Support NeoGAF

[Lethal01]

Thats why on PC you have different settings like low, medium, ultra. Etc.
But you can NOT have different settings based SSD. It doesn’t work like that.

The Unreal engines devs say the exact opposite, that with nanite it's easy to adjust the amount of data that's being streamed in to the point it can run on anything.

Well UE5 is another story. It's rendering methods are different from traditional.

But UE is tiny part of whole industry. PS5 will have lead in every UE game? Okey, no problem. But that like 2-3% of games.

You will see the tech of the Unreal Engine in the other major engines, it's not something that's exclusive to Epic at all. You will see it in Unity, You will see it from activision, You will see it from Ubisoft.

 

[geordiemp]

It is Amd smartshift, that's all there is to it. Look where Amd deploy smartshift and you get the same answer.

The hardware is still limited by smaller die and the choice to clock up to 2.23ghz.
It is this 2.23ghz, which gives 10.3tf, that we are concerned and suspicious of.

We you mean you are concerned.

OK lets try again. Its far easier to control heat if you can turn down the power predictively rather than waiting for the die to get over a thermal limit.

Thermal throttling is a slow PID loop where your reacting too late. Proportional integral and Derivative, look it up. Engineers will have poured statistics and thermal modelling / predictive control over thousands of engineering hours and modelled it to perfection for a consumer unit expected in 100 million volumes.

What if you can change clocks and power in milliseconds and can do it when edge cases of work to be done is likley to increase heat predictively ? What if you could do a 2 % drop to 10.1 TF in 1/8 of a frame time domain. when you have to every now and again....

What if that allowed you to do 10.28 TF for most of the frame, and drop 20 % of that heavy workload outlier frame time to 10.1 TF and that made the system much easier to cool ?

What if you could drop to 5 TF when nothing is happening for 1/10 of a frame, cooling the GPU even more, why run GPU hard if its heavy CPU time. Let GPU chill out when it can

Most are just ignorant and trolling, and wonder why its easier just to post a laugh GIF, its not worth the effort as posters dont even want to understand the concept of predictive control vs reactive.

 

[JTCx]

How is 12.1 theoretical?

Yeah...you're done

lmfao.

 

[longdi]

Sorry, but this is wilful ignorance. The post is very detailed but easy to understand. It even has a useful animated gif.

I'm not sure why you even bring up SmartShift or talk about the hardware being limited by being smaller die...?

Eh no.. because of how apu is design , you can do some load balancing like smartshift to eke out more performance in a thermal and power constrained closed system, but at the expense of some latency.

Sustained and variable can work hand in hand

I won't be surprised if Series X has the same feature, or MS can disable it if they find the latency penalty not worth.

It is always about the 2.23ghz figures, which gives 10.3tf, which gives the 'small' difference defence.

The question, of how often PS5 can stay at 2.23ghz, is the difference 2tf or more in realworld.

This suspicion is confounded by how Mark seems skirty to answer. And went in big circle to present smart shift as a big new feature. Why did he/Sony not strongly advertise PS5 as the first gpu running north of 2ghz? Would make headlines too!

Like 3950x is advertised as up to 4.7ghz, in practice, you only see that speed in a millisecond, on a single thread, using software reader. Who knows if AMD can also fudged it.

 

[Eliciel]

Oh dear, TF on all computers is the max calculated number if all the CUs are operating at the same time at 100 % efficiency. Never happens on anything.

Its like sayng my car operates at 180BHP and can go 150 mph..... A proper benchmark is 0-60 mph in 8 seconds on a flat road, no wind or rain..

So the actual real TF applied to any task will depend on how the power is used by the system for the tasks presented and any efficiencies in applying that power.

Mabe XSX will use more of its potential TF than Ps5, maybe the other way around in benchmarks....what will be the delta, will it increase more than the paper 15/18 % (Up/down) or will the difference be the smallest we have ever seen in a generation ?

What have third party devs indicated...they would be the ONLY people that would know this.

maybe others are helped with answering my question as well:

On my PC my AMD Ryzen 3600 is clocked at 3.6 ghz and 4.5 (Turbo-Frequency). If you check on programs like Core Temp or CPU-Z etc. you can see that frequency varies based on load from 3.6 - 4.5 GHZ but it literally NEVER stays there AND not all cores boost simoultaneously ALL the time.

