9800 GTX+/GTS 250

[AzN]

Originally posted by: BFG10K
What you call ?bandwidth happy? is actually the norm for DX10 parts because they generally have enough bandwidth so that it?s not the primary limiting factor. In essence you agree with me, but try to word it in a way so that it appears you disagree.

Again I?ll ask whether you consider the 4770 to be bandwidth happy? It must be according to your definition because it?s faster than the 4830 while experiencing a reduction in bandwidth.

Just because you own a GTX series or your prior Ultra doesn't mean all dx10 parts have generally enough bandwidth. G92 is one of those cards that lacks enough bandwidth.

No I don't consider 4770 to be a bandwidth happy card. You are obviously getting confused. It's faster than 4830 because it has faster shader clocks first off. Second it has higher texture fill and pixel fill because of core clocks. Bandwidth only determines how fast those pixels and textures go in and out of graphics memory a litttle more on higher resolutions or with AA. Just because it has slightly lower bandwidth does not mean it will be slower than 4830.

Again back to my explanation and here's an example.

HD4830
9200 Mpixels/s
18400 Mtexels/s
736 GFLOPS

HD4770
12000 Mpixels/s
24000 Mtexels/s
960 GFLOPS

Let say 4830 is bottlenecked by bandwidth by 10% and 4770 is bottlenecked by 20% and you would get something like this. Again there are variables in higher resolutions, settings, etc where those numbers would fluctuate...

HD4830
8280 Mpixels/s
16560 Mtexels/s
736 GFLOPS

HD4770
9600 Mpixels/s
19200 Mtexels/s
960 GFLOPS

While I?m not disputing the accuracy of your figures, they?re clearly an outlier. This is very obvious by the fact that the 8800 GTS 640 MB has more bandwidth that the 8800 GT, but it?s also slower overall, even with AA. If the 8800 GT was primarily limited by bandwidth, there?s no way it could be faster: http://www.behardware.com/arti...a-geforce-8800-gt.html

I guess my benchmarks don't count but yours do? You obviously don't seem to understand a single thing I tried to explain to you. Bandwidth doesn't drive anything. It only carries the driver. To say 8800gts 640mb should be faster because it has more bandwidth than 8800gt is absurd.

Although 8800gt is severely bottlenecked by bandwidth it can still beat 8800gts 640 mb because 8800gt has 112SP 56TMU while 8800gts has 96SP wtih 24TMU. Look at the easy picture like example above. That should explain why 8800gt is able to beat 8800gts. Now if 8800gt had more bandwidth it can stump 8800gts "further".

You appear to be typing responses that have nothing to do with what you quoted. Let?s go back to the beginning:

I stated the 4850 isn?t constrained by bandwidth because it beats the Ultra with AA despite having around half the bandwidth, and also that it?s about three times faster than the 3870 with AA, with less bandwidth again. Therefore, the 4850 is not hamstrung by bandwidth.

You then posted a slide showing ROP resolve being doubled and told us that is why the card does so well with AA.

I acknowledged this but then pointed out since improvements in ROPs are dramatically raising AA performance, this means bandwidth isn?t the limiting factor given it?s decreased from either the 3870 or the 8800 Ultra, yet the card is still faster with AA.

So again I?ll ask how an improvement to ROPs can show such performance gains with AA if the 4850 is limited by bandwidth like you claim? How can the reduction of bandwidth ? the aspect you claim the card is primarily limited by and influences AA performance - increase AA performance?

It?s a very simple question, so please address it instead of typing multiple sentences of irrelevant rhetoric. Thanks.

Obviously a clear example does not help you understand but confuses you even further. So I'll make it simpler by answering your question.

3870 AA is broken. It has less than 1/2 texture fillrate and shader performance compared to 4850. 3870 was never limited by bandwidth because it never had the fillrate for the need for more bandwidth. This was prevalent when ATI cut 3870 bandwidth by 50% and still perform exactly same as 2900xt. With the ultra I gave you a clear example which you can't seem to take in. Nothing I can do about that.

Why are you talking about made up numbers when we have actual benchmarks proving you wrong?

that's why it's called an example. you do know what an example is don't you?

Oh yes, the amazing 3DMark, which told us the 2900XT was faster than the 8800 GTX. Oh wait, that?s the complete opposite of reality when running games. Much like your theoretical figures which you made up.

IF you have a problem with techreport or their affiliates I suggest you take it up with them. Tell them who you are and where you write your articles and that you have problems with their conclusions. Oh BTW your partner has no problems with 3dmark as he uses this benchmark to determine his conclusions.

