• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Xbox Velocity Architecture - 100 GB is instantly accessible by the developer through a custom hardware decompression block

oldergamer

Member
Sony have stated the peak, its 22 GB/s with heavy compression.
So why is it some here are chasting anyone suggesting that Xbox can perform 2 to 3 times its effective i?O bandwidth with compression and other memory performance saving features, but when the same people are throwing around 22GB/s for sony they have nothing to complain about?

Kraken compression at best is a 30% saving.
 

Tripolygon

Banned
Can you show me where sony claimed 5.5gb/s sustained? I must have missed that.
Not interested in marketing and word play. No SSD can sustain peak performance in all IOPS. If you are writing 4K files, the performance drops to less than half. The high bandwidth is only reached when writing files larger than 64KB block size..
 

oldergamer

Member
giphy.gif
If its not claimed to be peak and it's not claimed to be sustained, then what is it? You guys can't have it both ways when it suits you.
 

oldergamer

Member
Not interested in marketing and word play. No SSD can sustain peak performance in all IOPS. If you are writing 4K files, the performance drops to less than half. The high bandwidth is only reached when writing files larger than 64KB block size..
Hence why manufacturers of SSD's list the sequential read/write numbers as how they measure performance. They look a lot better, or marketing speak as you mentioned.

So what are you saying, are you saying the performance numbers from Sony of 5.5gb/s is not peak, and its sustained?
 
Last edited:
Lots of different opinions in this thread and theories about the XSX SSD/IO setup, I decided to re-read eurogamers XBX reveal article and it's pretty clear what MS are trying to achieve and should bring in massive improvements to what we have today, to me it sounds like Sony & MS are aiming for similar goals with their SSD solutions, and admittedly Sony's looks faster but some posts seem to suggest MS have just panicked and stuck a SSD into the XSS late on which is crazy.
 

oldergamer

Member
Lots of different opinions in this thread and theories about the XSX SSD/IO setup, I decided to re-read eurogamers XBX reveal article and it's pretty clear what MS are trying to achieve and should bring in massive improvements to what we have today, to me it sounds like Sony & MS are aiming for similar goals with their SSD solutions, and admittedly Sony's looks faster but some posts seem to suggest MS have just panicked and stuck a SSD into the XSS late on which is crazy.
I don't think any posts suggest that. still some people are certainly discounting all other features of GPU performance suddenly to claim SSD is the most important thing (which i think is ridiculous). IMHO the most important thing can be a constant benefit to the performance of all types of games, NOT only those that are frequently streaming assets such as open world titles. I mean really, if SSD was this important to all games, what would be the point of people buying new video cards if i/o is what people consider the only way to unlock graphical detail ( i'm exaggerating btw ).
 
Last edited:

Redlight

Member
Ok, I am game. How am I downplaying the XSX SSD and I/O setup?

I didn't actually suggest that, I said 'downplaying good Xbox news' in an Xbox thread in the way you pointedly accused others of doing to Sony. But maybe you're right, how could I possibly get that impression? :)

Again reading XSX marketing spec read from a press pack.

Considering how much this magic 2-3 I/O bandwidth multiplier for XSX claims came back out of nowhere after UE5’s PS5 reatine demo kind of did just that... I guess it is a bit different than power of the cloud.

You are right...PS5 is better in that aspect ;).

Surely you are not trying to make those the only centres of attention and minimise other factors because your box WINS if looked only through that lens and you only care about winning a numbers war any way you can ?
 

Tripolygon

Banned
Hence why manufacturers of SSD's list the sequential read/write numbers as how they measure performance. The look a lot better, or marketing speak as you mentioned.

So what are you saying, are you saying the performance numbers from Sony of 5.5gb/s is not peak, and its sustained?
Yes that is what I'm saying.
 

Andodalf

Banned
so the question is, why would Cerny and Co. design it that way of is not going to matter this gen?