If you would try to explain this based on a basic CPU like 3600 on PC what does it mean in correlation to the XBOX and PS5 explanation:

This is what is being revealed about the XSEX processor:
AMD Zen 2 CPU with eight cores running at a nominal 3.8GHz, or when simultaneous multithreading (SMT) is used, at 3.6GHz.
(All information are drawn out of WIkipedia: Xbox Series X Hardware Wikipedia Aritcle)

This is what is being revealed about the PS5 processor:
AMD's 7nm Zen 2 microarchitecture with 8 CPU cores running at a variable frequency capped at 3.5 GHz.
(All information are drawn out of WIkipedia: PS5 Hardware Wikipedia Article)

--> How does this compare to:
AMD Ryzen 3600 clocked at 3.6ghz and 4.5 Turbo-Frequency.

So, again, please, could you do us all a favor and explian how does this compare to XSEX and PS5 based on that counter example from PC?
Anyone else feeling ready to answer this question and explain the difference and similarities, please go ahead and make this thread "read-worthy"..

 

[THE:MILKMAN]

longdi

Confirm you have read Liabe's post and then try to discuss and rebut what he says.

 

[FranXico]

maybe others are helped with answering my question as well:

On my PC my AMD Ryzen 3600 is clocked at 3.6 ghz and 4.5 (Turbo-Frequency). If I look in a tool like Core Temp I get following picture:
If you would try to explain this based on a basic CPU like 3600 on PC what does it mean in correlation to the XBOX and PS5 explanation:

This is what is being revealed about the XSEX processor:
AMD Zen 2 CPU with eight cores running at a nominal 3.8GHz, or when simultaneous multithreading (SMT) is used, at 3.6GHz.
(All information are drawn out of WIkipedia: Xbox Series X Hardware Wikipedia Aritcle)

This is what is being revealed about the PS5 processor:
AMD's 7nm Zen 2 microarchitecture with 8 CPU cores running at a variable frequency capped at 3.5 GHz.
(All information are drawn out of WIkipedia: PS5 Hardware Wikipedia Article)

--> How does this compare to:
AMD Ryzen 3600 clocked at 3.6ghz and 4.5 Turbo-Frequency.
If you check on programs like Core Temp or CPU-Z etc. you can see that frequency varies based on load from 3.6 - 4.5 GHZ but it literally NEVER stays there AND not all cores boost simoultaneously ALL the time.

So, again, please, could you do us all a favor and explian how does this compare to XSEX and PS5 based on that counter example from PC?
Anyone else feeling ready to answer this question and explain the difference and similarities, please go ahead and make this thread "read-worthy"..

On the PS5 CPU is also SMT, they configure it always on didn't even bother listing a frequency with it off.

 

[Elog]

Those textures need to be processed by the GPU and PS5 has a texture fillrate deficit to series X.

This is a conclusion that cannot be drawn - even though we do not know this since the silicon has not been shown, I would argue that right now the opposite is more likely to be true. If you look at the PS4 Pro vs XBox One X, the PS4 Pro has a texture filtrate advantage over Xbox One X despite having a Tflops disadvantage due to more silicon dedicated to this.

My bet is that with the PS5 they have followed in the footsteps of the PS4 Pro and will then have an oversized section of the GPU with render mapping/ shaders/ texture output and higher frequency to boot. That would be the logical design given what they are trying to achieve.

Overall comment (sorry if it is wordy):

I believe what many of you are missing is that within the PC framework there is only so much that can be done with texture resolution. As an example, a single model in avatar used 100-120 textures at 8K. Even at a 'measly' 4K this represents 5GB+ of uncompressed textures. Since the VRAM needs to contain all the textures that can be used within the load time of your I/O solution, a normal mid-range GPU card taps out even at the first model. It is undoable.

Within the PC framework we then try to work around this by using crap textures but smaller and smaller polygons for the rendering of the picture - i.e. we bandaid this problem with overall resolution and geometry that scales with TFLOPS.

Using Avatar the movie as an example, the problem is that you - as a gamer - would be blown away by the picture quality even if that movie was rendered at 1080p with those high quality texture assets. It would blow any 4K game you have seen out of the water.

From that we can conclude that graphics using any normal monitor or TV set-up scales - in terms of experience - much better with texture quality and amount than resolution from 1080p/1440p and upwards. The problem is that the GPU world has taught you not to think that way.

That is why I am so happy over the PS5 design and way of thinking. And while MS surely has increased the I/O capabilities of the XboxX there is nothing that indicates that the current setup has the bandwidth and latency required to really allow a significant step-up in texture quality.

Pleas note that the latency determines how much additional texture material you need in VRAM. Cerny stated that their goal was that the next 0.5sec would have to be in VRAM (i.e. any texture that might be needed in the next 0.5 sec based on the environment and the player move speed) need to be there. Given the amount of RAM that high-resolution textures occupies the latency really becomes the driver behind texture quality since the amount of textures you need in VRAM grows exponentially with the latency (as measured by 'how long does it take for from the time the hardware asks for a texture until it can be utilised by the GPU in VRAM').

