Next-Generation NVMe SSD and I/O Technology [PC, PS5, XBSX|S]

[Globalisateur]

Worth noting that the PC version features the same 3D audio and requires no hardware based audio unit to achieve it.

Nothing is free. 3D audio on PC has a CPU cost.

 

[I'mDudditz!]

Nothing. You. Posted. Refutes. What. He. Said.

Reread that a few times.

You've been childish throughout this entire exchange. Blocked.

 

[pjbliverpool]

No.

What exactly are trying to demonstrate with these videos? We already have the data points to support my post from Alex's Digital Foundry analysis. Specifically this:

He shows clearly that when all other factors are controlled, a much slower CPU and SSD make virtually no difference (to a point), and he also notes the same for GPU speed and PCIe bandwidth.

The video you posted falls in line with everything else that's been measured, i.e. ~34-37s for the NVMe PC load regardless of specs which impact the overall time in an incredibly minor way compared to the actual performance difference in those specs. Yes the PS5 is marginally faster at around 33 seconds but this is absolutely around the same speed in the context of the massive variation in hardware capabilities that are being tested.

Further reinforcing the point, here we have here a Ryzen 2700 + GTX 1060 with a 3.5GBs NVMe falling within that same load time window:

It's indisputable that the PC loading times are being held up by something other than the performance of the individual components (although they can influence the speed in a very minor way down to a certain threshold where the hardware genuinely starts to have a linear impact on load speed as per Alex's single core clock speed test above). And the obvious culprit for such a limiting factor would be a code limitation which logically would absolutely have to exist for the sake of future hardware. And also logically, would likely be capped at the PS5's sequence speed if for no other reason than to give the same artistic presentation and not require any changes to the animation.

Ok? That's great. What does PC have to do with anything I said about PS5 3D audio implementation? Seems like you want to inject PS5 vs PC comparisons at random here.

You mean like every one of your posts since you started posting here? The point I was making there is that the Tempest engine was being touted as a differentiating hardware advantage on the PS5 side. I think it's relevant in that case to point out that the same can be achieved without the Tempest engine. The mention of PC there is because that's the real world example, but it would likely carry over to Xbox.

 

[Remij]

You've been childish throughout this entire exchange. Blocked.

No loss whatsoever. Bye.

 

[I'mDudditz!]

He shows clearly that when all other factors are controlled, a much slower CPU and SSD make virtually no difference (to a point), and he also notes the same for GPU speed and PCIe bandwidth.

The video you posted falls in line with everything else that's been measured, i.e. ~34-37s for the NVMe PC load regardless of specs which impact the overall time in an incredibly minor way compared to the actual performance difference in those specs. Yes the PS5 is marginally faster at around 33 seconds but this is absolutely around the same speed in the contect of the massive variation in hardware capabilities that are being tested.

Further reinforcing the point, here we have here a Ryzen 2700 + GTX 1060 with a 2.5GBs NVMe falling within that same load time window:

You can’t simply compare the time it takes the portal sequence to complete without analyzing the underlying performance. PS5 manages to load required data from SSD to uncompressed format in memory while maintaining performance. The midrange and even upper midrange are not able to do that. You see performance tanking 50% because the data isn't loaded in time to generate subsequent frames. That is why I've been saying from the start the SSD isn't the most consequential aspect of PS5 i/o. The decompression hardware is much more important in this case due to the relatively small ssd bandwidth needed to get game data off the drive. Completely different from the next step of decompressing for gpu ready format. 33 vs 37 seconds isn't a big deal when booting into a game, but it is a major deal when we're discussing in game sequences and calculating frame time in milliseconds.

 

[davis.anthony]

You can’t simply compare the time it takes the portal sequence to complete without analyzing the underlying performance. PS5 manages to load required data from SSD to uncompressed format in memory while maintaining performance.

If you watch the Digital Foundry videos you will see that PS5 drops frames and has frame time spikes during the portal transitions.

So performance isn't as smooth as you're implying.

The midrange and even upper midrange are not able to do that.

Neither is PS5.

You see performance tanking 50% because the data isn't loaded in time to generate subsequent frames.

