How is this possible???

A

Atlantian

n00b
Joined
Nov 21, 2006
Messages
55
I've been looking to upgrade from my Nvidia 7950 GT Superclocked for awhile now. The one card I keep coming back to this card:
http://www.newegg.com/Product/Product.aspx?Item=N82E16814103044

But every online comparsion has it getting just knocked out by this card:
http://www.newegg.com/Product/Product.aspx?Item=N82E16814130093

Now the first card is the ATI 2900XT with 1gig of DDR4 ram, second is NVIDA 8800 ULTRA with 768 mb of DDR3 ram.

Going further the ATI card has 320 stream processors compared to NVIDIA's 128.

Then the Core clock on ATI is 825 as to NVIDIA's 655.

NVIDIA only beats out the ATI on memory clock by 150mhz, 2250mhz to 2100mhz.

Can someone please stop the insanity and tell me how the NVIDIA Ultra beats out this ATI card.

Thanks for the help
 
Because clock speeds dont realy mean much when compairing different cards.
 
I believe he's also confused with with 320SP vs the 128SP.

I'm limited with what I can find here at the office, but there are good reviews out to explain the difference between the two. Basically, the stream processors are not the same between the two cards.. call is good marketing on AMDs side.
 
It's really fairly simple.

A) The 2900's SPs aren't equal to the 8800's SPs. R600's (2900) SPs are vector -- specifically, they're VEC5 -- while G80's (8800) are scalar. G80's computational architecture is more efficient because the scheduler is more granular, and issuing isn't crippled by architectural constraints, so G80 can scratch closer to peak shading performance more often than R600. R600 can outclass G80 in raw shading performance, but it's outclassed by G80 in other areas.
B) The Ultra has a much higher shader clock than does the XT. G80's SPs are decoupled from the rest of the pipeline (TMUs and ROPs run at the advertised core clock speed, or 612 MHz in the case of the Ultra). R600's SPs run at the core clock speed.
C) G80 has dedicated AA logic, while R600 does not (R600 uses SPs to perform AA resolve always, while G80 only uses shader resolve in rare cases). This severely damages R600's performance with multisampling, which is that thing you enable when you buy a $400+ card :)
D) R6xx is a newer architecture that requires a great emphasis on shader compiler optimization. G80 is much less reliant on this kind of attention -- it performs well out of the box.
 
Atlantian said:
Can someone please stop the insanity and tell me how the NVIDIA Ultra beats out this ATI card.
Click to expand...

You are comparing sterile numbers that are meaningless without a deeper understanding of what they refer to. An Nvidia "stream processor" is not the same as an ATI "stream processor". The structure, efficiency and capabilities of each are different. Also, Nvidia's "stream processors" run at a much higher clock than ATI's - up to 1500Mhz on the Ultras. You're also comparing core clocks without knowing what exactly is running at those clocks! For example, on a clock-for-clock basis the Ultra's texturing engine is a lot more powerful than the 2900XT's.

To be honest, you really should do some research and then come back with questions if you need further clarification.
 
As for the memory, the ATi card has more raw bandwidth than the NV, even though the numbers on the ATi are smaller. That is because the ATi has a 512-bit memory bus, where the NV has a 384-bit bus. The wider bus allows for lower clocked RAM speed, while still generating more bandwidth.

It is also a common misconception that having more RAM will make the card faster, and that isn't the case under most conditions. As an example, you may see tests where the GTS 320MB card takes a hard nose dive in performance at higher settings where the 640MB card does not. That can happen when you don't have enough memory on the card to store textures, and they have to be retrieved from system RAM. However, if the 640MB card has enough onboard RAM, having 1GB won't make a difference.
 
Atlantian said:
Can someone please stop the insanity and tell me how the NVIDIA Ultra beats out this ATI card.Thanks for the help
Click to expand...

The best part is, nVidia beat ATI to the market by 6 months.

Un4given said:
As for the memory, the ATi card has more raw bandwidth than the NV, even though the numbers on the ATi are smaller. That is because the ATi has a 512-bit memory bus, where the NV has a 384-bit bus. The wider bus allows for lower clocked RAM speed, while still generating more bandwidth.
Click to expand...

Too bad nothing needs that bandwidth.
 
phide said:
It's really fairly simple.