You have no idea what tests I?ve done given I haven?t released all of the results to the public. Furthermore there?s a plethora of tests I can link to that were done by other reviewers that back my claims.

The problem is that you don?t acknowledge such a broad range of tests disproving you but instead cling to your own tests as the sole form of evidence. Again, my tests simply confirm what other reviewers are saying so that makes your tests the outlier, not mine.

I don't know what tests you've done but I do know what cards you've had. You had an ultra and a 4850 for a short time then GTX260 and finally GTX 285. Out of all those cards except for 4850 they were all bandwidth happy cards.

Your tests are correct and mine is wrong even though the card I tested was a G92 which has everything to do with this thread while your tests consist of G80, RV770, GTX series mostly bandwidth happy cards. Clearly you are the one that have a problem with accepting other people's findings unless it was already been done before by other web sites clearly that's all you ever write your articles on.

How do you know it could use 25% more memory? Where did this figure come from?

Again it?s not hard; just overclock the 4850?s core and witness an almost linear performance increase without touching the memory at all. Heck, ATi have done it for you with the 4870 where the performance gain in games basically matches the core performance improvement, thereby proving most of the extra bandwidth over the 4850 is simply not needed.

I'm just making a educated guess here considering 4870 performance doesn't improve with more bandwidth while 4850 does.

This might be too much for you to swallow. Here is gainward 4850 clocked @ 4870 core clocks and memory @ 1200mhz going up against a stock 4870.

http://www.xbitlabs.com/articl...1024mb-gs_7.html#sect1

Again 4870 is still considerably faster than 4850 at those clocks.

The 4850 beating the 4770 could be due to lots of reasons but you certainly can?t automatically infer it?s from more bandwidth. Actually the three cards slot in the order of their shader performance which is something similar with the G80 and G92 line.

OTOH after reducing the bandwidth on the 4770 it?s still faster than the 4830 with more bandwidth, you can safely conclude that the 4770 isn?t primarily limited by bandwidth.

But then again we are talking about AA benchmarks here where shader doesn't matter much even in your ultra AA core/shader/memory clock findings. Since 4770 push more pixels and you say 4770 isn't really bottlenecked by bandwidth why does it lose in AA benches?

 

[apoppin]

Originally posted by: SlowSpyder
Obviously the 8800GTX is dated now, but you could still game on it and play most anything I would think.

I have a feeling Nvidia initially wanted higher clocks than what the GTX2x0 cards launched at, but they just couldn't get there, or atleast not get enough of those parts to have decent yields.

Not really. It is pretty dated and doesn't handle DX10 games very well

i am doing a *comparison* - today

8800GTX vs. 9800GT/512MB vs. GTS250/512 MB

vs .. 2900XT [] and 4870/1GB

if you are interested in seeing pre-results, LMK

 

[AzN]

Originally posted by: Denithor

Originally posted by: Azn
G92 is not less powerful. With equal bandwidth as GTX260 it would easily be as fast as GTX260.

And in contrast, just clock the core and shaders on a GTX260 to 738/18xx. Raped ape

This is besides the point. You can clock anything and rape ape.

The original point of this thread was that a 9800GTX+ (128SP/738/1836) can nearly match a GTX 260/216 (216SP/576/1242) in some games. Think about that for a second. The 9800GTX+ is the fastest-clocked G92 core by far and it can barely match a much much lower clocked GTX 260/216 in certain games. Keys was saying if you set the core/shader clockspeeds equal (downclock 9800 or upclock 260) the higher SP count GTX 260 would rape the G92 card. Bandwidth has nothing to do with this - simply the amount of shader power available.

RE bandwidth - I see bandwidth like the hose on a pump. If you use a hose too small the pump isn't going to be able to do its job efficiently and flow rates will be lower than the pump is capable of producing (remember the old 7600GT anyone? severely bandwidth limited). Using a hose too big for pump capacity won't improve performance by itself - the extra space goes to waste because the pump cannot fill the pipeline (think 2900XT here - twice the bandwidth of any other card of its generation and still suck ass performance).

For memory bandwidth more is generally better but you obtain diminishing returns at some point. Once you have enough to keep the core(s) saturated and working constantly you don't need any more.

Except that GTX 260 can not do those clocks. A 9800gtx+ does overclock too you know. We are trying to determine memory bandwidth bottlenecks of G92. Not compare it with unachievable overclocks of GTX260 then compare it with as stock g92.

You definitely have the right idea about bandwidth. Something BFG seem to have a hard time understanding.

 

[Keysplayr]

Originally posted by: Azn

Originally posted by: Denithor

Originally posted by: Azn
G92 is not less powerful. With equal bandwidth as GTX260 it would easily be as fast as GTX260.

