Current Generation Hardware Speculation with a Technical Spin [post GDC 2020] [XBSX, PS5]

Status
Not open for further replies.
Cerny mentioned a more general freeing up of level design and asset storage. Levels could be designed knowing they didn't need to implement barriers or sequences catering to MB/s of bandwidth, and asset duplication would become much less necessary. I'm still not clear on where the latency figures are for this upcoming gen, which might influence how aggressive developers can be on buffering.

I'm just wondering if Sony's very fast version is necessary or if series x storage is good enough and if there was a specific reason they needed 5GBps.
 
I mainly only play first party games on consoles and I have a feeling they would support the audio, they bloody well better.
I'm confident first party will support it. Third party otoh isn't likely to put too much into it. MS tried that with Kinect, the implementations were mostly gimmicks and superficial. Given how few people have proper speaker systems in their homes,it isn't worth the trouble.
 
I'm just wondering if Sony's very fast version is necessary or if series x storage is good enough and if there was a specific reason they needed 5GBps.
I don't know if Cerny gave a specific reason why they chose the rate they settled on.
On the other hand, Microsoft has an API and hardware that tries to track what assets are in demand in hopes of more intelligently selecting what needs to load, so maybe it's possible to get similar results with less raw bandwidth?
 
The memory speed hardware specs actually seem to be really close, if you average the total bandwidth of the 16GB on Xbox you get 476GB/s vs the 448GB/s on the PS5. So it’s going to be very interesting to learn what the advantages of doing this end up being.
 
10 GB seems pretty generous for the high-speed partition. I presume that Microsoft's profiling indicated that there would be enough data that was not the critical footprint of the most demanding processes, at least for this stage in the generation's life cycle. If developers are able to move the majority of their bandwidth-limited work into that window, the average tilts significantly towards the maximum. At least until there are engines that aren't cross-gen and have been built to really push the architectures, it seems like they can generally do better than the average.
 
I'm really wondering why the SSD is limited to 2.4GB/s raw transfer speed. Don't they have a 4 lane PCIe4 connection or don't they use the max. performance because of expected heat/performance stability? Or do they think there is no real point to go beyond 2.4GB/s in practice?
 
Last edited:
The memory speed hardware specs actually seem to be really close, if you average the total bandwidth of the 16GB on Xbox you get 476GB/s vs the 448GB/s on the PS5. So it’s going to be very interesting to learn what the advantages of doing this end up being.

Averaging makes no sense here because it's not really "shared" bandwidth. They will just put the OS and app/certain data into the slower memory area while the performance relevant data will be in the 10TB memory space.
 
I'm really wondering why the SSD is limited to 2.4GB/s raw transfer speed. Don't they have a 4 lane PCIe4 connection or did they don't they use the max. performance because of expected heat/performance stability? Or do they think there is no real point to go beyond 2.4GB/s in practice?
They gave that value as a guarantee, subject to whatever preconditions that entails. Compression can be more or less effective, and conflicts or SSD maintenance issues could impact accesses. Weak SSD implementations can also falter in various corner cases like being full or if a long period of time has passed since data was written.
 
Did we get any details about the inner workings of the RT acceleration system? The absence of such info among the flood of info covering other parts is not a good sign IMO.
On PS5? He described exactly what the AMD RT patent had in it. Shaders send message to Intesection unit in the Compute Unit (it's integrated into the TMU complex as per patent even when Cerny just said it's in the CU) to do it's stuff and are free to work on other stuff meanwhile.
 
The memory speed hardware specs actually seem to be really close, if you average the total bandwidth of the 16GB on Xbox you get 476GB/s vs the 448GB/s on the PS5. So it’s going to be very interesting to learn what the advantages of doing this end up being.

You can't just average it like that, same way you don't average main and video memory speeds on PC. @3dilettante explained it well.

Also according to MS:

"Memory performance is asymmetrical - it's not something we could have done with the PC," explains Andrew Goossen "10 gigabytes of physical memory [runs at] 560GB/s. We call this GPU optimal memory. Six gigabytes [runs at] 336GB/s. We call this standard memory. GPU optimal and standard offer identical performance for CPU audio and file IO. The only hardware component that sees a difference in the GPU."
 
You can't just average it like that, same way you don't average main and video memory speeds on PC. @3dilettante explained it well.

Also according to MS:

"Memory performance is asymmetrical - it's not something we could have done with the PC," explains Andrew Goossen "10 gigabytes of physical memory [runs at] 560GB/s. We call this GPU optimal memory. Six gigabytes [runs at] 336GB/s. We call this standard memory. GPU optimal and standard offer identical performance for CPU audio and file IO. The only hardware component that sees a difference in the GPU."
MS memory is actually an elegant design, they are using one bus but have memory running at different speeds. It might create some additional complexity for developers but then again maybe there's something MS can provide tool wise, given the SSD speed it should be feasible.
 
MS memory is actually an elegant design, they are using one bus but have memory running at different speeds. It might create some additional complexity for developers but then again maybe there's something MS can provide tool wise, given the SSD speed it should be feasible.

Its not elegant its a cost cutting design, and whilst it is a smart way to drive down costs theres no performance benefit i can see to doing this.
 
On the other hand, Microsoft has an API and hardware that tries to track what assets are in demand in hopes of more intelligently selecting what needs to load, so maybe it's possible to get similar results with less raw bandwidth?

They gave that value as a guarantee, subject to whatever preconditions that entails. Compression can be more or less effective, and conflicts or SSD maintenance issues could impact accesses. Weak SSD implementations can also falter in various corner cases like being full or if a long period of time has passed since data was written.

Seems that the xsx is trying very much on the consistent side of things, not only for clocks but also ssd speeds.
 
Its not elegant its a cost cutting design, and whilst it is a smart way to drive down costs theres no performance benefit i can see to doing this.
They saved cost by not having two buses and two different memory chips to save cost there as well. Of course a single pool of memory would be better but it's a very good cost savings approach.
 
On the other hand, Microsoft has an API and hardware that tries to track what assets are in demand in hopes of more intelligently selecting what needs to load, so maybe it's possible to get similar results with less raw bandwidth?

Reading the whitepaper on MSFT SFS it looks like it's only good for PRT-like stuff (essentially whole mips of textures only)
And the cognitive load on developer is pretty high.
 
If asymmetrical bandwidth has as much value as some suggested I'd expect it to be everywhere. There's certainly an obvious reason most products don't design it that way.
 
I think the Sony target was: reload 2Gb while player turns camera 180. It was kind of hinted in the presentation, not said, though.
Assuming the Player does not play with high control Stick Sensitivity, or the game offers mouse Support. I do not think that was a good example.
 
If asymmetrical bandwidth has as much value as some suggested I'd expect it to be everywhere. There's certainly an obvious reason most products don't design it that way.
it has only one value - cost reduction 16gb vs 20gb
 
Status
Not open for further replies.
Back
Top