More pipelines, or faster speed?

Toggle sidebar Toggle sidebar
B

b0ng0

Distinguished
Jun 24, 2005
9
0
18,510
Hi I was wondering (for playing games), is it better to have a card like the 6600GT which has a faster clock speed than than the 6800, or is it better to have more pixel pipelines (the 6800 has more than the 6600GT)?

Thanks
 
"More pipelines, or faster speed?"

both, for now, that is...till ROPspeak becomes fashionable.


"was wondering (for playing games), is it better to have a card like the 6600GT which has a faster clock speed than than the 6800, or is it better to have more pixel pipelines"


6800 definately.
 
I was asking, mainly because I saw that the 6600GT has a clock speed of 500MHz and the 6800 only has 300MHz (or thereabouts).
 
It depends. I have seen benchmarks for both cards in the same review, and they are pretty much even. That being said, thats at resolutions up to 1024*768. After that, the 6800 will pull away given the more pipelines. Also, if you like AA and AF, the 6800 will win.

My System:
<A HREF="http://amdgamingrig.dyndns.org" target="_new">http://amdgamingrig.dyndns.org</A>
 
i was gonna ask same question about pipelines vs speed, but i was going to compare:

GPU Geforce 7800GTX

Core clock 450MHz
PixelPipelines 24
Memory Clock 1200MHz
Memory Size 256MB
Memory Interface 256-bit
Memory Type GDDR3
$545.00

vs

GPU Radeon X850XT PE

Core clock 540MHz
PixelPipelines 16
Memory Clock 1180MHz
Memory Size 256MB
Memory Interface 256-bit
Memory Type GDDR3
$479.00
 
it hurts my head just trying to figure out how to buy Graphics card, by Brand, Stability, Speed, Pipelines, and Price. ATI or NVIDIA Chip? is the 200 bucks worth it for the gain over another? i am just flooded with different opinions and what brand good what brand sux, speed dont matter or pixle pipelines dont matter much, this one stable this one not, this one hot this one noisy, overclock this one or this one already overlocked. its like trying to keep a database of opinions in your head and try to filter through. there needs to be search engine with a list of all vedio cards with all the attibutes and reviews and price that can bring up the best card with the price you can afford with best reviews and attributes. custom attributes can include like max overlocked speed and between 1 and 10 in stability and latest best driver. man that would be excelent way for newbs like me to find the card i need, but thats just a dream i guess cause of bias and bs and a storm of different opinions and differences without admitting the true best card that is in existence for the price. its like finding a needle in the heystack. what to believe?
 
Either of those cards will kick as(s)! Just whichever is in your budjet.


<pre><font color=red>°¤o,¸¸¸,o¤°`°¤o \\// o¤°`°¤o,¸¸¸,o¤°
And the sign says "You got to have a membership card to get inside" Huh
So I got me a pen and paper And I made up my own little sign</pre><p></font color=red>
 
The problem with that is that everyone's tastes and budgets are different.

But some points to help you sort through all the information:
1) At the begining, ignore vendors and stick to the chips. Performace numbers are essentially the same across all vendors. So, you won't get an instance where one vendor's 6800 puts out the same performance as another's 6800 Ultra.
2) Determine what games you want to play, with what options (resoution, detail settings, etc.), and how 'smooth' a a play experience you need. So for example, if the most recent game that I'm planning on playing is D2, there's no point in me even bothering to consider a 7800GTX. Conversely, if I'm planning on playing Doom 3 at 1600x1200, there isn't much point in looking at the ATI 9x00 or nVidia 5x00 series.

Just as a point of reference; under 20 fps is considered 'unplayable', 30+ is considered playable, but you could hit points of choppiness, and anything above 60 fps won't be truely noticable (people will argue with this, but that's only when you're at the extreme level).

3) Determine your budget; this is the biggest determining factor. Two price points to have: Your prefered budget, and you absolute limit.

Also, as part of your budget, you should have a general idea about how long you want the card to last and what you're time-fram for upgrades is. Next upgrade going to happen in 6 months or 2 years? The longer it's got to last, the higher up you may want to go now (assuming you want to play future games).

4) Read reviews. Now that you know the games you want to play, and the budget you have, start to hit up reviews of cards playing your games and see how they do. Build up a list of about 5 - 10 cards that you are considering that meet your requirements.

5) Price your cards. You are still ignoring vendors here. Just find a middle-ish price for each chip type onw your list, and expect about a $30 +/- variation on that price between vendors.

6) Pick your chipset. Take your budget, pricelist, and performance numbers and pick the GPU that's best for you

7) Now that you've figured out the GPU you want, go back and start pricing that GPU among each of the vendors. Who's got the cheapest card? What software comes with each, and which of that software do you want/need? What about pereferals (i.e. just found out that the XFX 7800 comes with a USB game pad 😀 T-shirts are also popular)? What about warranties?

