Support NeoGAF

[SonGoku]

Do you think is possible or at least worth it to implement in the new consoles with the new aggressive culling solutions in the same time with RT implementations like reflections ?

As for example a GE can remove geometry which doesn't exists in that time in the render window our reflection will be just partial so in part the RT waste part of
its potencial to make things which are not visible in camera affect the light in the scene.

I mean make senses as this kind of features (mesh shaders/primitive shaders) are more for its use in normal rasterization render than RT.

I think that if you start removing reflections from offscreen objects the results will be less accurate defeating the purpose of RY

Please read my earlier posts on what Project Acoustics really is. Basically they simulate the acoustics of all static environments of a game offline and create a simplified model out of the results by means of probes. This model is used in the XsX audio chip to get accurate room acoustics.

The PS5 will use RT for this purpose, yes. Although the question is how close the two solutions will be because raytracing does not deal with waves and probably needs some tricks to work properly.

I think we discussed this before?
The way you compared both isn't really apples to oranges since Project Acoustics is an API/engine devtool and the TE is an audio processor.
Also I don't think 3rd party devs are prohibited from implementing offline calculations done with Project Acoustics on PS5, similarly MS owns Havok but that doesn't stop devs from using it on competing consoles. Otherwise they would just use their inhouse solution for offline simulations across all platforms.

As far as RT hardware can there be any modifications on the API to do waves or similar bahavior?

 

[PaintTinJr]

You're both right. XSX can do more unique RT calculations per cycle in parallel (52 CUs) on (one case scenario) 10 GBs worth of game object data that's streamed in faster (560 GB/s). OTOH, PS5 does less unique RT calculations per cycle in parallel (36 CUs) on (one case scenario) 10 GB worth of game object data that's fed in slower (448 GB/s), but it can also do about 22% additional unique RT calculations in the same cycle on that same group of data because of the GPU's higher clock frequency.

Each approach has their advantages and disadvantages, depending on what effects devs want to achieve. But they essentially balance each other out in a lot of ways other than specific scenarios where one approach is better than the other in almost all of those cases (what those scenarios are we don't really know yet until we start seeing games in action).

Aside; I was reading some stuff and apparently MS has modified the group count for mesh shading on XSX to 256. The normal amount seems to be 32, and the other two are 64 and 128 (which is the max with Nvidia's support). Apparently the larger the group size the better, but I'm just going off what I read on other places like Beyond3D (they did screencap a MS Xbox team person who mentioned it in a video however (might've been a stream).

Seemed interesting to come across that. Hope we learn more about stuff like that for both systems, even if it's drip-fed.

What about the target resolution being no more than 1080p60? If the XsX couldn’t manage 1080p60 for the Minecraft RT demo it seems logical to assume that a wider workload of 4K - that would suit a wider CU count - isn’t on the cards, and at 1080p the narrow and faster GPU would fit the problem better.

 

[Tripolygon]

I have not made this claim. I have, however, explained that the primary goal of the Tempest Engine is to provide 3D Audio by means of HRTF and not room acoustics simulation. And I have also stated that any leftover computational power can be used by developers as they see fit. Can you now please stop this useless crusade?

What Darius87 is trying to tell you is that you are misunderstanding what Tempest Engine is.
Tempest engine is just a purpose built compute unit designed to be good at fast fourier transform which is how convolution is calculated. Simulating how sound moves, behaves and reacts in an environment. All these include sound occlusion, reverberance, decay etc.

Project Acoustics is wave acoustics engine that uses pre-computed forms of these effects that Tempest Engine can do in realtime.

 

[SonGoku]

What about the target resolution being no more than 1080p60? If the XsX couldn’t manage 1080p60 for the Minecraft RT demo it seems logical to assume that a wider workload of 4K - that would suit a wider CU count - isn’t on the cards, and at 1080p the narrow and faster GPU would fit the problem better.

I think thicc_girls_are_teh_best just worded it weirdly, i think we came to the same conclusion that the RT gap is the same as the compute gap which is to say 17-20%

 

[bitbydeath]

No, it did not.

If you insist that it did - please post a source.

I cant post a source since its pretty much impossible to link to something that does not exist.

PS5 and Xbox Series X GPU specs leak: how powerful is next-gen?

