• 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.

[NXG] Matrix UE5 PC Tech Analysis

3liteDragon

Member

Unreal Engine 5 has finally launched and brought with it the first look at Nanite, Lumen and the Matrix Demo running on PC. Does this show any signs of the future for game technology? How hard is it to equal console like performance? Does AMD or Nvidia have an edge? And what is going on with the GPU performance? Jack in to see how deep this rabbit hole goes!
 
Anticipation Popcorn GIF
 

SlimySnake

Flashless at the Golden Globes
winjer winjer NX gamer also says the game is 1080p on consoles. I stand corrected.

But my goodness, those GPU utilizations are awful. 60-70% his CPUs are leaving so much performance on the table. At first I thought it was due to the notoriously bad 2700z that he uses but the excellent 3600 is also holding back the 6800. Alex mentioned how the demo is bottlenecked by clockspeeds which could explain why my GPU utilzations for both the 2080 and 3080 are roughly around 97-100%. My CPU clocks are 4.6 ghz. Intel with their awful Rocket Lake lineup finally gets a win thanks to the high clock speeds. Who wouldve thought.

Anyway, the awful GPU utilization explains why the consoles are also running at 1080p despite the RDNA 2.0 cards matching RTX 20 series at 1440p if not exceeding them on PC when paired with high clock CPUs. Very disappointing to see. I wouldve figured Epic would've designed this around the PS5 and XSX clocks, and fully maxed out the GPU but it doesnt seem like they are doing this.
 

clintar

Member
winjer winjer NX gamer also says the game is 1080p on consoles. I stand corrected.

But my goodness, those GPU utilizations are awful. 60-70% his CPUs are leaving so much performance on the table. At first I thought it was due to the notoriously bad 2700z that he uses but the excellent 3600 is also holding back the 6800. Alex mentioned how the demo is bottlenecked by clockspeeds which could explain why my GPU utilzations for both the 2080 and 3080 are roughly around 97-100%. My CPU clocks are 4.6 ghz. Intel with their awful Rocket Lake lineup finally gets a win thanks to the high clock speeds. Who wouldve thought.

Anyway, the awful GPU utilization explains why the consoles are also running at 1080p despite the RDNA 2.0 cards matching RTX 20 series at 1440p if not exceeding them on PC when paired with high clock CPUs. Very disappointing to see. I wouldve figured Epic would've designed this around the PS5 and XSX clocks, and fully maxed out the GPU but it doesnt seem like they are doing this.
He explains that this is more than designing around ps5/xsx clocks and maxing out the gpu... It's more about shared pool of memory keeping it from doing work on cpu and moving it to gpu, making gpu wait. And I guess the storage helping on ps5.

Very impressive how well these consoles can run this compared to high end PCs.
 
Last edited:

Arioco

Member
Performance is surprisingly bad on PC right now. But didn't Mark Cerny talk about this very issue in his "The Road To PS5" presentation?

He said:

"At the SDD speeds we are talking about moving the data, meaning copying from one location to another, takes roughly an entire next gen CPU core. And that's just the tip of the iceberg, if all the overheads get a hundred times larger that will cripple the frame rate as soon as the player moves and that massive stream of data starts coming off the SDD".

Isn't that what is happening in The Matrix demo and what causes it to be so heavily CPU bound? The PC doesn't currently have any custom I/O hardware to help with compression/decompression, moving data, etc... like the one we find on the new consoles, so that would explain (at least partially) such bad performance when the character is moving. A PC has to bruteforce all those things and that can fully occupy the CPU.

I suspect this is what Tim Sweeney meant when he said PS5's SDD was years ahead of the PC (statement he was fiercely criticised for as you all remember 🙄), he wasn't talking about the SDD itself (you could already use an SDD on PS4 or Xbox One), but all the customizations needed to take full advantage of it.
 
Last edited:

adamsapple

Or is it just one of Phil's balls in my throat?
He explains that this is more than designing around ps5/xsx clocks and maxing out the gpu... It's more about shared pool of memory keeping it from doing work on cpu and moving it to gpu, making gpu wait. And I guess the storage helping on ps5.

Very impressive how well these consoles can run this compared to high end PCs.

