By using this site, you agree to our Privacy Policy and our Terms of Use. Close

Forums - Gaming Discussion - Ubisoft GDC Presentation of PS4 & X1 GPU & CPU Performance

walsufnir said:
darkknightkryta said:
Yeah, PS4's GPU is pretty beastly compared to the competition. Cell was also a beast of a processor. It's a shame there was never any incentive to progress the technology for Cell, I feel it scales very well.


Oh, there was. But not for consoles.

I don't think you read my statement properly...



Around the Network
ethomaz said:

Captain_Tom said:

The PS4 GPU won in that benchmark because they didn't use the X1's ESRAM, and the PS4 main memory is over twice as fast (Which made up for the X1's clock speed advantage).

I thought it won because they tested GPGPU (GPU compute) and PS4 have 8 ACEs, 64 queues for compute while Xbone only 2 ACEs, 8 queues... and of course the obvious difference in RAW power helps too (18 vs 12 CUs).


Damn I meant to say cpu lol.  My bad.



ethomaz said:

Captain_Tom said:

The PS4 GPU won in that benchmark because they didn't use the X1's ESRAM, and the PS4 main memory is over twice as fast (Which made up for the X1's clock speed advantage).

I thought it won because they tested GPGPU (GPU compute) and PS4 have 8 ACEs, 64 queues for compute while Xbone only 2 ACEs, 8 queues... and of course the obvious difference in RAW power helps too (18 vs 12 CUs).

I don't think the 64 queues and ACEs make the big difference here.  Those queues will make a difference when there are a lot of different GPU compute tasks contending for the GPU.  That's unlikely the case in a benchmark like this, where they have said they have combined their work into larger tasks.

Bandwidth could be the biggest bottleneck.  The presentation talks about mapping data between CPU & GPU, so the RAM speed and the coherency features likely make a big difference.

"didn't use the X1's ESRAM" - ESRAM is a limited resource so devs have to choose if they use it for backbuffer or textures or GPU compute data.  Can't have it all, since the XB1 has 8GB of main RAM and only 32MB of ESRAM, majority of a game's data is resident in RAM and needs to use RAM bandwidth to copy to/from ESRAM.



My 8th gen collection

Captain_Tom said:
HollyGamer said:
Captain_Tom said:
HollyGamer said:
Even if the CPU is weak they can utilize GPGPU for PS4, because that's the purpose of GPGPU for the AMD technology right? PS4 also benefit from HSA technology and more CU on the GPU.


Correct.  The PS4's GPU actually has a ton more optimizations and add-ons to the the GPU that make it dominate at GPGPU.  That is why in the end the gap will grow even more than people think it will.

I guess we have to wait for Naughty Dog to shows and smack every lazy developer face when Uncharted 4 comes :) .  I think when it comes to multyplatform games it's gonna be hard to shows PS4 graphic capability , because they trying to sell to all console fans, and parity clause is a nature thing for every third party developer. That's why console sales and sofware sales  is very important because the developer and publisher will see which console has a bigger fanbase and which fanbase buy most of their games, and determine which console will be  the priority. 


With a few exceptions (AC:U) practically every third party game has taken advantage of the extra performance.  Keep in mind that 1080p is 44% more pixels than 900p, and most games also have extra AA and effects on PS4.  

 

SoM for example was running at 1080p with high settings while the X1 was running 900p at medium settings.  That is pretty close to the difference (Minus the GPGPU advantage knowone has completily tackeled yet).

Yeah i mostly agree with u but they also still using midle ware like direct x and other stuff to. That said when developer like naughty Dogs able to code close to metal, i hope every deveolper who trully want to improvise learn from Naughty Dog in the future.



HollyGamer said:
Captain_Tom said:
HollyGamer said:
Captain_Tom said:
HollyGamer said:
Even if the CPU is weak they can utilize GPGPU for PS4, because that's the purpose of GPGPU for the AMD technology right? PS4 also benefit from HSA technology and more CU on the GPU.


Correct.  The PS4's GPU actually has a ton more optimizations and add-ons to the the GPU that make it dominate at GPGPU.  That is why in the end the gap will grow even more than people think it will.

I guess we have to wait for Naughty Dog to shows and smack every lazy developer face when Uncharted 4 comes :) .  I think when it comes to multyplatform games it's gonna be hard to shows PS4 graphic capability , because they trying to sell to all console fans, and parity clause is a nature thing for every third party developer. That's why console sales and sofware sales  is very important because the developer and publisher will see which console has a bigger fanbase and which fanbase buy most of their games, and determine which console will be  the priority. 


With a few exceptions (AC:U) practically every third party game has taken advantage of the extra performance.  Keep in mind that 1080p is 44% more pixels than 900p, and most games also have extra AA and effects on PS4.  

 

SoM for example was running at 1080p with high settings while the X1 was running 900p at medium settings.  That is pretty close to the difference (Minus the GPGPU advantage knowone has completily tackeled yet).

Yeah i mostly agree with u but they also still using midle ware like direct x and other stuff to. That said when developer like naughty Dogs able to code close to metal, i hope every deveolper who trully want to improvise learn from Naughty Dog in the future.

Hey, don't get me wrong - I completely agree that the grand majority of devs have been lazy with the extra performance.  Heck I will use ACU as an EXCELLENT example:

-Ubisoft says that the CPU's are not fast enough to handle 60 FPS with so many people on screen.  This is true if you only use the CPU for both of those things.  However large numbers of similar calculations are great for GPGPU. Hmmmmm, oh yeah Ubisofts hordes of stupid civilians would be perfect for GPGPU.  But they know that the X1's paltry GPGPU capabilities wouldn't make much of a difference, and so they were to lazy to take full advantage of the PS4 hardware.

-Yes Naughty Dog and Guerilla Games will come out and make ACU look practically last gen in comparison.  Don't kid yourself, third parties never want their games to look bad on even one platform and they will change there ways to catch up.  Just look at how ND and GG forced DICE to make BF3 take full advantage of the PS3.  It had higher res tectures, better lighting, more particle affects, and better AA than the 360 once they used the CELL's SPs effectivelly...And the PS3 was only 50% stronger than the 360 at best, so imagine what will happen now...



Around the Network
Captain_Tom said:
walsufnir said:
I always think these benchmarks funny. Remember the benchmark from last year that showed us PS4 CPU was faster than Xbox One? Now this shows the exact opposite.
Without any further data what was really tested (given these slides provide way more information than the graph last year) we can't infer *anything*, in my opinion.


The PS4 CPU won in that benchmark because they didn't use the X1's ESRAM, and the PS4 main memory is over twice as fast (Which made up for the X1's clock speed advantage).


Why wouldn't they? And where do they say they didn't? Any detail on the test for Xbone is missing. They are going from DX11 to PS4 optimization and PS4 details talk.



ICStats said:
walsufnir said:
ethomaz said:

mornelithe said:

You mean last year, before Microsoft bumped the clock up 10%?  And this year...after the bump, it shows differently?  No way!

Or after June SDK that free up the Kinect use of the CPU :D


1. The first benchmark arose when consoles already launched. So it was after the upclock.

2. Kinect freed gpu resources, not CPU. A shame ethomaz, as you want to sell us yourself as a hardware specialist and then you come up with this.

 

Either way, there is clearly data missing to explain numbers to us further.

This benchmark gives the performance for "5 milliseconds", in other words it is peak CPU performance without system overhead, and it is in line with the XB1's faster CPU clock speed.  The reason XB1 came short overall is because of system overhead (for Kinect voice recognition, multiple OS, snap, etc.).

 

What? So you say the system overhead fluctuates, let's say, every ten ms? That's most likely not true and I would be glad to be told otherwise with data.



walsufnir said:
Captain_Tom said:
walsufnir said:
I always think these benchmarks funny. Remember the benchmark from last year that showed us PS4 CPU was faster than Xbox One? Now this shows the exact opposite.
Without any further data what was really tested (given these slides provide way more information than the graph last year) we can't infer *anything*, in my opinion.


The PS4 CPU won in that benchmark because they didn't use the X1's ESRAM, and the PS4 main memory is over twice as fast (Which made up for the X1's clock speed advantage).


Why wouldn't they? And where do they say they didn't? Any detail on the test for Xbone is missing. They are going from DX11 to PS4 optimization and PS4 details talk.


I wouldn't because everyone knows that the ESRAM is ALWAYS going to be used by the GPU and not the CPU.



walsufnir said:

Why wouldn't they? And where do they say they didn't? Any detail on the test for Xbone is missing. They are going from DX11 to PS4 optimization and PS4 details talk.

I guess it is why you need the eSRAM for framebuffer... most devs will use eSRAM for that and it is already not enough for big 1080 framebuffer with post-processing stufs.

They don't have space on eSRAM to put GPU compute.

No dev will sacrifice framebuffer to use the eSRAM for other thing.



Gotta love the cell.

Not surprising really. As usual exclusives will be the ones to show the gap.



e=mc^2

Gaming on: PS4 Pro, Switch, SNES Mini, Wii U, PC (i5-7400, GTX 1060)