I get frame drops on my 4070ti at 100fps+

If I cap the game to 60fps the drops are at PS5 level.

That is why I've been saying from the start the SSD isn't the most consequential aspect of PS5 i/o. The decompression hardware is much more important in this case due to the relatively small ssd bandwidth needed to get game data off the drive.

Can you clarify 'SSD bandwidth'

Completely different from the next step of decompressing for gpu ready format. 33 vs 37 seconds isn't a big deal when booting into a game, but it is a major deal when we're discussing in game sequences and calculating frame time in milliseconds.

What if you're loading and doing more in those extra 4 seconds?

With the extra detail in the BVH and higher LOD's my PC certainly has more work to load and process than PS5 does during transitions.

So is 4 seconds extra loading not acceptable for the extra work being done?

 

[I'mDudditz!]

If you watch the Digital Foundry videos you will see that PS5 drops frames and has frame time spikes during the portal transitions.

So performance isn't as smooth as you're implying.

These are not performance dips, they are duplicate frames for image reconstruction at sub 4k. See below. Notice how the native 4k mode doesn't have these duplicate frames or related frame time spikes as often.

If you decide to contend can you please attach some evidence? You have a tendency to throw out claims with no accompanying evidence.

Neither is PS5.

No, the PS5 does not suffer performance dips. See above.

 

[pjbliverpool]

You can’t simply compare the time it takes the portal sequence to complete without analyzing the underlying performance. PS5 manages to load required data from SSD to uncompressed format in memory while maintaining performance. The midrange and even upper midrange are not able to do that. You see performance tanking 50% because the data isn't loaded in time to generate subsequent frames. That is why I've been saying from the start the SSD isn't the most consequential aspect of PS5 i/o. The decompression hardware is much more important in this case due to the relatively small ssd bandwidth needed to get game data off the drive. Completely different from the next step of decompressing for gpu ready format. 33 vs 37 seconds isn't a big deal when booting into a game, but it is a major deal when we're discussing in game sequences and calculating frame time in milliseconds.

Some systems can see major frame time spikes and stutters during that sequence for sure. But there are other factors at play here:

• VRAM limitations. The portal transition scene will place a heavy load on VRAM and on cards with 8GB or less which we know are limited in this game, they are likely running into issues there. I suspect the vast majority of the most serious stuttering we see in that sequence relates to that.
• The PC does seem to have some minor shader compilation stutter during that sequence which may have an impact on some of those video's if it's the first run through. The following video demonstrates both that, and how on a system that is not VRAM constrained (3080Ti 12GB) the sequence can be perfectly smooth at PS5 like settings (actually with a much higher average framerate and lows. Large frame rate swings are to be expected of course given the different environments on display.

 

[iroboto]

Ok but when discussing Nanite, it's generally understood we're analyzing it's capabilities with subpixel geometry rendering.

I’m not sure I agree with this fully. Nanite’s biggest advantage is being able to render pixel triangles, but that doesn’t necessarily mean the whole screen is completely made of it. That would be dependent on the game title, so its performance outside greater than 6pm triangles still matter.

Some major considerations here: a great deal of many games have triangles greater than 6px because anything smaller would choke the pipeline. So the fact that we’re seeing Nanite choose the 3D pipeline for larger triangles still shows how fast that pipeline is.

Mesh Shaders are still absolutely important and will play a role for games in the future.

 

[I'mDudditz!]

Some systems can see major frame time spikes and stutters during that sequence for sure. But there are other factors at play here:

• VRAM limitations. The portal transition scene will place a heavy load on VRAM and on cards with 8GB or less which we know are limited in this game, they are likely running into issues there. I suspect the vast majority of the most serious stuttering we see in that sequence relates to that.

Maybe, but I recall seeing GPUs with adequate VRAM falling victim to severe portal loading issues.

• The PC does seem to have some minor shader compilation stutter during that sequence which may have an impact on some of those video's if it's the first run through. The following video demonstrates both that, and how on a system that is not VRAM constrained (3080Ti 12GB) the sequence can be perfectly smooth at PS5 like settings (actually with a much higher average framerate and lows. Large frame rate swings are to be expected of course given the different environments on display.