A remarkable story unfolded last April where it seemed that an intrepid explorer of the 3DMark benchmark database stumb…

www.eurogamer.net

The Navi 10 Lite was from the Gonzalo leak. Github leak said nothing about Navi 10 Lite. It also seems like the Gonzales leak did not spell out "Navi 10 Lite" and is perhaps just speculation.

So - in summary - github leak did not say anything about "Navi 10 Lite".

Not sure if anyone provided it yet but this is from github.

 

[PaintTinJr]

They are still working on TV speakers & multi speaker setups. There's a inherent technical limitation of speakers because one ear can hear what is meant for the other ear.

It is true that you can't stop the left ear's audio wave reaching the right ear- and vice versa- but using superposition of waves and knowledge of the waves, the position of the listener (and the acoustics of the room) a powerful real-time computation could take place that ensured the wave reaching the wrong ear is diminished temporarily at the point it reaches the wrong ear - effectively stopping you hearing it.. In fact, just thinking about your comment has me thinking that the DualSense isn't getting its name from the haptic triggers, but instead by using dual mono-aural mics (like ears) to get the pad to feedback the results of the Tempest 3D audio at merely 0.5-1.0m displacement from the gamers head/ears.

 

[thismeinteil]

PlayStation Twitter : 18 million followers

• Number 1 Gaming account
• Number 1 Brand most followed
• 125th most followed account overall

PlayStation Instagram : 24 million followers

• Number 1 Gaming account
• 19th Brand most followed
• 203rd most followed account overall

Top 5 most liked gaming post on Instagram :

1. 5.371 m - PS5 logo reveal
2. 4.609 m - PS5's DualSense design reveal
3. 2.683 m - Fortnite x Avengers Endgame
4. 2.546 m - Fortnite Season 6
5. 2.544 m - Fortnite Season X

(Credit to ArmGunar)

PS is going to have the Top 3 once the console is revealed.

 

[PaintTinJr]

I think thicc_girls_are_teh_best just worded it weirdly, i think we came to the same conclusion that the RT gap is the same as the compute gap which is to say 17-20%

But in those circumstances it won't be. The PS5 will be faster than the XsX at RT because the cost of coordinating CU workloads for a narrow workload will further burden the CPU and memory bandwidth on the XsX in all likelihood. PC RTX card benchmarks don't show you how much extra RAM bandwidth or CPU utilisation occurs on the wider and slower card by any chance when Path Tracing? Do they give frame pacing stats? I'm pretty sure if they did, faster and narrow at 1080p would be the winner.

 

[bitbydeath]

Nope

Sorry to hear mate.
Keep fighting and get well soon.

 

[SenjutsuSage]

You are missing the point, PS5 has less RT units but each unit is 22% faster and can perform more intersections per cycle
Its the same calculation as the compute performance and same gap 17-20%

This is not related to the GPU hardware capability to be introduced with RDNA2 as he said GPU SF is one part
Their customization is related to the streaming from SSD process, to fall back to lower quality texture while the high quality texture streams from SSD

Probably same as this.

Inside Xbox Series X: the full specs

This is it. After months of teaser trailers, blog posts and even the occasional leak, we can finally reveal firm, hard …

www.eurogamer.net

Bespoke hardware within the GPU is available to smooth the transition between mips, on the off-chance that the higher quality texture arrives a frame or two later.

It also turns out it wasn't introduced with RDNA 2, but was actually introduced as "partially resident textures" way back before even the ps4 and xbox were released on the Radeon HD 7970. I posted details in the thread earlier. DirectX 12 is just now exposing the feature for use, and it was always available via an OpenGL extension.

 

[Insane Metal]

Nope

Damn man, you've been sick for a while now, right? Please take care and get well soon!

 

[Tetragrammaton_Knight]

Probably same as this.

Inside Xbox Series X: the full specs

This is it. After months of teaser trailers, blog posts and even the occasional leak, we can finally reveal firm, hard …

www.eurogamer.net

It also turns out it wasn't introduced with RDNA 2, but was actually introduced as "partially resident textures" way back before even the ps4 and xbox were released on the Radeon HD 7970. I posted details in the thread earlier. DirectX 12 is just now exposing the feature for use, and it was always available via an OpenGL extension.

Well to put it simple - Sampler Feedback is a further extension/development of Partially Resident Textures, but it's not the same thing. There were a lot of GPU features that started small and with limited functionality but then grew, became more flexible while providing greater benefits. A recent example could be VRS move from tier 1 to tier 2.

