Interesting X-Bit Results

[Gamingphreek]

While the X1800XT leads 99% of the time with any sort of eye candy features on, it is interesting that the 7800's have much higher minimum framerates in many games for some reason.

Aside from the inferior AA technique and the older lack of a 512bit interconnect, what is causing the 7800's to have such high minimum framerates but lag behind the XT in maximum framerates.

Article

-Kevin

 

[Drayvn]

Very good Observation never noticed the min framerates.

Hmmm. Dont really know. It is pretty odd to have such a combination really.

 

[Gamingphreek]

That is what i was thinking as i was going through it. I other reviewers list the XT as far superior with AA and AF enabled, however, while Nvidias max frames take a huge plummit, their minimum frames dont suffer nearly as much.

-Kevin

 

[mwmorph]

more memory bandwith means better aa/af performance(what is it? 1.5 or 1.6ghz combined with 256bit memory bandwith combined with 512bit ring bus right?)

the 7800gtx will hold better min framerates since the 24pipes will handle many more demanding effects at the same time better during heavy action so as to keep frames up..

 

[CPlusPlusGeek]

Clock speeds, in the areas that are less intensive on the pipes its all clock speed.... the x1800 XT wins that.

 

[Gamingphreek]

Well the high memory clock shouldn't give ATI much a performance advantage as it merely makes up for the lack of 8 Pixel Pipelines.

As for the Ring Bus that has nothing to do with Memory.

WHat areas would be less intensive on pipes but more intensive on clockspeed??

-Kevin

 

[meelk]

one word:
shimmer.

 

[mwmorph]

Originally posted by: Gamingphreek
Well the high memory clock shouldn't give ATI much a performance advantage as it merely makes up for the lack of 8 Pixel Pipelines.

As for the Ring Bus that has nothing to do with Memory.

WHat areas would be less intensive on pipes but more intensive on clockspeed??

-Kevin

i didnt mean clockspeed, but AA is much more dependant on Memory Bbandwith.

 

[Keysplayr]

Originally posted by: meelk
one word:
shimmer.

W....T.....F......???? Great. Another class act added to the forum.

 

[CPlusPlusGeek]

Originally posted by: meelk
one word:
shimmer.

One Response:
Barely any, I never notice it, and certainly not any worse than ATi

 

[hans030390]

I'm pretty sure the extra pipes in the 7800gt/x give it higher minimum framerates

 

[nRollo]

Originally posted by: Drayvn
Very good Observation never noticed the min framerates.

Hmmm. Dont really know. It is pretty odd to have such a combination really.

I don't know about that- it seemed to me with my X800XT PE the Doom3 fps were similar to my 6800GT, but the GT felt smoother.

Also, with the 9700Pro, when I'd push it for settings, the averages were still high, but the counter seemed to really plummet at times. (and IIRC there were articles about this)

I've always thought of ATI cards as having lower minimums and higher maximums, just assumed it was an idiosynchrasy of their design or drivers.

 

[CPlusPlusGeek]

Maybe the only reason ATi has higher average framerates is due to the maximum framerates shooting higher and bumping up the avg.?

 

[Gamingphreek]

Originally posted by: meelk
one word:
shimmer.

You can take that crap right out of my thread...thanks.

mwmorph i understand the AA is more dependant on memory bandwidth but clockspeed and pipes are directly related, and directly affect memory bandwidth. Does the added clockspeed reduce latencies more; i guess i am asking what are the advantages of a higher memory clock in contrast to a greater number of pipelines.

-Kevin

 

[v8envy]

How very interesting. Time to design benchmarks which report the area under the FPS curve, rather than maxima.

Optimizing a graphics card to provide 30000 fps when you're staring at a wall at the expense of slowing down below 30 fps when the action gets heavy (and you NEED the speed) is a lose. Great if all you play is 3dmark, but a deal killer for twitch FPS.

The more I hear about ATI's new offering the less I like it.

 

[mwmorph]

Originally posted by: Gamingphreek

Originally posted by: meelk
one word:
shimmer.

You can take that crap right out of my thread...thanks.

mwmorph i understand the AA is more dependant on memory bandwidth but clockspeed and pipes are directly related, and directly affect memory bandwidth. Does the added clockspeed reduce latencies more; i guess i am asking what are the advantages of a higher memory clock in contrast to a greater number of pipelines.

-Kevin

well memory clockspeed, memory bus width relate to memory bandwith.
i dont understand how you got pipes into the equation. the ring bus contribues to general bandwith in the gpu which can sorta therotically affect memory bandwith, but pipes and core clcokspeed are more pixel fill rate than anything right?

 

[CPlusPlusGeek]

Where are the min FPS shown? I dont see them in this link...

 

[Pabster]

Originally posted by: meelk
one word:
shimmer.

Can you say ATi Employee / Fanboy?

 

[nRollo]

Originally posted by: CPlusPlusGeek
Where are the min FPS shown? I dont see them in this link...

I don't know about this link, I only read the last page so far, but HardOCP is a good one for minimums and Firing Squad has been showing it more lately too.

To me, on cards at this level the minimums are the key factor- averages are pretty high on most games already.

 

[jiffylube1024]

Originally posted by: Gamingphreek
Well the high memory clock shouldn't give ATI much a performance advantage as it merely makes up for the lack of 8 Pixel Pipelines.

