Current Generation Hardware Speculation with a Technical Spin [post launch 2021] [XBSX, PS5]

Status
Not open for further replies.
Direct X 12 U on the XSX has an ever lower level of access than Direct X 12 on the PC. You can't compare DX 12 U on PC with Vulkan and then put that onto the XSX.
There are some additional commands that are console specific, and some additional options that don't exist on DX12 because of hardware customization. But outside of that, I don't think DX12 is closer to the metal in that sense. I haven't seen any traces of information that xbox titles can access memory or insert commands at an additional level lower than specified by public API, outside of console specific instructions.
 
There are some additional commands that are console specific, and some additional options that don't exist on DX12 because of hardware customization. But outside of that, I don't think DX12 is closer to the metal in that sense. I haven't seen any traces of information that xbox titles can access memory or insert commands at an additional level lower than specified by public API, outside of console specific instructions.
I think it was the hot chips talk where the MS guys were asked if there was any difference between DX12 for XSX and PC and they answered there was console differences that allowed closer to the metal than PC DX12 did.
 
I think it was the hot chips talk where the MS guys were asked if there was any difference between DX12 for XSX and PC and they answered there was console differences that allowed closer to the metal than PC DX12 did.

From what I remember of the hot chips talk it was somewhat ambiguous about whether the 'closer to the metal' coding was in reference to the code that a game developer would write and the code that the directX team wrote. its entirely possibly that they could streamline the driver implementation on the xbox consoles, resulting in a dx12 implementation that was more efficient/faster/overall better, and 'closer to the metal' while still not giving developers anything closer to the metal
 
DX12U for XS does have some additional settings and features that are exposed.
But from a general discussion I don't think there's anything wrong in using the general DX12U spec as the basis. Especially as we don't know all the specific changes.

XS DX12U aren't that far removed from eachother. In fact wouldn't even call them simply siblings, more like twins. That's how close they are.
 
With Sony adding Cache Scrubbers onto the PS5, do we really have any idea of potential performance savings that it can expect?
From what Sony have said, AMD were also able to use them in the RDNA 2 cards, but I'm not sure if they did.

I realise Sony are not saying anything more at this point, which is disappointing, but what do we actually know?

I was talking to a computer engineer who loves gaming and he said that because the PS5 GPU was so fast that there is inherent problems with Cache getting clogged up (his basic wording to me), and the Cache Scrubbers were required to keep the GPU running efficiently at those speeds.
That's the only thing I have heard, and that's not from Sony, so what do we know at this point?
I'm not even aware of any patents around them.

So what's up with those scrubbers?
 
With Sony adding Cache Scrubbers onto the PS5, do we really have any idea of potential performance savings that it can expect?
From what Sony have said, AMD were also able to use them in the RDNA 2 cards, but I'm not sure if they did.

I realise Sony are not saying anything more at this point, which is disappointing, but what do we actually know?

I was talking to a computer engineer who loves gaming and he said that because the PS5 GPU was so fast that there is inherent problems with Cache getting clogged up (his basic wording to me), and the Cache Scrubbers were required to keep the GPU running efficiently at those speeds.
That's the only thing I have heard, and that's not from Sony, so what do we know at this point?
I'm not even aware of any patents around them.

So what's up with those scrubbers?

My understanding is that they're located at the CPU end, by the secondary CCX.

In other words, they're hanging on the passenger side of their best friend's ride.

Sorry. I would actually like to know more about them too.
 
With Sony adding Cache Scrubbers onto the PS5, do we really have any idea of potential performance savings that it can expect?
From what Sony have said, AMD were also able to use them in the RDNA 2 cards, but I'm not sure if they did.

I realise Sony are not saying anything more at this point, which is disappointing, but what do we actually know?

I was talking to a computer engineer who loves gaming and he said that because the PS5 GPU was so fast that there is inherent problems with Cache getting clogged up (his basic wording to me), and the Cache Scrubbers were required to keep the GPU running efficiently at those speeds.
That's the only thing I have heard, and that's not from Sony, so what do we know at this point?
I'm not even aware of any patents around them.

So what's up with those scrubbers?

I believe the cache scrubbers were a cost-saving method for keeping silicon costs low and chip sizes manageable (especially for future shrinks). Instead of larger and costlier cache taking up die space, create something (in this case cache scrubbers) that can evict old unwanted assets/data quickly, rather than needing huge chunks of costly cache to store assets/data being swapped in/out at a given moment.
 
