Stream Processing Units

Z

zer0_c00l

Active Member
OK MY 4870 HAS 800 Stream Processing Units AND My evga gtx 260 has 216. my question is will my 260 outprform my 4870 or the other way around? 512mb version.......... I havnt had a nvidia card in a while and impulse bought the 260 and was going to give the 4870 to my brother
 
Last edited:
Hmm its a tricky one. When I was buying my new card I couldnt decide between the two, although I would have gone the 1GB 4870 instead of 512MB. I believe you would find the GTX 260 would beat it but if you had a 1GB version I wouldnt know.

Seeing as you have both cards its an excellent opportunity to do some benchmarking of your own! Do it on the exact same system and run 3DMark06/Vantage at different performance settings and see which is better, also play some games and get average fps scores. It would also be interesting to do the same tests but with both cards overclocked to the max to see which is the better overclocker.
 
Awesome, post the results up on here when your done, I'm keen to see them :)
Make sure you take screenshots for proof ;)
 
well, the 48701gb puts up a small fight to the GTX260, but with AA and AF turned on, the GTX wins against any 512mb card.
 
No, the Stream Processing Units in ATI graphics chips are not the same as the Processing Cores of the nVidia chips. They both accomplish the same tasks overall but in very different ways.

Apples/Oranges here.
 
Zatharus said:
No, the Stream Processing Units in ATI graphics chips are not the same as the Processing Cores of the nVidia chips. They both accomplish the same tasks overall but in very different ways.

Apples/Oranges here.
Click to expand...

Yeah, you cannot directly compare the numbers in each card. The GTX 260 core 216 though is superior to the 4870, especially once you get into higher resoloutions and aa/af.
 
I think ATI divides the processors into groups to do different codes and Nvidias processors do all the same. I guess you could say Nvidia goes at it the brute force way.
 
Close, but not quite. nVidia's architecture is a multi-threaded approach that can operate without specific "stream" programming. ATI's approach is more or less a set of vector processing units eached grouped under "master" shader unit if you will. This is really a case of MIMD (nVidia) vs SIMD (ATI). Both shader processing architechtures have "groups" of transisters that process similar data, they just process it differently.

A very general explanation would be:

nVidia's Cores = 1 complete processing unit each
ATI's Stream Processors = 1 part of a complete processing unit (say roughly 5 SPUs = 1 core but this is not a very good comparison)
 
Sorta, ATI divide them into 5/6 pairs. Each one in the pairs handles a different code. Nvidias processors will run as many as needed and can all run the same code.

All in all, if the games were coded more for ATI they would probable be the better architecture.
 
In the Sum of FPS Benchmarks Totals the 4870 is behind the 260, though I can’t tell if it’s the 512 or 1 GB version.
Though at the higher resolutions (without AA), the 4870 does better than the 260, 1920 X 1200, 1680 X 1050, and 1280 X 1024. Also in 3DMark06, (Default). Interesting though that while the 260 scores higher overall in the Sum of FPS Benchmarks Totals, a pair in SLI scores a little lower that a pair of 4870s in crossfire.
 
Bodaggit23 said:
To be fair, you asked if there was an advantage, not for a diagram. :cool:
Click to expand...

Taking this one a little too far? Yes...

His was a question that deserved answering not nitpicking.


Advantages in each processing mode - a summary:

MIMD

- More flexible in code execution. Better at working with multiple instructions in parallel.

- More efficient and typically faster in code execution. MIMD units can work independently. SIMD units must wait for all processes to finish in many circumstances.

- Can be cheaper than SIMD processors because of the lack of a control unit.


SIMD

- SIMD is easier to program for and usually much simpler to debug.

- SIMD processors are synchronized by design. MIMD based processors may need to have synchonization procedures in place which will usually come at the cost of some overhead. This implicit synchronization in SIMD can also allow for more efficient inter-processor communications.

- SIMD processors can be fully utilized by a single program whereas MIMD processors generally require multiple programs.
 
You must log in or register to reply here.
Back
Top