Maybe this was recorded prior to patch updates but the highest mip textures aren't loading in for the ship deck at the end, similar to Alex original video. Regardless, the frame time activity is what's more interesting to me. The latency can be ignored in instances like the rift sequence because it's on rails and maintenance of high enough frame rate levels make it more imperceptible. What happens when instead you have a sequence where the player is still in primary control and instead of 1.6gb, say 2.6gb is requested? I assume it is perceived much worse, which is why I'm curious about how a demon souls port would play out based on the developer's claims. And the idea of having to make multiple play passes for new areas in a game doesn't sound acceptable to me at all.

 

[I'mDudditz!]

I’m not sure I agree with this fully. Nanite’s biggest advantage is being able to render pixel triangles, but that doesn’t necessarily mean the whole screen is completely made of it. That would be dependent on the game title, so its performance outside greater than 6pm triangles still matter.

Some major considerations here: a great deal of many games have triangles greater than 6px because anything smaller would choke the pipeline. So the fact that we’re seeing Nanite choose the 3D pipeline for larger triangles still shows how fast that pipeline is.

Mesh Shaders are still absolutely important and will play a role for games in the future.

Ok.

 

[iroboto]

Ok.

I think on this one: RE primitive shaders and mesh shaders, it will happen over a longer period of time. Costly to switch game engine front ends especially during a time many companies fell behind during covid.

A bit like compute shaders, it took some time last generation to switch a large portion of those engines over to compute.

I think Gen 2 and Gen 3 games should signal the switch over with mesh shaders being standard in games by next generation. That’s not an indication it’s a hype feature, but I don’t think the adoption will be that fast.

 

[davis.anthony]

These are not performance dips, they are duplicate frames for image reconstruction at sub 4k. See below. Notice how the native 4k mode doesn't have these duplicate frames or related frame time spikes as often.

Your video literally shows a frame time spike in fidelity mode during the portal transition as well as a frame drop.

So the problem is stil there, it's just hidden better because of the low frame rate.

But I didn't think consoles players used 30fps now when a game has 60fps.

If you decide to contend can you please attach some evidence? You have a tendency to throw out claims with no accompanying evidence.

Didn't you recently claim PS5 Pro would easily beat a 7900XT/X in RT?

No, the PS5 does not suffer performance dips. See above.

Again, your video shows frame time spikes, the low frame rate masks it a little better.

I'm sure if I capped my 4070ti to 30fps it would hide everything (And make my eyes bleed at the same time)

 

[Allandor]

Nothing is free. 3D audio on PC has a CPU cost.

At ps360 times it was around 5% of one core. That was the time e.g. the last 3d audio chips died (still have my xfi but no longer in my PC). As more and more cores were available that idled around 3d audio was even less of a problem. Even the consoles had chips to save some CPU resources, but even than they were not really used that much.
Todays 3d audio is not that much different from the possibilities we had in the past. At the end more or less your speakers decide how the audio sounds to you, but the sound channels, reverb, .... was already there for many years now. But it is even harder to make sound better than graphics because for sound we reached a peak long ago and everything that goes beyond that is more a thing of your ears, feeling and sound setup.

And yes, consoles have still extra audio hardware, but that is more or less only used for headset "enhancements". Nothing that really improves the sound calculation for a game.

But wasn't this topic about the SSDs and impossible PC ports (that work quite well)?

 

[I'mDudditz!]

Your video literally shows a frame time spike in fidelity mode during the portal transition as well as a frame drop.

The performance mode drops from 60 to 59. The fidelity mode graph drops from 30 to 29. These are duplicate frames, not i/o or gpu drops.

Why can't you offer any video evidence of any PC stuttering that is limited to single frame drops? I've asked you multiple times now to provide evidence.

 

[davis.anthony]

The performance mode drops from 60 to 59. The fidelity mode graph drops from 30 to 29. These are duplicate frames, not i/o or gpu drops.

