I’m a physicist/prof…I quite literally teach data analysis. You have no clue what you are talking about. You aren’t even able to keep straight which claims you are trying to ‘evaluate’. Stahp.
Sony hyped the SSD, and RIGHTFULLY SO! SSD’s are legit game changers this gen. The issue being conflated is Sony’s SSD marketing and R&C’s use of the SSD. MS boasts about the powerful GPU it has (and RIGHTFULLY SO!), but we don’t feign shock when games come out that don’t fully utilize that tech/power because we surely all understand that it takes time to build games around these tech advancements and during a cross gen phase devs aren’t positioned to do that yet. Even on next gen native titles like R&C, devs did not have the time to fully build the game around the SSD (regardless of their marketing suggestions). All Sony said was the game shows off the SSD (and it does!).
In fact, Insomniac even said their game was just the tip of the iceberg wrt SSD use cases! So any suggestion or framing of the DF video as proving they are lying here is nonsense. One noes not lead to the other. And btw, Sony HAS straight up lied numerous times already this gen so don’t imagine I am adverse to calling them out on it. 
There is a mismatch between the actual claims that Sony/Insomniac made and the claims many here are ‘evaluating’ as false. The arguments here are full of nonn-sequiturs*. Ppl here seem eager to react to the DF vid and throw it in the face of Sony fanboys online. And that is fair game, but don’t put the fanboy extrapolations of what Sony actually said on Sony’s shoulders here.
*example: Sony saying SSD’s are revolutionary is true. MS and devs all agree on this. Just because R&C does not require 5.5 GB/s to run does not in any way force us to conclude that these claims (which the whole industry agrees on) is somehow false.
1 Like
That’s the thing though. Sony didn’t hype the SSD. They hyped their SSD. They implied again and again that their superior SSD would be a complete game changer. And I do hold a fair bit of the rabid fan response on Sony. They knew exactly what they were doing by talking such a big game with their SSD.
If all Sony did was hype up SSD tech, then that would be cool. That’s what I was really excited about coming in to the generation. But instead they made sure to focus on how their faster SSD would cause a gap between PS5 vs Series X, not just in the context of last gen vs this gen.
Again, Sony claimed their SSD was revolutionary. They tried to spin it as only they would have the truly massive upgrade, when that couldn’t have been further from the truth. So yeah, calling them out on this is perfectly fine imo.
9 Likes
And MS hyped up their XVA as revolutionary too, even gave their I/O soln a snazzy name!..come on, this isn’t a serious counterpoint, is it?
Their velocity architecture? Surely you don’t think ms hyped that up even nearly as much as Sony did their SSD?
2 Likes
I would take that as a serious counterpoint, yes. XVA hyped up a bit but talked about as something coming over the first few years. Magic SSD hyped up beyond belief the instant it was announced. They’re not really the same.
3 Likes
Are you kidding me? You’ve jumped the bad-faith shark at this point that your counterpoint to the discussion at hand (which was a legitimate criticism of Sony/Insomniac’s hyping of the SSD, backed by data points) is a whataboutism about… checks notes… naming a function of a system’s architecture and putting out a glossary, then discussing the function in podcasts? That’s absolutely fine to do, and given how many people scoffed at Smart Delivery as purely a marketing scheme (when we know for a fact it wasn’t at this point), I’m sure that absurd complaint will age well.
This thread is absolutely embarrassing when two people come in and dictate what is/isn’t valid conversation with their holier-than-thou attitudes and perverse notion of what console warring is because they’re trying to appear balanced. I have a whole tirade saved about the misuse and misrepresentation of data to prove that their isn’t a power delta (which there absolutely is, even without the optimizations of the new GDK/DX12U and hardware/software efficiencies) but at this point, it’s clear that it would fall on two-pairs of deaf ears that will just create circular arguments with strawmen and whataboutisms without addressing any of the actual points people are trying to discuss.