I believe the cache scrubbers were a cost-saving method for keeping silicon costs low and chip sizes manageable (especially for future shrinks). Instead of larger and costlier cache taking up die space, create something (in this case cache scrubbers) that can evict old unwanted assets/data quickly, rather than needing huge chunks of costly cache to store assets/data being swapped in/out at a given moment.

Cerny talks about it at 19m15s in the road to PS5 video.
The GPU's cache scrubbers seem to actually be controlled by the coherency engines in the I/O complex. It seems they're just small controllers next to the caches that put zeroes into ranges of memory addresses given to them by the coherency engines.

That said, they're probably not discernible in the SoC's x-ray pictures as they're likely tiny blocks in the chip. It also makes sense that AMD wouldn't be interested in putting them into the RDNA2 PC GPUs as they won't be paired with anything resembling the PS5's coherency engines.
 
Cerny talks about it at 19m15s in the road to PS5 video.
The GPU's cache scrubbers seem to actually be controlled by the coherency engines in the I/O complex. It seems they're just small controllers next to the caches that put zeroes into ranges of memory addresses given to them by the coherency engines.

That said, they're probably not discernible in the SoC's x-ray pictures as they're likely tiny blocks in the chip. It also makes sense that AMD wouldn't be interested in putting them into the RDNA2 PC GPUs as they won't be paired with anything resembling the PS5's coherency engines.
Well, a question that comes to my mind is, what do cache scrubber actually add.
If they increase efficiency of the cache (hit rate) than the GPU is less idle, which also means it might be more efficient. But on the other hand, they have a power-limit for the gpu, so if the gpu is more efficient, the GPU has less idle time and therefore needs more power which might end in less frequency. Less frequency means, the cache-scrubbers, the cache and other parts are slow so less efficient (less than with higher frequency). So it is really get's complicated. A shame that there is no easy way to experiment with those chips to better understand what they are doing.
What I also do not understand there, doesn't it add latency to the "cache refresh" if the cache-scrubbers first check if everything is up to date in the cache. It still has to read the data and compare it part by part with the data in main memory. So it actually might read a bit more but can write faster if not everything has changed. What I mean is (and I have experienced something like this often in my carrier) sometimes it is faster just to do something instead of first checking if it must be done.
I know, they wouldn't have made it if it wouldn't be somehow faster/more efficient. But I want to understand why.
 
Cerny talks about it at 19m15s in the road to PS5 video.
The GPU's cache scrubbers seem to actually be controlled by the coherency engines in the I/O complex. It seems they're just small controllers next to the caches that put zeroes into ranges of memory addresses given to them by the coherency engines.

That said, they're probably not discernible in the SoC's x-ray pictures as they're likely tiny blocks in the chip. It also makes sense that AMD wouldn't be interested in putting them into the RDNA2 PC GPUs as they won't be paired with anything resembling the PS5's coherency engines.

Sure. But the main purpose of cache scrubbers is to automatically evict/scrub data that is no longer required by the GPU. However, from my understanding developers do have a certain level of control on flagging what data should or should not be evicted, or keep certain data at an 'always available state', without wasting precious compute cycles/time.
 
However, from my understanding developers do have a certain level of control on flagging what data should or should not be evicted, or keep certain data at an 'always available state', without wasting precious compute cycles/time.
I think Cerny said the I/O complex's functionality (which would then include the cache scrubbers) is transparent to the developers. At the end of the cache scrubbers' description, he says something along the lines of "as a developer, the best part is you don't need to know any of this". He says all they need to do is tell the system which assets they want to use.
 
I think Cerny said the I/O complex's functionality (which would then include the cache scrubbers) is transparent to the developers. At the end of the cache scrubbers' description, he says something along the lines of "as a developer, the best part is you don't need to know any of this". He says all they need to do is tell the system which assets they want to use.

I'm not following or maybe misunderstanding what you're trying to present. I understand what Cerny has stated, but developers do have access to the CSs on making sure certain data isn't being evicted too early or keeping certain data in an idle state. If the developer chooses none of these things (no granular control), the CSs act as Cerny stated.
 
I'm not following or maybe misunderstanding what you're trying to present. I understand what Cerny has stated, but developers do have access to the CSs on making sure certain data isn't being evicted too early or keeping certain data in an idle state. If the developer chooses none of these things (no granular control), the CSs act as Cerny stated.