I am really looking forward to this and I am looking forward to the PC world to also get access to this since it is clear - once you think this through - that I/O drives graphical fidelity more than TFLOPS. It is just that in the PC world no single company/body sits on the I/O - everyone sits on it so change has historically been very slow here.

 

[Utherellus]

The Unreal engines devs say the exact opposite, that with nanite it's easy to adjust the amount of data that's being streamed in to the point it can run on anything.



You will see the tech of the Unreal Engine in the other major engines, it's not something that's exclusive to Epic at all. You will see it in Unity, You will see it from activision, You will see it from Ubisoft.

Highly doubt it but we will see, yes.

 

[Marlenus]

Sorry, but this is wilful ignorance. The post is very detailed but easy to understand. It even has a useful animated gif.

I'm not sure why you even bring up SmartShift or talk about the hardware being limited by being smaller die...?

Smaller die = higher heat density which requires a cooling solution above what the tdp would suggest.

 

[Marlenus]

This is a conclusion that cannot be drawn - even though we do not know this since the silicon has not been shown, I would argue that right now the opposite is more likely to be true. If you look at the PS4 Pro vs XBox One X, the PS4 Pro has a texture filtrate advantage over Xbox One X despite having a Tflops disadvantage due to more silicon dedicated to this.

My bet is that with the PS5 they have followed in the footsteps of the PS4 Pro and will then have an oversized section of the GPU with render mapping/ shaders/ texture output and higher frequency to boot. That would be the logical design given what they are trying to achieve.

Overall comment (sorry if it is wordy):

I believe what many of you are missing is that within the PC framework there is only so much that can be done with texture resolution. As an example, a single model in avatar used 100-120 textures at 8K. Even at a 'measly' 4K this represents 5GB+ of uncompressed textures. Since the VRAM needs to contain all the textures that can be used within the load time of your I/O solution, a normal mid-range GPU card taps out even at the first model. It is undoable.

Within the PC framework we then try to work around this by using crap textures but smaller and smaller polygons for the rendering of the picture - i.e. we bandaid this problem with overall resolution and geometry that scales with TFLOPS.

Using Avatar the movie as an example, the problem is that you - as a gamer - would be blown away by the picture quality even if that movie was rendered at 1080p with those high quality texture assets. It would blow any 4K game you have seen out of the water.

From that we can conclude that graphics using any normal monitor or TV set-up scales - in terms of experience - much better with texture quality and amount than resolution from 1080p/1440p and upwards. The problem is that the GPU world has taught you not to think that way.

That is why I am so happy over the PS5 design and way of thinking. And while MS surely has increased the I/O capabilities of the XboxX there is nothing that indicates that the current setup has the bandwidth and latency required to really allow a significant step-up in texture quality.

Pleas note that the latency determines how much additional texture material you need in VRAM. Cerny stated that their goal was that the next 0.5sec would have to be in VRAM (i.e. any texture that might be needed in the next 0.5 sec based on the environment and the player move speed) need to be there. Given the amount of RAM that high-resolution textures occupies the latency really becomes the driver behind texture quality since the amount of textures you need in VRAM grows exponentially with the latency (as measured by 'how long does it take for from the time the hardware asks for a texture until it can be utilised by the GPU in VRAM').

I am really looking forward to this and I am looking forward to the PC world to also get access to this since it is clear - once you think this through - that I/O drives graphical fidelity more than TFLOPS. It is just that in the PC world no single company/body sits on the I/O - everyone sits on it so change has historically been very slow here.

There are 4 TMUs per Compute Unit so texture fill rate scales the same way shader performance does which gives the xbox an 18% advantage over PS5.

EDIT: Ps4 pro has a pixel fillrate advantage over xbox one x due to having 64 rops vs the 32 rops in the one x. The one X has more CUs and higher clocks than the ps4 pro so has a texture fillrate advantage.

 

[geordiemp]

Sorry, but this is wilful ignorance. The post is very detailed but easy to understand. It even has a useful animated gif.

I'm not sure why you even bring up SmartShift or talk about the hardware being limited by being smaller die...?

Its more than that though, his post simplifies and there will be a benefit in the time domain in ms being predictive in 2 ms

maybe others are helped with answering my question as well:

On my PC my AMD Ryzen 3600 is clocked at 3.6 ghz and 4.5 (Turbo-Frequency). If you check on programs like Core Temp or CPU-Z etc. you can see that frequency varies based on load from 3.6 - 4.5 GHZ but it literally NEVER stays there AND not all cores boost simoultaneously ALL the time.

