When comparing GF FX to 9700PRO, why dont benchmarkers do this....

[BoomAM]

Hi,
I was wondering why, dont benchmarkers:
1) Clock the memory`s so that they both have the same amount of theoretical maximum bandwidth (ie, lowering the 9700pro`s, memrory speed)
2) Clock the GPU/VPU`s so that they are both the same speed.

This would then allow us to see which is faster propely. See which core is the faster one, which has a more efficient memory system, ect;

Do you see where im going with this people.

 

[Deeko]

I see where you are going, but its stupid. Who really cares about that? What matters is which is better in the way it is used.

 

[chizow]

Although its not stupid notion, its certainly pointless. This is the whole apples-to-apples approach 3DMarketing is taking, which is why its fundamentally flawed. The two cards have different architectures and therefore render differently. They were more/less efficient at certain things b/c of design differences, and compensate accordingly. It would be like saying "We should underclock a P4 to see how it does vs. an XP". Design differences force the 2 cards to do what is most efficient to produce the end product. Which one gets there faster is the point of benchmarking. That's why 3DMiss is a broken benchmark; it synthetically "levels the playing field", producing artificial results that simply don't occur in reality.

Chiz

 

[BoomAM]

Stupid, no.
Pointless, not exactely.

For one, it`ll settle any disputes of which core is better.
And another, it`d be interesting.
We are seeing the FX lose to the 9700PRO cos of bandwidth problems. If the cards bandwidth were equal, which would then be better?
We could also see which core was more efficient as well.

You`ve gotta admit it would be a interesting read. Come one, you know you want to. Admit it.

 

[chizow]

See thats the thing, if there was a single agreed upon better way of doing things, everyone would do it the same way. What is drawn up in blue prints based on theory doesn't always translate to real-world performance. It would be a good exercise if you felt there was only 1 to do things, but efficiency has never been a good indicator of performance.

Chiz

 

[BoomAM]

Chiz,
All im saying is that it`d be interesting to see which core would come out on top, if the bandwidths were the same.
Which has the better memory optimisation technologys.
Which core is faster now that the bandwidths the same.

It`d just be interesting to read, thats all, im sure you can appriciate that that.

 

[chizow]

Of course, but it would be unfair to one and an advantage to another. All indications show the FX is bandwidth limited, but since there is no 256-bit option on an FX that test is not an option. We already know what happens to a 9700pro when you cut its bandwidth down; you get a 9500pro. What I'm saying is that all the design factors in a graphics card compensate for other shortcomings. The FX relies on high fill rate and fast clock speeds for its performance, but suffers in shader intensive and bandwidth intensive applications. Likewise, the 9700pro requires extra bandwidth with more complex core processes at the expense of clock speed to get what it needs to get done. The reason I have a major issue with 3DMark2k3 is because it claims to be comparing apples-to-apples by forcing different cards to render in the same manner. But in reality, there is more than 1 way to do the same thing, thereby penalizing the non-conforming parts.

Chiz

 

[bunnyfubbles]

No they wouldn't be the same. Theorectically you'd have to lower the R9700's memory to 250MHz in order to achieve the same bandwidth the 500MHz DDRII supplies to the FX Ultra. Why? The memory architecture of the 9700 allows for a 256 bit memory bus whereas the FX is still at 128. (256bits) 32 bytes * 250 MHz * 2 (DDR) = 16000 MB/sec memory bandwidth while (128 bits) 16 bytes * 500 MHz * 2 (DDR) = 16000 MB/sec memory bandwidth. Because of the difference the theorectical bandwidths would be the same but there is a great chance for inefficiency to come in to play due to the rather drastically different approaches.

Besides, you cannot compare them apples to apples because nVidia designed the FX to have fast core and memory speeds so you run them as they are meant to be. What would be fair is running the products as fast as you can with equal cooling environments (this most likely means overclocking). This would probably show that the 9700 most likely has far more room to overclock than the FX/Ultra and would gain more ground on the product. Ofcourse the FX Ultra's core speed and memory speed are faster, it was designed to have fast core and memory clocks.

