Is the RSX even based on the 7800?

[BenQ]

According those these specs.....

The 7800 GTX can do 34.4 GLFOPS ( Vertex Shaders ) and 278.6 GFLOPS ( Pixel Shaders ) for a total of 313 GFLOPS.

Or is that the total? Is there anything in the 7800 GTX that would be performing FLOPS other than the pixel and vertex shaders?

And then I compared that number with the RSX, which is reported to be able to manage approx 1,700 GFLOPS ( 1.7 TFLOPS - the Cell managing 200 to 300 GFLOPS ).

I also understand that the RSX operates at a higher clock frequency than the GTX which would account for some of these FLOPS. If the 7800 GTX were running at 600mhz, the total FLOPS would be 437 GFLOPS ( 48 GFLOPS for vertex Shaders and 389 GFLOPS ).

So the RSX can manage well over 1 TFLOPS more than a 7800 GTX ( even if that GTX were running at 600 mhz ).

Is my math wrong, because if the RSX can manage MORE than an extra TFLOP over and above the 7800 GTX, wouldn't than mean that the RSX was way Way WAY more powerful?

 

[BenQ]

Oh, I got the 7800 GTX specs fro here - http://www.pcper.com/article.php?aid=150&type=expert&pid=3

 

[jvd]

And then I compared that number with the RSX, which is reported to be able to manage approx 1,700 GFLOPS ( 1.7 TFLOPS - the Cell managing 200 to 300 GFLOPS

The tflop numbers for the rsx are total bullshit . I'm sorry but they used that so they can claim 2tflops trumping ms's 1tflop.

Both numbes are bull .

I def believe the rsx is based on the g70 tech. IT could be a g71 modified or a g75 or whatever , but i believe it shares the same tech as the 7800 series .

What really makes me believe this is when deanoc mentioned using int 8 hdr instead of fp 16 hdr . Thats not really what did it but when he mentioned that that they could use msaa with int 8 and so it was a big win .

See the nv40 and the g70 can't do hdr floating point with msaa and it seems the rsx shares that with it . SO i highly doubt its a new tech as I"m sure in thier next generation nvidia would want to support that seeing how ati's new cards do

 

[ector]

nVidia probably count every single color or texcoord interpolation as a multiply-add

 

[BenQ]

The tflop numbers for the rsx are total bullshit . I'm sorry but they used that so they can claim 2tflops trumping ms's 1tflop.

Both numbes are bull .

I def believe the rsx is based on the g70 tech. IT could be a g71 modified or a g75 or whatever , but i believe it shares the same tech as the 7800 series .

What really makes me believe this is when deanoc mentioned using int 8 hdr instead of fp 16 hdr . Thats not really what did it but when he mentioned that that they could use msaa with int 8 and so it was a big win .

See the nv40 and the g70 can't do hdr floating point with msaa and it seems the rsx shares that with it . SO i highly doubt its a new tech as I"m sure in thier next generation nvidia would want to support that seeing how ati's new cards do

Hmmmm. I find it hard to believe that Sony would lie THAT much about the RSX. They are claiming essentially 5X's more FLOPS for the RSX than the 7800 GTX.

But would it really be a lie? Is there not some way you could calculate the RSX to approx 1.7 TFLOP with some funny math that might not be "technically" lying?

 

[flick556]

The terflop numbers include fixed function stuff.

Maybe rsx and 7800 can do HDR + AA as fallows:

Render one frame with HDR, show on screen

Render next frame with AA, blend wih previous frame and show on screen

repeat

or

Render one frame with HDR

Render same frame with AA

Blend together, and show on screen

I think the first method will have less of a performance hit and may show good results provided things don't change to quickly from one frame to the next and a high frame rate is maintained.

It would be unrealistic for pc devs to do this since it will only benefit one particular card but in a closed box design it does not seem too bad.

 

[jvd]

You'd have to draw 2 frames for everyone u display and the rendering time will be vastly diffrent for both.

Aside from that the final image will look fugly imo opinon , not to mention you'd prob get the same problem that ati has with temporal fsaa

 

[flick556]

You'd have to draw 2 frames for everyone u display and the rendering time will be vastly diffrent for both.

Aside from that the final image will look fugly imo opinon , not to mention you'd prob get the same problem that ati has with temporal fsaa

I have heard lots of good things about TAA, only down side is it falls apart when the framerate is low.

 

[dukmahsik]

we know both ati and nvidia rate their flops differently, aren't historically, ati cards always had lower flop ratings? but put out higher iq at higher res

 

[BenQ]

I just don't understand how you can fudge FLOPS numbers. Are not all FLOPS "created equal"?

Or I guess I could understand a slight variance. But with the 7800GTX rated at less than a half a TFLOP and the RSX rated at awhopping 1.7 TFLOPS...... how can this be accounted for?