And in contrast, just clock the core and shaders on a GTX260 to 738/18xx. Raped ape

This is besides the point. You can clock anything and rape ape.

The original point of this thread was that a 9800GTX+ (128SP/738/1836) can nearly match a GTX 260/216 (216SP/576/1242) in some games. Think about that for a second. The 9800GTX+ is the fastest-clocked G92 core by far and it can barely match a much much lower clocked GTX 260/216 in certain games. Keys was saying if you set the core/shader clockspeeds equal (downclock 9800 or upclock 260) the higher SP count GTX 260 would rape the G92 card. Bandwidth has nothing to do with this - simply the amount of shader power available.

RE bandwidth - I see bandwidth like the hose on a pump. If you use a hose too small the pump isn't going to be able to do its job efficiently and flow rates will be lower than the pump is capable of producing (remember the old 7600GT anyone? severely bandwidth limited). Using a hose too big for pump capacity won't improve performance by itself - the extra space goes to waste because the pump cannot fill the pipeline (think 2900XT here - twice the bandwidth of any other card of its generation and still suck ass performance).

For memory bandwidth more is generally better but you obtain diminishing returns at some point. Once you have enough to keep the core(s) saturated and working constantly you don't need any more.

Except that GTX 260 can not do those clocks. A 9800gtx+ does overclock too you know. We are trying to determine memory bandwidth bottlenecks of G92. Not compare it with unachievable overclocks of GTX260 then compare it with as stock g92.

You definitely have the right idea about bandwidth. Something BFG seem to have a hard time understanding.

But the 9800GTX+ CAN do GTX260 clocks.

Ok, so that everything is clear, let me ask you a few questions rather that play ring around the rosie.

Are you saying that if a 9800GTX+ had the same memory bandwidth of a GTX260, it would exceed GTX260 performance?

If so, does this apply to shader intensive games or non shader intensive games?

In turn, if the GTX260 had the same memory bandwidth of a 9800GTX+ do you think it would perform under a 9800GTX+?

And lastly, the two scenarios above having the same shader clocks (either both cards clocked at core/mem 9800GTX+ or GTX260 speeds, what would you perceive the outcome to be?

Keys

 

[AzN]

Originally posted by: Keysplayr
But the 9800GTX+ CAN do GTX260 clocks.

Ok, so that everything is clear, let me ask you a few questions rather that play ring around the rosie.

I don't understand why you are trying to compare a card that has 28ROP, 72TMU, 216SP with 448bit bus and then compare it to a card that has 16ROP, 64TMU, 128SP with 256bit bus wanting to downclock to the level of the bigger chip with more of everything. Isn't the answer so damn obvious? G92 does well because it's clocked much faster.

Are you saying that if a 9800GTX+ had the same memory bandwidth of a GTX260, it would exceed GTX260 performance?

If so, does this apply to shader intensive games or non shader intensive games?

In turn, if the GTX260 had the same memory bandwidth of a 9800GTX+ do you think it would perform under a 9800GTX+?

And lastly, the two scenarios above having the same shader clocks (either both cards clocked at core/mem 9800GTX+ or GTX260 speeds, what would you perceive the outcome to be?

Keys

The answer to your questions have already been answered in few posts above when I replied to sickbeast.

Let me ask you this. What if GTX260 had the same memory bandwidth as GTS250? What do you think the outcome will be? Will the chip still have the same performance as stock GTX260 or will it drop in performance and perform near GTS250 levels?

I can easily drop my GTX260 memory clocks and benchmark to show you how that memory bandwidth is effected. Considering GTS is only 10% slower in raw frame rates and 30% in AA benches I don't doubt my GTX260 will snoop to GTS250 levels.

 

[Keysplayr]

Originally posted by: Azn

Originally posted by: Keysplayr
But the 9800GTX+ CAN do GTX260 clocks.

Ok, so that everything is clear, let me ask you a few questions rather that play ring around the rosie.

I don't understand why you are trying to compare a card that has 28ROP, 72TMU, 216SP with 448bit bus and then compare it to a card that has 16ROP, 64TMU, 128SP with 256bit bus wanting to downclock to the level of the bigger chip with more of everything. Isn't the answer so damn obvious? G92 does well because it's clocked much faster.

Are you saying that if a 9800GTX+ had the same memory bandwidth of a GTX260, it would exceed GTX260 performance?

If so, does this apply to shader intensive games or non shader intensive games?

In turn, if the GTX260 had the same memory bandwidth of a 9800GTX+ do you think it would perform under a 9800GTX+?