 

[SonGoku]

It is true that you can't stop the left ear's audio wave reaching the right ear- and vice versa- but using superposition of waves and knowledge of the waves, the position of the listener (and the acoustics of the room) a powerful real-time computation could take place that ensured the wave reaching the wrong ear is diminished temporarily at the point it reaches the wrong ear - effectively stopping you hearing it.. In fact, just thinking about your comment has me thinking that the DualSense isn't getting its name from the haptic triggers, but instead by using dual mono-aural mics (like ears) to get the pad to feedback the results of the Tempest 3D audio at merely 0.5-1.0m displacement from the gamers head/ears.

Well... they are still working on it, the ceveat is that is harder to implement and considering computation resources are finite on consoles the amount of audio sources might be more limited compared to headphones

But in those circumstances it won't be. The PS5 will be faster than the XsX at RT because the cost of coordinating CU workloads for a narrow workload will further burden the CPU and memory bandwidth on the XsX in all likelihood. PC RTX card benchmarks don't show you how much extra RAM bandwidth or CPU utilisation occurs on the wider and slower card by any chance when Path Tracing? Do they give frame pacing stats? I'm pretty sure if they did, faster and narrow at 1080p would be the winner.

Interesting take from F Fafalada

Probably same as this.

That's just richard referring to hardware accelerated

It also turns out it wasn't introduced with RDNA 2, but was actually introduced as "partially resident textures" way back before even the ps4 and xbox were released on the Radeon HD 7970. I posted details in the thread earlier. DirectX 12 is just now exposing the feature for use, and it was always available via an OpenGL extension.

Yes I read your post, theres different tiers of support, RDNA2 implementation should be more advanced to meet the full support criteria

 

[Rusco Da Vino]

It's basically Bugatti Veyron vs Chiron - sure, the latter has 50% more HP, more VMax etc., but that doesn't mean Veyron is slow by any means.

But on the other hand, I'm pretty sure any dev whenever asked would take extra 2TF instead of half, extra 8GB RAM instead of 256MB, additional 8 threads instead of 2, and so on, because ultimately that's what they are working on, an actual hardware with actual specs, not on percentages or X-times multipliers.

Those 2TF might be as worthy as 2 PS4, enough to render TLoU2 and GoW4 for example, that's a lot if you ask me, not "just XX%". But then again, Sony might still stick to 30FPS, whereas MS clearly wants to deliver 60 in their games, so in simple math Sony's games can potentially have more TF/frame, and as a result have actually better visuals at the cost of framerate.

You missed this real dev talking about the PS5/XSX and the 1.8tf of difference, didnt you?

 

[psorcerer]

What about the target resolution being no more than 1080p60? If the XsX couldn’t manage 1080p60 for the Minecraft RT demo it seems logical to assume that a wider workload of 4K - that would suit a wider CU count - isn’t on the cards, and at 1080p the narrow and faster GPU would fit the problem better.

I think RT will be bottlenecked by random memory access.
Real life benchmarks by NV themselves could not extract more than ~3-4Grays/sec from the card with the claimed 10Grays/sec intersection capacity.

 

[SonGoku]

I think RT will be bottlenecked by random memory access.
Real life benchmarks by NV themselves could not extract more than ~3-4Grays/sec from the card with the claimed 10Grays/sec intersection capacity.

I remember reading APUs had an advantage due to everything being closer and less latency
thoughts?

 

[psorcerer]

I remember reading APUs had an advantage due to everything being closer and less latency
thoughts?

Nope, random memory access is slow and you can do nothing about that.
Better data locality can be achieved with a more tiled architecture and a ring bus (cell, wink wink), but modern GPUs will suck at it.

 

[SonGoku]

Nope, random memory access is slow and you can do nothing about that.
Better data locality can be achieved with a more tiled architecture and a ring bus (cell, wink wink), but modern GPUs will suck at it.

Kugoson redeemed?

 

[psorcerer]

Kugoson redeemed?

The guy is not writing like an engineer. As fake as it gets.

 

[Gediminas]

It's basically Bugatti Veyron vs Chiron - sure, the latter has 50% more HP, more VMax etc., but that doesn't mean Veyron is slow by any means.

But on the other hand, I'm pretty sure any dev whenever asked would take extra 2TF instead of half, extra 8GB RAM instead of 256MB, additional 8 threads instead of 2, and so on, because ultimately that's what they are working on, an actual hardware with actual specs, not on percentages or X-times multipliers.

