NullZ3r0
Banned
40% more CUs too. So it will eat PS5s lunch in RT performance.
-
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.
Microsoft Explains The Xbox Series X's High-Speed Secret Sauce
- Thread starter Vasto
- Start date
40% more CUs too. So it will eat PS5s lunch in RT performance.
ZehDon
Gold Member
You've posted this multiple times. It's laughably wrong.
The "infinite power of the cloud" was laughed at, rightly, because the practical limitations prevented it from ever working. That is self-evident for anyone with a lick of technical understanding - even Microsoft's own tech demos were very specific use cases that never applied to any game released. The difference with Microsoft's I/O stack changes here is that the same people who laughed at the "infinite power of the cloud" are the same people highlighting that Microsoft's approach to the XSX's I/O is interesting and results in a practical improvement to loading data from the SSD.
So, please, stop repeating yourself.
supernova8
Banned
Aha very kind of you to keep with my shitty painter analogy
That makes sense. So would I be correct in thinking that PS5 will be better at loading new levels/totally new environments quickly but XSX should be better at sustained visuals where the 'things' (models, environments themselves don't necessarily change so much?
As a concrete example, perhaps XSX would struggle to do the very quick dimensional changes seen in Ratchet and Clank, but PS5 might not be able to sustain the same graphical performance if the 'settings' were maxed out in something like FIFA, or a Battlefield game where the map itself is already loaded in (I think?) and therefore the super fast SSD wouldn't be as useful.
Again I could be misunderstanding. Trying to take a balanced view.
You have to be kidding me
He in explained in precise detail how the SSD was designed, how it differs from standard SSD's, how it's superior than current SSD's, how bottlenecks were removed, how latency were removed, why they chose 12 channels 6 priority levels instead of the normal 8 channel 2 priority levels of every other SSD. How they customized the I/O with features no other SSD has. He did this all without using BS misleading fancy prettied up words based on software.
Where's the MS equivalent of this? Where is the in depth breakdown of their SSD? All i'm seeing are words for software made stuff to fool people into thinking their fairly standard SSD is on par or better that PS5's clearly better SSD.
Pretty embarrassing that a, as you claim, mediocre game developer outclassed an entire company that, as you also claim, made the computer what it is today.
Trainsmove
Banned
Now that we know that the hell blade 2 trailer was running in real time there is no doubt that series X is the most powerful console on the planet in visuals alone. And said here by many A SSD speed will not save you.
Sorry about the mis spelling.
Sorry about the mis spelling.
Last edited:
supernova8
Banned
No, I think you'll find that's the Jedi Council.
thicc_girls_are_teh_best
Gold Member
You can't remove latency altogether, even SRAM has latency. You can only reduce it, and I'm sure Sony have done this. But there is some evidence pointing to MS having done this moreso. And you don't necessarily need extremely low latency for bandwidth-happy solutions and we know Sony have a particular focus on maximizing bandwidth saturation, MS less so.
MS haven't done a singular video presentation wrapping all their XvA stuff into a nice 40-so minute video file, but they have mentioned a lot of things about it already. Some of it you have to search around though, such as various Twitter comments (and other things in certain interviews, like with the GPU mesh shader level support, which is 256. Conversely, Nvidia mesh shader level support is 128).
Most of what they have mentioned actually has technical research and implementations that support it, so it's not necessarily words of PR fluff (it is being presented in a way to enable good PR though, obviously. But Sony does this as well, Cerny even did a bit of it in Road to PS5 presentation). If you pay close attention, Cerny did clearly spin some mentions when it came to even certain parts of the SSD I/O, such as terminology like cache coherency engines. Using engines in that context is an emotive appeal, to convey power. It's certainly powerful for what it is, but an exaggerative term like engine needn't be used. There was more of this in their GPU portion though particularly with downplaying certain aspects of GPU design they knew a certain competitor had an advantage in (while hoping those who were listening would ignore patterns in vast majority of GPU benchmarks).
If you're waiting for a full system architecture dissertation on Series systems, you'll most likely have to wait until August 17th, when the Hot Chips presentation is held.
Last edited:
Where is the details of the variable clocks outside of the cerny marketing speak a couple, most of the time. Near ect. Every company does marketing speak. I guess nvidia is helping Microsoft cover up the lies right? They even mentioned some of the benefits of the velocity coming to PC.
supernova8
Banned
Maybe it's just me but I don't give a shit about the velocity architecture or whatever Sony is doing.