And lastly, the two scenarios above having the same shader clocks (either both cards clocked at core/mem 9800GTX+ or GTX260 speeds, what would you perceive the outcome to be?

Keys

The answer to your questions have already been answered in few posts above when I replied to sickbeast.

Let me ask you this. What if GTX260 had the same memory bandwidth as GTS250? What do you think the outcome will be? Will the chip still have the same performance as stock GTX260 or will it drop in performance and perform near GTS250 levels?

I can easily drop my GTX260 memory clocks and benchmark to show you how that memory bandwidth is effected. Considering GTS is only 10% slower in raw frame rates and 30% in AA benches I don't doubt my GTX260 will snoop to GTS250 levels.

No, please directly answer my questions first. My questions directly. Then I will be more than happy to answer the questions you asked to answer my questions. LOL.

 

[AzN]

Originally posted by: Keysplayr
No, please directly answer my questions first. My questions directly. Then I will be more than happy to answer the questions you asked to answer my questions. LOL.

Now you are just being silly.

I don't need to hear your feedback when I already know the answer to which I can do the benches myself to show you how bandwidth is effected.

1920x1080

Crysis 4xAF High settings dx10

Stock GTX 260 216 576/1350/999
33.14 fps

GTX 260 216 576/1350/629 70.4gb/s of bandwidth same as GTS 250.
31.34 fps

8800GTS @ 756/1836/1026 just shy of GTS250 1100mhz memory.
30.14fps


4xAF 4xAA High settings dx10

Stock GTX 260 216 576/1350/999
27.42 fps

GTX 260 216 576/1350/629
23.39 fps






1680x1050
Crysis 4xAF high settings

8800gts @ 756/1836/1026
34.43 fps

CORE REDUCTION 590/1836/1026
32.13 fps -7.2% difference

BANDWIDTH REDUCTION 756/1836/800
29.72 fps -15.8% difference

 

[Keysplayr]

Originally posted by: Azn

Originally posted by: Keysplayr
No, please directly answer my questions first. My questions directly. Then I will be more than happy to answer the questions you asked to answer my questions. LOL.

Now you are just being silly.

I don't need to hear your feedback when I already know the answer to which I can do the benches myself to show you how bandwidth is effected.

1920x1080

Crysis 4xAF High settings dx10

Stock GTX 260 216 576/1350/999
33.14 fps

GTX 260 216 576/1350/629 70.4gb/s of bandwidth same as GTS 250.
31.34 fps

8800GTS @ 756/1836/1026 just shy of GTS250 1100mhz memory.
30.14fps


4xAF 4xAA High settings dx10

Stock GTX 260 216 576/1350/999
27.42 fps

GTX 260 216 576/1350/629
23.39 fps






1680x1050
Crysis 4xAF high settings

8800gts @ 756/1836/1026
34.43 fps

CORE REDUCTION 590/1836/1026
32.13 fps -7.2% difference

BANDWIDTH REDUCTION 756/1836/800
29.72 fps -15.8% difference

totally uncooperative. thanks. You know all the answers, so I guess you don't need to be in a forum? Why talk about anything at all if you know all?

 

[apoppin]

i am testing a GTS 250 right now

anything you want to see while i have it ?
- i am comparing it to 9800GT

 

[AzN]

Originally posted by: Keysplayr
totally uncooperative. thanks. You know all the answers, so I guess you don't need to be in a forum? Why talk about anything at all if you know all?

Like you can't look at the few posts up? I just did the benchmarks after I asked you the question. I might not need the forum but people like you need guys like me.

 

[AzN]

Originally posted by: apoppin
i am testing a GTS 250 right now

anything you want to see while i have it ?
- i am comparing it to 9800GT

Can you compare with a GTX260? Get to the bottom of this thread.

 

[SickBeast]

Originally posted by: Azn

Originally posted by: SickBeast
Actually, I'm pretty sure that the entire reason why the 4850 is slower than the 4870 is a memory bandwidth bottleneck. The performance differential there should give you an idea as to how a GTS250 would perform with a 512-bit bus or else GDDR5 memory, in terms of a percentage.

Obviously when you consider they are the exact same chip just using faster memory.

GTS250 in other hand has more texture fillrate than GTX260 216SP. Hell it might even be a little faster depending on the games of course.

That's exactly what I was getting at. The chips are the same. Even if they are clocked differently at stock, they will both overclock to the same level. At equal clocks, the 4870 *is* faster, meaning that there is indeed a bottleneck. The testing methodology will largely determine how much this bottleneck will affect the card. At high resolutions with AA and AF enabled, the bottleneck will be much more significant.