If you would try to explain this based on a basic CPU like 3600 on PC what does it mean in correlation to the XBOX and PS5 explanation:

This is what is being revealed about the XSEX processor:
AMD Zen 2 CPU with eight cores running at a nominal 3.8GHz, or when simultaneous multithreading (SMT) is used, at 3.6GHz.
(All information are drawn out of WIkipedia: Xbox Series X Hardware Wikipedia Aritcle)

This is what is being revealed about the PS5 processor:
AMD's 7nm Zen 2 microarchitecture with 8 CPU cores running at a variable frequency capped at 3.5 GHz.
(All information are drawn out of WIkipedia: PS5 Hardware Wikipedia Article)

--> How does this compare to:
AMD Ryzen 3600 clocked at 3.6ghz and 4.5 Turbo-Frequency.

So, again, please, could you do us all a favor and explian how does this compare to XSEX and PS5 based on that counter example from PC?
Anyone else feeling ready to answer this question and explain the difference and similarities, please go ahead and make this thread "read-worthy"..

Your boosting is based on first boosting and then thermal throttling back when temps get over a set point or whatever te PID loop is, as a control loop that is slow. IF you cannot imagine that measuring heat after its been generted is slower loop then I give up.

If it was predictive based, it would be 4,5 Ghz and drop a little predictively when sets of instructions would generate excess heat and do it for 2 millisceonds or and then back up to 4.5 Ghz.

One control loop is slower based on heat AFTER the heat is already generated and is in seconds. One control loop turns down as the work load comes in just as excess heat is about to be generated.

Predictive vs reactive. One is in ms, one is in seconds - what do you think is more efficient ?

I am not even getting into the new consoles APU will be on the latest 2020 node for power efficiency (RDNA 2), likely EUV critical layer lithography has been suggested indirectly by AMD...so we dont know the GHz optimums for RDNA2 APUs yet as there are none on the market.

I bet this predictive down clock to be thing going forward as its logically efficient, unless Sony have that concept patented ........

Predictive work load downclock is NOT smart shift, they are different before you ask.

 

[jimbojim]

It is Amd smartshift, that's all there is to it. Look where Amd deploy smartshift and you get the same answer.

The hardware is still limited by smaller die and the choice to clock up to 2.23ghz.
It is this 2.23ghz, which gives 10.3tf, that we are concerned and suspicious of.

Worst case scenarios in games happens rarely, so, 2.23 is there, most of the time. And if worst case scenarios happens, it only needs couple ( i presume 2 % ) percentages to drop power consumption for 10%.

 

[Trueblakjedi]

You forgot to mention that PS5 has 20% faster clock and unified VRAM in your summary... I'm sure it was just an oversight and not the "minimization" that is so grievous to you.

2Tflops magnitude difference is irrelevant - unless you want to keep playing last gen games that is.

Games will be developed and scale to take advantage of what power is available - there won't be anything "left on the table". So it's only relative performance - not absolute magnitude difference - that is significant.

They both have unified ram and even if they didn't, it doesn't matter because all the memory is on the same die. End story.

Even at variable 20% clock, PS5 is literally still SLOWER. As in the XSX still does more work per clock across all of it work units. 12.1 vs 10.2

Pixel fillrate may be the only category that would be a "clock" win if the XSX and PS5 both have 64 ROPs.

There's no way for you to dispute that. What could you possibly be arguing here?

 

[jimbojim]

maybe others are helped with answering my question as well:

On my PC my AMD Ryzen 3600 is clocked at 3.6 ghz and 4.5 (Turbo-Frequency). If you check on programs like Core Temp or CPU-Z etc. you can see that frequency varies based on load from 3.6 - 4.5 GHZ but it literally NEVER stays there AND not all cores boost simoultaneously ALL the time.

If you would try to explain this based on a basic CPU like 3600 on PC what does it mean in correlation to the XBOX and PS5 explanation:

This is what is being revealed about the XSEX processor:
AMD Zen 2 CPU with eight cores running at a nominal 3.8GHz, or when simultaneous multithreading (SMT) is used, at 3.6GHz.
(All information are drawn out of WIkipedia: Xbox Series X Hardware Wikipedia Aritcle)

This is what is being revealed about the PS5 processor:
AMD's 7nm Zen 2 microarchitecture with 8 CPU cores running at a variable frequency capped at 3.5 GHz.
(All information are drawn out of WIkipedia: PS5 Hardware Wikipedia Article)

--> How does this compare to:
AMD Ryzen 3600 clocked at 3.6ghz and 4.5 Turbo-Frequency.

