More GPU Core, or more Pixel Pipelines?

I have been thinking about this question ever since my killer 6800LE at 12x1, 6vp. I was very upset when I saw most people having their LEs unlocked to the full 16 pipelines, but for some reason most of them couldn't touch my 3dMark scores. I had my Leadtek 6800LE set at the 1.4v bios mod, and topped out the card at 425/950 (12x1, 6vp). My 3dMark03 score was up around 12,300, and that is more powerful than a 6800GT (16x1, 6vp) at stock (350/1000). Other people with the 6800LE at 16 pipes weren't touching my score, since their clock speeds were average (roughly) 375/850. Is a significantly more powerful core that much more powerful than the extra 4 pipes?

Now we see two new cards come out, the X1800 (still @ 16 pipes) series cards on the high-end, and the 6800GS (12 pipes) for mid-range.

Currently, the X1800XT is competing against the monstrous 24 pipeline GTX. How is this I've wondered? I know ATI's clock is entirely different than nVidia's clock, but putting that aside we see that ATI's 16 pipe card is in the 600mhz gpu range (stock), and nVidia's 24 pipe card is in the 400mhz gpu range (stock). Alright, so the beefier core speaks up.

Remember what I said about my LE... now look at the 6800GS at 12 pipes. The 6800GS has a core speed of 425mhz (stock), and a GDDR3 memory speed of 1000mhz (stock). The 6800GT has a core speed of 350mhz, and a GDDR3 memory speed of 1000mhz (stock). Alright, we can put the memory aside since they are equal. It is said that the 6800GS is like the performance of the GT... well, this is a PERFECT comparison. Both cards are made by nVidia, so this is the most direct comparison that can be made (remember that you cannot directly compare ATI's clock for nVidia's clock, but in this case we have the same card pretty much to compare).

Alright, so I know we still have to wait for benchmark comparisons between the GS and the GT, but this goes to show that the 75mhz extra on the GPU core is almost equalivalent (or shall we say similar) to an extra 4 pipelines.

I know I'm speaking very "in general," and I'm not showing benchmarks, but from my experience between having a 6800GT and the 6800LE, I choose the LE (my GT wasn't a good clocker anyways).

Yes and the core on a review hit 525 ! which is much faster than any 6800gt or ultra core I have ever seen.

With the 6800GS nvidia now owns the midrange market for $200 what a performer, they pretty much killed the sales of the 6800GT and ultra though unless consumers are clueless.

Well, if you think about it 525mhz should be considerable. Most cards have room for 50-150mhz OC (depending on cooling). A lot of people were breaking 500mhz core on the 7800, so with nVidia's new cores coming out I guess a considerable OC is more likely.

Would you mind linking to that review, and also telling me how they cooled it (and if it was voltmodded)?

Overall I think it depends on the situation or the luck of the card getting the higher clock to make up for the lack in pipes, or just how the cards designed also. Might just scale very well with higher clocks then lower clocks which makes it come into the lead.

Now my cards nothing too special yet X1800XL @ XT speeds, but im working on it. I know my card scales very well if I oc the memory more then the core. Comparing to the 7800GTX 256meg version well it does have a lower clock on both mem and core, yet it does have a 8 pipe advantage over me, but my card has roughly a 33% faster core clock and 20% faster memory. Yet its suppositly suppose to hold rougly the same scores overall though.

Now its 2 different architectures I'm talking about but yet they score similar in tests with similar speed cpus at im guessing defualt settings.

Taking this off the orb say keeping similar percentage differences. This is what I'd come with sorta.

This 7800GTX scored 9k without CPU score, it had a 556/1350 clock
http://service.futuremark.com/compare?3dm05=1153323

Now taking into that acount lets up my card
739.5/810 Now I know already at that speed its well above 9k, and nearing 10k actually. With just 640/800 I was almost at 9k just a few points shy of it. And people running 700/800 are right around 9.6/9.7k.

Now just throw away the names of the cards. And just look at them as a 16pipe and a 24pipe card. If taking that into account you can see its faster to run higher clocks then run more pipes.

Now that you bring this up Im curious if my old X800Pro Vivo I modded to 16 pipes right away would of clocked better just leaving it as the 12 pipes. Oh well.

g0dM@n said:

Well, if you think about it 525mhz should be considerable. Most cards have room for 50-150mhz OC (depending on cooling). A lot of people were breaking 500mhz core on the 7800, so with nVidia's new cores coming out I guess a considerable OC is more likely.

Would you mind linking to that review, and also telling me how they cooled it (and if it was voltmodded)?

Click to expand...

http://www.ocforums.com/showthread.php?t=422598&highlight=6800gs

from the first post about the 6800gs, linky for review included. looks like all stock goodness 525/1230 ....
maybe was a cherry ?

deathman20 said:

Now that you bring this up Im curious if my old X800Pro Vivo I modded to 16 pipes right away would of clocked better just leaving it as the 12 pipes. Oh well.

Click to expand...

That is precisely what I'm trying to get at. Think about it, I have seen X800Pros at 12 beat X800/X850 16 pipe cards. I usually only see this when it's an unflashed card.

i.e. A card at 12 vs. the same card at 16 that was successfully unlocked. I haven't seen someone try both 12 and 16 on the same card in hand. Perhaps someone with a fully unlockable 6800LE should try the same amount of pipes as me at 12 pipes, and then OC... then see if the scores are better that way.

My locked X800 Pro is one of the best (was the best officially for a while) clockers. It took 6800GT's and X800XL's to the cleaners, but it couldn't challenge anything but the worst clocking X800XT's. In the X800 Pro's case, the 16 pipe will almost always win. A 12 pipe ATi needs to be clocked a good 100MHz higher than a 16 pipe to compete. The unlocking doesn't take off anywhere close to that when its possible.

Gautam said:

The unlocking doesn't take off anywhere close to that when its possible.

Click to expand...

What do you mean, that it doesn't clock as high when it's unlocked? I didn't understand your statement.

Unlocking the pipelines of X800 Pro's would reduce the overclockability of the core speed, yeah. The pipes didn't come for free.

Yeah, so my point is...

deathman20 said:

Now that you bring this up Im curious if my old X800Pro Vivo I modded to 16 pipes right away would of clocked better just leaving it as the 12 pipes. Oh well.

Click to expand...

Could a card get a better score in 3dmark by going for the lower pipes instead?

As in... OC to its limit at 12 pipes, and then OC to its limit at 16 pipes... is it possible to get more outa the card at the lower pipes (keeping in mind that both settings use the same cooling, same other components, and same voltages). Has anyone tried this and gotten a better score with lower pipes? I'm interested in trying, but I don't know if my X800GTO² will be a good example. I wish my 6800LE could fully unlock cuz then I could try it out on that.

g0dM@n said:

Could a card get a better score in 3dmark by going for the lower pipes instead?

Click to expand...

No chance for the X800 Pros (or GTO2's for that matter) is what I'm saying. The pipes are worth far, far too much. Might be different for the 6800LE's.

Well, I guess I'll find some benchmarks on the 6800GT (16 pipe) vs the 6800GS (12 pipe), and then we'll know how it is for nVidia, since you're telling me with ATI it doesn't really hold true.