• 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.

Digital Foundry: Deathloop PC vs PS5, Optimised Settings, Performance Testing + More

Mr Moose

Member
Microsoft have never claimed the XSX runs way better then the PS5.

PS5 games ran better when XSX tools were half baked. Its been a while since a PS5 game has ran better, apart from the raytracing bug in little nightmares 2 on xsx.
It's been 3 weeks.

F1 2021.

What examples, the hitman 3 one shows the opposite of cernys claims.
There is no example of the PS5 performing significantly better then a 5700xt in games without Raytracing.
I think you mean Hitman 2.
 
Last edited:

Md Ray

Member
If only cerny showed some demos demonstrating his point😟
We have results showing 32 CU GPU being on par or better than a 40 CU GPU with less bandwidth. What more demo do you want?

If Sony had ended up with a 48 CU GPU clocked at 1673 MHz, I'm certain the PS5 with this config would have been slower in many gaming workloads than with the 36 CU @ 2230 MHz config we have in the final unit.
 
Last edited:

Sosokrates

Report me if I continue to console war
We have results showing 32 CU GPU being on par or better than a 40 CU GPU with less bandwidth. What more demo do you want?

If Sony had ended up with a 48 CU GPU clocked at 1673 MHz, I'm certain the PS5 with this config would have been slower in many gaming workloads compared to the 36 CU config we have now in the final unit.
I think the 6600xt and 5700xt is actually a bad comparison in this circumstance because theres to many other variables such as the memory setup and superior tflop number of the 6600xt.

Imo the better test to see if higher clocks give a significantly more performance beyond the tflops is something like this:

 

Md Ray

Member
I think the 6600xt and 5700xt is actually a bad comparison in this circumstance because theres to many other variables such as the memory setup and superior tflop number of the 6600xt
There are instances where 6600 XT with 5-6% more TF had up to 11% higher avg fps while having a lot less bandwidth than 5700 XT. I think it's a good comparison and proves Cerny was right.
 
Last edited:

Sosokrates

Report me if I continue to console war
There are instances where 6600 XT with 5-6% more TF had up to 11% higher avg fps. I don't think it's a bad comparison.

The problem is you dont know where the improvement is coming from. It could be the memory setup, higher tflops or higher clockspeed.
 
I think the 6600xt and 5700xt is actually a bad comparison in this circumstance because theres to many other variables such as the memory setup and superior tflop number of the 6600xt.

Imo the better test to see if higher clocks give a significantly more performance beyond the tflops is something like this:


I can run my GPU at higher clock then I do because the higher clock actually reduces performance.

Edit second thought it's probably because boosting.
 
Last edited:

Md Ray

Member
The problem is you dont know where the improvement is coming from. It could be the memory setup, higher tflops or higher clockspeed.
Edited my post.

The 6600 XT is at a disadvantage when it comes to bandwidth/mem setup compared to 5700 XT.

Any improvement you see on 6600 XT is coming from the higher clock speed, so Cerny was right all along.

Please just say 5% better. I know 11 sounds better.
No, I won't. You have a problem?
 
Last edited:

Sosokrates

Report me if I continue to console war
Edited my post.

The 6600 XT is at a disadvantage when it comes to bandwidth compared to 5700 XT.

Any improvement you see on 6600 XT is coming from the higher clock speed, so Cerny was right all along.


No. You have a problem?


The 6600xt has about 1-5% better performance then the 5700xt, so I dont know how this proves Cenry correct.
1-5% is not significant and that improvement could be because the 6600xt is caperble of more calculations(Tflops)

Amyway mr md ray i need to put my chicken and potatoes in the oven 👨‍🍳
 
Last edited:

FireFly

Member
The card physically has less BW than 5700 XT. It's just that it's more evident at 4K.
It also has an Infinity Cache, which it can "hit" ~55% of the time at 1080p. When the data it needs is in the cache, it gets up to 3.25x more bandwidth.