Why can't you offer any video evidence of any PC stuttering that is limited to single frame drops? I've asked you multiple times now to provide evidence.

There are stutters and frame time spikes on PS5, so it's not perfectly smooth despite its super duper I/O hardware.

You also said the duplicate frames are due to upscaling, which the 30fps mode doesn't do which is why they're not present in that mode, but now you're saying they exist in the 30fps mode? Make your mind up please.

And you claim you've asked me multiple times for evidence of single frame drops on PC.

Can you show me these multiple requests?

I've never even said they don't or do exist on PC anyway, so it seems to merely be a desperate attempt to go off a tangent which is a common tactic of yours and try and formulate some kind of a win.

Stutters in R&C exit on both PS5 and PC, so can you please stop implying they're perfect on PS5 because they're not.

 

[I'mDudditz!]

You're taking this way too personal. It's time for both of us to move on.

 

[function]

What is your definition of reasonably modest?

It's all a bit arbitrary as you indicate, but for me "reasonably modest" would be something from the last two or three years, definitely below enthusiast level, and around where you're targetting bang for buck.

So I'd guess for me something like a Ryzen 3600, 16+ GB of ram (my ancient PC has had 32GB of DDR3 for years and years, but that wasn't typical), and a 2~3 GB/s NVMe drive. All pretty reasonable stuff for a mainstream gamer build, even a couple of years ago.

This will forever be impossible to test, so we just have to go with on paper specs; The PS5 i/o is 130% faster than Series consoles therefore we can only assume Sony first party games would run at least twice as fast on a hypothetical, never-to-happen, xbox port during i/o limited situations.

Well for "so far" (which is what I was talking about) we've seen that a PC with IO performance significantly lower than Series S/X - and with significantly higher overheads - copes very capably. Anything to come along so far would be fine on Series consoles with room to spare. Down the line there may be games where that isn't the case, though I would expect section of a game (any game) where series consoles weren't able to match PS5 would represent a tiny fraction of gameplay. For most situations Series consoles IO solutions are and will remain overkill. But it's fine for X/S and PS5 to have overkill solutions because the transistor costs are small, the power savings are big and the impact when they're delivering is big.

Interestingly, the Series consoles hardware can be connected to two 3.5 GB/s drives at once, and iirc from a Codemasters interview, in early development kits 5+ GB/s SSDs were installed (this was before the decompression block was working). The Series silicon is able to perform beyond what the standard spec SSDs can deliver - I've no doubt they can take a 7.3 GB/s PCIe gen 4 SSD like PS5 and like any other PCIe 4 system. MS did their maths and made a bang for buck choice. Cerny made a different choice with the PS5, but at the end of the day they are all just choices on the same spectrum and for the same reasons.

Well, there is this little game called Spider-Man 2 set to release in a couple of months that should provide some answers.

That will certainly be the next big port that I'm interested in looking at from the SSD/IO perspective. And by that time - hopefully - Nixxes and Nvidia/AMD will have their DS 1.2 game fine tuned (or dropped in favour of DS 1.1).

 

[Flappy Pannus]

That will certainly be the next big port that I'm interested in looking at from the SSD/IO perspective. And by that time - hopefully - Nixxes and Nvidia/AMD will have their DS 1.2 game fine tuned (or dropped in favour of DS 1.1).

We may be on DS 2.0 by then. I'm not expecting that port until 2 years after the PS5 launch, perhaps even later.

 

[Flappy Pannus]

No, the PS5 does not suffer performance dips. See above.

During the portal transitions? It absolutely does. From your own linked video:



These are not simply duplicate frames that you get in cutscene transitions to establish the TAA, you can see there are multiple stutters, and you can feel it. The Native 4K mode is already starting out at a much lower fps so of course it's not going to exhibit the same drops in absolute framerate terms if the the overheard of the I/O is the same as in the performance mode, as it would be.

There are two segments further on in the game that have a similar jump between several worlds in a short span, and they stutter just as much, if not more. Significant? Not really. Better than a high-end PC? Perhaps, they certainly stutter less than on my PC (12400f, 3060). But, they do stutter on PS5, and it's not just a single frame.