Why did they make the Cell when only a handful of games ever really pushed it beyond what the 360 could do? Sony wants to push emergent tech, and probably has some ideas of what they can do with it, even if others don't want to
 

FranXico

Member
admittedly Sony's looks faster but some posts seem to suggest MS have just panicked and stuck a SSD into the XSS late on which is crazy.
No. Really not. Both companies received similar feedback from developers, and had similar goals as a result. MS did not smack an SSD into the system last minute. It would seem that they are improving compression to narrow the speed gap, but both solutions were well thought out. Sony did not just slap a faster SSD either. They designed the rest of the system around the IO, to the point of sacrificing GPU/CPU budget
 

Tripolygon

Banned
Why did they make the Cell when only a handful of games ever really pushed it beyond what the 360 could do? Sony wants to push emergent tech, and probably has some ideas of what they can do with it, even if others don't want to
SSD is not Cell. Give developers more bandwidth and they will saturate it with data. The IO from storage to RAM has seen the slowest improvement in capability over generations. This is the first time we will have a massive jump in storage IO. You can never have too much bandwidth because the faster you can move data around, the faster you can work on it.

Lets put things in perspective. The RAM has almost 100x more bandwidth than the SSD.
 
Last edited:

ToadMan

Member
My guess is multiplats will see SeX running higher1800p (vs 1440p), with better framerates and sharper RT effects.
1st parties will be identically beautiful. SeX will probably have an edge in crazy effects you cant think off now.

Well the PS4 had a greater power advantage (some 40%) over xb1. That translated to identical frames and one step higher resolution.

18% isn’t enough for what you predict. Not to mention devs will target PS5 and port to Xsex giving them no incentive to improve performance.

This will be a dull gen for console warriors - 1st party will be where the actual power of these consoles is found.
 
Last edited:

oldergamer

Member
Yes that is what I'm saying.
That doesn't seem possible IMO. That's why i mentioned they still have to work within the confines of the available hardware to them, and clearly can't manufacture their own custom NAND chips. Sony would have to be using NAND chips with an MT performance number that doesn't currently exist?

I'm speculating here, but:

- If it takes 12 lanes for them to hit 5.5gb/s (as was posted in a translation a few pages back) and its possible to hit 5.0gb/s using only 8 lanes ( Phison and other manufacturers have done this at increased cost )
- if there is truly a suboptimal performance degradation when using a single 64MB NAND chip on a single lane

if you go by Phisons numbers they are achieving 0.625gb per 8/lane(s)
I'm not sure what the MT rating is on the NAND sony is using, but I'd assume its 1200MT as a higher rating is more expensive
PS5 would be achieving 0.458gb per 12/lane(s)

I don't know what speed NAN chips sony would need, but for 5.5gb/s to be sustained or Random read speed, they would need to be including hardware in PS5 that doesn't exist. This is why i'm thinking thier number of 5.5gb is peak sequential read. only slightly above other NVME drives you can get at retail. Keeping in mind there are only so many manufacturers ( 6 or 7? ) of NAND memory out there. They would have to buy chips from one of them.
 

Chankoras

Member
Why did they make the Cell when only a handful of games ever really pushed it beyond what the 360 could do? Sony wants to push emergent tech, and probably has some ideas of what they can do with it, even if others don't want to
Maybe so, but I believe Cerny has a different philosophy than Kutaragi, which is ease of development, also I was referring to a performance metric, the person I quoted claimed that I would not be reached, but I doubt Cerny is going to add an unnecessary expense to the system that will not be used..
 

ToadMan

Member
I wonder why people are doubling down on the SSD being the thing the most important next gen??

whereas if it was reversed,the notion that the SSD was going to change anything would be ridiculed to high heaven.

SSD in the ps5 has become this gens power of the cloud. Until the games prove otherwise.

I’m genuinely surprised that someone who frequents a gaming enthusiast web site and is in a thread to talk about tech, can simultaneously be so misinformed about the way the technology works.

But anyway, putting aside the likelihood you are trolling for a moment (hopefully it is trolling), let’s just think about it for one moment shall we?

Systems have to be balanced. Taking a gpu from this gen and putting it in a decade old system would do nothing because the older system would be limited by all its other components and indeed the design of its games.

It is a specious argument to theorise about performance gains of one component - the SSD - in a system not designed to make use of it. You wish to play down the technology - that’s up to you.

Here’s a nice thought experiment.

The x360 had a built in HDD and the slow transfer rate of that device meant games had to be designed and built with its slow transfer rate in mind. Even the 360 was compromised by its low storage transfer rate.