"With up to 128MB of ground-breaking AMD Infinity Cache acting as a massive bandwidth amplifier, get up to 3.25x the effective bandwidth of 256-bit 16 Gbps GDDR6 memory"


So 45% of the time it has access to 57% of the bandwidth of a 5700 XT, and the other 55% of the time, it has access to up to 3.25x the bandwidth. So I make that up to double the effective bandwidth of the 5700 XT. Obviously it's up to, so we don't know what the average speedup is. But we can't just assume it is less than 1.35x because we want it to have less effective bandwidth than the 5700 XT.
 

Md Ray

Member
The 6600xt has about 1-5% better performance then the 5700xt, so I dont know how this proves Cenry correct.
1-5% is not significant and that improvement could be because the 6600xt is caperble of more calculations(Tflops)
The point that Cerny made was that a lower CU GPU with 4.6 TF can be just as fast as a higher CU GPU with a lower clock speed arriving at the same 4.6 TF number.

And looking at the above 6600 XT (32 CU) and 5700 XT (40 CU) results, he was right. 6600 XT has a 5-6% higher TF, which also translates to 6% higher fps on avg.
 

Sosokrates

Report me if I continue to console war
The point that Cerny made was that a lower CU GPU with 4.6 TF can be just as fast as a higher CU GPU with a lower clock speed arriving at the same 4.6 TF number.

And looking at the above 6600 XT (32 CU) and 5700 XT (40 CU) results, he was right. 6600 XT has a 5-6% higher TF, which also translates to 6% higher fps on avg.

No, he said that the narrow and fast version has significant performance gains over the wide + slow, even though they are the same Tflops.

Cernys comparison would be like if there were a 44cu 6600xt @1885mhz , the 32cu varient at 2.6ghz would have significant performance gains.
 

Md Ray

Member
It also has an Infinity Cache, which it can "hit" ~55% of the time at 1080p. When the data it needs is in the cache, it gets up to 3.25x more bandwidth.



"With up to 128MB of ground-breaking AMD Infinity Cache acting as a massive bandwidth amplifier, get up to 3.25x the effective bandwidth of 256-bit 16 Gbps GDDR6 memory"


So 45% of the time it has access to 57% of the bandwidth of a 5700 XT, and the other 55% of the time, it has access to up to 3.25x the bandwidth. So I make that up to double the effective bandwidth of the 5700 XT. Obviously it's up to, so we don't know what the average speedup is. But we can't just assume it is less than 1.35x because we want it to have less effective bandwidth than the 5700 XT.

128 MB infinity cache and 256-bit applies to 6800 XT and up.

What's the effective bandwidth of 6600 XT? It has significantly less cache at 32MB and a much narrower 128-bit bus width, not 256-bit. It also clearly says: "the hit rate for small $ is poor" in that tweet. 4K results are your evidence that 6600 XT has poor mem BW than 5700 XT.
 

Md Ray

Member
No, he said that the narrow and fast version has significant performance gains over the wide + slow, even though they are the same Tflops.

Cernys comparison would be like if there were a 44cu 6600xt @1885mhz , the 32cu varient at 2.6ghz would have significant performance gains.
No...

I went back and watched the vid. He did not use the words "significant performance gains"
 
Last edited:
Edited my post.

The 6600 XT is at a disadvantage when it comes to bandwidth/mem setup compared to 5700 XT.

Any improvement you see on 6600 XT is coming from the higher clock speed, so Cerny was right all along.


No, I won't. You have a problem?
As I said a few posts ago. In cyberpunk the 5600xt benches just a bit lower than its tflop count would show against the 5700xt which has more bandwidth. Cus and higher clocks.
The card physically has less BW than 5700 XT. It's just that it's more evident at 4K.
And not evident at 1080p in most games.
And looking at the above 6600 XT (32 CU) and 5700 XT (40 CU) results, he was right. 6600 XT has a 5-6% higher TF, which also translates to 6% higher fps on avg.
Thanks

Edit
Gonna add that AMD heavily markets the 6600xt as a 1080p card. If the memory bandwidth effected it at 1080p they would be wasting money on cu's and that's something I doubt they are doing.
 