Impressive compared to relative PC hardware but in a vacuum 30 FPS w/ drops doesn't really sound impressive, right ?
 
So if I'm understanding this correctly what NXGamer NXGamer is saying is that:

The Matrix demo is throwing around a lot of data, both from disk to RAM and across the CPU and GPU caches. The PC is bottlenecked here because of the split nature of the RAM pools and lack of dedicated hardware and optimisations, however the consoles (PS5 and Series X/S) don't suffer from this problem because the entire data streaming process is being offloaded onto dedicated I/O hardware and memory sub-systems. This is one of the reasons for a lot of the CPU overhead on the PC version, as the entire data streaming (textures and geometry) is being handled by dedicated silicon on the consoles.

The PC version is also struggling to push GPU utilisation, this is because there's a lot of stale data present in the GPU caches which is causing performance issues and hitches and especially the random spikes in CPU clocks, again the consoles dedicated hardware such as the PS5's cache scrubbers mitigate this issue.

He re-iterates that it's not an issue of data bandwidth but more so the efficiency in which the data is being moved from the disk, into memory, and then into the CPU/GPU.

He concludes that the engine is still poorly optimised across both PC and console and will improve over time, especially now that major third party developers are adopting the engine.
 
Last edited:

SlimySnake

Flashless at the Golden Globes
He explains that this is more than designing around ps5/xsx clocks and maxing out the gpu... It's more about shared pool of memory keeping it from doing work on cpu and moving it to gpu, making gpu wait. And I guess the storage helping on ps5.
Eh. thats a stretch. While it's possible that my higher clocked CPU is helping with decompression, my system and every single PC system out there is limited by the same split ram architecture. The game isnt even vram bound. tops out at around 5GB. way less than the 8GB vram on these cards.

I think he's putting too much emphasis on the PS5's memory management. The PS5 GPU is NOT performing up to its fullest potential because of the same CPU bottlenecks. The framerates drop to 20 fps all the same at 1080p.

You can look at the PC benchmarks below. The 12900k shows as 3.2 Ghz but it can go up to 5.2 Ghz. And thats it. It's really that simple. The same cards in NX gamer's video perform much better here when paired with a better CPU.

NNqLaaV.jpg

nu3SLl7.jpg
 

M1chl

Currently Gif and Meme Champion
Eh. thats a stretch. While it's possible that my higher clocked CPU is helping with decompression, my system and every single PC system out there is limited by the same split ram architecture. The game isnt even vram bound. tops out at around 5GB. way less than the 8GB vram on these cards.

I think he's putting too much emphasis on the PS5's memory management. The PS5 GPU is NOT performing up to its fullest potential because of the same CPU bottlenecks. The framerates drop to 20 fps all the same at 1080p.

You can look at the PC benchmarks below. The 12900k shows as 3.2 Ghz but it can go up to 5.2 Ghz. And thats it. It's really that simple. The same cards in NX gamer's video perform much better here when paired with a better CPU.

NNqLaaV.jpg

nu3SLl7.jpg
People still think that the I/O block which mainly consisting of the SSD memory subsystem and fetch into V/RAM is enough to feed the data to screen. Cute. Even with all of that advancement, it's like 40x slower than it needs to be, not even counting the latency.

So in conclusion it's using ole Master/Slave architecture, where the "director" is one/two CPU cores (1C/2T) and it's spawning work for other cores. I expected more, especially after ID Tech 7, which runs on anything basically due to it's almost unmatched ability to max out the available performance. If you do it old time style, you are going to stress the caches and there is going to be a problem then, directly on the silicon.

And it's just a code. You would expect they would get this right...
 

Clear

CliffyB's Cock Holster
People REALLY need to wrap the heads around the importance of pipeline efficiency, both in hardware and software.

In simple terms UE5 performs as well as it does visually by smart data organization. But the cost of that optimization is that its super-demanding on tight synchronization of all parts of the system; CPU, GPU, and I/O. Because every time one part is sat idle waiting for another to finish its workload, performance is lost.

Modern consoles are not only easier to calibrate for this due to them having fixed spec hardware, but they are built with a new paradigm in mind wherein the I/O subsystem works intelligently to expedite communication. On PC, this is much more difficult because not only are system specs highly variable but architecturally they pre-date the paradigm shift.