Those 2TF might be as worthy as 2 PS4, enough to render TLoU2 and GoW4 for example, that's a lot if you ask me, not "just XX%". But then again, Sony might still stick to 30FPS, whereas MS clearly wants to deliver 60 in their games, so in simple math Sony's games can potentially have more TF/frame, and as a result have actually better visuals at the cost of framerate.

it didn't age well

 

[Fake]

I remember reading APUs had an advantage due to everything being closer and less latency
thoughts?

Correct. I remember reading an article made by AMD when they first launch their APU after ATI acquisition.

 

[Grodiak]

In fact, just thinking about your comment has me thinking that the DualSense isn't getting its name from the haptic triggers, but instead by using dual mono-aural mics (like ears) to get the pad to feedback the results of the Tempest 3D audio at merely 0.5-1.0m displacement from the gamers head/ears.

While this could THEORETICALLY work, I have little to no faith in that ever working practically. That 0.5-1m displacement, while it sounds small, is enough to have a lot happen. And what if I am laying down on my couch vs sitting up? What if I am slanted slightly to the left, leaning on my elbow? Too many variables. Hard to see it work even in a well-controlled environment.

Hell, if they somehow DO pull it off on a chip that is only a small part of the total cost of the box, they should sell that technology to soundstages around the world - and utilize such precise calibration capabilities on every single home-theatre system out there.

 

[Panajev2001a]

Its literally the other way around.

I do not believe nor do I see evidence of this being the case sorry.

 

[DaGwaphics]

What about the target resolution being no more than 1080p60? If the XsX couldn’t manage 1080p60 for the Minecraft RT demo it seems logical to assume that a wider workload of 4K - that would suit a wider CU count - isn’t on the cards, and at 1080p the narrow and faster GPU would fit the problem better.

The PS5 in that scenario would perform worse, the workload would be the same as XSX with less GPU compute for RT. Neither one of these consoles will be doing much full path tracing. That's a demo, shouldn't be a real expectation, neither is strong enough.

 

[megreotsugua]

They said it again.

VERY HIGH RESOLUTION TEXTURES AND ASSETS WILL FAVOR THE PS5.

Texture resolution/details > Screen output resolution.

Edit: Oh they said PS5 faster GPU will favor higher polygon count for more details in the game.

 

[BluRayHiDef]

Does anyone think that Sony and Microsoft will add more RAM to their boxes between now and their release dates?

 

[Shmunter]

While this could THEORETICALLY work, I have little to no faith in that ever working practically. That 0.5-1m displacement, while it sounds small, is enough to have a lot happen. And what if I am laying down on my couch vs sitting up? What if I am slanted slightly to the left, leaning on my elbow? Too many variables. Hard to see it work even in a well-controlled environment.

Hell, if they somehow DO pull it off on a chip that is only a small part of the total cost of the box, they should sell that technology to soundstages around the world - and utilize such precise calibration capabilities on every single home-theatre system out there.

Even if not ideal, it still would potentially deliver results leagues ahead of static systems like DTS, Dolby or Atmos.

 

[Metroiddarks]

Does anyone think that Sony and Microsoft will add more RAM to their boxes between now and their release dates?

Not GDDR6 but maybe PS5 will use the same as in PS4 and add little more DDR4 in this case only for the OS.

 

[megreotsugua]

You missed this real dev talking about the PS5/XSX and the 1.8tf of difference, didnt you?

They won't even listen to a rendering engineer from Crytek. It'll be too much to expect for them to listen to that dev.

 

[B_Boss]

You missed this real dev talking about the PS5/XSX and the 1.8tf of difference, didnt you?

While Jason there isn’t exactly saying what Salehi said, there seems to be an almost similar tone, a sort of excitement about the PS5 that they (and reportedly other devs.) share.

 

[megreotsugua]

Nope, random memory access is slow and you can do nothing about that.
Better data locality can be achieved with a more tiled architecture and a ring bus (cell, wink wink), but modern GPUs will suck at it.

Looking at RT on and off comparisons video on youtube, I would say RT benefit is minimal at best, and sometimes it even looks worse.

1. Is there a benefit to even implement RT at all in the next-gen systems then?
2. Why waste computational resources on an inefficient process?
3. What are your expectations on AA and AAA games next-gen regarding RT?
4. Those games that will try to push better looking games as much as possible, better realism and more eye-popping graphics, do you see them implementing RT?
5. Also, am I correct in my understanding that devs can use offline RT to approximate the kind of lighting they need to bake so that lighting in-game will not be too computational intensive? Same effect but less GPU resources?