Assuming a HDD transfer rate in an x360 of 100mb/s, the Xsex is 48 time’s faster at transferring compressed data. That’s a big bump from 360 days.

Except, the Xsex is 50 times more powerful than an x360.

It turns out that all that has happened is that the storage tech in the Xsex has increased in line with cpu and gpu gains.

MS are busy working away to try and improve that 4.8gb/s compressed data transfer rate even higher with software - presumably they’re concerned enough about data bandwidth to invest all that effort to get even more.
 

ToadMan

Member
That doesn't seem possible IMO. That's why i mentioned they still have to work within the confines of the available hardware to them, and clearly can't manufacture their own custom NAND chips. Sony would have to be using NAND chips with an MT performance number that doesn't currently exist?

I'm speculating here, but:

- If it takes 12 lanes for them to hit 5.5gb/s (as was posted in a translation a few pages back) and its possible to hit 5.0gb/s using only 8 lanes ( Phison and other manufacturers have done this at increased cost )
- if there is truly a suboptimal performance degradation when using a single 64MB NAND chip on a single lane

if you go by Phisons numbers they are achieving 0.625gb per 8/lane(s)
I'm not sure what the MT rating is on the NAND sony is using, but I'd assume its 1200MT as a higher rating is more expensive
PS5 would be achieving 0.458gb per 12/lane(s)

I don't know what speed NAN chips sony would need, but for 5.5gb/s to be sustained or Random read speed, they would need to be including hardware in PS5 that doesn't exist. This is why i'm thinking thier number of 5.5gb is peak sequential read. only slightly above other NVME drives you can get at retail. Keeping in mind there are only so many manufacturers ( 6 or 7? ) of NAND memory out there. They would have to buy chips from one of them.

Sony are using 4 lanes, not 12.
 

DaGwaphics

Member
The x360 had a built in HDD and the slow transfer rate of that device meant games had to be designed and built with its slow transfer rate in mind. Even the 360 was compromised by its low storage transfer rate.

Actually, the x360 was not affected by HDD speeds. Games were not allowed to require the drive (though, texture packages and other upgrades were allowed). 360 played from the DVD and I'm sure was very constrained by the the performance of the DVD drive, but not the HDD.
 
That's not 1-on-1 in scale to each other. That's purely to properly explain what's part of the I/O complex

It's still taking up die space that could be used for more GPU CUs for example.

Anyways do you know if the PS4s APU has something similar to the I/O complex?

I'm asking about the PS4s I/O system because it can help me understand how the PS5s I/0 is an improvement besides just higher read and write speeds.
 

oldergamer

Member
That is the max sequential read/write speed of the controller assuming you are using NAND modules capable of achieving those speed. Microsoft has stated their raw speed as 2.4GB/s so they are probably using 4 256GB NAND modules at ~800MT/s to achieve a theoretical 3.2GB/s and accounting for overhead and their sustainable throughput brings them down to 2.4GB/s.

The bus is not the limiter, the NAND module is. The controller supports 1200MT/s per channel and they are using PCIE 4.0 which is really fast.
My bad i missed this post earlier
 

Leyasu

Banned
I do not think people are saying it will magically make things better, maybe I am not reading the right crazies in this instance.

Not sure what the two wrongs not make a right comment was for: the “power of the cloud” comment had physics and implementation details issues back in 2013 and we saw nothing in 7.5 years backing it up. The SSD comments, which all apply also to the XSX (compared to the average PC and previous generation consoles), are not comparable to that.

Essentially it seems to boil down to “number where XSX is superior == something useful” vs “number where XSX is not superior == either the extra performance is likely worthless or XSX has secret trump card that closes the gap completely”.
Two wrongs was a reference to xbox fanboys now popping up with xbox secret bandwidth sauce.

The PS5 and XsX both have tangible advantages. One being compute and a smidge on the CPU side plus RAM bandwidth, the other side has the SSD speed and transfer rate. Neither is capable of doing something the other is not capable of.

That is all. The SSD will not magically overcome anything and make games look prettier on the PS5 and the differences on the XsX are not going to night and day. Faster loading and less obvious LOD on one side, and a bump in resolution on the other.
 