The end result is that while its possible to mitigate the shortfall by increasing the capability of CPU, GPU and I/O, the generalist nature of the way these parts are connected on PC, means that its likely to be disadvantaged proportionately to how much stress is placed on synchrony between them.
 

Guilty_AI

Member
People REALLY need to wrap the heads around the importance of pipeline efficiency, both in hardware and software.

In simple terms UE5 performs as well as it does visually by smart data organization. But the cost of that optimization is that its super-demanding on tight synchronization of all parts of the system; CPU, GPU, and I/O. Because every time one part is sat idle waiting for another to finish its workload, performance is lost.

Modern consoles are not only easier to calibrate for this due to them having fixed spec hardware, but they are built with a new paradigm in mind wherein the I/O subsystem works intelligently to expedite communication. On PC, this is much more difficult because not only are system specs highly variable but architecturally they pre-date the paradigm shift.

The end result is that while its possible to mitigate the shortfall by increasing the capability of CPU, GPU and I/O, the generalist nature of the way these parts are connected on PC, means that its likely to be disadvantaged proportionately to how much stress is placed on synchrony between them.
I feel all these optimizations become kinda pointless when that extra performance gain can be matched or surpassed by simply using slightly better specs



(These parts are actually cheaper than the ones he's using too)

Its not really worth sacrificing the versatility of the platform just to squeeze every last drop of performance from the hardware.
It IS, however, worth it for a piece of hardware that'll have to survive in a ever evolving industry for the next 7 years, so those types of optimizations make sense on consoles.

That said, i think this demo is not really properly optimized on either PC platforms or consoles.
 
Last edited:
It's going to be interesting to see how the next batch of upgraded engines compare. Idtech without a doubt is going to adopt all the Direct X12 Ultimate extensions such as Mesh Shader, SFS and Direct ML. To be brutally honest, lumen cannot hold a candle (Pun intended) to Ray Traced lighting. If engines like idtech and Forzatech integrate all those additions and keep the industry leading CPU optimisation intact then it's going to be an interesting comparison.
 
Last edited:
Get over yourself. You've been corrected countless times on this topic by myself and others, so why are you still being hard-headed?
Corrected? So why don't you try correcting the creators of Nanite?

They literally told you that nanite doesn't "stream data in and out on the fly". that its not dependent on a "fast ssd". That the valley of the ancient demo was 2x more performance heavier than the PS5 2020 lumen demo.

The Nanite on Valley of the ancient pc demo was literally the worse case scenario.
This is why Valley of the ancient ran 1080p 30fps on PS5 and XSX.

They literally tell you on video. Timestamp 42:00



"Thats called overdraw and when that happens alot it can be VERY EXPENSIVE.
Its rare in most use cases but in the Valley of the ancient, the one case that can cause that to happen is actually quite prevalent throughout the map...If you have that happen just once, it will be more expensive but it won't be that bad. But if you have lot of that happening like in this demo. The overdraw can get quite a bit more EXPENSIVE. IN this demo, what we see versus other content we have tested in the past. Nanite is 2x more expensive. This demo in general ends up being about TWICE AS EXPENSIVE than what we have seen in other previous content example. "

Literally Brian debunked all of this. Nanite doesn't "stream data in and out on the fly".Watch the videos and STOP spreading misinformation.
Brian also literally said, nanite cost 2x in valley of the ancient than in the PS5 demo. That Valley of the ancient NOT the ps5 demo pushed nanite to its limits.

How many times does UE engineers have to say that the 2020 PS5 demo is the WEAKEST demo out of the three nanite wise before you people accept it?

"In this demo, what we see versus other content we have tested in the past. Nanite is 2x more expensive. This demo in general ends up being about TWICE AS EXPENSIVE than what we have seen in other previous content example. " Brian Karis Creator of Nanite/UE5

Once you go through an area, it goes into memory. So if you play through the demo once, it keeps everything as it goes.
Whether you are on PC, Xbox or PS5. Thats how nanite works.

NXGamer clearly has no idea how nanite works and his theories have been debunked by the creator of Nanite and multiple UE engineers.

 
Last edited:

ChiefDada

Gold Member
Corrected? So why don't you try correcting the creators of Nanite?