4 Likes
All this does is show that you dunno what ‘whataboutism’ actually refers to. Whatsboutism is deflecting by accusing another of something related to what you did…which is not at all what my post was about. Unlike you guys, I am not accusing Sony of anything nefarious on the back of them simply promoting their product. It can’t be whatboutism when there is nothing wrong with what either entity is doing in that regard. Please stop being so reactionary and engage with the points as they are presented.
The argument being put forward from Arsal was that because Sony talked up its own soln to the I/O and RAM challenge, therefore somehow that means they were suggesting only their SSD was viable as a soln to said challenge. That is not a serious nor thoughtful argument. There is no evidentiary support to back up that framing. Cerny himself talked up SSD’s broadly in his Road to PS5 talk ffs.
Sony never once said nor suggested that was the case that only a 5.5 GB/s SSD was capable of changing how games were designed. They talked about their soln because they like their soln. Companies promote their solns because they believe those solns are good fits for the challenge at hand. R&C is a good showcase for their SSD, not because it should be imagined to demonstrate the full potential of it, but because it does things impossible at the existing fidelity of modern AAA games if you were to remove the SSD (unless you had an enormous RAM pool).
And Mike and I have been discussing tech in this thread alongside many others for going on a year now, so spare me this silly framing as if we leapt out of a bush and did drive by trolling or something. And lest you forget, it is quite literally his job to moderate discussion and keep it on track, on topic and to keep hyperbolic, reactionary discourse from overtaking the thread.
Both Mike and I and countless others here have talked up the XSX tech and power in probably hundreds of posts over the course of the past year. This is not new to us. Stop insinuating that we are anti-Xbox or pro-Sony or whatever other delusions are occupying your head. There are countless posts in this exact thread demonstrating otherwise.
Sony never hyped their SSD as being magical. Only its fans did. THAT is the point Mike was making before, where it comes off as you guys reacting to what fanboys on twitter or whatever claim instead of what Sony actually said, but then pinning that on Sony’s shoulders.
Neither company did anything wrong by highlighting their soln to the RAM and I/O challenges they sought to address. They just outlined their thinking and why they went with the soln they chose. BOTH companies did this. And it was perfectly appropriate to do so in both cases.
Sony did not do nearly as much to ‘hype up’ their SSD as you guys imagine they did. They did promote it for sure, but MS promoted their XVA even more than Sony did their SSD (MS had blog posts, dev interviews, a tech glossary web page, demos and various promo trailers). The difference is Sony has legions of fanboys online who were emotionally invested in the SSD being a secret sauce magic bullet and THEY hyped it to hell and back. Not Sony. Mark Cerny was, for instance, totally reserved in his comments in his presentation. They were not hyperbolic at all.
You should consider the possibility that maybe the reason we hear so much of PS5’s SSD is less due to Sony themselves and more due to its fans clinging for the only straw in sight. They don’t exactly have anything else to grasp at and yes, even that (imho) is a weaker overall soln than what MS has in XVA.
Can we please take a breath and cool down a bit? I admit I came in strong at first but I’ve been trying to de-escalate the tone and carry actual conversations, so I’d appreciate some cooperation here. 
Hmmmm I’m not sure…maybe? It was months ago where a clip was floating around Twitter and the person I saw retweeting it was generally a reliable person. I’ll have to see if I can find the quote or video but you may be right. If you are, I appreciate the correction.
I went down the video list in DF and pulled the last 5 new gen comparisons that weren’t using any form of BC or had odd bugs that may effect the results (admittedly I should say I tried to as I was also working and now see I somehow missed Scarlet Nexus, so I’ll add that as an extra data point below). If key numbers weren’t given, I then went to VG Tech to hopefully get more solid information. I was trying to be as accurate as possible and I wasn’t cherry picking games either nor do I think that makes sense when trying to determine the frequency of a claim, in this case a large gap in IQ to the extent of millions of pixels. Keep in mind that I wasn’t trying to make any claims of there never being a gap, there not currently being a gap, or that there won’t be any gaps in the future. It was to touch on the claim of “almost every game” had a gap of millions of pixels.