Last edited:

FireFly

Member
128 MB infinity cache and 256-bit applies to 6800 XT and up.

What's the effective bandwidth of 6600 XT? It has significantly less cache at 32MB and a much narrower 128-bit bus width, not 256-bit. It also clearly says: "the hit rate for small $ is poor" in that tweet. 4K results are your evidence that 6600 XT has poor mem BW than 5700 XT.
1.) The 3.25x claim is on the general Radeon RX 6000 page, so it looks to be a property of the SRAM itself, rather than being specific to the 6800 XT.


2.) The smaller cache means it will get "hit" less often which reduces its effective bandwidth, but it still gets used 55% of the time at 1080p. You can't just pretend that doesn't happen.
3.) The 6600 XT has 256.0 GB/s of memory bandwidth through 128-bit bus, compared to 448 Gb/s for the 5700 XT's 256-bit bus. That's 57%. 0.57 * 0.45 + 0.55 * 3.25 = 2.044. So the 6600 XT has up to double the effective bandwidth at 1080p.
4.) Yes, the hit rate is poor at 1440p and 4K especially. That's why the card does worse as the resolution increases. But I was talking about 1080p performance, not 1440p or 4K performance. The way the IC works means the card will have less effective bandwidth as you increase the resolution.1080p is where we see the biggest performance advantage, and where the card may be benefiting from the IC in comparison to the 5700 XT.
 
Last edited:

Sosokrates

Report me if I continue to console war
No...

I went back and watched the vid. He did not use the words "significant performance gains"
He never said that.

Sorry my mistake, he said that when referring to the boost clock implementation.

However he did say higher clocks results in rasterization and others running faster.
You know "a rising tide lifts all boats".

He gave the impression that narrow and faster is better then wider and slower when tflops are the same.

While there will be some benefits to running clocks higher I think he was exaggerating its benefits over a gpu with the same tflops but wider and slower.
I mean the seriesS is far slower yet its still about 40% of the PS5s power, if higher clocks make the gpu overall faster why arent we seeing greater disparity?
 

Keihart

Member
I think it looks great. What people are overlooking is the quality of the environment designs, the art, the geometry details.
If You Say So Reaction GIF by Identity
 

Loxus

Member
Sorry my mistake, he said that when referring to the boost clock implementation.

However he did say higher clocks results in rasterization and others running faster.
You know "a rising tide lifts all boats".

He gave the impression that narrow and faster is better then wider and slower when tflops are the same.

While there will be some benefits to running clocks higher I think he was exaggerating its benefits over a gpu with the same tflops but wider and slower.
I mean the seriesS is far slower yet its still about 40% of the PS5s power, if higher clocks make the gpu overall faster why arent we seeing greater disparity?

f:id:keepitreal:20200329151011j:plain


"Another set of issues for the GPU involved size and frequency.

How big do we make the GPU and what frequency do we run it at.

This is a balancing act, the chip has a cost and there's a cost for whatever we use to supply that chip with power and to cool it.

In general I like running the GPU at a higher frequency. Let me show you why.

f:id:keepitreal:20200329151024j:plain


Here's two possible configurations for a GPU roughly of the level of the PlayStation 4 Pro. This is a thought experiment don't take these configurations too seriously.

If you just calculate teraflops you get the same number, but actually the performance is noticeably different because teraflops is defined as the computational capability of the vector ALU.

That's just one part of the GPU, there are a lot of other units and those other units all run faster when the GPU frequency is higher. At 33% higher frequency rasterization goes 33% faster, processing the command buffer goes that much faster and other caches have that much higher bandwidth and so on.

About the only downside is that system memory is 33% further away in terms of cycles. But the large number of benefits more than counterbalanced that.

As a friend of mine says a rising tide lifts all boats.