Two wrongs was a reference to xbox fanboys now popping up with xbox secret bandwidth sauce.

The PS5 and XsX both have tangible advantages. One being compute and a smidge on the CPU side plus RAM bandwidth, the other side has the SSD speed and transfer rate. Neither is capable of doing something the other is not capable of.

That is all. The SSD will not magically overcome anything and make games look prettier on the PS5 and the differences on the XsX are not going to night and day. Faster loading and less obvious LOD on one side, and a bump in resolution on the other.

Good post, we are getting two very similar impressive looking consoles, i just wish I had more time to play them, then I could justify owning both!!
 

Leyasu

Banned
I’m genuinely surprised that someone who frequents a gaming enthusiast web site and is in a thread to talk about tech, can simultaneously be so misinformed about the way the technology works.

But anyway, putting aside the likelihood you are trolling for a moment (hopefully it is trolling), let’s just think about it for one moment shall we?

Systems have to be balanced. Taking a gpu from this gen and putting it in a decade old system would do nothing because the older system would be limited by all its other components and indeed the design of its games.

It is a specious argument to theorise about performance gains of one component - the SSD - in a system not designed to make use of it. You wish to play down the technology - that’s up to you.

Here’s a nice thought experiment.

The x360 had a built in HDD and the slow transfer rate of that device meant games had to be designed and built with its slow transfer rate in mind. Even the 360 was compromised by its low storage transfer rate.

Assuming a HDD transfer rate in an x360 of 100mb/s, the Xsex is 48 time’s faster at transferring compressed data. That’s a big bump from 360 days.

Except, the Xsex is 50 times more powerful than an x360.

It turns out that all that has happened is that the storage tech in the Xsex has increased in line with cpu and gpu gains.

MS are busy working away to try and improve that 4.8gb/s compressed data transfer rate even higher with software - presumably they’re concerned enough about data bandwidth to invest all that effort to get even more.
I don't know if I should thank you for this post or be angry at you for what you implied in the first 2 sentences lol.
 

oldergamer

Member
Based on the above
what configuration of chips would give sony 825 GB?

13 x 64MB NAND = 832GB
7 x 128 NAND = 896GB

What number of lanes would they need, and what performance level on the chips? i've seen 1600MT is available, but that would be expensive.
 
Based on the above
what configuration of chips would give sony 825 GB?

13 x 64MB NAND = 832GB
7 x 128 NAND = 896GB

What number of lanes would they need, and what performance level on the chips? i've seen 1600MT is available, but that would be expensive.

It's probably 12 64 GiB (Gibibyte) modules, for 768 GiB, or 825 GB (Gigabyte). Gibibyte is about 1.074x a Gigabyte (I've seen some other conversion rates but none of them make sense considering the 825 GB Sony's given and knowing what capacities NAND modules are usually made at, unless Sony is counting some other reserved NAND in there that isn't accessed by the controller?).

If the raw sequential read speed is 5.5 GB (Gigabyte) per second, then each module has to provide around 458.3 MB (Megabyte) per second of bandwidth, assuming 5.5 GB/s refers to sustained sequential read speeds.

Sony are using 4 lanes, not 12.

I think they got lanes confused with channels.

seems thicc_girls_are_teh_best thicc_girls_are_teh_best double standard is at work :
expecting a worst example to be the best case in one case.
expecting the best example to be the worst on the other.

You mind telling me what you're referring to and where you think the issue is? I've made a lot of posts so I can't recall which one you're specifically referring to.

That is the max sequential read/write speed of the controller assuming you are using NAND modules capable of achieving those speed. Microsoft has stated their raw speed as 2.4GB/s so they are probably using 4 256GB NAND modules at ~800MT/s to achieve a theoretical 3.2GB/s and accounting for overhead and their sustainable throughput brings them down to 2.4GB/s.

The bus is not the limiter, the NAND module is. The controller supports 1200MT/s per channel and they are using PCIE 4.0 which is really fast.

Overhead on PCIe 4.0 would account for nowhere near a 800 MB drop, as 4.0 uses 128b/130b encoding (3.0 and earlier used 8b/10b encoding). Just taking the encoding factor into account the most I'd expect from raw overhead on PCIe 4.0 is ~ 5 MB (if that) lost on a 2 GB/s transceiver lane consider on 3.0 1 GB/s provided more like 985 MB/s actual (15 MB/s lost due to overhead through 8b/10b encoding).