 

[SickBeast]

Originally posted by: BFG10K

Originally posted by: SickBeast

It only makes sense that current graphics cards are being held back by a 256-bit interface.

No, they really aren't. Again there are numerous examples of DX10 parts with less bandwidth beating other DX10 parts with more bandwidth because they have higher processing capability (e.g. 4850 vs 3870, 8800 GT vs 8800 GTS 640, 4770 vs 4830, etc).

Actually, I'm pretty sure that the entire reason why the 4850 is slower than the 4870 is a memory bandwidth bottleneck.

How do you figure that? The 4870 has a 20% higher core clock and around double the bandwidth of a 4850, but it?s only around 20%-30% faster in most situations. If memory bandwidth was the primary limitation then we should see much higher performance gains that that.

Again it?s not hard to test; just overclock the 4850?s core without touching the memory and witness an almost linear performance increase. Now do the same with memory and you?ll see very little performance change.

The performance differential there should give you an idea as to how a GTS250 would perform with a 512-bit bus or else GDDR5 memory, in terms of a percentage.

Based on current and past trends with DX10 parts, it would likely add very little performance.

I have no doubt that a 256-bit interface is the 'sweet spot' right now for current graphics cards. The point that I am making is that it is indeed a bottleneck in many circumstances (high res w/ AA+AF). With newer cards like the upcoming 5870 and NV300, I'm saying that a 256-bit bus will not be enough in all likelihood.

 

[BFG10K]

Originally posted by: Azn

G92 is one of those cards that lacks enough bandwidth.

Actually it doesn?t because I?ve demonstrated how a 8800 GT is faster than a 8800 GTS 640 despite having less bandwidth. Therefore the G92 is not primarily limited by bandwidth. Do you understand that if I reduce bandwidth but performance still goes up, it means bandwidth isn?t the limitation?

For that matter why do you constantly ignore the plethora of evidence that proves you wrong?

No I don't consider 4770 to be a bandwidth happy card. You are obviously getting confused. It's faster than 4830 because it has faster shader clocks first off. Second it has higher texture fill and pixel fill because of core clocks. Bandwidth only determines how fast those pixels and textures go in and out of graphics memory a litttle more on higher resolutions or with AA. Just because it has slightly lower bandwidth does not mean it will be slower than 4830.

Again a total skirt of the issue. If the 4770 isn?t bandwidth happy then by your admission bandwidth isn?t the primary limitation because it has less of it than the 4830, but it?s still faster.

Again do you understand the concept that if I reduce bandwidth but performance still goes up, it means bandwidth isn?t responsible for that increase?

I guess my benchmarks don't count but yours do?

My benchmarks largely show the same thing as Derek?s where the core clock showed the biggest difference. So in one corner we have mine, Derek?s figures, and a plethora of DX10 benchmarks that mostly show memory bandwidth isn?t the primary limiting factor.

In another corner we have Azn and his results; your results are a clear outlier.

To say 8800gts 640mb should be faster because it has more bandwidth than 8800gt is absurd.

Why? You?re the one claiming the G92 is primarily limited by bandwidth so I demonstrated how the 8800 GT (with less bandwidth) is faster than the 8800 GTS 640, even with AA. Therefore your claims are nonsensical given real-world results contradict them.

Although 8800gt is severely bottlenecked by bandwidth it can still beat 8800gts 640 mb because 8800gt has 112SP 56TMU while 8800gts has 96SP wtih 24TMU. Look at the easy picture like example above. That should explain why 8800gt is able to beat 8800gts. Now if 8800gt had more bandwidth it can stump 8800gts "further".

So in other words you?re claiming the G92 is bottlenecked primarily by bandwidth, but it doesn?t matter because the core is faster? Again how can that be if bandwidth is the major component holding back performance like you claim, but the part has even less of it of the 8800 GTS 640, yet is still faster?

Again I?ll ask for the third time whether you understand the concept of reducing bandwidth means the performance increase can?t possible be coming from it?

Obviously a clear example does not help you understand but confuses you even further. So I'll make it simpler by answering your question.

It?s like Keys says, you can?t provide a straight answer to anything that disputes your theories.

3870 was never limited by bandwidth because it never had the fillrate for the need for more bandwidth. This was prevalent when ATI cut 3870 bandwidth by 50% and still perform exactly same as 2900xt.

The 4850 isn?t primarily limited by bandwidth either. Again this is trivial to prove by simply moving the sliders one at a time and seeing which has the biggest impact on performance (hint: it?s not memory).

With the ultra I gave you a clear example which you can't seem to take in. Nothing I can do about that.