Edit: Also, is baked lighting (eg. lightmaps) data-driven technique? I mean, the better the baked lighting, the bigger the data?


Sorry if my questions are too basic.

 

[DaGwaphics]

Pre-baked lighting can be very "expensive" in a lot of ways. I think that next-gen AAAs will use RT extensively, but in combination with existing techniques. RT as the only light model won't be a thing, IMO.

 

[Grodiak]

Even if not ideal, it still would still potentially deliver results leagues ahead of static systems like DTS, Dolby or Atmos.

Yeah, for sure. Don't get me wrong - I am beyond ecstatic that so much weight is being put into the sound side of things! It is great that so much innovation is being put into the audio side! People just tend to not understand how hardware-intensive some real-time audio calculations are though. I mean, it is "just" sound, right?

Atmos' brand in consumer electronics has, however always been a bit lost to me. Please, someone smarter explain. Again, I live in the film world - have mixed in Atmos. But clearly theatrical Atmos and using that brand in consoles, for instance, have very different meanings. I doubt people have an array of speakers mounted on their ceilings? And soundbars that suggest they bounce sounds from your ceiling is nothing but a gimmick.

 

[thicc_girls_are_teh_best]

I think thicc_girls_are_teh_best just worded it weirdly, i think we came to the same conclusion that the RT gap is the same as the compute gap which is to say 17-20%

Yeah basically. I just wanted to illustrate with a rough example how the performance of both would work in practice when it comes to RT effects, with a standardized scenario, to maybe help some folks better visualize it instead of going off the numbers alone.

 

[StreetsofBeige]

Looking at RT on and off comparisons video on youtube, I would say RT benefit is minimal at best, and sometimes it even looks worse.

1. Is there a benefit to even implement RT at all in the next-gen systems then?
2. Why waste computational resources on an inefficient process?
3. What are your expectations on AA and AAA games next-gen regarding RT?
4. Those games that will try to push better looking games as much as possible, better realism and more eye-popping graphics, do you see them implementing RT?
5. Also, am I correct in my understanding that devs can use offline RT to approximate the kind of lighting they need to bake so that lighting in-game will not be too computational intensive? Same effect but less GPU resources?

Sorry if my questions are too basic.

To sum up, it gives graphic artists a way to show their skills making every surface shiny like a mirror.

 

[BluRayHiDef]

Some of these numbers are confirmed but not all.

Here is unconfirmed numbers from Techpowerup (https://www.techpowerup.com/):

PS5:
Shading Units 2304 (Stream Processors?)
TMUs 144 (Texture mapping units)
ROPs 64 (Raster operators?)
Compute Units 36 (Compute units)
L2 Cache4 MB

Xbox Series X:
Shading Units 3328
TMUs 208
ROPs 80
Compute Units 52
L2 Cache 5 MB

The only thing we know for sure as far as I know is the CU count.

PlayStation 5's GPU:
2,230 Mhz (frequency)
2304 SUs (shading units)
144 TMUs (texture mapping units)
64 ROPs (render output units)
36 CUs (compute units)
4MB of L2 Cache

Xbox Series X's GPU:
1,825 Mhz (frequency)
3,328 SUs (shading units)
208 TMUs (texture mapping units)
80 ROPs (render output units)
52 CUs (compute units)
5MB of L2 Cache

-------------------------- Shading Rate Difference

PS5's GPU:
2304 SUs x 2,230 x 1000 = 5,137,920,000 shading operations per second

XSX's GPU:
3,328 SUs x 1,825 x 1000 = 6,073,600,000 shading operations per second

Calculation of Percentage Difference: (5,137,920,000 shading operations per second) / (6,073,600,000 shading operations per second ) = 0.845943098 -> 0.845943098 x 100 = 84.5943098% = ~ 84.59%

The PlayStation 5's shading rate is 84.6% of the Xbox Series X's shading rate.

-------------------------- Fillrate Difference

PS5's GPU:
144 TMUs x 2,230 x 1000 = 321,120,000 texels per second

XSX's GPU:
208 TMUs x 1,825 x 1000 = 379,600,000 texels per second

Calculation of Percentage Difference: (321,120,000 texels per second) / (379,600,000 texels per second ) = 0.845943098 -> 100 x 0.845943098 = 84.5943098% =~ 84.6%

The PlayStation 5's fill rate is 84.6% of the Xbox Series X's fill rate.