As for the Ring Bus that has nothing to do with Memory.

WHat areas would be less intensive on pipes but more intensive on clockspeed??

-Kevin

Well, the ATI card does have 8 fewer pipes, but it has the same number of shaders and runs at a higher clockspeed to compensate.

Although the argument does follow to reason that the extra pipes improve the minimum fps, while the extra memory bandwidth gives it better AA/AF performance at high resolutions.

 

[Corporate Thug]

Originally posted by: meelk
one word:
shimmer.

one word: troll

 

[Gamingphreek]

Originally posted by: mwmorph

Originally posted by: Gamingphreek

Originally posted by: meelk
one word:
shimmer.

You can take that crap right out of my thread...thanks.

mwmorph i understand the AA is more dependant on memory bandwidth but clockspeed and pipes are directly related, and directly affect memory bandwidth. Does the added clockspeed reduce latencies more; i guess i am asking what are the advantages of a higher memory clock in contrast to a greater number of pipelines.

-Kevin

well memory clockspeed, memory bus width relate to memory bandwith.
i dont understand how you got pipes into the equation. the ring bus contribues to general bandwith in the gpu which can sorta therotically affect memory bandwith, but pipes and core clcokspeed are more pixel fill rate than anything right?

Ooh i mispoke. I sort of combined Pixel Fill Rate and Memory bandwidth i think. Well they both have a 256bit memory bus, but the X1800XT has a higher memory clock. So higher memory bandwidth. The only thing is, i thought cards were becoming less and less bandwidth dependant and more and more fill rate dependant.

-Kevin

Edit: Guys stop commenting on the troll. Feeding him will only make it worse

 

[mwmorph]

Originally posted by: Gamingphreek

Originally posted by: mwmorph

Originally posted by: Gamingphreek

Originally posted by: meelk
one word:
shimmer.

You can take that crap right out of my thread...thanks.

mwmorph i understand the AA is more dependant on memory bandwidth but clockspeed and pipes are directly related, and directly affect memory bandwidth. Does the added clockspeed reduce latencies more; i guess i am asking what are the advantages of a higher memory clock in contrast to a greater number of pipelines.

-Kevin

well memory clockspeed, memory bus width relate to memory bandwith.
i dont understand how you got pipes into the equation. the ring bus contribues to general bandwith in the gpu which can sorta therotically affect memory bandwith, but pipes and core clcokspeed are more pixel fill rate than anything right?

Ooh i mispoke. I sort of combined Pixel Fill Rate and Memory bandwidth i think. Well they both have a 256bit memory bus, but the X1800XT has a higher memory clock. So higher memory bandwidth. The only thing is, i thought cards were becoming less and less bandwidth dependant and more and more fill rate dependant.

-Kevin

Edit: Guys stop commenting on the troll. Feeding him will only make it worse

I think due to recent events with optiomizations on both sides as well as architectural differences, pixel fill rate is completely worthless as a stat. think. the X800xt was clocked something like 500mhz to compete with a geforce 6800ultra at 400mhz. both with identical pixel pipes. or the X800xl at 400mhz v.s. the 6800gt at 350mhz. same # of pipes.

 

[CP5670]

Which benchmarks is this about? I could only find a couple of games in that xbit review where the minimums were reported and the numbers seemed to be either comparable or in favor of the XT.

Originally posted by: Rollo
I don't know about this link, I only read the last page so far, but HardOCP is a good one for minimums and Firing Squad has been showing it more lately too.

To me, on cards at this level the minimums are the key factor- averages are pretty high on most games already.

Same here. I don't really care about averages at all actually, only the minimums. It would be nice if more sites would show those in their articles.

 

[Pete]

Originally posted by: Gamingphreek
Well the high memory clock shouldn't give ATI much a performance advantage as it merely makes up for the lack of 8 Pixel Pipelines.

As for the Ring Bus that has nothing to do with Memory.

WHat areas would be less intensive on pipes but more intensive on clockspeed??

High core clock makes up for a pipelines deficiency. High memory clock helps with all bandwidth-limited functions, like AA.

The ring bus has everything to do with memory. I'm not sure what you mean by this.

Pipes and clockspeed are interrelated, but it's possible that the GTX's greater number of pipes contribute to their higher min framerates. It's also possible that ATI has some tweaking to do with their new memory controller.

As for your main point, the minimum framerate, two things. One, I believe most games show just the lowest framerate achieved throughout the whole demo. That's potentially a single frame at that framerate. More useful would be something like Halo's benchmark results, where you could see what percentage of frames fell into what framerate decile. HOCP, tho their framerate chart would be ideal to settle this issue, usually doesn't have framerate timelines with comparable settings, in keeping with their more subjective methodology.

Two, I'm not seeing this lower min. yet higher avg. framerate across the board--or even most of the time--especially with the XT, and not so much with the XL. I see them roughly comparable to nV cards, overall.

Look at the games that show min fps: FEAR MP demo, Pariah, Project Snowblind, PoP:WW, SC:CT, CMR2005, Pac Fighters, LOMAC, Perimeter, and WH40koW. In fact, only DoW and PF show comparatively low mins, and PF may well be due to its being OGL (and, thus, drivers, or the devs' [code's] purported propensity to favor nV h/w).

So, I don't see an issue. What did you base your theory on?