on next gen would be ace.
 
I still think it might be a small ssd (like 256GB) with a 2TB hdd for main storage. This continues to allow both internal and external expansion.

The main storage is normally compressed in LZ which is not exactly easy to decompress at many GB/s. So they could have an uncompressed ssd storage, with some algo for automatic space management, or an sdk for devs to manage what they need to be copied there. Or otherwise go wild on a very fast fixed function LZ decoder.

Pro or high end models can still have a big ssd and no hdd. And more importantly devs get a guaranteed performance from the ssd, regardless of which drive is replaced or added externally. It allows users to change the internal drive for a 5TB drive inexpensively. (if they have a slot with 15mm clearance this time).

This sounds like a messy design, especially when smaller units are likely planned for the future.

I wonder if this has more to do with AMD's Infinity Fabric (or a variation of it) being integrated into the systems overall I/O communications. I believe (not a 100% sure) Sony wants better or more efficient coherence between all their I/O related components and devices, especially the relations dealing with VR-gaming (PSVR2). Anyone (like myself) whom are into VR-gaming can tell you, slow streaming assets can break immersion really quickly.
 
SSDs are hitting 3.5 GB/s in the PC space.

Whatever Sony are proposing, it's going to be interesting. Need to be able to play with automatic handling of external HDD, and now can outperform anything in SSD land.

Wonder if a three level solution (HDD -> flash -> slow ram cache) could work out. Under some circumstances, perhaps it could be better than simply a larger super fast SSD?
 
Some of us already guessed as to the likely size of the GPU. And it worked out to something like 52 active CU's if you assume a 360mm^2 die size.

I think some argued that you couldn't use 7nm Vega GPU as a 1-to-1 comparison to estimate the apprx CU size because a bunch of hardware features would be stripped out on a Navi/console. But if they are adding dedicated RT hardware to the console GPU's then it may not be that much difference.
 
Some of us already guessed as to the likely size of the GPU. And it worked out to something like 52 active CU's if you assume a 360mm^2 die size.

I think some argued that you couldn't use 7nm Vega GPU as a 1-to-1 comparison to estimate the apprx CU size because a bunch of hardware features would be stripped out on a Navi/console. But if they are adding dedicated RT hardware to the console GPU's then it may not be that much difference.
I doubt it has dedicated RT hardware. It’s probably just RT via compute, which is how everything does RT in DXR except for Turing.

Vega VII is 64 CUs with a 4096 bit HBM2 interface and hardware for INT8 and FP64. All in 330mm^2. Zen 2 chiplet is 70mm^2. 56 CUs to yield 52 may be fairly close.
 
Great news! I'm extremely excited about the SSD bit. Finally.

What about the same-as-PS4 CPU core count, though? I fear that it will hurt PS5 in the long run...
 
Great news! I'm extremely excited about the SSD bit. Finally.

What about the same-as-PS4 CPU core count, though? I fear that it will hurt PS5 in the long run...

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.
 
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.
We don’t know if SMT will be enabled, or if games will have access to it.
 
We don’t know if SMT will be enabled, or if games will have access to it.

True, but the base hardware is capable of it and the 360's CPU used it, so it's not exactly unprecedented. Just having the capability is an indication of the processing units having an increased capacity for work relative to Jaguar.
 
Status
Not open for further replies.
Back
Top