So, again, please, could you do us all a favor and explian how does this compare to XSEX and PS5 based on that counter example from PC?
Anyone else feeling ready to answer this question and explain the difference and similarities, please go ahead and make this thread "read-worthy"..

How it compares? Easily, it's different. Why GPU cards on PC have a higher TDP then consoles with everything included? Well, that's why there is console designs under limited budget and bunch of optimizations. Just 5700XT alone has TDP over 200W, isn't it?

 

[Marlenus]

Worst case scenarios in games happens rarely, so, 2.23 is there, most of the time. And if worst case scenarios happens, it only needs couple ( i presume 2 % ) percentages to drop power consumption for 10%.

The fact the GPU can hit 2.23Ghz in a console thermal/power envelope is pretty staggering to be honest.

Since Cerny said it holds its clockspeeds in most cases i tend to believe him but i am interested what scensrios cause clock drops.

 

[jimbojim]

They both gave unified ram. End story.

Is it unified or something else? Otherwise, RAM speed wouldn't be separated and if game on XSX needs more than 10 GB, RAM speed drops. I've said also, RAM speed per TF is similar on both

 

[longdi]

longdi

Confirm you have read Liabe's post and then try to discuss and rebut what he says.

I read it and don't conflict what i said about smartshift

 

[Exodia]

I think the Sony I/O inputs into EPIC started a while ago with UE4 and days gone, not that it makes any difference to the Narrative, just interesting. Statement from B3D mod who states sony dev who worked on it :

Improvements in UE4 build improves all games...but the history is interesting and he goes on to say thats why ARK struggled. with frame rate

This is wrong. Every studio makes their own custom changes to the engine. Gears also made changes to the I/O. State of Decay did.
Recore did. Its completely false to call that "Sony I/O inputs". Days gone changes NEVER came to UE4. Just like other studios custom changes never came. This is exactly what am talking about how everything will become "because of the SSD". These false narratives are getting out of hand.

 

[Trueblakjedi]

You forgot to mention that PS5 has 20% faster clock and unified VRAM in your summary... I'm sure it was just an oversight and not the "minimization" that is so grievous to you.

2Tflops magnitude difference is irrelevant - unless you want to keep playing last gen games that is.

Games will be developed and scale to take advantage of what power is available - there won't be anything "left on the table". So it's only relative performance - not absolute magnitude difference - that is significant.

What you wrote is unintelligible. Its relative performance versus absolute difference?

No matter what the PS5 does at full optimized clock, the X can do 1.8 TRILLION more calculations before you throw in RT. LOL

It has 44% more units doing work...before RT and ML... that scales.

You sound crazy

 

[Elog]

There are 4 TMUs per Compute Unit so texture fill rate scales the same way shader performance does which gives the xbox an 18% advantage over PS5.

EDIT: Ps4 pro has a pixel fillrate advantage over xbox one x due to having 64 rops vs the 32 rops in the one x. The one X has more CUs and higher clocks than the ps4 pro so has a texture fillrate advantage.

You are right - should be pixel fill rate advantage - my mistake. The point I was trying to make though was that already with PS4 Pro we started to see a design path where Sony started to look at the GPU components that are required for higher resolution textures rather than the TFlops only. I am very curious to see if that is where they end up this time around as well (which I assume - but obviously do not know).

 

[quest]

The fact the GPU can hit 2.23Ghz in a console thermal/power envelope is pretty staggering to be honest.

Since Cerny said it holds its clockspeeds in most cases i tend to believe him but i am interested what scensrios cause clock drops.

So 51%? There is a reason Sony won't put this on paper and are sweeping it under the rug. If it was awesome they put out the specs like the SSD. Instead max clocks and marketing speak. Be great if they posted the complete specs and white paper. It would clear up lots of confusion but leave less wiggle room for astroturfers. Don't see it happening the media won't push Sony for answers like need to be.

 

[Trueblakjedi]

Is it unified or something else? Otherwise, RAM speed wouldn't be separated and if game on XSX needs more than 10 GB, RAM speed drops. I've said also, RAM speed per TF is similar on both

Its unified. The cpu can see and access everything but the GPU only cares about 10 GB.

The "separate" speeds is about the lanes x 56GBPs lane speed which the units are assigned. Its not as if the tracings are cut.

 

[jimbojim]

What you wrote is unintelligible. Its relative performance versus absolute difference?

No matter what the PS5 does at full optimized clock, the X can do 1.8 TRILLION more calculations before you throw in RT. LOL

It has 44% more units doing work...before RT and ML... that scales.

You sound crazy

You can't ignored clocks just like that. Bur whatever.

 

[longdi]

People seems confused with variable and sustained.

You can have variable and sustained, same as you can have variable and not sustained. I bet modern gpu are variable, like for a long time already.

I wish MS will make clear in their next round of presentation.

 

[jimbojim]

Its unified. The cpu can see and access everything but the GPU only cares about 10 GB.

The "separate" speeds is about the lanes x 56GBPs lane speed which the units are assigned. Its not as if the tracings are cut.

If GPU needs more than 10, RAM speed will decrease. Don't ignore RAM Speed per TF which is a fact ( even at XSX's 560 GB/s ) which is similar on both

 

[Trueblakjedi]

Yeah...you're done

lmfao.

You make zero sense.

 

[THE:MILKMAN]

Smaller die = higher heat density which requires a cooling solution above what the tdp would suggest.

I understand this (though we have no idea what the TDP is) but I was asking him why he even brought it up given my original post to him was just asking him to check out Liabe's post.

For me the TLDR is that this strategy is more efficient and they get more useful work out of the APU 99% of the time. Similar with how RAM is only 2x over this gen but with the efficiencies of working more of the RAM more of the time, it means an effective 5-6x increase.

The proof will be seen (or not) when we see games.

I read it and don't conflict what i said about smartshift

OK, but SmartShift wasn't mentioned in Liabe's post. Mark Cerny said "while we're at it we use AMD's SmartShift technology to squeeze a few more pixels from the GPU".

I took that to mean it was an additional feature applied over and above their variable frequency strategy?

Like I said above though, the ultimate proof of whether any of this stuff is meaningful to games is when we lay eyes on them.

 

[Md Ray]

Coming from PCMR, the difference between PS5 and Series X is akin to 2070S vs 2080Ti.

??
Care to explain this? How is the difference between PS5 and SX is akin to 2070S vs 2080 Ti?

 

[Trueblakjedi]

You can't ignored clocks just like that. Bur whatever.

I didnt ignore clocks. The parts of the gpu components affected by clocks are considered which is why i wrote that there would be a PS5 pixel fill advantage under a 64 ROPs equivalence in both systems.

The ps5 had to clock up to 2.23 to get over 10 TF... which is characterization of the work per cycle it can do.

This isnt hard why are you making seem like the clock is more important than the work. It isnt.

 

[Panajev2001a]

So 51%? There is a reason Sony won't put this on paper and are sweeping it under the rug. If it was awesome they put out the specs like the SSD. Instead max clocks and marketing speak. Be great if they posted the complete specs and white paper.

They talked about it at length in their presentation and details out there allow for some meaningful and comprehensive analysis: https://www.resetera.com/threads/pl...ep-dive-ot-secret-agent-cerny.175780/page-340

It would clear up lots of confusion but leave less wiggle room for astroturfers

Nature/Astroturfers will find a way... their job is to create confusion where it makes sense and seed stretchable half truths that can be weaponised in songster forum/social media discussions.

For those old enough to remember, a clear case was Sony giving out a full HotChip detailed Emotion Engine presentation and lots of documentation about the Emotion Engine (CELL even had a hardware simulator and full docs available BEFORE PS3 came out) and since PS had better numbers than the competition (Dreamcast at the time) the rabid fans/astroturfers of the competing platform were trying to seed doubt about BS their ethical vs sustained numbers, etc... so there is NO winning by engaging those people on those terms. They are not trying to discuss or to appeal to rational arguments but to feed / and hype their faction into frenzy and spread F.U.D. About what they do not support...

 

[longdi]

OK, but SmartShift wasn't mentioned in Liabe's post. Mark Cerny said "while we're at it we use AMD's SmartShift technology to squeeze a few more pixels from the GPU".

I took that to mean it was an additional feature applied over and above their variable frequency strategy?

Like I said above though, the ultimate proof of whether any of this stuff is meaningful to games is when we lay eyes on them.

Yes proof will be in pudding.

And yes thats what Mark alluded to, he is using amd smart shift AND their rich library of PS games data, to further fine tuned smart shift. So much developers can hands off with the load balancing if they want.

But its still smartshift, it still incurs latency, it still questionably 'sustained'

 

[Panajev2001a]

Yes proof will be in pudding.

And yes thats what Mark alluded to, he is using amd smart shift AND their rich library of PS games data, to further fine tuned smart shift. So much developers can hands off with the load balancing if they want.

But its still smartshift, it still incurs latency, it still questionably 'sustained'

He said they ALSO included smartshift and I am willing to believe that is also customised over what is on the market already, but let’s not lose any chance to try to make other non XSX platforms look worse, shall we?