Also it's easier to fully use 36CUs in parallel than it is to fully use 48CUs. When triangles are small it's much harder to fill all those CUs with useful work."


Mark Cerny had 2 PS5 configurations.
One 36 CUs @ 1GHz and the other 48CU @ 0.75. And in terms of games, the 36 CUs config performs better.

In order to call Cerny a liar, you would either have to have 2 PS5s (36CUs vs. 48CUs) or 2 RDNA 2 GPUs (36CUs vs. 48CUs) and they would have to be running a game design around the hardware.

Comparing GPUs from different generations/architectures and custom GPUs to dedicated PC GPUs can only give inaccurate information.
 
Last edited:

Sosokrates

Report me if I continue to console war
f:id:keepitreal:20200329151011j:plain


"Another set of issues for the GPU involved size and frequency.

How big do we make the GPU and what frequency do we run it at.

At this is a balancing act the chip has a cost and there's a cost for whatever we use to supply that chip with power and to cool it.

In general I like running the GPU at a higher frequency. Let me show you why.

f:id:keepitreal:20200329151024j:plain


Here's two possible configurations for a GPU roughly of the level of the PlayStation 4 Pro. This is a thought experiment don't take these configurations too seriously.

If you just calculate teraflops you get the same number, but actually the performance is noticeably different because teraflops is defined as the computational capability of the vector ALU.

That's just one part of the GPU, there are a lot of other units and those other units all run faster when the GPU frequency is higher. At 33% higher frequency rasterization goes 33% faster, processing the command buffer goes that much faster the 2 and other caches have that much higher bandwidth and so on.

About the only downside is that system memory is 33% further away in terms of cycles. But the large number of benefits more than counterbalanced that.

As a friend of mine says a rising tide lifts all boats.

Also it's easier to fully use 36CUs in parallel than it is to fully use 48CUs. When triangles are small it's much harder to fill all those CUs with useful work."


Mark Cerny had 2 PS5 configurations.
One 36 CUs @ 1GHz and the other 48CU @ 0.75. And in terms of games, the 36 CUs config performs better.

In order to call Cerny a liar, you would either have to have 2 PS5s (36CUs vs. 48CUs) or 2 RDNA 2 GPUs (36CUs vs. 48CUs) and they would have to be running a game design around the hardware.

Comparing GPUs from different generations/architectures and custom GPUs to dedicated PC GPUs can only give inaccurate information.

You can disagree with it, but in comparison to the Xbox series consoles we are not seeing the performance of the PS5 differ more then the specs would suggest. You would think based on what cerny said regarding the PS5 it would punch above it weight against the seriesX, but it doesent.
The seriesX and seriesS have shown that performance scales with the Tflops and memory bandwidth regardless of what the clock speed.

People touting the PS5s higher clock speed is some kinda golden goose beyond what the tflops are is some secret sauce terrority.

These consoles are not that special, gone are the days of custom exotic hardware. They are basically PCs. And judging by the PS5s games we are seeing a console that has performance like a console with a 18% weaker GPU, and about 15% less ram bandwidth and a ssd/io which is about 2.5x faster compared to the xsx.

And there absolutely nothing wrong with this, its a great console. But people who try and say its more then what it is should be called out.

Same thing happened with the xbox one, with its move engines, DX12, BRAD wardell, drawcalls etc etc
 
Last edited:

Md Ray

Member
He gave the impression that narrow and faster is better then wider and slower when tflops are the same.

While there will be some benefits to running clocks higher I think he was exaggerating its benefits over a gpu with the same tflops but wider and slower.
I mean the seriesS is far slower yet its still about 40% of the PS5s power, if higher clocks make the gpu overall faster why arent we seeing greater disparity?
What exaggeration? Do the math. With 48 CU @ 1.67 GHz, you get the same TFLOPs, but rasterization rate, pixel fillrate along with cache bandwidths - they all go down compared to 36 CU @ 2.23 GHz config. It's a fact. I don't understand your issue here.
 