I want both companies to stop fucking blabbering on about the nitty gritty hardware and just show us the games.
I want them to just blow their loads in terms of games in one big (2 hour if necessary) gameplay blowout rather than this drip-feed bullshit.
I want Microsoft to get a fucking a move on. It's been 3 months since the hardware teardown from DigitalFoundry and even that was only mildly entertaining.
July 23? Fucking hell. I'd love it if they just released all the trailers to their games as individual 4K60 YouTube videos that we can watch at our own leisure.
I would honestly laugh if Sony decided to announce their pricing on July 22 and open pre-orders the day before Xbox's first party games reveal.
Serves them right.
I want both companies to stop fucking blabbering on about the nitty gritty hardware and just show us the games.
I want them to just blow their loads in terms of games in one big (2 hour if necessary) gameplay blowout rather than this drip-feed bullshit.
I want Microsoft to get a fucking a move on. It's been 3 months since the hardware teardown from DigitalFoundry and even that was only mildly entertaining.
July 23? Fucking hell. I'd love it if they just released all the trailers to their games as individual 4K60 YouTube videos that we can watch at our own leisure.
I would honestly laugh if Sony decided to announce their pricing on July 22 and open pre-orders the day before Xbox's first party games reveal.
Serves them right.
JustP_Gaming
Banned
Yes MS has said they can use the 100gb SSD as RAM.
StreetsofBeige
Gold Member
Sounds good. Problem is what if your SSD is filled with games?
JustP_Gaming
Banned
Good question.
Bogroll
Likes moldy games
I would guess a 80GB game would say requires 180 GB on the box etc.
Alex Scott
Member
Triceratops are good for measuring non graphic workloads not games. Having a higher teraflops doesn't mean better performance in fps, might in resolution but not in fps. You can't compare teraflops and see which is better.
cireza
Banned
I know this. See my other post right after the one you quoted.
JerryinSoCal
Member
More CU's but also slower and I'm not betting on RT being amazing for the consoles either way, we just saw how Watch Dogs Legion had to be set at 1080p 30fps to get ultra settings with RT enabled on a high end graphics card.
Last edited:
ZywyPL
Banned
So as usual consoles will have to settle for medium-high settings and everything will run just fine. Or use some tricks like GT7 with quarter resolution for RT effects.
Last edited:
Really people don't understand what the "magic" 2.5 multiplier means.
This is always a reference to classic rendering (e.g. GCN based hardware). With the old hardware, textures had always been loaded in a package and as whole texture, because texture tiling was not that great with the HDD because general bandwidth was just to low. That changes with SSDs. With the SSDs texture tiling is finally useable and allows game engines just to request parts of a texture that are really visible. So only those small parts of a texture must be loaded into memory. That's were the factor 2.5 came from.
Sony can also use texture tiling because the hardware allows this feature, but it has also some negative side effects, that MS "fixed" with another custom hardware feature so texture pop in should not be visible.
There is not magically more bandwidth or memory available but texture tiling (or tiling resources) allows it to use the available bandwidth and memory much more efficient. That's where the multiplier come into play but just when comparing old tech vs new tech.
Sony can use the same software side tech to save memory and bandwidth (if there are no patents) but will have problems with texture pop in. Therefore the Sony solution is more like brute forcing with SSD bandwidth.
Both solutions have their pros and cons.
The other thing MS states that they drastically reduced latency to the SSD memory. Sony also did something like that with their priority management. But because Sony also will support "standard" m.2 SSDs (well some at least) the reduction in latency they reached can't be that significant brlecause it ends where the inserted SSD comes into play. MS seems to have also optimised that part by partnering with Seagate. I really hope they ripped of all that is useless for the console and made the modules as cheap for customers as possible.
This is always a reference to classic rendering (e.g. GCN based hardware). With the old hardware, textures had always been loaded in a package and as whole texture, because texture tiling was not that great with the HDD because general bandwidth was just to low. That changes with SSDs. With the SSDs texture tiling is finally useable and allows game engines just to request parts of a texture that are really visible. So only those small parts of a texture must be loaded into memory. That's were the factor 2.5 came from.