 

[geordiemp]

This is wrong. Every studio makes their own custom changes to the engine. Gears also made changes to the I/O. State of Decay did.
Recore did. Its completely false to call that "Sony I/O inputs". Days gone changes NEVER came to UE4. Just like other studios custom changes never came. This is exactly what am talking about how everything will become "because of the SSD". These false narratives are getting out of hand.

I posted my source, and your source is again ?

First I said Days gone totally rewrote the IO and you called rubbish, do you want me to link your post to remind everyone of your bullcrap ? I then gave the source...

And you immediately reposte with so did Gears and SOD...as well so there...

They are not mutually exclusive events anyway.

I gave you the days gone source, they took 6 months.

Where is your link for Gears 5 rewriting UE4 IO, Source ?

You like to claim false on everything that does not fit your deluded xbox fanboy discord narrative and NEVER give any source of backing other than I said so.

 

[geordiemp]

Eh no.. because of how apu is design , you can do some load balancing like smartshift to eke out more performance in a thermal and power constrained closed system, but at the expense of some latency.

Sustained and variable can work hand in hand

I won't be surprised if Series X has the same feature, or MS can disable it if they find the latency penalty not worth.

It is always about the 2.23ghz figures, which gives 10.3tf, which gives the 'small' difference defence.

The question, of how often PS5 can stay at 2.23ghz, is the difference 2tf or more in realworld.

This suspicion is confounded by how Mark seems skirty to answer. And went in big circle to present smart shift as a big new feature. Why did he/Sony not strongly advertise PS5 as the first gpu running north of 2ghz? Would make headlines too!

Like 3950x is advertised as up to 4.7ghz, in practice, you only see that speed in a millisecond, on a single thread, using software reader. Who knows if AMD can also fudged it.

You have no idea what your talking about, you are full of it.

 

[longdi]

He said they ALSO included smartshift and I am willing to believe that is also customised over what is on the market already, but let’s not lose any chance to try to make other non XSX platforms look worse, shall we?

Yes it is down to how much money Sony paid to make hardware custom to reach and sustained 2.23ghz.

History have shown you can make gpu run high clocks with tweaks, like hd4870 to 4890 or kepler to Maxwell.

Im just concerned that Mark did not make more about the 2.23ghz and seem evasive even, that the hardware customisation aint that big.

 

[geordiemp]

Yes it is down to how much money Sony paid to make hardware custom to reach and sustained 2.23ghz.

History have shown you can make gpu run high clocks with tweaks, like hd4870 to 4890 or kepler to Maxwell.

Im just concerned that Mark did not make more about the 2.23ghz and seem evasive even, that the hardware customisation aint that big.

Phil would be embarrassed you posting with his picture.

 

[Shmunter]

Has this been posted? Complete 180 on the PS5 I/o. Simply nothing out there in the same class....

Edit: apparently very late to the party on this one

 

[longdi]

The next set of questions DF should ask

MS : does Series X use AMD smart shift with variable load balancing?

Sony : how much APU hardware is customed, did you need to add/remove any hw registers and whatnot from the base 40CU rDNA2 to achieve 2.23ghz on the same 7nm?

 

[THE:MILKMAN]

Yes proof will be in pudding.

And yes thats what Mark alluded to, he is using amd smart shift AND their rich library of PS games data, to further fine tuned smart shift. So much developers can hands off with the load balancing if they want.

But its still smartshift, it still incurs latency, it still questionably 'sustained'

For sure AMD and Sony worked on this power control together but SmartShift, as I understand it, shifts power to the GPU from the CPU if the current load allows and vice versa. The knock-on effect is one clocks down and the other clocks up but this is a good thing. Maybe we could call the PS5 version SmartShift Turbo or V2.0?

I'm not sure where 'sustained' is coming from? Mark Cerny was clear clocks are variable or continuous boost but will spend "most of the time at or close to" the quoted maximums. I'll take clocking down a little if it means more effective utilisation of the APU.

 

[Physiognomonics]

They both have unified ram and even if they didn't, it doesn't matter because all the memory is on the same die. End story.

Even at variable 20% clock, PS5 is literally still SLOWER. As in the XSX still does more work per clock across all of it work units. 12.1 vs 10.2

Pixel fillrate may be the only category that would be a "clock" win if the XSX and PS5 both have 64 ROPs.

There's no way for you to dispute that. What could you possibly be arguing here?