Last edited:

Sosokrates

Report me if I continue to console war
What exaggeration? Do the math. With 48 CU @ 1.67 GHz, you get the same TFLOPs, but rasterization rate, pixel fillrate along with cache bandwidths - they all go down compared to 36 CU @ 2.23 GHz config. It's a fact. I don't understand your issue here.

Well the only way we can know is cernys word or how the PS5 compares to xbox series and PCs.
Ive stated my opinion and if we disagree.... we disagree, pretty simple really.
 
Last edited:

Md Ray

Member
but in comparison to the Xbox series consoles we are not seeing the performance of the PS5 differ more then the specs would suggest.
Are we not? Like I've said multiple times on this thread and other threads. It depends.

You would think based on what cerny said regarding the PS5 it would punch above it weight against the seriesX, but it doesent.
The seriesX and seriesS have shown that performance scales with the Tflops and memory bandwidth regardless of what the clock speed.
And you would think based on looking at TF and BW numbers alone the XSX would always have an 18-25% advantage, but it doesn't. PS5 does punch above its weight.

A game isn't bound by one part of the GPU. Different game engines and even different scenes within a game can be bound by any number of things like ALU or fillrate, rasterization or bandwidth or a combination of things for example.

In control, this scene here shows a 16% difference in favor of XSX as one may expect:
K4pxRj2.png


Well, here in these other scenes at the same settings and res, we see that advantage is reduced to 0-3% because there are parts of the PS5 GPU that have higher triangle throughput, fillrate than XSX GPU due to higher clock speed which is what's allowing it to produce fps on par 12TF console in some of these scenes/games. There's also a scene from RE8 below:

LBiUgVM.png

JOqj3Wi.png

fAQPIfy.jpg


Once again the above results prove Cerny correct.
 
Last edited:
f:id:keepitreal:20200329151011j:plain


"Another set of issues for the GPU involved size and frequency.

How big do we make the GPU and what frequency do we run it at.

This is a balancing act, the chip has a cost and there's a cost for whatever we use to supply that chip with power and to cool it.

In general I like running the GPU at a higher frequency. Let me show you why.

f:id:keepitreal:20200329151024j:plain


Here's two possible configurations for a GPU roughly of the level of the PlayStation 4 Pro. This is a thought experiment don't take these configurations too seriously.

If you just calculate teraflops you get the same number, but actually the performance is noticeably different because teraflops is defined as the computational capability of the vector ALU.

That's just one part of the GPU, there are a lot of other units and those other units all run faster when the GPU frequency is higher. At 33% higher frequency rasterization goes 33% faster, processing the command buffer goes that much faster and other caches have that much higher bandwidth and so on.

About the only downside is that system memory is 33% further away in terms of cycles. But the large number of benefits more than counterbalanced that.

As a friend of mine says a rising tide lifts all boats.

Also it's easier to fully use 36CUs in parallel than it is to fully use 48CUs. When triangles are small it's much harder to fill all those CUs with useful work."


Mark Cerny had 2 PS5 configurations.
One 36 CUs @ 1GHz and the other 48CU @ 0.75. And in terms of games, the 36 CUs config performs better.

In order to call Cerny a liar, you would either have to have 2 PS5s (36CUs vs. 48CUs) or 2 RDNA 2 GPUs (36CUs vs. 48CUs) and they would have to be running a game design around the hardware.

Comparing GPUs from different generations/architectures and custom GPUs to dedicated PC GPUs can only give inaccurate information.

Looking back at this. One of the reasons why you would want less CUs over more CUs is that they can make the APU smaller. Which means you could in theory produce more chips per wafer. Which obviously leads to a higher production and more PS5s being sold.

I know that really isn't on topic but it's a nice benefit of going with a smaller APU than a larger one that offers the same performance.

Please correct me if im wrong though.
 
Last edited:

Sosokrates

Report me if I continue to console war
Are we not? Like I've said multiple times on this thread and other threads. It depends.


