Help me understand something, please

[CPA]

I know all of the reviews, both editorial and customer, state that the 8800GT is better than the HD 3870 at gaming, but what I don't understand and the reviews never explain is why. Why is the 8800GT better at gaming, especially in light of the following spec comparisons:

8800 GT HD 3870
Core Clock 700mhz 775mhz
Memory Clock 2000mhz 2400mhz
Memory Size 512MB 512MB
Memory Type DDR3 DDR4
Processors 112 320

My comparison is between an EVGA 8800GT SSC (it had the fastest Core Clock) and a Diamond HD 3870. They specs were pulled from Newegg.

So, I'm not a graphics card expert, but I just don't get how the 8800GT provides better performance. Is it solely because of the drivers? If so, eventually could the 3870 blow the 8800GT out of the water if AMD gets the drivers streamlined? Or am I missing something more important?

 

[Stoneburner]

Probably lack of ROP's

 

[CPA]

Originally posted by: Stoneburner
Probably lack of ROP's

ROP?

 

[clickynext]

Plenty of possible reasons, hardware architecture is fairly complex. The difference between NVIDIA and ATI cards is fairly fundamental, I would guess. It's better to just look at the benchmarks and not the specs.

 

[requiem1]

the reason is numbers don't mean everything. real world performance does. basically, they have different architecture - so u should not view the specs equally. 1mhz on nvidia doesnt equal 1 mhz on ati.

 

[Stoneburner]

http://en.wikipedia.org/wiki/Render_Output_unit

For some reason, ATI never increased the number of ROP's since the x1800 series.

 

[Fallen Kell]

The same way that the 8800 GTX wiped the floor with the ATI high end card. The ATI "stream" processors can not perform all the same operations that an Nvidia "stream" processor can. In the original battle Nvidia vs ATI DX10, the ATI part only have 1 out of every 8 "stream" processors have a floating point math unit, which meant that all those operations could only occur on those "steam" processors. The Nvidia system had ALL its stream processors exactly the same, no specialty ones that could only perform certain functions, every single one could perform the exact same thing as the other ones. ATI is continuing this practice for some reason (costs and development time), and the problem still exists on this new rendition. They are bottle-necked for certain types of functions. All it takes is one bottle-neck to put the overall performance down the drain... They may have "faster", higher bandwidth memory, but that isn't their problem. Their problem is they can't do enough floating point operations to calculate the image.... Sure they have the data they need to do the operation, but there is no available processor unit to perform the operation, so it queues up and waits, and while those wait, the image frame itself has to wait since it isn't complete.

 

[CPA]

Originally posted by: Fallen Kell
The same way that the 8800 GTX wiped the floor with the ATI high end card. The ATI "stream" processors can not perform all the same operations that an Nvidia "stream" processor can. In the original battle Nvidia vs ATI DX10, the ATI part only have 1 out of every 8 "stream" processors have a floating point math unit, which meant that all those operations could only occur on those "steam" processors. The Nvidia system had ALL its stream processors exactly the same, no specialty ones that could only perform certain functions, every single one could perform the exact same thing as the other ones. ATI is continuing this practice for some reason (costs and development time), and the problem still exists on this new rendition. They are bottle-necked for certain types of functions. All it takes is one bottle-neck to put the overall performance down the drain... They may have "faster", higher bandwidth memory, but that isn't their problem. Their problem is they can't do enough floating point operations to calculate the image.... Sure they have the data they need to do the operation, but there is no available processor unit to perform the operation, so it queues up and waits, and while those wait, the image frame itself has to wait since it isn't complete.

Interesting. That helps. Thanks.