Sony can also use texture tiling because the hardware allows this feature, but it has also some negative side effects, that MS "fixed" with another custom hardware feature so texture pop in should not be visible.
There is not magically more bandwidth or memory available but texture tiling (or tiling resources) allows it to use the available bandwidth and memory much more efficient. That's where the multiplier come into play but just when comparing old tech vs new tech.
Sony can use the same software side tech to save memory and bandwidth (if there are no patents) but will have problems with texture pop in. Therefore the Sony solution is more like brute forcing with SSD bandwidth.
Both solutions have their pros and cons.
The other thing MS states that they drastically reduced latency to the SSD memory. Sony also did something like that with their priority management. But because Sony also will support "standard" m.2 SSDs (well some at least) the reduction in latency they reached can't be that significant brlecause it ends where the inserted SSD comes into play. MS seems to have also optimised that part by partnering with Seagate. I really hope they ripped of all that is useless for the console and made the modules as cheap for customers as possible.
Panajev2001a
GAF's Pleasant Genius
I think this is why they require a bit more bandwidth from drives that do not support the same number of priority levels for example.
The rest of the work to minimise latency is actually in the I/O complex they packed with SRAM, I/O co-processors and coherency engine, the cache scrubbers (part of the same setup to lower latency perceived by the GPU and maximise cache hits), the Kraken decompressor, and more.
D
Deleted member 775630
Unconfirmed Member
12GB/s is what's needed to match their performance with a 4.8GB/s SSD because they use 2.5x less memory bandwidth.
I didn't want to downplay what Sony did with the latency. But it really ends where a new SSD is placed into the system. There everything can only get optimized for bandwidth and not really latency because Sony has no influence on the inserted SSDs. They can only influence it by supporting only certain SSDs. They also gave to maintain the m.2 protocols and can't optimize there.
But the custom SSD extension allows MS to go further because everything is custom. Even a custom pricing
Last edited:
CurtBizzy
Member
I'm pretty sure it's 4.8 instead of 12
D
Deleted member 775630
Unconfirmed Member
I'm not saying SSD performance, I mean console performance. Using the XSX SSD combined with SFS they are able to get way more detail into a scene, 2.5x as much. Systems that don't use SFS would need a 12GB/s SSD to match that.I'm pretty sure it's 4.8 instead of 12
pawel86ck
Banned
12TF and 12 GB/s I/O together can go a long way
.But before we will celebrate there's one last thing that bothers me. Some people here have asked a very good question. If HW decompression chip can only process 6GB/s in the end then it should cancel SFS 2.5x multiplier benefit. I wonder if 6GB/s decompression number refers to input data (6GB/s input data can translate into 12GB/s depending on compression ratio), or output data (output data caped at 6GB/s will cancel SFS bebefits). If it refers only only to the input data, then SFS 2.5x multipier will work and we can celebrate
.
Last edited:
CurtBizzy
Member
SFS being 2.5x more efficient does not translate 2.5x more data
Last edited:
It is output. Because the input is limited by the SSD speed. But even than, every tile can be compressed on its own. So the multiplier still applies. But it should be very rare that textures get compressed more than another 50% because there is still the native texture compression format the gpu can work with.12TF and 12 GB/s I/O together can go a long way
But before we will celebrate there's one last thing that bothers me. Some people here have asked a very good question. If HW decompression chip can only process 6GB/s in the end then it should cancel SFS 2.5x multiplier benefit. I wonder if 6GB/s decompression number refers to input data (6GB/s input data can translate into 12GB/s depending on compression ratio), or output data (output data caped at 6GB/s will cancel SFS bebefits). If it refers only only to the input data, then SFS 2.5x multipier will work and we can celebrate
If you look on the unreal demo with it's 768 MB/s both consoles have more than enough bandwidth to handle that.
geordiemp
Member
Show us. I dont believe you.
What hardware have MS done to reduce SSD access latency ? Lot or words and PR yes and ?
Last edited:
D
Deleted member 775630
Unconfirmed Member
Compared to other systems it does.