They literally told you that nanite doesn't "stream data in and out on the fly". that its not dependent on a "fast ssd". That the valley of the ancient demo was 2x more performance heavier than the PS5 2020 lumen demo.

The Nanite on Valley of the ancient pc demo was literally the worse case scenario.
This is why Valley of the ancient ran 1080p 30fps on PS5 and XSX. Yet the matrix demo with a full blown living breathing open world ran at a higher resolution and fps on PS5 and XSX.

They literally tell you on video. Timestamp 42:00



"Thats called overdraw and when that happens alot it can be VERY EXPENSIVE.
Its rare in most use cases but in the Valley of the ancient, the one case that can cause that to happen is actually quite prevalent throughout the map...If you have that happen just once, it will be more expensive but it won't be that bad. But if you have lot of that happening like in this demo. The overdraw can get quite a bit more EXPENSIVE. IN this demo, what we see versus other content we have tested in the past. Nanite is 2x more expensive. This demo in general ends up being about TWICE AS EXPENSIVE than what we have seen in other previous content example. "

Literally Brian debunked all of this. Nanite doesn't "stream data in and out on the fly".Watch the videos and STOP spreading misinformation. The demo was already in RAM because he played through it while they were waiting. He literally said that.
Brian also literally said, nanite cost 2x in valley of the ancient than in the PS5 demo. That Valley of the ancient NOT the ps5 demo pushed nanite to its limits.

How many freaking times does UE engineers have to say that the 2020 PS5 demo is the WEAKEST demo out of the three nanite wise before you people accept it?

"In this demo, what we see versus other content we have tested in the past. Nanite is 2x more expensive. This demo in general ends up being about TWICE AS EXPENSIVE than what we have seen in other previous content example. " Brian Karis Creator of Nanite/UE5

Once you go through an area, it goes into memory. So if you play through the demo once, it keeps everything as it goes.
Whether you are on PC, Xbox or PS5. Thats how nanite works. Did you even bother to read or watch the Nanite white paper? What wrong with you people and spreading misinformation?

NXGamer clearly has no idea how nanite works and his thoeries have been debunked by the creator of Nanite and multiple UE engineers.


Here We Go Again GIF
 

Dream-Knife

Banned
winjer winjer NX gamer also says the game is 1080p on consoles. I stand corrected.

But my goodness, those GPU utilizations are awful. 60-70% his CPUs are leaving so much performance on the table. At first I thought it was due to the notoriously bad 2700z that he uses but the excellent 3600 is also holding back the 6800. Alex mentioned how the demo is bottlenecked by clockspeeds which could explain why my GPU utilzations for both the 2080 and 3080 are roughly around 97-100%. My CPU clocks are 4.6 ghz. Intel with their awful Rocket Lake lineup finally gets a win thanks to the high clock speeds. Who wouldve thought.

Anyway, the awful GPU utilization explains why the consoles are also running at 1080p despite the RDNA 2.0 cards matching RTX 20 series at 1440p if not exceeding them on PC when paired with high clock CPUs. Very disappointing to see. I wouldve figured Epic would've designed this around the PS5 and XSX clocks, and fully maxed out the GPU but it doesnt seem like they are doing this.
What GPU are you using? What FPS were you getting with the 2080 and 3080?
 

SlimySnake

Flashless at the Golden Globes
What GPU are you using? What FPS were you getting with the 2080 and 3080?
Depends on the resolution. For 2080, I was getting 10-13 fps at native 4k. 23-25 in 1440p and around 35-40 in 1080p. basically 10-12 fps gain per resolution drop.

for the 3080, i didnt even bother going below native 4k since i get around 40-48 fps while driving around. In DLSS Quality, it averages 48 fps. Demo hardly gets over 50 fps even if i use performance dlss so its kinda pointless to reduce the resolution on a 3080.

I think you meant to ask which CPU i am using. I am using the i7-11700kf. it consumes 140 watts on its own while running the game though i did not see it run at 5.0 ghz like it typically does during loading screens. Pretty much sits at around 4.6 ghz during gameplay.