Maybe I'm just being a noob but I saw those numbers and I thought "RSX = way more powerful.'

 

[Arun]

Are not all FLOPS "created equal"?

NO!
Hppe that helped. And the 512MB GTX is also in the 1.7TFlops region; you counted the "programmable" flops for the GTX only. It has the same number of special-purpose flops as the RSX, which you have to add to the programmable number. Although, TBH, only counting the programmable flops would be even fairer, in which case it is also equal. Or at least, that's what the E3 info said. It could be outdated by now.


Uttar

 

[BenQ]

NO!
Hppe that helped. And the 512MB GTX is also in the 1.7TFlops region; you counted the "programmable" flops for the GTX only. It has the same number of special-purpose flops as the RSX, which you have to add to the programmable number. Although, TBH, only counting the programmable flops would be even fairer, in which case it is also equal. Or at least, that's what the E3 info said. It could be outdated by now.


Uttar

So, both the RSX and the 7800 GTX 512, have "roughly" 400 Programmable GFLOPS, and then over 1 TFLOP of "Special Purpose" FLOPS.

OK, ya got me. What's the difference between programmable and special purpose FLOPS?

 

[Gholbine]

So, both the RSX and the 7800 GTX 512, have "roughly" 400 Programmable GFLOPS, and then over 1 TFLOP of "Special Purpose" FLOPS.

OK, ya got me. What's the difference between programmable and special purpose FLOPS?

Programmable FLOPs can do whatever the coder wishes, but fixed-function FLOPs are completely hardware based, I believe. They're there for one purpose and one purpose only, HDR for example. They can do HDR and nothing else.

 

[Guilty Bystander]

So, both the RSX and the 7800 GTX 512, have "roughly" 400 Programmable GFLOPS, and then over 1 TFLOP of "Special Purpose" FLOPS

That's not true.
The GTX does:
5FMACs x 2Flops/FMAC x 56ALU's x 550MHz : 1000 = 308GFlop/s of programmable GFLOP/s.

 

[jvd]

I have heard lots of good things about TAA, only down side is it falls apart when the framerate is low.

Its great but u have to have framerates above your refresh rate . If frames dip for one second things start to tear in the back .

Along with that anything lower than 75hz and you see alot of artifacts in the image .

75hz or fps is the lowest i would use it at . I personaly perfer a 90hz refresh rate and 90fps steady to use it .

Of course when its used right its beautifull . However your still using two images very similar to each other . In his example u will be using to vastly diffrent images

 

[Bill]

That's not true.
The GTX does:
5FMACs x 2Flops/FMAC x 56ALU's x 550MHz : 1000 = 308GFlop/s of programmable GFLOP/s.

Is this counting the mini-ALU's though?

I think the reason Nvidia has been so strong lately is the mini-alu's.

It increased their power like 20% on a per clock, per pipe basis from NV40. It's sort of been the secret to their success.

 

[Eleazar]

FLOPS are arbitrary and do not reflect real world performance. Anyone who takes FLOPS seriously is out of their mind. Not only that, they do not even specify how they got that number for the FLOPS. So it could be something as simple as adding two numbers together. FLOPS carry about as much weight as saying all 300 horsepower cars are faster than all 200 horsepower cars. In fact, FLOPS carry even less weight than this statement.

 

[!eVo!-X Ant UK]

Should'nt this be in the console tech section??

 

[weaksauce]

FLOPS are arbitrary and do not reflect real world performance. Anyone who takes FLOPS seriously is out of their mind. Not only that, they do not even specify how they got that number for the FLOPS. So it could be something as simple as adding two numbers together. FLOPS carry about as much weight as saying all 300 horsepower cars are faster than all 200 horsepower cars. In fact, FLOPS carry even less weight than this statement.

Uhm well a 300 horsepowered car is more powerful than the 200, even if it doesn't come up in the same top velocity.

But I don't know, it's more powerful in the way that it has higher clocks and maybe more pipes, but it has a lower memory bandwidth.
Also, you can't really compare it to a pc card. Even if they are equal it will perform better in a console, eh..

 

[Alpha_Spartan]

How Sony got 2TFLOP. It's extremely complex code, so you might need to read it twice:

#include stdMSPRbullshit
#include stdSYPRbullshit

main(){

int x360Gflops = MSPR();

string e3Presentation = pS3Specs( x360Gflops );
print( e3Presentation );

return 0;
}

/* basically returns some PR bullshit to feed to the press and fan drones
/  this should be followed by a mix of realtime graphics and prerendered shit to give
/  the illusion that the PS3 is orders of magnitude more powerful than the Xbox 360
*/
string pS3Specs( int x ){

int y = 2 * x;
return SNYPR(y);

}