A) The 2900's SPs aren't equal to the 8800's SPs. R600's (2900) SPs are vector -- specifically, they're VEC5 -- while G80's (8800) are scalar. G80's computational architecture is more efficient because the scheduler is more granular, and issuing isn't crippled by architectural constraints, so G80 can scratch closer to peak shading performance more often than R600. R600 can outclass G80 in raw shading performance, but it's outclassed by G80 in other areas.
B) The Ultra has a much higher shader clock than does the XT. G80's SPs are decoupled from the rest of the pipeline (TMUs and ROPs run at the advertised core clock speed, or 612 MHz in the case of the Ultra). R600's SPs run at the core clock speed.
C) G80 has dedicated AA logic, while R600 does not (R600 uses SPs to perform AA resolve always, while G80 only uses shader resolve in rare cases). This severely damages R600's performance with multisampling, which is that thing you enable when you buy a $400+ card :)
D) R6xx is a newer architecture that requires a great emphasis on shader compiler optimization. G80 is much less reliant on this kind of attention -- it performs well out of the box.
Click to expand...

Also to add, the G80 has more texture units and ROPs, and the stream processors on the Ultra run at 1.5 GHz, while they are coupled on the R600 running at the same speed as the core, 825MHz. The R600 requires an extreme amount of precision on the part of the scheduler, the G80 isn't so reliant on that for optimum performance.

So to sum up for the original poster, very different architectures that you cannot compare apples-to-apples.
 
like everyone else said.

In the end, clocks between nvidia and ati cant be compared. And the amount of ram just dont matter all that much.
Few years back, the more ram, the better, but thats not the case here.
These days, the only time the extra ram comes into play is at very extremely high resolutions or with excessively high textures.
We see this when we compare the 8800GTS 320mb to the 8800GTS 640mb...The 320mb is just as fast, sometimes faster, except for a few rare situations.

The 1GB on the 2900xt is more of a marketing ploy to sucker people in with big numbers.
 
Porsche 911Turbo - 6 cylinders, 3.6Liters -> 480 hp, 505lb.-ft.
Mustang GT - 8 cylinders, 4.6Liters -> 300 hp, 320 lb.-ft.

Oh yeah small detail but
Mustang -> $35,349. 0-60mph ???? But I dont think it beats the 911Turbo.
Porsche -> $170,700. 0-60mph 3.7secs

So engineering\design\architecture has something to do with it.
 
SpeedyVV said:
Porsche 911Turbo - 6 cylinders, 3.6Liters -> 480 hp, 505lb.-ft.
Mustang GT - 8 cylinders, 4.6Liters -> 300 hp, 320 lb.-ft.

Oh yeah small detail but
Mustang -> $35,349. 0-60mph ???? But I dont think it beats the 911Turbo.
Porsche -> $170,700. 0-60mph 3.7secs

So engineering\design\architecture has something to do with it.
Click to expand...

I call BS, your a liar. Years of design an engineering dont pay off, thats an old wives tale, stop posting FUD.

/sarc:D

For the record, i cant stand mustangs. Theyre nice, but everyone has on. 911 Turbo = sex, although the new Nissan GTR is going to give it a good run for its money. Only 80k and faster.
 
nissanztt90 said:
For the record, i cant stand mustangs. Theyre nice, but everyone has on. 911 Turbo = sex, although the new Nissan GTR is going to give it a good run for its money. Only 80k and faster.
Click to expand...

Sure the new GTR is gonna be fast but it's ugly as sin. Whereas the 911 looks as amazing as always.

I would get back on topic but I can't really say anything that hasn't been mentioned already. I seem to remember reading somewhere (maybe MaximumPC) that ATI didn't build the 2900 for the purpose of crushing the 8800 in gaming performance but for all I know that could just be ATI's PR spin on it still catching a beat down from the 8800.
 
Retne said:
Sure the new GTR is gonna be fast but it's ugly as sin. Whereas the 911 looks as amazing as always.

I would get back on topic but I can't really say anything that hasn't been mentioned already. I seem to remember reading somewhere (maybe MaximumPC) that ATI didn't build the 2900 for the purpose of crushing the 8800 in gaming performance but for all I know that could just be ATI's PR spin on it still catching a beat down from the 8800.
Click to expand...

same thing they said when the X1900 came out right? that it wasn't made to compete with the 7900GTX.....it's simple marketing CYA
 
You must log in or register to reply here.
Back
Top