We know they both have 64 rops as number of rops is dictated by the number of shader array (and both have 4). But PS5 has other parts about 20% stronger than XSX, basically everything that is not in the CUs as both should have the same number of (because again those are determined by the number of shader array, not the number of CUs, and we know XSX has 2 Shader Engines and 4 shader arrays):

- Geometry processor
- Primitive units
- Graphics command processor
- Rasterizer
- Rops
- ACEs (Async Compute Engines)
- The new L1 cache (compared to GCN)

 

[quest]

They talked about it at length in their presentation and details out there allow for some meaningful and comprehensive analysis: https://www.resetera.com/threads/pl...ep-dive-ot-secret-agent-cerny.175780/page-340



Nature/Astroturfers will find a way... their job is to create confusion where it makes sense and seed stretchable half truths that can be weaponised in songster forum/social media discussions.

For those old enough to remember, a clear case was Sony giving out a full HotChip detailed Emotion Engine presentation and lots of documentation about the Emotion Engine (CELL even had a hardware simulator and full docs available BEFORE PS3 came out) and since PS had better numbers than the competition (Dreamcast at the time) the rabid fans/astroturfers of the competing platform were trying to seed doubt about BS their ethical vs sustained numbers, etc... so there is NO winning by engaging those people on those terms. They are not trying to discuss or to appeal to rational arguments but to feed / and hype their faction into frenzy and spread F.U.D. About what they do not support...

Where the ethics came from was Sony using numbers in their pr they knew would never be remotely be used in games 75 million polygons per second. Was it dirty probably but these are billions dollar corporations they don't play fair. As they say if your not cheating your not trying. I remember the time well it was maybe the golden era of gaming.

Has this been posted? Complete 180 on the PS5 I/o. Simply nothing out there in the same class....

Only 50 times lol.

 

[Panajev2001a]

Im just concerned that Mark did not make more about the 2.23ghz and seem evasive even, that the hardware customisation aint that big.

Key word “concern”... ... why are you concerned? It does not scream life long fan not day one purchaser nor it seems evidence based...

 

[Panajev2001a]

Where the ethics came from was Sony using numbers in their pr they knew would never be remotely be used in games 75 million polygons per second. Was it dirty probably

Those were the HW specs, context was provided, white papers and tech specs were provided, there was even a down to the metal cheap dev kit you could buy... see what I was saying that no matter what you do if people want to find a chink in the armour and amplify it they will ?

 

[Shmunter]

Only 50 times lol.

Always the bridesmaid, never the bride

 

[longdi]

Key word “concern”... ... why are you concerned? It does not scream life long fan not day one purchaser nor it seems evidence based...

Basically if their custom work has allowed the world first 2.23ghz gpu, that's an achievement to be proud of, in the same veins as their custom ssd work, imo. Did they forget to put the clockspeed achievement up the same pedestal as the ssd?

 

[Psykodad]

This is wrong. Every studio makes their own custom changes to the engine. Gears also made changes to the I/O. State of Decay did.
Recore did. Its completely false to call that "Sony I/O inputs". Days gone changes NEVER came to UE4. Just like other studios custom changes never came. This is exactly what am talking about how everything will become "because of the SSD". These false narratives are getting out of hand.

UE5 is being developed specifically around PS5 and it's SSD though.

 

[geordiemp]

The next set of questions DF should ask

MS : does Series X use AMD smart shift with variable load balancing?

Sony : how much APU hardware is customed, did you need to add/remove any hw registers and whatnot from the base 40CU rDNA2 to achieve 2.23ghz on the same 7nm?

So the new Xbox discord tactics in a thread entitled matt weighs in on PS5 IO and were back to TF arguments.

Have discord conceded the ps5 IO superiority as not in the same league so we are derailing threads on the TF and sustained argument again. ?

Let me show you what you wont see on XSX at this detail

UE5 is being developed specifically around PS5 and it's SSD though.

UE4 will still work on UE5, EPIC have already said that.

Its just that Nanite subset has been optimsied FIRST on Ps5 and will clearly be more performant on Ps5 and why Ps5 was chosen.

 

[Bryank75]

(10*560 + 6*336)/16 = (5600+2016)/16= 7616/16 = 476gbps

Thats called a weighted average.

476>448

QED

Yeah except you are counting 3 GB that will never be used for games on the XBox SX, so you calculated it wrong!

 

[Bryank75]

this is something completely different though. That was a midgen refresh.
cant compare it, when there is a completely new gen.
Also, you can still do a lot more work with 2TF's RDNA2 than you can with 500GF's of GCN, since RDNA 2.0 is much more efficient.
Anyway, we will see as soon as the games arrive then we can finally stop the bullshit ssd-Talking.

PS5 is RDNA 2 also, so the ratio is actually smaller, much smaller.