Once you're done that, you should have the right card for you.
 
same goes for everything else i suppose but yep i hear you...rule of thumb is to first and foremost cover all bases when doing your own research/budgeting...reviews/opinions are always clouded with bias/perceptions and can only go so far as to offer you some grounding between comparos, then make the necessary concessions/provisions and just let the product speak for itself.
 
Two things make the 6800 out-perform the faster clocked 6600GT. First is the 12-pipes vs. 8-pipes, but second is the 6800 has a 256-bit memory interface, while the 6600GT is 128-bit. The reason it isn't much faster, is the clock speed difference.

<A HREF="http://service.futuremark.com/compare?3dm05=658042" target="_new">3DMark05</A> <A HREF="http://service.futuremark.com/compare?2k3=3781954" target="_new">3DMark03</A>
 
You can measure relative performance in Cleevemarks.

For the card with fewer pipelines, calculate it's benchmark thusly:

6600GT: 8 pipes x 500 Mhz = 4000 Cleevemarks

Now for the second card. Because cards with a wider memory interface are inherently more efficient, give the card with the larger interface a bonus. Do the same calculation to start, then Divide the greater interface by the lesser interface. Divide that number by ten, add one, and multiply the result with the first result:

6800: 12 pipes x 300 Mhz = 3600
256 bits / 128 bits = 2
2/10 = 0.2 +1 = 1.2
1.2 x 3600 = 4320 Cleevemarks

You see?

Geforce 6600GT:
4000 Cleevemarks

Geforce 6800:
4320 Cleevemarks

The 6800 is clearly superior, by approximately 8%.

Later, I will show you how to calculate the color of the pants worn by the videocard's engineers by counting the number of transistors on the GPU.

________________
<b>Geforce <font color=red>6800 Ultra</b></font color=red>
<b>AthlonXP <font color=red>~3300+</b></font color=red> <i>(Barton 2500+ o/c 412 FSB @ 2266 Mhz)</i>
<b>3dMark05: <font color=red>5,275</b>
 
More pipes = more information done per cycle. Clock speeds dont mean crap these days.

<i><font color=red>Only an overclocker can make a computer into a convectional oven.</i></font color=red>
 
LOL. You better OC your card because stock mine out Cleevemarks yours:

16 * 425 = 6800
6800 * 1.2 = 8160 Cleevemarks!

Now if I could keep those blasted X850XTpe owners like Eden away, I'd be the Cleevemark Champion. (LOL, close call on the edit, as I started to type I'd be on top of Cleeves ORB 😱 ).

<A HREF="http://service.futuremark.com/compare?3dm05=658042" target="_new">3DMark05</A> <A HREF="http://service.futuremark.com/compare?2k3=3781954" target="_new">3DMark03</A>
 
ROFL!
I tried to guess the color of the pants of the engeneer, ended up with Pink, think I skipped a few transistor, else its disturbing!

Asus P4P800DX, P4C 2.6ghz@3.25ghz, 2X512 OCZ PC4000 3-4-4-8, MSI 6800Ultra stock, 2X30gig Raid0
 
I disagree, addiarmadar... an 8-pipe card at 400 Mhz is doing the same amount of pipeline work per cycle as a 16 pipe card at 200 Mhz.

Clockspeeds are still quite important...

________________
<b>Geforce <font color=red>6800 Ultra</b></font color=red>
<b>AthlonXP <font color=red>~3300+</b></font color=red> <i>(Barton 2500+ o/c 412 FSB @ 2266 Mhz)</i>
<b>3dMark05: <font color=red>5,275</b>
 
Ok could you just review what you just posted and and say that again...

To say that 16 pipe at 200mhz does the same as 8 pipe at 400mhz just proves my point. It means that the 16 pipe option does the same thing with less work.

What would that same 16 pipe card do if its clock was at 400mhz, just like the 8 pipe option? One must factor the variables, but pipeline will have the biggest influence if the differential is big enough.

<i><font color=red>Only an overclocker can make a computer into a convectional oven.</i></font color=red>
 
Well, your point was that "clockspeeds don't matter"... so it doesn't jive with that.

An 8-pipe card can do more work than a 16-pipe card if clockspeeds are sufficiently higher. Therefore, clockspeed DOES matter... does it not? If clockspeed didn't matter, and only the number of pipelines did, then regardless of clockspeed a 16-pipe card would ALWAYS do more work per clock than an 8-pipe card.

But that isn't the case, if clockspeeds are sufficiently different... I mean, which is more important? Double the pipes, or double the clockspeed? The result is identical, either is equally important as far as work per clock...

________________
<b>Geforce <font color=red>6800 Ultra</b></font color=red>
<b>AthlonXP <font color=red>~3300+</b></font color=red> <i>(Barton 2500+ o/c 412 FSB @ 2266 Mhz)</i>
<b>3dMark05: <font color=red>5,275</b>
 
Youve heard my argument, Ive heard yours. It up to the poeple on who to believe.