Regarding the figures…
Scarlet Nexus: 4K60 on both. DRS is used but both have the same range from 1440p to 4K and it’s said both hold closer to the native 4K figures.
Doom in RT targets 1800p with DRS. They say the resolution figures bottoms out at 50% but usually averages around 70% of 1800p. These percentages were given for both consoles btw. They also note that the Series X holds a higher resolution on average but no actual numbers were given, so that’s when I went to VG Tech. VG Tech also gave the highest and lowest resolutions possible on both systems between 1800p and 1275p. The single comparison difference they provided was the 1680p vs 1800p, so those are the numbers I went with.
In 120Hz mode, they gave the figures of 1800p vs 1584p, so I went with those figures with the comparison. I should note that the lowest figures for 120Hz mode are 1275p on the Series X and 1120p on the PS5, which has a lower difference than the upper bound figures I originally used, by about half the pixel difference, so that should show I wasn’t purposely selecting the smallest resolution gaps possible to push any kind of narrative.
With Balanced mode, the target is 4K60 on both and DF said that it hits the higher resolutions by and large, so I just chalked this mode as a wash.
In regards to Metro, DF determined the PS5 was around 80% of the SX resolution on average, or a range of 1512p-1728p on the SX vs 1296p-1512p on the PS5. Since no exact figures were given, I again went to VG Tech and saw the widest figure they provided was 2560x1440 on the PS5 and 2844x1600 on the SX, so I went with that. Again I didn’t try to cherry pick any numbers to push a narrative.
I honestly don’t even care which system has higher resolutions and such, at least in a direct comparison kind of way. Unless the port is trash on the SX (which it won’t ever be), I’ll always get the SX version since that’s my preferred system and I trust MS to honor that purchase in the future more than Sony. The only reason why I listed a handful of games was to dispell the claim made that almost every game has a gap of millions of pixels. That’s the only reason I even bothered making that post. And again, it wasn’t to claim there isn’t a gap or delta or that the gap won’t grow in the future. On the contrary I know it’ll grow in the future, which is why I’ve been saying forming definitive conclusions based on the games today does not make a whole lot of sense.
I agree with almost everything you say. However I just can’t agree that we should put that much weight into what marketing people say when the subject should be about technology, especially if the reaction is influenced based on how a small but loud group of fanboys act on the internet due to this marketing. I’ve seen fanboys referenced a number of times and I have to wonder why? Why give them any level of validity to the point that they are going to influence how much weight marketing has in a topic of game technology. Do any of us really believe they would be any less insufferable if the SSD was hyped up any less? Who cares what those idiots think or say? And this goes for fanboys of any system since we do see similar things on twitter about true RDNA2, 12 TF, XVA,and etc. In my opinion none of these people should even be a factor in this thread and the moment reactions or views are influenced by those kids, it’s at that point an emotional reaction and that can influence how conversations play out, points are intrepreted, and narratives driven.
On top of that, I don’t agree that we should pick and choose what to be critical about based on how loud a marketing department/campaign is. The claims themselves is what drives the controversies and every company has made claims that haven’t come to fruition yet and that’s fine because it’s going to take time to best take advantage of these systems. That’s the crux of what I’ve been trying to say. I addressed it earlier to how I think it’s fair to include marketing into the conversation since those marketers use technology in loose ways to promote their products. So to be fair, they are inviting themselves to be part of that conversation. I just don’t think they should be given as much weight as they are given because they don’t know what they are talking about really and they aren’t there to really inform the customers. They are all there to best highlight their produce in the best light possible.
If people want to talk about Sony’s marketing of their systems or marketing in general, cool. I don’t think I’m being irrational when I express that it should not be a focal point in this thread. Instead another thread can be made to have those discussions.
When someone has little to no engagement in this thread and pops in to take a jab at a company, that isn’t a valid conversation, it’s a drive by post. I’m honestly not trying to dictate anything, I’ve just been trying to express that I think forming any final conclusions now with the games we have today makes no sense. I don’t think that’s really an irrational opinion.