But you still haven?t addressed my question:

How is it that the 4850 competes with the 8800 Ultra with 4xAA and beats it with 8xAA when it?s bandwidth limited like you claim, but has around half the bandwidth of the 8800 Ultra?

How is it that the ROP improvements increase AA performance if you claim memory is holding things back?

How is it that the 4850 shows bigger increases from core clocks when you claim memory is the primary limiting factor?

IF you have a problem with techreport or their affiliates I suggest you take it up with them. Tell them who you are and where you write your articles and that you have problems with their conclusions. Oh BTW your partner has no problems with 3dmark as he uses this benchmark to determine his conclusions.

I don?t have a problem with Tech-Report, I have a problem with you putting 3DMark?s theoretical figures and your guesses ahead of real benchmark results that directly disprove your claims.

As for my ?partner?, he?s free to use any testing tool he pleases, as am I. I personally don?t use 3DMark because a tool that claims a 2900XT is faster than a 8800 GTX simply can?t be trusted.

I don't know what tests you've done but I do know what cards you've had. You had an ultra and a 4850 for a short time then GTX260 and finally GTX 285. Out of all those cards except for 4850 they were all bandwidth happy cards.

Wonderful, so we?ve established you don?t consider either the 4850 or the 4770 to be bandwidth happy, and you also think they?re bandwidth limited. Then again for the fourth time I?ll ask how you can claim this when a reduction of bandwidth still shows performance gains with AA?

If these parts are primarily bandwidth limited like you claim then performance should be going down with reduced bandwidth, but it isn?t. That?s the whole point; we?re reducing bandwidth, but performance is still going up. Therefore the primary performance bottleneck is elsewhere.

Your tests are correct and mine is wrong even though the card I tested was a G92 which has everything to do with this thread while your tests consist of G80, RV770, GTX series mostly bandwidth happy cards. Clearly you are the one that have a problem with accepting other people's findings unless it was already been done before by other web sites clearly that's all you ever write your articles on.

I don?t have a problem accepting other peoples? findings. On the contrary; other peoples? findings show me the 8800 GT is faster than the 8800 GTS 640 despite having less bandwidth, thereby telling me bandwidth isn?t the primary limiting factor as it currently ships on the G92.

This might be too much for you to swallow. Here is gainward 4850 clocked @ 4870 core clocks and memory @ 1200mhz going up against a stock 4870.

http://www.xbitlabs.com/articl...1024mb-gs_7.html#sect1

Again 4870 is still considerably faster than 4850 at those clocks.

I?m not saying memory makes zero difference, only that it isn?t the primary limiting factor i.e. other clocks make much more difference.

And actually in many most of those titles the card is either tied with the 4870, or the 4870 is less than 10% faster, with many of gaps narrowing even further when the resolution is increased, despite having much less bandwidth courtesy of GDDR3.

Since 4770 push more pixels and you say 4770 isn't really bottlenecked by bandwidth why does it lose in AA benches?

Again I?m not saying memory bandwidth makes zero difference, only that it?s not the primary limiting factor with most DX10 parts. We can see this with the 4770 reducing bandwidth over the 4830, yet still being faster with AA.

 

[Keysplayr]

Originally posted by: Azn

Originally posted by: Keysplayr
totally uncooperative. thanks. You know all the answers, so I guess you don't need to be in a forum? Why talk about anything at all if you know all?

Like you can't look at the few posts up? I just did the benchmarks after I asked you the question. I might not need the forum but people like you need guys like me.

This assumption of yours it what I had the most problem with.

"G92 is not less powerful. With equal bandwidth as GTX260 it would easily be as fast as GTX260."

We cannot reduce the bit width of the GTX260 to bring it down to the level of bandwidth of the 9800GTX+ right? Okay. What would the memory clock have to be on a GTX260 to bring down the memory bandwidth to that of the level of the 9800GTX+?

I'm terrible at this sort of math. Would someone care to?

Apoppin, do you have a GTX260? Preferably the 192sp version? I have a GTX280 but I think that is quite an improbable experiment unless I could disable shaders and ROP's. Which I doubt.

 

[toyota]

Originally posted by: Keysplayr

Originally posted by: Azn

Originally posted by: Keysplayr
totally uncooperative. thanks. You know all the answers, so I guess you don't need to be in a forum? Why talk about anything at all if you know all?

Like you can't look at the few posts up? I just did the benchmarks after I asked you the question. I might not need the forum but people like you need guys like me.

This assumption of yours it what I had the most problem with.

"G92 is not less powerful. With equal bandwidth as GTX260 it would easily be as fast as GTX260."