 

[BoomAM]

Chiz, you obviously dont like 3D mark, but who says that you have to use 3D mark to do the benchmark.

bunnyfubbles, I know that they wouldnt be the same, but you are missing my point slightly.
What i was saying was, that under identical bandwidth/clock speed situations, which core is better, more efficient, has better memory optimisations, is better at AA, is better at AF, is better at AA & AF, ect;

Comparing a 325mhz core with a 500mhz core, under the same bandwidth scenarios, would be unfair to the 325, but have them both clocked at say, 350, and then see which is faster.

Its not a case of which is faster by OCing, its a case of which core is actually better under the exact same conditions.

 

[EdipisReks]

Originally posted by: BoomAM

Its not a case of which is faster by OCing, its a case of which core is actually better under the exact same conditions.

but who gives a flying f$ck?

 

[BoomAM]

ME for starters. And im sure that there are others as well.
Think of it, as in the name of research.

Look at it like this.
think of the 9700pro and the gf fx as Heatsinks.
Same system configuration,
same system ventilation fans,
same Heatsink fan.
Which will do better?

Im not saying that the benchmarkers should do this with every video card review, Just as a one off every time a new card gets released, or, when two performance cards out perform each other in 50% of the benchmarks, just as a comparison.

 

[gururu]

sorry BoomAM, but it would certainly be a nonproductive analysis. the others have argued well and for the right reasons.
you have to ask yourself what the results would tell you:

Would they really tell you which has the faster core? Maybe, but the specs say the FX is (500MHz vs. 310MHz).
Would they tell you which was more efficient? Certainly not. There are too many factors to jump to such a conclusion when the parameters are altered out of spec (chizow is on the money).
Would it tell you which card to buy? No, because the results wouldn't demonstrate real world performance.
Would it tell you which card had higher quality components? Not any better than a standard comparison.
Would it tell you which had a better memory subsystem? Maybe, but it is clear that the FX is better based on spec (DDRII vs. DDRI).
Would it tell you whether ATI engineers were as good as Nvidia engineers? No, because the product must be considered with the sum of its parts as it was designed to be run (chizow again).

seriously, though it may be interesting to some, I'd find it trivial and not very enlightening.

 

[BoomAM]

I suppose, but imo, it would be interesting.
I`d argue some more, but i cant be arsed, so i`ll just agree with you. lol

 

[Bacinator]

Originally posted by: BoomAM
Hi,
I was wondering why, dont benchmarkers:
1) Clock the memory`s so that they both have the same amount of theoretical maximum bandwidth (ie, lowering the 9700pro`s, memrory speed)
2) Clock the GPU/VPU`s so that they are both the same speed.

This would then allow us to see which is faster propely. See which core is the faster one, which has a more efficient memory system, ect;

Do you see where im going with this people.

It was stated many times in this thread, but under lengthly/biased/ignorant explanations. Simple:

Apples and oranges. Fruit, yes. Can you honestly compare them using the same scale. Green apples are better than red. But how do you determine the scale when you get a red apple, green apple, and orange, using the same criteria. Does a nectarine fit into that scale?

Rotory car engines can handle many more RPM's than traditional 4-strokes. Does that make them better?

Intel uses high FSB speeds, AMD uses less... 2 methods, different techniques for testing... Some are similar... But they are different.

Optical mice truly better than mechanical?

Contacts better than glasses?

All of these things work on different scales. There are multiple ways to do the same thing, and just because you are researching different ways to do the same thing, doesn't mean you should post ignorant posts.

 

[Keysplayr]

Originally posted by: Deeko
I see where you are going, but its stupid. Who really cares about that? What matters is which is better in the way it is used.

Deeko, just wondering if you o/c your stuff.?

 

[kylebisme]

Originally posted by: chizow
Although its not stupid notion, its certainly pointless. This is the whole apples-to-apples approach 3DMarketing is taking, which is why its fundamentally flawed.

Chiz

i don't realy see how it is pointless at all, also i don't see how it compares to 3dmark, a am interested in dissussing both of this things but first i wan't to figure out what you base your openion that 3dmark is "fundamentally flawed" on; which i never got a responce from you here.



BoomAM, i am interested to see such a test as well, granted it is how the thing plays games that realy matters but such invstagation can be inlighteining as well. on the real games note i also want to see q3 timedemos at 1600x1200 with max aa, af and everything else. on my 9700pro i get 65fps in timedemo1, i have to wonder how the gefocefx would fair considering the limmitations on bandwith and piplines it has in compasion to the 9700pro.

