Velocity Architecture vs PS5 SSD

Now that both consoles are out, and there is a fair few games out on them, some next gen only, some multigen, and some back comp, we should get an idea of how these two competing technologies compare in real world results.

The theory told us that the PS5 SSD was more than twice as fast as the Xbox SSD, and there was even some talk that with Ooodle texture compression the PS5 could do up to 17gbs.

On the other hand Microsoft has said the Xbox SSD can do 2.4gbs uncompressed, and 4.8 compressed. Microsoft has expressed a number of time that this is sustained throughput, and does not drop when it gets hot. The other time MS highlighted something similar was when they released the XSX specs, and they kept hammering home that the APU speeds were consistent, and not boost, knowing all along that Sony had variable clocks. So are we sure that the PS5 SSDs speeds are also sustained regardless of heat, or could MS be alluding to the fact that the PS5 SSD isn’t sustained? Sony hasn’t yet stated their SSD speeds are constant, so maybe they do drop.

The main difference between the two is that Sony went straight for a hardware solution, while MS innovated the Velocity Architecture to boost performance. Microsoft has hinted that the actual performance of the Xbox SSD may well be higher.

So the real world results have been far from what we expected from the specs. The PS 5 has far from dominated loading times, and infact the XSX has beaten the PS 5 in a number of cases. At a minimum we can take away that the PS5 isn’t showing a 2 x SSD speed over the XSX.

You can’t use the argument that these games arnt coded for the new hardware, because we have next gen only games also mirroring these results.

So what’s up? Is the Velocity Architecture acting as a multiplier for the SSD? Is Sony’s SSD not as quick as they have stated (Sony has a long history of putting out bullshit specs on their consoles)?

I have also thought about this

My guess the lack of a clear ps5 advantage boils down to two things

  • velocity architecture is obviously a great design
  • sony probably doesnt give us every detail. Like you mentioned, i suspect they give some sort of peak performance in the specs for their SSD (like they do with their gpu/cpu). I suspect their ssd performance varies and is not 100% constant peak speed

There are a lot of reasons for this result:

  • it is the start of the gen => faster hardware is useless if software isn’t fast enough
  • loading assets is only one part of the asset pipeline and in some cases maybe not the limiting one => halving the part making up 20% of the complete time is not a significant improvement
  • Sony focused on the SSD because they have a obvious numerical advantage there and fanboys in the internet quickly embraced the narrative => reality is more nuanced

I expect some interesting results in the coming years. In some cases maybe the Series S will load the fastest because 1080p assets are way smaller than 4k ones.

3 Likes

I think a real comparison can be done the moment we see pure next gen versions of multiplatform games. At the moment there is too much legacy support that may or may not hinder the actual capabilities of what the new architectures are about beyond the raw speed of the ssd storage devices.

1 Like

But arnt we just dealing with the speed data transfer to the RAM? For the initial boot up anyway. We are looking at how long its taking the RAM to fill up, and then the RAM to the GPU. I wonder if the superior RAM bandwidth is helping the XSX compete. Maybe the PS5s RAM is a bit of a bottleneck compared to the XSX?

It is too early to tell. We don’t have fully next gen yet on Xbox for example. And no purely next gen multiplatform games either.

1 Like

When sampler feedback streaming is enabled and adopted it’d going to be interesting to see what the final head to head will be. Though that’s mainly a texture related boost, so might not factor into loading.

1 Like

I honestly believe that the speed of PS5 is well past the point of diminishing returns.

2 Likes

next gen engines this is when you see true next gen so hellblade 2 for xbox and for PS maybe ND New game or days gone devs game

This thread is a year or two too early. Data structures need to be designed and engines need to be optimized for the IO features of both systems. There is absolutely nothing to go on now.

4 Likes

Well, we do know that the PS5 SSD speeds are not sustained (as opposed to the Xbox’s SSD which is sustained). What we’re seeing here is that the PS5’s SSD “advantage” will probably not amount to any meaningful in practice.

Prior to the consoles releasing I was in the camp that said the PS5 would be twice as fast in loading games and rebooting once you were killed than the XSX, but that the PS5s SSD wouldn’t give it any graphical advantage. If anything the PS5s RAM bandwidth would be a bottleneck for it.

When it comes to loading up a level there is more to it than loading in assets. A lot of times there’s the creation of data structures and such, and that’s CPU intensive. Note how every single game uses the exact same amount of RAM allocated for games and how the all differ in loading times. Fighting games are quick and straightforward while RPGs take a good while.

Let’s just assume that the PS5 reads at 4 GBps. If it were as simple as loading in assets, every game should take 4 seconds to load. That’s obviously not the case.

But streaming in assets definitely benefits from a fast SSD. Because of levels loaded, all the data structures are created, and now you’re just loading in files. Ratchet & Clank probably loads in GBs of assets to change the level in 2-3 seconds, something not possible on the XSX.

1 Like

On point.

We dunno how either will really perform in proper next gen games yet. I personally expect XVA to be able to slightly outperform the PS5 solution, but also kinda feel that both solutions swamp out the potential bottleneck on I/O so we might not even be positioned to compare beyond loading times.

We will at least see games shown this year that will have engines built around this tech and devs will surely make claims in interviews about no load times or whatnot. But yeah we probably won’t be playing any major games in 2021 that have this stuff implemented unless we are super lucky and MS lets us stream stage demos of this stuff or releases a tech demo kinda thing. Phil suggested they were looking into something like that last year iirc (don’t remember if it was specified as normal demos or stage demos but streaming was noted).

1 Like

Too early for this. Plus we always knew it would be the first party PS5 games that would benefit most from this.

2 Likes

Are you sure? If we’re talking about purely raw speeds, I agree, but devs have seen demos of 10GB of data loading into memory in just 2 seconds using direct storage.

The level they load while falling through a portal/rift is an on-rails and fully scripted intermediate level. The playable level you actually end up transitioning to is after WAY more than 2-3 sec of loading.

I checked the demo they showed and it took ~3 sec to transition to the first intermediate level where you can only lean/jump on rails, then after ~5 sec of that you hit another portal (another ~3 sec loading) and after that you again are in an on rails/lean/jump scenario for 5 sec, then another few secs to load new area where you are (again) fully on rails for only 4-5 sec before yet another portal takes you to ANOTHER on rails sections with no real gameplay, then another 3 sec of loading and FINALLY after around 35 seconds you end up in a fully playable scene.

It seems they are really spreading out their loading across all these back to back intermediate scenes.

2 Likes

That’s a good point. On XSX the compressed speeds allow for 9.6 GB+ in 2 sec. That is less time than you’d need for the portal load transition to an intermediate level. So in that time you already have the RAM filled up on both consoles.