Ascend
Member
Remember that SFS doesn't load unneeded textures/mips. That means that the unneeded textures don't need to be decompressed either.12TF and 12 GB/s I/O together can go a long way
But before we will celebrate there's one last thing that bothers me. Some people here have asked a very good question. If HW decompression chip can only process 6GB/s in the end then it should cancel SFS 2.5x multiplier benefit. I wonder if 6GB/s decompression number refers to input data (6GB/s input data can translate into 12GB/s depending on compression ratio), or output data (output data caped at 6GB/s will cancel SFS bebefits). If it refers only only to the input data, then SFS 2.5x multipier will work and we can celebrate
In practice it doesn't. Effectively it does.
Last edited:
eastwood333
Banned
So Cerny can make any statement and its considered fact, no PR ever!!!!
But if MS says something, its automatically PR?
I know some of you guys will die on the sword with PS5 having to have the SSD advantage given the XSX is superior in every other category but this is getting ridiculous.
Any positive news about XSX SSD, Velocity ,BC Pack etc and you guys come in with pitchforks. It's really funny to watch the damage control
Last edited:
Panajev2001a
GAF's Pleasant Genius
I didn't want to downplay what Sony did with the latency. But it really ends where a new SSD is placed into the system. There everything can only get optimized for bandwidth and not really latency because Sony has no influence on the inserted SSDs. They can only influence it by supporting only certain SSDs. They also gave to maintain the m.2 protocols and can't optimize there.
But the custom SSD extension allows MS to go further because everything is custom. Even a custom pricing
Sure, you made very good points, I was just pointing out that they thought about it hence their higher requirements and certifications for external SSD’s and to be fair you add an SSD, you do not trace the existing one. Would not be surprised to see them do something funky, but they did address latency minimisation concerns in the GDC talk.
MS will not need to require faster than internal SSD’s to compensate for the latency hiding techniques applied to the internal drive though. That is an advantage but yeah I do not think they will pass cost savings to consumers there
.While 6 priority levels is a customisation that standard .m2 SSD drives do not generally have, that is something they feel confident to be able to address as a bandwidth problem. The rest of the latency optimisations I think are all in the I/O processing complex. Low latency is as key if not more than bandwidth if you treat it as a way to expand RAM and I doubt they did not factor how to integrate external additional drives.
Having their own proprietary SSD that is exactly the same as the interns one slows them to reduce the logic they need to have to minimise/hide latency. Supporting more requests in flight, adding more SRAM cache in the I/O complex (makes best case/cache hits even lower latency), and reduce latency everywhere else in the pipeline (dedicated processors to map and I map memory, manage memory coherency, etc... the less you need to involve the CPU the less latency you have for requests)
Panajev2001a
GAF's Pleasant Genius
*compared to games that make no use of tiles resources / virtual texturing schemes.
Zoro7
Banned
Lol no. That’s not how it works.
Ascend
Member
With tiled resources you would still be loading three times what you would actually use, because there was no sampler feedback.
geordiemp
Member
I asked for the link where it is believed XSX SSD is lower latency ? Where is this stated ?
MS we are 2.5 x faster .....and instant. That is not scientific. Instant is not a scientific term.
Also your using strawman.
Cerny said 100 x faster than Ps4, 5.5 Gbs up to 22 Gbs max decompress, average 8-9. That is correct terminology.
MS said 2.4 Gbs and 4.8 Gbs with textures possible, max 6 Gbs, and maybe 2.5 x faster than something something.
You cannot answer as you dont know, so continue with teh attacks, strawman first, then make it personal.
Yawn.
Last edited:
D
Deleted member 775630
Unconfirmed Member
Not sure if this is true. I mean, we don't know this. They didn't clarify this, did they?
Yes it is, but feel free to proof me wrong.
Which is what most of the memory bandwidth is being used for...SFS is only being used for textures
Ascend
Member
Yeah... "Only" textures... What other files are larger than textures and need to be transferred more often than textures...?SFS is only being used for textures
CurtBizzy
Member
Those are really really small compared to textures.
CurtBizzy
Member
Says Lockhart
???Says Lockhart
Zoro7
Banned
This is like saying my car that only goes 100mph can actually go 200mph because it’s so efficient. It’s bollocks.
D
Deleted member 775630
Unconfirmed Member
Can you say why something is bollocks, or are you just using random and wrong metaphors to troll?
Kortan
Neo Member
No, but a more apt analogy is that a car that goes 100mph can go further if the road is straight and smooth. The same car on a road full of turns and potholes won't be used to it's full potential.