-------------------------- Render Output Rate Difference

PS5's GPU:
64 ROPs x 2,230 x 1000 = 142,720,000 rendering operations per second

XSX's GPU:
80 ROPs x 1,825 x 1000 = 146,000,000 rendering operations per second

(142,720,000 rendering operations per second)/(146,000,000 rendering operations per second) = 0.9775342466 -> 0.9775342466 x 100 = 97.75342466 =~ 97.75%

The PS5's render output rate is 97.75% of the XSX's render output rate.

-------------------------- Compute Rate Difference

PS5's GPU:
36 CUs x 2,230 x 1000 = 80,280,000 computations per second

XSX's GPU:
52 CUs x 1,825 x 1000 = 94,900,000 computations per second

(80,280,000 computations per second)/(94,900,000 computations per second) = 0.845943098 -> 0.845943098 x 100 = 84.5943098% =~ 84.59%

The PS5's computation rate is 84.59% of the XSX's computation rate.

-------------------------- L2 Cache Bandwidth Difference

PS5:
4MB x 2,230 x 1000 = 8,920,000 MB/s

XSX:
5MB x 1,825 x 1000 = 9,125,000 MB/s

(8,920,000 MB/s)/(9,125,000 MB/s) = 0.9775342466 -> 0.9775342466 x 100 = 0.9775342466 =~ 97.75%

The PS5's L2 Cache Bandwidth is 97.75% of the XSX's L2 Cache bandwidth

________

So, ah, are these calculations correct?

 

[StreetsofBeige]

PlayStation 5's GPU:
2,230 Mhz (frequency)
2304 SUs (shading units)
144 TMUs (texture mapping units)
64 ROPs (render output units)
36 CUs (compute units)
4MB of L2 Cache

Xbox Series X's GPU:
1,825 Mhz (frequency)
3,328 SUs (shading units)
208 TMUs (texture mapping units)
80 ROPs (render output units)
52 CUs (compute units)
5MB of L2 Cache

-------------------------- Shading Rate Difference

PS5's GPU:
2304 SUs x 2,230 x 1000 = 5,137,920,000 shading operations per second

XSX's GPU:
3,328 SUs x 1,825 x 1000 = 6,073,600,000 shading operations per second

Calculation of Percentage Difference: (5,137,920,000 shading operations per second) / (6,073,600,000 shading operations per second ) = 0.845943098 -> 0.845943098 x 100 = 84.5943098% = ~ 84.59%

The PlayStation 5's shading rate is 84.6% of the Xbox Series X's shading rate.

-------------------------- Fillrate Difference

PS5's GPU:
144 TMUs x 2,230 x 1000 = 321,120,000 texels per second

XSX's GPU:
208 TMUs x 1,825 x 1000 = 379,600,000 texels per second

Calculation of Percentage Difference: (321,120,000 texels per second) / (379,600,000 texels per second ) = 0.845943098 -> 100 x 0.845943098 = 84.5943098% =~ 84.6%

The PlayStation 5's fill rate is 84.6% of the Xbox Series X's fill rate.

-------------------------- Render Output Rate Difference

PS5's GPU:
64 ROPs x 2,230 x 1000 = 142,720,000 rendering operations per second

XSX's GPU:
80 ROPs x 1,825 x 1000 = 146,000,000 rendering operations per second

(142,720,000 rendering operations per second)/(146,000,000 rendering operations per second) = 0.9775342466 -> 0.9775342466 x 100 = 97.75342466 =~ 97.75%

The PS5's render output rate is 97.75% of the XSX's render output rate.

-------------------------- Compute Rate Difference

PS5's GPU:
36 CUs x 2,230 x 1000 = 80,280,000 computations per second

XSX's GPU:
52 CUs x 1,825 x 1000 = 94,900,000 computations per second

(80,280,000 computations per second)/(94,900,000 computations per second) = 0.845943098 -> 0.845943098 x 100 = 84.5943098% =~ 84.59%

The PS5's computation rate is 84.59% of the XSX's computation rate.