We cannot reduce the bit width of the GTX260 to bring it down to the level of bandwidth of the 9800GTX+ right? Okay. What would the memory clock have to be on a GTX260 to bring down the memory bandwidth to that of the level of the 9800GTX+?

I'm terrible at this sort of math. Would someone care to?

Apoppin, do you have a GTX260? Preferably the 192sp version? I have a GTX280 but I think that is quite an improbable experiment unless I could disable shaders and ROP's. Which I doubt.

I have the 192sp version of the gtx260 if you need me to test anything.

 

[BFG10K]

Originally posted by: Keysplayr

We cannot reduce the bit width of the GTX260 to bring it down to the level of bandwidth of the 9800GTX+ right? Okay. What would the memory clock have to be on a GTX260 to bring down the memory bandwidth to that of the level of the 9800GTX+?

GTX260: 111.9GB/s, 999MHz (or 1998 MHz).
9800 GTX+: 70.4GB/s, 1100MHz (or 2200 MHz).

The 9800 has 41.5 GB/sec less bandwidth, which is a 37% reduction over the GTX260.

So the GTX260 would have to be clocked 37% lower, which is 629 MHz (or 1258 MHz).

 

[SlowSpyder]

Originally posted by: apoppin

Originally posted by: SlowSpyder
Obviously the 8800GTX is dated now, but you could still game on it and play most anything I would think.

I have a feeling Nvidia initially wanted higher clocks than what the GTX2x0 cards launched at, but they just couldn't get there, or atleast not get enough of those parts to have decent yields.

Not really. It is pretty dated and doesn't handle DX10 games very well

i am doing a *comparison* - today

8800GTX vs. 9800GT/512MB vs. GTS250/512 MB

vs .. 2900XT [] and 4870/1GB

if you are interested in seeing pre-results, LMK

Haha... I've mentioned it before in the past, I'll do it again. I think you and I have very differing tastes when it comes to what is acceptable. It's hardly the juggernaut it was when it released, but it beats the hell out of integrated too.

I'd actually really like to see those results... I have a sweet spot for the 2900 cards. I bought a 2900 Pro (that easily overclocked to 2900XT levels and beyond like I'm willing to bet all of them did) that I got for a very good price at the time, back when 8800GTX 320 cards were still ~$300, and I loved that card. It really held it's place with new hardware as Nvidia really didn't release anything new (8800 -> 9800 series) and AMD really didn't do much (a 30MHz increase in clock speed with a 3870 and less memory bandwidth) as far as performance go. I ended up pairing with a Phenom 1 eventually. It was kind of a second class rig that made me very happy with it's performance actually. I say second class mostly because most people here wouldn't want to touch that hardware.

Anyway, I'd like to see the results simply because I remember so many people saying at the 2900 XT launch how we just should wait for the DX10 games, then we'll see it's true potential. I think we can tell by how competitive Nvida and AMD are now with much faster parts that are built on the same architecture (more or less) that the whole DX10 being the 2900 XT's sweet spot just wasn't the case. But I'd be curious to see how the somewhat horsepower-lacking 2900 core does with it's over abundance of memory bandwidth and current drivers (my guess is it'll still be the slowest card out of those you mentioned... maybe once in a while giving the 9800GT a run for it's money). Thanks!

 

[apoppin]

Originally posted by: Azn

Originally posted by: apoppin
i am testing a GTS 250 right now

anything you want to see while i have it ?
- i am comparing it to 9800GT

Can you compare with a GTX260? Get to the bottom of this thread.

i don't have one and i am still waiting on GTX 275

i will temporarily trade someone for it; maybe for a 4870/1GB
-BFG10K has one and and a 4850 but you don't like his results
======================

Originally posted by: SlowSpyder

Originally posted by: apoppin

Originally posted by: SlowSpyder
Obviously the 8800GTX is dated now, but you could still game on it and play most anything I would think.

I have a feeling Nvidia initially wanted higher clocks than what the GTX2x0 cards launched at, but they just couldn't get there, or atleast not get enough of those parts to have decent yields.

Not really. It is pretty dated and doesn't handle DX10 games very well

i am doing a *comparison* - today

8800GTX vs. 9800GT/512MB vs. GTS250/512 MB

vs .. 2900XT [] and 4870/1GB

if you are interested in seeing pre-results, LMK

Haha... I've mentioned it before in the past, I'll do it again. I think you and I have very differing tastes when it comes to what is acceptable. It's hardly the juggernaut it was when it released, but it beats the hell out of integrated too.