Zoom in to see the numbers. Notice how it maxes out the GPU utilization. Even when it dips on occasion, it dips to around 85% and comes back up immediately. I notice it happens during stutters which are caused by shader compiling. It never goes to 70% or 60% like you see in NX Gamer's videos because its CPU clock bound like Alex suggested, and those CPUs pretty much top out at 4.0 ghz.

XO2iPVa.jpg
 

DenchDeckard

Moderated wildly
He explains that this is more than designing around ps5/xsx clocks and maxing out the gpu... It's more about shared pool of memory keeping it from doing work on cpu and moving it to gpu, making gpu wait. And I guess the storage helping on ps5.

Very impressive how well these consoles can run this compared to high end PCs.
It's not impressive how this engine is running. It's a bit of a mess.
 

sinnergy

Member
It's not impressive how this engine is running. It's a bit of a mess.
But the good thing is , you can help EPIC out 🤣🤡 as you know how to fix this cutting edge render technology. Seeing that other 3rd and 1st parties already have better solutions..

You can become a millionaire man!
 

DenchDeckard

Moderated wildly
But the good thing is , you can help EPIC out 🤣🤡 as you know how to fix this cutting edge render technology. Seeing that other 3rd and 1st parties already have better solutions..

You can become a millionaire man!
I'll leave it to the third and first party devs who will help get this engine up to speed, just like UE3 and 4. :D
 

sendit

Member
Regardless, price to performance is absolute garbage on the PC. Will have to wait until the 4000 series Nvidia cards get released.
 

Clear

CliffyB's Cock Holster
I feel all these optimizations become kinda pointless when that extra performance gain can be matched or surpassed by simply using slightly better specs

Likely true in most cases because, as I said, the impact really depends on how much stress is being placed on I/O. The Matrix demo is obviously being an unusually heavy case.

Also, its undeniable that having a performance surplus on pure CPU or GPU isn't an equally valid area of advantage when that resource can be tapped.

The bottom line though is that what UE5 offers is genuinely impressive, the geometry virtualization particularly, so the data-heavy approach is definitely the future.
 

azertydu91

Hard to Kill
People still think that the I/O block which mainly consisting of the SSD memory subsystem and fetch into V/RAM is enough to feed the data to screen. Cute. Even with all of that advancement, it's like 40x slower than it needs to be, not even counting the latency.

So in conclusion it's using ole Master/Slave architecture, where the "director" is one/two CPU cores (1C/2T) and it's spawning work for other cores. I expected more, especially after ID Tech 7, which runs on anything basically due to it's almost unmatched ability to max out the available performance. If you do it old time style, you are going to stress the caches and there is going to be a problem then, directly on the silicon.

And it's just a code. You would expect they would get this right...
I was wondering if using direct storage would mitigate some of those issues?
Or will the cost on that GPU be too much , if you have any idea about this problem I'd gladly hear your opinion about it.
 

M1chl

Currently Gif and Meme Champion
I was wondering if using direct storage would mitigate some of those issues?
Or will the cost on that GPU be too much , if you have any idea about this problem I'd gladly hear your opinion about it.
Direct Storage by itself no, with RTX IO probably. Honestly it's more important to look at what is actually being done on CPU, if it's decompression of the assets, then these techniques would be beneficial, if it's some other processes, in that case the improvement would be marginal.
 

winjer

Gold Member
winjer winjer NX gamer also says the game is 1080p on consoles. I stand corrected.

But my goodness, those GPU utilizations are awful. 60-70% his CPUs are leaving so much performance on the table. At first I thought it was due to the notoriously bad 2700z that he uses but the excellent 3600 is also holding back the 6800. Alex mentioned how the demo is bottlenecked by clockspeeds which could explain why my GPU utilzations for both the 2080 and 3080 are roughly around 97-100%. My CPU clocks are 4.6 ghz. Intel with their awful Rocket Lake lineup finally gets a win thanks to the high clock speeds. Who wouldve thought.

Anyway, the awful GPU utilization explains why the consoles are also running at 1080p despite the RDNA 2.0 cards matching RTX 20 series at 1440p if not exceeding them on PC when paired with high clock CPUs. Very disappointing to see. I wouldve figured Epic would've designed this around the PS5 and XSX clocks, and fully maxed out the GPU but it doesnt seem like they are doing this.