So I don't expect MS's API solution (well, not in the final system anyway) to be to the point where they are losing over 790 MB of throughput due to software API overhead.
 
Last edited:

Thirty7ven

Banned
I wish that I too could justify owning both with the time to play them.

Just buy the console that will offer you the experience you prefer. Gamepad, System Interface, Services, Exclusives.

If that's not what drives you, and you don't care about the digital library you already have, then I guess multplatform games performance is what drives your purchase and you will only know this when the games come out. If this is you, then don't buy either of the systems on day 1, because you won't be able to know the answer by then.
 

BrentonB

Member
Loading 2.4GB (Raw) and 4.8GB (compressed) of data per second in tandem with decompression hardware that can handle up to 6GB/s is excellent performance. Series X games will load in about 6 seconds. That's a huge improvement over current gen and should improve the gaming experience immensely. Please look forward to next gen.
 
Last edited:

Leyasu

Banned
Just buy the console that will offer you the experience you prefer. Gamepad, System Interface, Services, Exclusives.

If that's not what drives you, and you don't care about the digital library you already have, then I guess multplatform games performance is what drives your purchase and you will only know this when the games come out. If this is you, then don't buy either of the systems on day 1, because you won't be able to know the answer by then.
I’ll be buying the XsX day one. Not even a question. I would get a ps5 as well if I had the time to play them both
 
Well the PS4 had a greater power advantage (some 40%) over xb1. That translates to identical frames and one step higher resolution.

18% isn’t enough for what you predict. Not to mention devs will target PS5 and port to Xsex giving them no incentive to improve performance.

This will be a dull gen for console warriors - 1st party will be where the actual power of these consoles is found.
Where did you hear that devs would be targeting the PS5? It would make more sense to target the PC and port those titles to consoles. If the PS5 SSD solution isn't found anywhere else what happens when a game is built around a 9 GB/S SSD that isn't found anywhere else? It makes more sense to make a game around a standard HDD and if the SSD on the PS5 is an advantage it will show in the games performance. Just like Red Dead Redemption 2 wasn't developed for the XBOX One X but that platform had better performance because of the inherit advantage of the platform.
 

Tripolygon

Banned
Overhead on PCIe 4.0 would account for nowhere near a 800 MB drop, as 4.0 uses 128b/130b encoding (3.0 and earlier used 8b/10b encoding). Just taking the encoding factor into account the most I'd expect from raw overhead on PCIe 4.0 is ~ 5 MB (if that) lost on a 2 GB/s transceiver lane consider on 3.0 1 GB/s provided more like 985 MB/s actual (15 MB/s lost due to overhead through 8b/10b encoding).

So I don't expect MS's API solution (well, not in the final system anyway) to be to the point where they are losing over 790 MB of throughput due to software API overhead.
I'm not talking about encoding factor or Microsoft's API. I'm talking about NAND to Logical to Physical mapping and ECC overhead. They are also using a DRAM-less controller so the paging table also has to reside in the NAND unless they are stored in RAM during boot up. All these add up to reduce total bandwidth from the peak capability of the NAND modules they are using which i would wager is in excess of 3GB/s. Hence their reported 2.4GB sustained speed that is given. Likewise I'm sure they took thermal throttle into account in their stated specs as well.
 
Last edited:

martino

Member
You mind telling me what you're referring to and where you think the issue is? I've made a lot of posts so I can't recall which one you're specifically referring to.
i don't find your post and rereading mine it was not well worded
But you recently underline habit of some gaffer to have double standard dealing with the unknown and i think the post i quoted was a perfect of that
 
Last edited:

rntongo

Banned
I think the main confusion is where/what the 2-3x multiplier applies to. To give a previous similar example with Xbox One there was the PRT/tiled resources API+hardware that combined the data move engines and eSRAM that could take 6GB of textures and fit them into the 32MB and then stream them in as and when with Granite. This would be nearly a 200x multiplier!