Would you like to compare the the 9600 to the 9800 or the 6600 to the 6800? The lower ends have a much higher core clock.....

<i><font color=red>Only an overclocker can make a computer into a convectional oven.</i></font color=red>
 
Well graphic task being heavily parrallel, its certain the more pipeline will bring more than more mhz!

I dont see any "limits" as how many pipeline you can put on a card, smalle manufacturing process should allow you too cramp more pipes almost all the time. I wonder at what poitn you will have enough of them...
As far as clockspeed is concerned its important but we saw Intel hit a brick wall there and in the graphic industry, that hasnt moved so fast either...

I think both of you guys are right!

Asus P4P800DX, P4C 2.6ghz@3.25ghz, 2X512 OCZ PC4000 3-4-4-8, MSI 6800Ultra stock, 2X30gig Raid0
 
Of course # of pipes is important in todays cards, I agree with you. But if clock speeds meant nothing, then a X800XL would = a X850XTpe. Both 16-pipe, both 256-bit mem bus, ones clocked higher and therefore performs better.

Or for a good example, go back. Why do the 9600XT and 9500 pro perform close? Both have 128-bit memory. The 4 pipe 9600XT is clocked much higher to make up for the 4 less pipes.




<A HREF="http://service.futuremark.com/compare?3dm05=658042" target="_new">3DMark05</A> <A HREF="http://service.futuremark.com/compare?2k3=3781954" target="_new">3DMark03</A>
 
Yup its like saying that now, with dual core, clock speed is not important at all, it does remain a factor, but the more parrallel the architechture is, the less you depend on clockspeed to bring performance. That said if its too underclocked, it wont be enough!

Asus P4P800DX, P4C 2.6ghz@3.25ghz, 2X512 OCZ PC4000 3-4-4-8, MSI 6800Ultra stock, 2X30gig Raid0
 
Would you like to compare the the 9600 to the 9800 or the 6600 to the 6800? The lower ends have a much higher core clock.....
Click to expand...
I would most certainly compare a 600 Mhz 9600 (4-pipe) with a 300 Mhz 9800 (8-pipe). If they both had 256 bit memory busses, the performance would be virtually identical.

In fact, this is exactly the scenario Paul pointed out, the 9600XT (4-pipe) is the equal to the 9500 PRO (8-pipe), both having a 128-bit bus, the 9600XT having about twice the clockspeed of the 9500 PRO.

If you have a 300mhz 8-pipe card you could double the clockspeed to 600 Mhz, or double the number of pipes to 16... the resulting GPU power is theroetically identical in either case.

The result is exactly the same. Therefore, they are equally important.... saying it's not as important is like saying an airplane's engine is more important than it's wingspan to make it take off. But in reality either factor is equally important in getting the plane off the ground.


________________
<b>Geforce <font color=red>6800 Ultra</b></font color=red>
<b>AthlonXP <font color=red>~3300+</b></font color=red> <i>(Barton 2500+ o/c 412 FSB @ 2266 Mhz)</i>
<b>3dMark05: <font color=red>5,275</b>
 
Would you like to compare the the 9600 to the 9800 or the 6600 to the 6800? The lower ends have a much higher core clock.....
Click to expand...
The lower ends have 1/2 the memory interface also. 128-bit vs 256-bit memory totally throws off the comparison. We can see that buy comparing the 9500 pro to 9700 pro, or the 128-bit 9800 pro to a 256-bit 9800 pro.


We can't do Cleeves 256-bit comparison, but we can do a 128-bit comparison: 128-bit 9800 vs 9600XT.

Basically with both being 128-bit memory at 600MHz (effective), we are comparing an <A HREF="http://www.digit-life.com/articles2/digest3d/0405/itogi-video-fc1-wxp-1600-agp.html" target="_new">8-pipe 380MHz core to a 4-pipe 500MHz core</A>. As expected the 9600XT loses as it doesn't have twice the core clock. Now if only someone had both cards and underclocked the 9800 core to 300 MHz and overclocked the 9600XT to a 600MHz core, then we would have a straight comparison. I wouldn't say both cards would always perform identical as there are other small factors, but I'd say in general it would be pretty even supporting what Cleeve is saying.

My feeling is (for now anyway), we have to look at the # of pipes, the core speeds, and the memory interface. It's the basics that All theoretical comparisons have to take all three into account for starters(ie, there is more to it). Now with all this talk of redefining pipes, etc. I'm sure things will change and we'll have more learning to do. Will R520 "learn" us a thing or two?


<A HREF="http://service.futuremark.com/compare?3dm05=658042" target="_new">3DMark05</A> <A HREF="http://service.futuremark.com/compare?2k3=3781954" target="_new">3DMark03</A>
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom