journalism + tech *spin*

[shredenvain]

I just love how some game sites are taking this Pdf and running stories saying that it is a GPU performance benchmark. They somehow are taking it as proof that one console's gpu is 2x as fast as the other. This presentation is in no way representative of how an actual game will perform on the consoles. Presentations like this are great for those interested game development, but I wish there were some way of keeping them out of the hands of click bait flamewar articles on the internet

 

[Shortbread]

I just love how some game sites are taking this Pdf and running stories saying that it is a GPU performance benchmark. They somehow are taking it as proof that one console's gpu is 2x as fast as the other. This presentation is in no way representative of how an actual game will perform on the consoles. Presentations like this are great for those interested game development, but I wish there were some way of keeping them out of the hands of click bait flamewar articles on the internet

Why do you even care? If you're happy with the XB1 as you claim, then this shouldn't bother you.

 

[iroboto]

Why do you even care? If you're happy with the XB1 as you claim, then this shouldn't bother you.

While this is true, that isn't the right way to look at it. Misinformation is misinformation and some people get really aggravated by it and some people are able to just ignore it.

 

[Deleted member 11852]

While this is true, that isn't the right way to look at it. Misinformation is misinformation and some people get really aggravated by it and some people are able to just ignore it.

If a person can't tolerate misinformation then they want to be giving most of the internet a miss

 

[iroboto]

If a person can't tolerate misinformation then they want to be giving most of the internet a miss

Lol true.

 

[TheWretched]

While this is true, that isn't the right way to look at it. Misinformation is misinformation and some people get really aggravated by it and some people are able to just ignore it.

I am at odds that this is misinformation. It's information. People interpret it one way or another. Not saying it's good or bad, but keeping it away from people will do more harm than good, either way.

And... to a degree, it IS a gpu benchmark (what shredenvain doesn't want it to be). It's showing the GPU used to run physics on the GPU to be displayed. Doesn't really matter if it's dancers or... for example particles, like they used the GPU for in Resogun. Having half the particles taken away (or halving the framerate) will make the graphics suffer. Even if it isn't strictly "graphics" that is suffering.

 

[iroboto]

GDC paper on compute-based Cloth Physics including CPU performance

I am at odds that this is misinformation. It's information. People interpret it one way or another. Not saying it's good or bad, but keeping it away from people will do more harm than good, either way.



And... to a degree, it IS a gpu benchmark (what shredenvain doesn't want it to be). It's showing the GPU used to run physics on the GPU to be displayed. Doesn't really matter if it's dancers or... for example particles, like they used the GPU for in Resogun. Having half the particles taken away (or halving the framerate) will make the graphics suffer. Even if it isn't strictly "graphics" that is suffering.

That's true, but you can't use a single case (with no knowledge of how it's written) and shoe horn it into evidence of a very general statement. It's like saying global temperatures are down this year compared to last so global warming doesn't exist. (What actually annoys him is the misuse of the information, not the data itself)

 

[TheWretched]

I surely understand that. But it's just a natural comparison to make.

One GPU has X amount of CUs, the other has Y. X is bigger than Y, thus, X can calculate stuff faster. I don't really get the problem.

In his original post, he's saying:

They somehow are taking it as proof that one console's gpu is 2x as fast as the other. This presentation is in no way representative of how an actual game will perform on the consoles.

Benchmarks are never representative of games. Doesn't matter if it's compute or "actual" graphics. It's comparing "scene a" between the two machines. Nothing more, nothing less. If there's a "dancers" game, it'd run about twice as fast on one console compared to another. It doesn't have to be dancers, could be something completely different. Some other compute code could lead to even faster PS4 times (say, very heavy main RAM usage and ESRAM doesn't help in that case... hypothetically). Does that make the game it theoretically uses "invalid" as well?

Of course it's not a full on review of the consoles. It's one case, a vs. b. And it doesn't show every nook and cranny each console excells at. Not a single benchmark ever does. But it doesn't invalidate the original findings, either. It's simply twice as many dancers.

 

[iroboto]

Yup you are correct. If we just look at the data we are indeed seeing twice the number of dancers.

He'd sleep better at night just not caring about what or how people think as shortbread said.

 

[function]

As a Dreamcast owner and fan (it's still the best damn console of all time) I have some sympathy for those who are upset at seeing disinformation and fuckwits harm a platform that you like and that you feel has something to give.

It's good to feel passionate about your passion, so long as you don't say goodbye to reason.

 

[shredenvain]

I surely understand that. But it's just a natural comparison to make.

One GPU has X amount of CUs, the other has Y. X is bigger than Y, thus, X can calculate stuff faster. I don't really get the problem.

In his original post, he's saying:



Benchmarks are never representative of games. Doesn't matter if it's compute or "actual" graphics. It's comparing "scene a" between the two machines. Nothing more, nothing less. If there's a "dancers" game, it'd run about twice as fast on one console compared to another. It doesn't have to be dancers, could be something completely different. Some other compute code could lead to even faster PS4 times (say, very heavy main RAM usage and ESRAM doesn't help in that case... hypothetically). Does that make the game it theoretically uses "invalid" as well?

Of course it's not a full on review of the consoles. It's one case, a vs. b. And it doesn't show every nook and cranny each console excells at. Not a single benchmark ever does. But it doesn't invalidate the original findings, either. It's simply twice as many dancers.

I have no problem with this gpgpu physics benchmark or it's results. In all actuality it is pretty interesting. Oh and @Shortbread it doesnt have a damn thing to do with what console I own. I own several consoles and plan on getting the ones I dont own soon.
It isnt about system insecurity nor about one system being more powerful than the other.
My problem is certain people from certain game sites take the information and use it as ammunition for an unusually hate filled console war. On top of that they take the info and spin it to fit their agenda claiming it as proof one system has a 2x real world gpu advantage.
The actual reality of the situation is no where near as dramatic. As most people with any sense or even the time to actually research the subject knows. I feel the same way about a certain site whose name should not be mentioned that trys to say the X1 is secretly filled with a hidden 2.4 to 5 tflops of gpu alu. The difference is when I google news feeds about either system you dont see many sites trying to rip the Sony machine a new A-hole with needless unfounded misinterpreted crap. I guess I am just a little sick of the flamewar.

 

[steveOrino]

I guess I am just a little sick of the flamewar.

Man, its been this way since the Genesis/Snes (6800 @7.67mhz vs Ricoh 5A22 @ 3.58mhz), + or - the easy access internet. Its never going away. People need a justification for their choices and civil discourse is always a casualty.

 

[Lager]

As a Dreamcast owner and fan (it's still the best damn console of all time) I have some sympathy for those who are upset at seeing disinformation and fuckwits harm a platform that you like and that you feel has something to give.

It's good to feel passionate about your passion, so long as you don't say goodbye to reason.

The only fuckwits who "harmed" the Dreamcast were the people working at SEGA.

 

[function]

No, it was also the paid fuckwits in the UK gaming press who reported things like the DC not being internet capable but the ps2 was out of the box, the DC not being able to play audio cds, the ps2 having twice as much memory, the ps2 being more than 20 times as powerful, etc etc

And that brings me back to the earlier point about the way this particular GDC presentation is being used. There's often no attempt to interpret information presented in articles like this, hit hungry websites pull out a couple of figures they don't understand and boom an article.

But this is very OT, so I'll cut to the chase and just leave it with: We can all be fuckwits. But its nothing to celebrate, and it's especially annoying from professional games journalists.

 

[Nesh]

The only fuckwits who "harmed" the Dreamcast were the people working at SEGA.

Some gaming publishers were "fuckwits" to be honest

The people working at SEGA during the Saturn era did fuck up Saturn. But the DC was agreat execution. Although I have no idea what SEGA was thinking when they released the DC with just 1 analog stick (and if I am not mistaken just two triggers without shoulder buttons?)

the console was flooded with AAA games and had a very very succesfull launch. It deserved to live longer but it died prematurely for reasons not necesarilly blamed on SEGA. SEGA couldnt do anything to save it.

The DC was probably the first console to have an online service and the hardware was online-ready. It also provided 50hz-60hz options for most games whereas the PS2 didnt except for a few. It also had VGA support. The PS2 never had any.

The Playstation brand was a monster and the console promised better visuals along with next gen versions of its most succesful franchises whereas the DC started out fresh. Games like MGS2 being shown on PS2 at E3 killed all expectations for DC standing up against Sony's visual and physics capabilities.

Piracy, media and lack of support by huge publishers such as EA meant the console would miss lots of important franchises. It didnt see a single Fifa or Need for Speed release.

Anyways, since I mentioned MGS2, that game was super impressive due to its crazy detail to physics and interaction. Water, rain, air, ice melting, bullet proof shields reacting realistically to bullet impacts, impressive AI etc. MGS2 even had bullet speed calculations. I have yet to see a game on the current gen consoles that have set the bar as high as MGS2 did back then.

I am not really sure if they can indeed set new standards. Do they really pack the processing power to unleash physics and interactions that make the worlds more believable? We have yet to see such a game

 

[onQ]

I'm not even sure what's going on in this thread anymore.

 

[shredenvain]

I'm not even sure what's going on in this thread anymore.

I guess you can blame me for that. I apologize if I messed up the thread.

 

[iroboto]

We're derailed.

They mentioned that having a very long shader increases GPU utilization and brings down CPU utilization. If the shader is too short, then there is imbalance between over CPU usage and not enough GPU usage.

But for something like async GPU I imagine you can't have ridiculously long shader code because eventually you want to go back to the priority items. But with shorter shader code (which is then made up of lots of CPU requests) we see there isn't much point.

While I imagine it's going to be different for each case, it does paint a picture of some of the restrictions that need to be considered for effective use cases for async requests for GPGPU processing.

Are there any examples to look at async compute much like this one does for regular computer shaders?

 

[Ethatron]

"Long" for GPU shader programs is relative, a single specific instance of a compute shader running 1 millisecond is an eternity, and could equate to say 100k instructions (including -+ streaming mem latency etc. pp.). That's an incompileable shader if explicitly written out and not looped, at least in debug (with program verification).

You have the number of dancers per ms, and the frequency, you can calculate the number of cycles per dancer (and divide by 10 to get mem lat. out, like me).

 

[NRP]

I'm not even sure what's going on in this thread anymore.

It's about fuckwits.