I think the 2-3x multiplier is overall efficiency of RAM like we saw in the PS5 RAM utilization slides:

WBwdSLvz8Gto7pmcnf6Awb-1200-80.jpg

ps5ram.0.jpg


I will also repeat again that too much emphasis is being put on the "SSD" itself when it is far more the hardware and software that comes after that.

The thing I'm most skeptical about from what Mark Cerny said is that devs don't need to know or do much at all about all the hardware. Does it virtually automatically.

I await the receipts!

Not at all! The 2-3x comes from something else(although it's still about increasing efficiency of RAM). That picture is titled RAM UTILIZATION because it's talking about using RAM differently by putting most of what's needed for an instance of gameplay in RAM. It's simply because next gen consoles will have SSDs so you can load more data that needs to be used intantly.

On the other hand, the 2-3x multiplier form SFS is due to more efficient texture streaming. So for example another system would need 2-3x the physical RAM or it could stream the equivalent of a system with 2-3x the SSD speed. How does it do this, these other systems would waste RAM and SSD I/O bandwidth by streaming textures that are not going to be used in a scene or never at all.
 

THE:MILKMAN

Member
Not at all! The 2-3x comes from something else(although it's still about increasing efficiency of RAM). That picture is titled RAM UTILIZATION because it's talking about using RAM differently by putting most of what's needed for an instance of gameplay in RAM. It's simply because next gen consoles will have SSDs so you can load more data that needs to be used intantly.

On the other hand, the 2-3x multiplier form SFS is due to more efficient texture streaming. So for example another system would need 2-3x the physical RAM or it could stream the equivalent of a system with 2-3x the SSD speed. How does it do this, these other systems would waste RAM and SSD I/O bandwidth by streaming textures that are not going to be used in a scene or never at all.

Let's wait and see what Microsoft have to say about it. I think Jason Ronald Tweeted they'll talk soon.
 

Three

Member
You're quoting a random twitter user who got the 2-3x figure from the official microsoft xbox series x page to discredit the figure?
No I'm not. I've said that that figure is compared to loading the entire texture. So no PRT or SF. What you're doing though is using that figure to link to another official twitter user mentioning something 'custom' and pretending this is some secret sauce that would give it 2x or 3x multiplier compared to the competition. This isn't the case. When that other official twitter user was asked about this figure and it being custom or not they didn't say it was and said they don't know the figure.
 
Last edited:

ToadMan

Member
Actually, the x360 was not affected by HDD speeds. Games were not allowed to require the drive (though, texture packages and other upgrades were allowed). 360 played from the DVD and I'm sure was very constrained by the the performance of the DVD drive, but not the HDD.

You know I’d totally forgotten that.

My 360 rrod about 10 years ago and I just put up with PS3 and never replaced it, so those kind of details have slipped from my mind!
 

THE:MILKMAN

Member
Those are 12 channels as the graphic says. That’s “inside” the custom SSD. Between the SSD controller and the SoC it’s 4x Lanes.

Like I said the poster was clearly talking about the 12 NAND chips and lanes/channels are synonyms.

NXGamer put up this video today. I think it relevant to this discussion.

 

ToadMan

Member
Where did you hear that devs would be targeting the PS5? It would make more sense to target the PC and port those titles to consoles. If the PS5 SSD solution isn't found anywhere else what happens when a game is built around a 9 GB/S SSD that isn't found anywhere else? It makes more sense to make a game around a standard HDD and if the SSD on the PS5 is an advantage it will show in the games performance. Just like Red Dead Redemption 2 wasn't developed for the XBOX One X but that platform had better performance because of the inherit advantage of the platform.

That was how it worked in gens past.

PCs are never the target for console based games.

Its easier to develop for the most limited platform and port that code to others. PS5 and Xsex have very similar architecture and performance so the porting should be straightforward.

For SSD matters, multiplats can just use downgraded textures where necessary if Xsex should have trouble keeping up. Personally I don’t think it will.

The UE5 demo is already a demonstration of that Principle for multiplats - target PS5, port to Xsex. The main 3rd party platform chose PS5 for its first demo of the gen. They’re optimising for PS5 in the expectation ports to Xsex will achieve similarly good performance.

PCs versions will just up the minimum RAM required.
 
Top Bottom