And you would think based on looking at TF and BW numbers alone the XSX would always have an 18-25% advantage, but it doesn't. PS5 does punch above its weight.

A game isn't bound by one part of the GPU. Different game engines and even different scenes within a game can be bound by any number of things like ALU or fillrate, rasterization or bandwidth or a combination of things for example.

In control, this scene here shows a 16% difference in favor of XSX as one may expect:
K4pxRj2.png


Well, here in these other scenes at the same settings and res, we see that advantage is reduced to 0-3% because there are parts of the PS5 GPU that have higher triangle throughput, fillrate than XSX GPU due to higher clock speed which is what's allowing it to produce fps on par 12TF console in some of these scenes/games. There's also a scene from RE8 below:

LBiUgVM.png

JOqj3Wi.png

fAQPIfy.jpg


Once again the above results prove Cerny correct.

Lol you think im going to agree with you... I dont. You can say the PS5 punches above its weight, that great, thats your opinion, we disagree, the xbox one also had parity in certain games, going by your logic that means the x1 also punched above its weight.

The specs suggest 20-25% improvement for series X?

Well hitman 3 has a 30% higher resolution then ps5 and metro exodus has about 24% so....
 

Sosokrates

Report me if I continue to console war
Don't care.

Show me a game where X1 matched PS4 frame rates at identical settings and res.
If you dont care whats the point in posting to me?
X1 games which had parity were most of the fifas and other sports games, Ac unity im sure there others.
I see you ignored the 24% and 30% resolution advantage i stated.

Actually I made a mistake 2160p is actually 44% higher then 1800p.
Ironically its the same percentage difference of 900p and 1080p.
 
Last edited:

Md Ray

Member
If you dont care whats the point in posting to me?
X1 games which had parity were most of the fifas and other sports games, Ac unity im sure there others.
I see you ignored the 24% and 30% resolution advantage i stated.
Well, as I said it depends from game engine to game engine. XSX will have an advantage in some and in some PS5 is able to perform on par 12TF console due to higher clock speed, you ignored all the screenshots I posted.
Oh, and Hitman 3 has higher res at the cost of frame-rate dips.
 

Sosokrates

Report me if I continue to console war
Well, as I said it depends from game engine to game engine. XSX will have an advantage in some and in some PS5 is able to perform on par 12TF console due to higher clock speed, you ignored all the screenshots I posted.
Oh, and Hitman 3 has higher res at the cost of frame-rate dips.

Well, I dont agree that the reason the PS5 sees parity is because of its high clock speed making up for its tflop deficit.
I think in controls case the reason they are so close is because 1.they just want to get the ungrade out. 2.it was easier to just cap the seriesX version at 30fps rather then add more work for specific seriesX enhancemens, interestingly DF suggested a 1080p60fps Raytracing mode on the seriesX for control would be possible, but not so on the PS5.
 

hlm666

Member
There are instances where 6600 XT with 5-6% more TF had up to 11% higher avg fps while having a lot less bandwidth than 5700 XT. I think it's a good comparison and proves Cerny was right.
The 6600xt has 32mb infinity cache so that makes it much harder to say less bandwidth doesn't matter and higher frequency is why its faster.
 

SlimySnake

Flashless at the Golden Globes
No, he said that the narrow and fast version has significant performance gains over the wide + slow, even though they are the same Tflops.

Cernys comparison would be like if there were a 44cu 6600xt @1885mhz , the 32cu varient at 2.6ghz would have significant performance gains.
You know whats funny? Cerny was talking about a GPU similar to a PS4 pro in his comparison, right? Well, lets take the PS4 Pro 36 CU GPU at 911 mhz and the X1X 40 CU GPU at 1172 mhz and see the results. Because with just an 11% increase in CUs and a 28% increase in clockspeeds, X1X was able to run several games at native 4k including the best looking game of all time, RDR2. While the PS5 had to settle for literally half the resolution.

The tflops difference between the two consoles was only 42%, but several games were native 4k on the x1x. Rise of the Tomb Raider, Shadow of tomb raider, DMCV were all half of the resolution of the X1X versions. Nearly all x1x exclusives had native 4k modes. FH4, Gears 5, Halo 5 all run at native 4k. PS4 exclusives had to settle for 1440p at times because 4kcb was too much.

I think Cerny looked at that and learned that higher clocks might actually help him keep the silicon costs down and get higher performance.

Then there is this:

s0n39Hi.png


The XSX column numbers are directly from MS's hot chips presentation from last year extrapolated to PS5 using their calculations. As you can see, XSX has an advantage in ray tracing and texture fillrate calculations, but the PS5 GPU succeed in operations that rely heavily on clockspeeds. Again, this is straight from MS's own slides. If these things werent so important, MS wouldnt be mentioning them in technical presentations.

202008180221421_575px.jpg

202008180228051.jpg
 
Last edited:

Sosokrates

Report me if I continue to console war
You know whats funny? Cerny was talking about a GPU similar to a PS4 pro in his comparison, right? Well, lets take the PS4 Pro 36 CU GPU at 911 mhz and the X1X 40 CU GPU at 1172 mhz and see the results. Because with just an 11% increase in CUs and a 28% increase in clockspeeds, X1X was able to run several games at native 4k including the best looking game of all time, RDR2. While the PS5 had to settle for literally half the resolution.

The tflops difference between the two consoles was only 42%, but several games were native 4k on the x1x. Rise of the Tomb Raider, Shadow of tomb raider, DMCV were all half of the resolution of the X1X versions. Nearly all x1x exclusives had native 4k modes. FH4, Gears 5, Halo 5 all run at native 4k. PS4 exclusives had to settle for 1440p at times because 4kcb was too much.

I think Cerny looked at that and learned that higher clocks might actually help him keep the silicon costs down and get higher performance.

Then there is this:

s0n39Hi.png


The XSX column numbers are directly from MS's hot chips presentation from last year extrapolated to PS5 using their calculations. As you can see, XSX has an advantage in ray tracing and texture fillrate calculations, but the PS5 GPU succeed in operations that rely heavily on clockspeeds. Again, this is straight from MS's own slides. If these things werent so important, MS wouldnt be mentioning them in technical presentations.

202008180221421_575px.jpg

202008180228051.jpg

No doubt cost was a big factor.

But cerny did say narrow + fast provide a noticeable difference in performance compared to wide + slow but same TF.
I suppose a small difference is still noticeable. But cerny makes it sound like something not small.

[/Cerny] If you just calculate teraflops you get the same number, but actually the performance is noticeably different because teraflops is defined as the computational capability of the vector ALU.

Also people said the same thing about the X1 having better triangle culling + better pixel fillrate, but that did not make any difference beyond what the tflops + ram bandwidth suggested.
 
Last edited:

Loxus

Member
Looking back at this. One of the reasons why you would want less CUs over more CUs is that they can make the APU smaller. Which means you could in theory produce more chips per wafer. Which obviously leads to a higher production and more PS5s being sold.

I know that really isn't on topic but it's a nice benefit of going with a smaller APU than a larger one that offers the same performance.

Please correct me if im wrong though.
No, your correct.
I heard somewhere, every 3 XBSX chips, you get 4 PS5 chips.

It's also a balancing act with heat and cooling also. Larger the chip, the lower the clocks and the higher the clocks on a large chip can produce more heat. So Cerny settled with the PS5 we have now.
 
Last edited:

Loxus

Member
You can disagree with it, but in comparison to the Xbox series consoles we are not seeing the performance of the PS5 differ more then the specs would suggest. You would think based on what cerny said regarding the PS5 it would punch above it weight against the seriesX, but it doesent.
The seriesX and seriesS have shown that performance scales with the Tflops and memory bandwidth regardless of what the clock speed.