 

[BoomAM]

What is is with ppl here lately, everyone seems to be throwing insults like they`re going out of fashion.

What alot of you lot seem to be missing, is that its just for research and stuff.
Like, would the 9700pro be better than a GF FX, given the same bandwidth limitations?
Its all very well comparing my suggestion to apples & engines, but you are way off my inital point.

 

[cowdog]

I don't really have the video card technical know-its to discuss this in direct terms, but I think I'm getting one general point.

It sounds like such a comparison would be similar to comparing two high-end sport cars. One may rely on high rpm, while the other relies on power at low rpm (very over simplified, but the basic point is there). To reduce the power of the latter to the level of the former would not really make any sense for a comparative purpose. Each vehicle will have parts and subsystems designed to meet the overall approach to performance, and the only meaningful comparison would be to compare each as a whole (as each is designed). Now obviously each approach has pros and cons, but that's the reality of taking any particular design approach. In this vehicle example, I don't care if the fuel injection systems perform very differently and have different domains of stable function (they are likely designed to function differently), but I do care about the overall performance. Isn't this basicually the case with ATI and nVidia?

(edit: oops, I just restated some else's point. Suppose I should read all posts before posting. Hehe. But what the hey, it's all just internet play. I sure hope that nVidia and ATI duke it out in a bloody all-out competition, because I made a poor choice in picking up this Saphire 8500LE with its slow, overclock unfriendly ram. If any comparisons help to fuel that battle, I'm all for it.)

 

[kylebisme]

i know it was a rhetorical question, but sence you posed it i want to respond. i think it has a lot to do with human nature, tool has a wonderful song about the situation which you can find the lyrics to here. its a bit abstract and realy comes off better with the music but hopefuly you all might still se the relvence.

 

[jiffylube1024]

Originally posted by: keysplayr2003

Originally posted by: Deeko
I see where you are going, but its stupid. Who really cares about that? What matters is which is better in the way it is used.

Deeko, just wondering if you o/c your stuff.?

Deeko was one of the earlier guys here to get his P4 1.6A overclocked to 2.7+ GHz, so to answer your question, yes, I believe he does .

 

[Sunner]

I think it would be interesting to see.

Not that it would change my opinions about either card, since that's not the way they're meant to run, but still, it would all be in good fun

 

[BoomAM]

Originally posted by: Sunner
I think it would be interesting to see.

Not that it would change my opinions about either card, since that's not the way they're meant to run, but still, it would all be in good fun

Thats what ive been trying to say.
Its just a bit of fun, in the name of research. I still think that the GF FX is a multi-use, glorified vacuum cleaner though.

 

[sash1]

"Comparing a 325mhz core with a 500mhz core, under the same bandwidth scenarios, would be unfair to the 325, but have them both clocked at say, 350, and then see which is faster."

You've totally missed the bill on this one. People have already said it; video cards were meant to be run at the speeds they are set at, not neutered so you can compare the cores.

Chizow pretty much nailed it: "It would be like saying "We should underclock a P4 to see how it does vs. an XP"." and to pretty much summarize what I would say on this topic: "The FX relies on high fill rate and fast clock speeds for its performance, but suffers in shader intensive and bandwidth intensive applications. Likewise, the 9700pro requires extra bandwidth with more complex core processes at the expense of clock speed to get what it needs to get done."

"I still think that the GF FX is a multiuse, glorified vacuum cleaner though."

A 7dB fan is a vacuum cleaner?

"Thats what ive been trying to say.
Its just a bit of fun, in the name of research."

Well, "in the name of research," you can just look at the technologies of the cards and determine for yourself which is best.

~Aunix

 

[BoomAM]

Only the Gainward GF FX is 7dBa, and thats probably in 2D mode.
Well, "in the name of research," you can just look at the technologies of the cards and determine for yourself which is best.
I already know which is best. IMO, its the card i brought, the 9700PRO.

OK then, what about keeping the core speed the same, and only changing the memory bandwidth so they`re equal?

 

[sash1]

"Only the Gainward GF FX is 7dBa, and thats probably in 2D mode."

umm... NO! It's max 7dB: see here

~Aunix