I’m also not trying to appear balanced at all, I don’t try to appear in any specific way. Just because I don’t jump to opportunities to take pot shots at one company or get overly riled up when someone doesn’t shower another with praise, that doesn’t mean I’m trying to appear balanced. Being able to disconnect marketing from specs and not letting what some people say about things like the SSDs or smart delivery bother me is not being balanced.
I’ve been trying to address your points, I made a lot of effort to have an actual conversation with you but have only gotten crickets so far. You don’t seem to be even understanding my points if you think I was trying to prove there isn’t a power delta. Especially when I have said multiple times that I’m not making any point or claim that there is no gap/delta:
I don’t know if it’s blinders, or emotions, or a stressful day at work, but if you think I was saying there was no GPU delta, then you haven’t been reading what I’ve been typing. I’d prefer it if we could have an actual conversation and not lose the plot because we’re too focused on proving why each other is wrong and get lost in those weeds. I’m not using whataboutisms or strawman arguments, I don’t even want to argue. My point is it’s too early to make any final claims on whatever the deltas are because we just don’t know yet. You’re free to disagree and great, let’s discuss that, I’m curious on your thoughts, but please don’t try to spin my intentions just so it’s easy for you to dismiss my points. I want to be able to disagree but still respect and have conversations.
1 Like
And on this day, the 23rd of September in the year of our lord Jon Bon Jovi 2021 it was declared to “Take it to DM’s”
3 Likes
Hadn’t seen staff post, so removed.
SFS actually helps loading the correct LOD based on the size the surface occupies on screen (it’s actually the center of the implementation).
Without SFS what you do is try to load the correct LOD, but with less data you often either load a higher or lower LOD then needed.
SFS uses sampler feedback data, that gives at a detailed mapping on how the texture was used, even if the same surface is big enough to use more than 1 lod of the same texture so you are more likely to load the correct LOD.
And there’s also one big advantage: If SFS does miss the prediction it has a mechanism in place to use the incorrect LOD and load the correct one in 1-frame delay. This is done at the texture unit hardware level and according to Ms the code to handle this in software is super complex. Being done in hardware also allowed them to use some tricks to smooth the transition between the incorrect LOD and the right one.
tl;dr: SFS actually reduces the amount of data you load, even if you use virtual texturing to try to always load the correct LOD all the time (Ms has been touting SFS as a 3x memory multiplier because basically they need to load and maintain a lot less data than even using an SSD without SFS).
BTW, they showed a tech demo for developers:
3 Likes
At work so can’t watch now, but I have watched presentations on SFS before and read articles about it.
My understanding was that it was to do with texture detail – disposing of higher res mipmaps that aren’t being used to free up memory?
When I said ‘LODs’ I meant geometry/foilage transparencies etc. Where developers swap in a billboard, for a low poly model, for a higher poly model, all the way down to LOD 0, which is a highly detailed model, the closer to the camera the objects are. Part of why they set their ‘swap out’ distances where they do is down to memory availability (although I think there are other factors).
Perhaps it relates to geometry as well, will watch the video later, but it’s ancillary to the point I am making.
LOD detail was just an example I pulled off the top of my head of what the developer might do with the memory saving SFS had given them. My general point was that they would always use the memory SFS freed up for something else, not leave it empty to pocket a faster load time.
Not just discarding but also loading, but overall, yeah.
Ah OK, I thought you were using lod instead of mip level and ran with it 
In that case, it can be related with SFS as well.
We usually see jumps because the hardware tessellation in the current gpus are only suitable to geometry defined in only a few expected formats. But sometimes the source format has a different one, so we usually have a few discrete LOD levels which are swapped and some tessellation level is applied to them if at all.
There is some advantages in how geometry is dealt with on rdna2/sx and that is that the hardware that process geometry part is thrown away now that the compute shaders are faster enough to out perform them. So with that you get more flexibility and you can run any program on the gpu that will extract the geometry from your source data.