People touting the PS5s higher clock speed is some kinda golden goose beyond what the tflops are is some secret sauce terrority.

These consoles are not that special, gone are the days of custom exotic hardware. They are basically PCs. And judging by the PS5s games we are seeing a console that has performance like a console with a 18% weaker GPU, and about 15% less ram bandwidth and a ssd/io which is about 2.5x faster compared to the xsx.

And there absolutely nothing wrong with this, its a great console. But people who try and say its more then what it is should be called out.

Same thing happened with the xbox one, with its move engines, DX12, BRAD wardell, drawcalls etc etc


This is next-gen: see Unreal Engine 5 running on PlayStation 5
The 'Lumen in the Land of Nanite' demo includes a close-up on a statue built from 33 million triangles with 8K textures. It's displayed at maximum fidelity within the scene, with no developer input required. Moving into the next room, the demo wows us with almost 500 of those same statues in place (485 to be precise), all displayed at the same maximum quality. That's 16 billion triangles in total, running smoothly in-scene.

This is most likely one of the main reason for high clocks/low CU count when Mark Cerny was, deciding which PS5 configuration to use (36CU vs. 48CU) and the 36CU had better results. It's that simple.

"Also it's easier to fully use 36CUs in parallel than it is to fully use 48CUs. When triangles are small it's much harder to fill all those CUs with useful work." - Mark Cerny


So far, we haven't seen this level of fidelity on Xbox or PC.



All we have for Xbox and PC is this.
The giant robot has 50 million triangles and that's not even close to a billion, let alone 16 billion.

Maybe high clock/Low CU count = better in-game textures/triangles.
And Low clock/high CU count = higher resolution.

Low quality textures @ 4K still looks like shit as seen in Halo Infinite.
 

Loxus

Member
Still says it when I click on Amazon.

1p8sd0u.jpg
Read the fine print.
Introducing Xbox Series X - our fastest, most powerful console ever.

our
/ˈou(ə)r,är/
determiner
  1. belonging to or associated with the speaker and one or more other people previously mentioned or easily identified.
    "Jo and I had our hair cut"
  2. used by a writer, editor, or monarch to refer to something belonging to or associated with himself or herself.
    "we want to know what you, our readers, think"
 


This is next-gen: see Unreal Engine 5 running on PlayStation 5
The 'Lumen in the Land of Nanite' demo includes a close-up on a statue built from 33 million triangles with 8K textures. It's displayed at maximum fidelity within the scene, with no developer input required. Moving into the next room, the demo wows us with almost 500 of those same statues in place (485 to be precise), all displayed at the same maximum quality. That's 16 billion triangles in total, running smoothly in-scene.

This is most likely one of the main reason for high clocks/low CU count when Mark Cerny was, deciding which PS5 configuration to use (36CU vs. 48CU) and the 36CU had better results. It's that simple.

"Also it's easier to fully use 36CUs in parallel than it is to fully use 48CUs. When triangles are small it's much harder to fill all those CUs with useful work." - Mark Cerny


So far, we haven't seen this level of fidelity on Xbox or PC.



All we have for Xbox and PC is this.
The giant robot has 50 million triangles and that's not even close to a billion, let alone 16 billion.

Maybe high clock/Low CU count = better in-game textures/triangles.
And Low clock/high CU count = higher resolution.

Low quality textures @ 4K still looks like shit as seen in Halo Infinite.


Where do you think developers make your PS5 games/demos?

 
Last edited:

hlm666

Member
All we have for Xbox and PC is this.
The giant robot has 50 million triangles and that's not even close to a billion, let alone 16 billion.

Maybe high clock/Low CU count = better in-game textures/triangles.
And Low clock/high CU count = higher resolution.

Low quality textures @ 4K still looks like shit as seen in Halo Infinite.
Hey champ not sure where the 16 billion triangles are coming from but the first ue5 demo isn't doing that either....... Oh yeh it also did run on pc.

 
Top Bottom