Sorry to just copy paste my post from another thread, but NXGamer probably screwed up his review, very badly:

His results can't be right. His frame rate is way too low, compared to mine.
I do have a slightly better PC than him. But not that much.
I have a 3700X and an RTX 2070S. He has a 2700X and an RTX 2070 non super.
But the 3700X is just 7% faster in games, than a 2700X.

My 2070S is clocked at 1920Mhz, resulting in 9.8 TFLOPs.
His 2070 is clocked at 2040, resulting in 9.4 TFLOps. This is a difference of 4.2%.

On the same spot, I get 38 fps. He gets 23 fps. This is a difference of 65% in performance. In a demo that is CPU bound.
While the 2 CPUs should have a difference of around 7% in performance.
Either he f***d up really bad with his packaged demo. Or he has some performance issues with his PC, which would also explain why in so many of his analysis he gets lower results, than on consoles.

Then he speaks about how when he is driving, he gets to 10-14 fps.
In my PC, I get 30-31 fps. With lows of 26 fps when crashing.

Here is a screenshot, in the demo, settings at 3, resolution 1440p with TSR at 75%, meaning it's rendered at a base of 1080p.
LZQ5zBt.jpg


Here is his result, on the exact same spot. He is rendering at 1080p, settings at 3.
QkaIqWu.jpg
 
Very interesting results winjer winjer . For one I noticed that NXG RAM usage is 13gb Vs yours of only only 9gb. The frametime graph looks much more volatile on NXG too. Could it be something as simple as NXG testing while shaders are still being compiled? Alex from DF mentioned the first runs were really stuttery due to shader compilation, which eventually cleared up later on.

NXGamer NXGamer very interested to know your opinion on these variances. The differences between the systems are marginal and definitely should not results in such a big performance delta..
 

winjer

Gold Member
Very interesting results winjer winjer . For one I noticed that NXG RAM usage is 13gb Vs yours of only only 9gb. The frametime graph looks much more volatile on NXG too. Could it be something as simple as NXG testing while shaders are still being compiled? Alex from DF mentioned the first runs were really stuttery due to shader compilation, which eventually cleared up later on.

NXGamer NXGamer very interested to know your opinion on these variances. The differences between the systems are marginal and definitely should not results in such a big performance delta..

Alex showed footage while shaders were compiling and after 4 runs, when shaders where compiled.
He got similar average frame rates on the first run. But with those massive spikes in between.
I got similar results to Alex. With the first run having high frame rates, punctuated with frame dips when shaders where compiling.
Shader compilation barely affects average frame rates. But it makes 1% lows much lower.
 
Last edited:

DaGwaphics

Member
Impressive compared to relative PC hardware but in a vacuum 30 FPS w/ drops doesn't really sound impressive, right ?

IKR, he almost makes it sound as if the consoles are purring along at 60fps or something. It looks like you can match console performance relatively easily with a 5600x and a RTX 3060 or thereabouts. It's the crowd that wants 60fps or 120fps that are pressed at the moment.
 

GHG

Gold Member
Sorry to just copy paste my post from another thread, but NXGamer probably screwed up his review, very badly:

His results can't be right. His frame rate is way too low, compared to mine.
I do have a slightly better PC than him. But not that much.
I have a 3700X and an RTX 2070S. He has a 2700X and an RTX 2070 non super.
But the 3700X is just 7% faster in games, than a 2700X.

My 2070S is clocked at 1920Mhz, resulting in 9.8 TFLOPs.
His 2070 is clocked at 2040, resulting in 9.4 TFLOps. This is a difference of 4.2%.

On the same spot, I get 38 fps. He gets 23 fps. This is a difference of 65% in performance. In a demo that is CPU bound.
While the 2 CPUs should have a difference of around 7% in performance.
Either he f***d up really bad with his packaged demo. Or he has some performance issues with his PC, which would also explain why in so many of his analysis he gets lower results, than on consoles.

Then he speaks about how when he is driving, he gets to 10-14 fps.
In my PC, I get 30-31 fps. With lows of 26 fps when crashing.