And one very common way of storing high amounts of geometry is saving it as a texture representing the distance each point in the object surface has from the center of the object. This for example is the format Nanite uses on UE5, when you import the model to the engine they create the textures representing the geometry at the highest level, and at runtime parse that structure generating just the right amount of vertices needed for the position the object is in.
That’s how they achieved that insane level of detail in close ups and no popin during transitions.
And the good news is that for that specific format, SFS would provide similar gains.
But for other geometry data that is not stored as texture, no it wouldn’t affect directly, but saving needed memory from textures open up more for other data, including geometry. And same for loading since the SSD needs to load less texture data due SFS, there’s more bandwidth left to load geometry data.
And for the point you make, that’s reasonable to expect, but keep in mind that the whole point of the SSD is that since seek times are super low and bandwidth is high is that you don’t need to pre-load that much in advance, so you can only load what’s going to be immediable needed and that’s also a memory multiplier and loading time saver.
What SFS (or the Ps5 IO speeds) are trying to achieve is just to load so much data that your immediate needs are just what are exactly in front of you. As the SFS demo shows, they can literally load GBs of texture data on demand as you turn around in the room. So the amount of data you need to start by is so low, you can actually load the scene super quickly (they reload the scene in less than 0.19s in that demo)
1 Like
Thanks. So that last sentence is what I was really trying to get at here:
“since the SSD needs to load less texture data due SFS, there’s more bandwidth left to load geometry data”
I’m no expert in video game design, so there might be a better example of what a developer might do with the saving than my suggestion of geometry data, but the broader point was this:
That I don’t see SFS lowering load speeds significantly, as developers will just pocket the SFS memory saving and use it to push out some other aspect of the game design. So rather than faster loading, we will get better graphics in some other area instead. Could be wrong, that’s just my assumption.
I tried to answer that in the edit but perhaps I wasn’t direct enough.
But with the way the IO in those systems are designed you are able to load a large amount of data very quickly, so much that we are seeing they loading the scenario behind the player as quickly as the player turns.
So since the seek times are so low and the bandwidth is so high developers can load less at a single time resulting in shorter loading times, even if the fidelity is increased tremendously from past gen.
For example, that’s exactly what the second UE5 demo did when she entered the portal
Yeah I get that, but going back to where we started the conversation, my point was that a) the PS5 would always be a little faster in any given scenario b) the difference would be marginal and insignificant to most gamers.
I stand by that.
SFS could, on paper, reduce the gap. But I believe (and actually would rather) the developer using SFS will take that memory saving and pay it forward into an improvement in some other memory-bound constraint – making the Xbox version better rather than saving a few seconds on loading when the load speeds were so insignificant in the first place.
It isn’t “on paper” gains with SFS. Real world, in-game test scenarios show 3x I/O gain on texture streaming which amounts to about 2.5x overall.
SFS gets XSX’s overall I/O to around 12 GB/s (this is what we should expect as a typical bw, not a peak theoretical). The huge boost to texture I/O means those can be streamed into RAM faster than PS5’s SSD can accomplish the analogous process (because it has to bring the entire texture in, not just parts of it). That opens up more time that can be spent on streaming in anything else devs want like geo assets, extra texture assets, etc.
The other important element about SFS nobody talks about is that since you can pack relevant textures into RAM much more effectively than on PS5 (since ya only bring in ~1/3 of each texture) you also are streaming less often in like for like scenarios compared to PS5.
1 Like
Exactly.
And also it’s where all comes together: They developed bcpack because it offered great texture compression to all texture formats used by the games, AND it offered partial loading even for compressed textures.
This can be a big differentiator from PS5 because its hardware decompressors don’t support any streaming format, in some of them you even have to load more than just the data you want because they are compressed together. And the formats that do support that need to use either the gpu or the cpu for decompression.
It shouldn’t be a problem with the raw speed from their ssd, but you can see how Ms struck a balance between hardware and software for the best ration of results/cost.
1 Like