-------------------------- L2 Cache Bandwidth Difference

PS5:
4MB x 2,230 x 1000 = 8,920,000 MB/s

XSX:
5MB x 1,825 x 1000 = 9,125,000 MB/s

(8,920,000 MB/s)/(9,125,000 MB/s) = 0.9775342466 -> 0.9775342466 x 100 = 0.9775342466 =~ 97.75%

The PS5's L2 Cache Bandwidth is 97.75% of the XSX's L2 Cache bandwidth

________

So, ah, are these calculations correct?

You're assuming PS5 can run at max cpu and gpu ghz at all times.

 

[BluRayHiDef]

You're assuming PS5 can run at max cpu and gpu ghz at all times.

This isn't about the CPU but only the GPU, and Digital Foundry has revealed that developers have been running the GPU at its max frequency.

 

[lepermessiah]

Muh juice be bubbling.

 

[Shmunter]

-------------------------- Render Output Rate Difference

PS5's GPU:
64 ROPs x 2,230 x 1000 = 142,720,000 rendering operations per second

XSX's GPU:
80 ROPs x 1,825 x 1000 = 146,000,000 rendering operations per second

(142,720,000 rendering operations per second)/(146,000,000 rendering operations per second) = 0.9775342466 -> 0.9775342466 x 100 = 97.75342466 =~ 97.75%

The PS5's render output rate is 97.75% of the XSX's render output rate.

......

So, ah, are these calculations correct?

I Believe the 80 vs 64 ROPS on XsX is still under contention

 

[ethomaz]

Some of these numbers are confirmed but not all.

Here is unconfirmed numbers from Techpowerup (https://www.techpowerup.com/):

PS5:
Shading Units 2304 (Stream Processors?)
TMUs 144 (Texture mapping units)
ROPs 64 (Raster operators?)
Compute Units 36 (Compute units)
L2 Cache4 MB

Xbox Series X:
Shading Units 3328
TMUs 208
ROPs 80
Compute Units 52
L2 Cache 5 MB

The only thing we know for sure as far as I know is the CU count.

ROPs and L2 cache are the only doubts I have.

If Xbox GPU has 2 Shader Engines / 4 Shader Arrays then I believe it will have 64 ROPs.

 

[BluRayHiDef]

I Believe the 80 vs 64 ROPS on XsX is still under contention

In regard to what exactly? That the PS5 has more ROPs or that the XSX has less?

 

[ethomaz]

In regard to what exactly? That the PS5 has more ROPs or that the XSX has less?

Both 64.

It is 16 ROPs per Shader Array.

 

[SlimySnake]

You're assuming PS5 can run at max cpu and gpu ghz at all times.

DF already confirmed that devs are running at max gpu frequency, and cerny himself stated that just a few percent drop in frequency can save 10% power reduction. the ps5 cpu is probably running at 3.2 ghz when gpu clocks are max.

 

[StreetsofBeige]

Both 64.

It is 16 ROPs per Shader Array.

Where does it say both systems have 64 rops?

 

[BluRayHiDef]

Both 64.

It is 16 ROPs per Shader Array.

Since the corresponding components for each console's GPU amount to less units in that of the PS5, wouldn't the PS5 have a lower number of shader arrays and therefore a lower number of ROPs?

 

[BluRayHiDef]

Does anyone know how many cores of each console's CPU is dedicated to each console's OS?

 

[SonGoku]

Does anyone know how many cores of each console's CPU is dedicated to each console's OS?

XSX 1
PS5 we don't know but reasonable assumption is 1 too

 

[ethomaz]

Since the corresponding components for each console's GPU amount to less units in that of the PS5, wouldn't the PS5 have a lower number of shader arrays and therefore a lower number of ROPs?

Because increasing the number of Shader Array increase the complexity and power draw / heat of the chip.

I can only see a single option:

- 4 Shader Arrays with 7 WGP each

How can you match 56 CUs with 6 SAs or 8 SAs? 6 SAs with 5 WGP = 60 CUs, 6 SAs with 4 WGP = 48 CUs, 8 SAs with 4 WGP = 64 CUs, 8 SAs with 3 WGP = 48 CUs.

IMO only 4 SAs with 7 WGP makes sense.

PS5 is 4 SAs with 5 WGP each... identical to Navi 10.... AMD already showed more WGP per SA like Navi 14 that has 2 SAs with 6 WGP each.

 

[ethomaz]

Where does it say both systems have 64 rops?

If Xbox APU is 4 Shader Arrays then it indeed has 64 ROPs.

I will say more if it is not 64ROPs then there is way more chances to be 96ROPs.

80 ROPs is really weird.