Here is a screenshot, in the demo, settings at 3, resolution 1440p with TSR at 75%, meaning it's rendered at a base of 1080p.
LZQ5zBt.jpg


Here is his result, on the exact same spot. He is rendering at 1080p, settings at 3.
QkaIqWu.jpg

Do you not understand the fact that when you hit a CPU bottleneck it can disproportionately tank your performance?
 

sircaw

Banned
He explains that this is more than designing around ps5/xsx clocks and maxing out the gpu... It's more about shared pool of memory keeping it from doing work on cpu and moving it to gpu, making gpu wait. And I guess the storage helping on ps5.

Very impressive how well these consoles can run this compared to high end PCs.
Don't you dare say such things, a wildfire will roar through these forums on that note. "lollipop_disappointed:
 

Loxus

Member
I know right.
Dude thinks PC running the demo in editor mode equals to PS5 environment.

While ignoring quotes from people like Nick Penwarden, the VP of engineering at Epic Games.
Epic Games had to rewrite parts of Unreal Engine to keep up with the PS5's SSD
"The PlayStation 5 provides a huge leap in both computing and graphics performance, but its storage architecture is also truly special," Nick Penwarden, VP of engineering at Epic Games told us.
"The ability to stream in content at extreme speeds enables developers to create denser and more detailed environments, changing how we think about streaming content. It’s so impactful that we’ve rewritten our core I/O subsystems for Unreal Engine with the PlayStation 5 in mind," he added.


SSD speed or how many MBs moved is not what NX Gamer is talking about but the latency of the CPU decompression.

As we saw in NX Gamer video,
System Memory utilization = 12 GB at one point (this is game assets stored before or after decompression to help reduce latency of having to go back to storage and decompress again constantly.)
GPU Memory utilization = 7 GB (most game assets working on the games behalf)
That a total of 19-20 GB
The consoles only has around 13 GB of RAM to play with, 14 GB the most. So he speculates the I/O fast latency to get something from storage to RAM is what is helping with frame rate in some cases. Not everything can fit in RAM and CPU resources are freed up.

He implies more assets should be in GPU Memory to help with the latency because his GPU with 16 GB isn't fully being utilized.

Imo, RTX IO will better utilize GPU Memory and 8 GB wouldn't be enough. Decompression can also limit the SSD potential.
gB1HPzm.jpg


PC environment with latency from CPU decompression.
pv56VYI.png


PS5 environment with Custom I/O.
sseGaPi.jpg

rJokUyZ.jpg



NX Gamer also thinks Nanite constantly updating the Nanite clusters (which the CPU handles) when moving is the reason for frame rate drops also. So Alex is right about higher clocked CPU is better, the Nanite clusters can be updated faster. At least that's how I interpreted it from both videos.
 

GHG

Gold Member
Why is his system hitting these bottlenecks when everyone else in the world isn’t with the same graphics cards and CPus.

Look at results from other people, and you'll understand what I said.

Have we got benchmarks with the 2700x for this demo? All I know is that the 2xxx series of CPU's from AMD are far more susceptible to encountering bottlenecks when gaming due to the CCX layout. Any average % differences in CPU's go out of the window when that happens.
 

winjer

Gold Member
Have we got benchmarks with the 2700x for this demo?
Yes.

All I know is that the 2xxx series of CPU's from AMD are far more susceptible to encountering bottlenecks when gaming due to the CCX layout. Any average % differences in CPU's go out of the window when that happens.

Zen+ has an identical CCX configuration to Zen2.
 

GHG

Gold Member
Yes.



Zen+ has an identical CCX configuration to Zen2.

The layout is the same the latencies are not. They made a huge improvement from the 2xxx series to the 3xxx series. Hence emulation went from being a joke on any 2xxx CPU's to almost on par with Intel on the 3xxx series. You can read details on the changes here:


Also, if yes then show them. A quick google turned up nothing for the 2700x.
 
Last edited:

winjer

Gold Member
The layout is the same the latencies are not. They made a huge improvement from the 2xxx series to the 3xxx series. Hence emulation went from being a joke on any 2xxx CPU's to almost on par with Intel on the 3xxx series. You can read details on the changes here:


Also, if yes then show them. A quick google turned up nothing for the 2700x.

Mate, just look at the benchmarks done with a dozen cpus. The 2700x is just 5fps slower than a 3800x.
The 2700x is an old CPU, but it's not that bad.
 
Top Bottom