What I gather from Cerny's statements is the developers only get to say which assets need to be in the RAM at any given time. They don't access the cache scrubbers directly as those only take commands from the coherency engine.
I don't know if the developers have access to the coherency engine, but if it works as advertised I don't know why they would want to.
 
Looking at the specs deeper, what is the biggest upgrade over last gen? CPU, SSD, or the ecosystem? At first glance the CPU seems to standout well above the Jaguar.
 
Thinking about it, I don't really see as much of an opportunity for a mid gen refresh this time around. "4k" hit pretty big around the middle of last time and was as good a business reason as you'd find for such a thing. Now the two "big" consoles are "4k" ready anyway, and it'll take a while for the vast majority of consumers to get a 4k hdr tv. Thus the existence of the Series S to begin with.

What I don't see is any immediate display tech on the horizon to justify another such optional upgrade. HDR will advance slowly, but that's somewhat coverable. 8k is cool looking, but I feel like the diminishing returns aren't going to make it as exciting as 4k was, movies and shows are still catching up with "4k" as it is. I can see controllers advancing, at least optionally. Microsoft would be smart to put out a feature parity controller to the PS5's, but the survey they put out shows they're possibly too stupid to do so (you don't ask the average consumer if they want to pay money for something they've never tried personally.) I can also see minor updates. Wireless charging for controllers if that ever gets long range and convenient, bigger SSDs as costs come down, newer Wifi and mediate outputs, slimmer cases and new colors, a lot of the classic stuff, but maybe not rendering upgrades.
 
Thinking about it, I don't really see as much of an opportunity for a mid gen refresh this time around.

Potential for much better RT. This generation of consoles has AMD's first gen RT, it's likely they'll be able to increase the capabilities and performance of their RT significantly with what they learn from their first gen. implementation. This is especially true as they get feedback from developers on what works, what doesn't work, and what developers would like to have available in hardware.

WRT RT, it's possible a mid gen refresh could represent a similar or larger jump than from previous gen to current gen. It just depends on how R&D on RT hardware assist on GPUs progress.

With both Sony and especially MS, basically having the foundation for rolling generations, this allows both companies to incorporate new technologies or advancements in technology without having to drop support for the older console.

Regards,
SB
 
Potential for much better RT. This generation of consoles has AMD's first gen RT, it's likely they'll be able to increase the capabilities and performance of their RT significantly with what they learn from their first gen. implementation. This is especially true as they get feedback from developers on what works, what doesn't work, and what developers would like to have available in hardware.

WRT RT, it's possible a mid gen refresh could represent a similar or larger jump than from previous gen to current gen. It just depends on how R&D on RT hardware assist on GPUs progress.

With both Sony and especially MS, basically having the foundation for rolling generations, this allows both companies to incorporate new technologies or advancements in technology without having to drop support for the older console.

Regards,
SB

Saying "better raytracing" doesn't seem more interesting to the average consumer than saying better anything else. I think the "4k!!!" advertising for the PS4 Pro and One X was pretty telling for what people bought it for, and that's not happening again. It's not as easy to sell better raytracing either, there's no "bigger number" you can easily point to for the average consumer. I can see an optional "Boost mode" maybe, like with the PS4 Pro, on a PS5 Slim. Let's face it, a PS5 "Slim" or synonym would be pretty welcome. And if it's say, built on TSMC N4 then you could have an optional 20% boost mode to get titles that dip from 60fps up, or unlocked titles higher, etc.

But I'd judge that the average consumer that really, really cares about "better raytracing, moar pixels!" has a huge overlap with consumers that build gaming PCs anyway. Not saying it won't happen, but with the PS4 outselling the PS4 Pro even when they were both out, and no opportunity as big in the awareness of the average consumer coming, I just don't see it as an obvious business case versus say, that PS5 slim with a boost mode.
 
Looking at the specs deeper, what is the biggest upgrade over last gen? CPU, SSD, or the ecosystem? At first glance the CPU seems to standout well above the Jaguar.
SSD is by far the biggest upgrade.
We're going from raw 40MB/s to raw 2500MB/s <-> 5500MB/s plus higher compression ratios.

Nothing else is getting a 100x jump. In fact, save for maybe the SNES -> N64 performance in 3D rasterization I don't know of any other subsystem getting a jump of a similar order of magnitude between console generations.
 
Status
Not open for further replies.
Back
Top