I'd actually really like to see those results... I have a sweet spot for the 2900 cards. I bought a 2900 Pro (that easily overclocked to 2900XT levels and beyond like I'm willing to bet all of them did) that I got for a very good price at the time, back when 8800GTX 320 cards were still ~$300, and I loved that card. It really held it's place with new hardware as Nvidia really didn't release anything new (8800 -> 9800 series) and AMD really didn't do much (a 30MHz increase in clock speed with a 3870 and less memory bandwidth) as far as performance go. I ended up pairing with a Phenom 1 eventually. It was kind of a second class rig that made me very happy with it's performance actually. I say second class mostly because most people here wouldn't want to touch that hardware.

Anyway, I'd like to see the results simply because I remember so many people saying at the 2900 XT launch how we just should wait for the DX10 games, then we'll see it's true potential. I think we can tell by how competitive Nvida and AMD are now with much faster parts that are built on the same architecture (more or less) that the whole DX10 being the 2900 XT's sweet spot just wasn't the case. But I'd be curious to see how the somewhat horsepower-lacking 2900 core does with it's over abundance of memory bandwidth and current drivers (my guess is it'll still be the slowest card out of those you mentioned... maybe once in a while giving the 9800GT a run for it's money). Thanks!

i have completed 2900xt results and also 9800GT - working on the 250 GTS and then 8800GTX after that

2900xt would be like the old GTS-640 .. a real dog by any DX10 standards - even at 14x9
- it does OK at medium details with DX9

i can give you specifics
- you know the games i use - make a request

 

[Keysplayr]

Originally posted by: BFG10K

Originally posted by: Keysplayr

We cannot reduce the bit width of the GTX260 to bring it down to the level of bandwidth of the 9800GTX+ right? Okay. What would the memory clock have to be on a GTX260 to bring down the memory bandwidth to that of the level of the 9800GTX+?

GTX260: 111.9GB/s, 999MHz (or 1998 MHz).
9800 GTX+: 70.4GB/s, 1100MHz (or 2200 MHz).

The 9800 has 41.5 GB/sec less bandwidth, which is a 37% reduction over the GTX260.

So the GTX260 would have to be clocked 37% lower, which is 629 MHz (or 1258 MHz).

Thanks BFG.

And thanks Toyota, but I think for this test, somebody needs to have both cards and use them on the same platform. CPU/mem/res/drivers, etc.

This way, there is no room for dispute ("Wait, they are using different CPU's, more memory, different drivers.") type of thing. That is most annoying. Can't have that.

 

[apoppin]

why is my Galaxy GTS 250/512MB consistently outperforming my Palit 9800GT/512MB ?
- i thought they were fairly close .. is the 9800GTX that much faster than the GT?

 

[SSChevy2001]

Originally posted by: apoppin
why is my Galaxy GTS 250/512MB consistently outperforming my Palit 9800GT/512MB ?
- i thought they were fairly close .. is the 9800GTX that much faster than the GT?

Are you serious?

GTS250 738/1836/1100 128SPs 16ROPs 64TMUs
9800GT 600/1500/900 112SPs 16ROPs 56TMUs

22% increase in Memory Bandwidth
40% increase in FLOPS
23% increase in Pixel Fill Rate
41% increase in Texture Fill Rate

 

[AzN]

Originally posted by: Keysplayr

Originally posted by: Azn

Originally posted by: Keysplayr
totally uncooperative. thanks. You know all the answers, so I guess you don't need to be in a forum? Why talk about anything at all if you know all?

Like you can't look at the few posts up? I just did the benchmarks after I asked you the question. I might not need the forum but people like you need guys like me.

This assumption of yours it what I had the most problem with.

"G92 is not less powerful. With equal bandwidth as GTX260 it would easily be as fast as GTX260."

We cannot reduce the bit width of the GTX260 to bring it down to the level of bandwidth of the 9800GTX+ right? Okay. What would the memory clock have to be on a GTX260 to bring down the memory bandwidth to that of the level of the 9800GTX+?

I'm terrible at this sort of math. Would someone care to?

Apoppin, do you have a GTX260? Preferably the 192sp version? I have a GTX280 but I think that is quite an improbable experiment unless I could disable shaders and ROP's. Which I doubt.

I have GTX260 216 and I did the benchmark already. GTX 260 memory has to be 629mhz which I posted the results. Here it is again...


1920x1080

Crysis 4xAF High settings dx10

Stock GTX 260 216 576/1350/999
33.14 fps

GTX 260 216 576/1350/629 70.4gb/s of bandwidth same as GTS 250.
31.34 fps

8800GTS @ 756/1836/1026 just shy of GTS250 1100mhz memory.
30.14fps


4xAF 4xAA High settings dx10

Stock GTX 260 216 576/1350/999
27.42 fps

GTX 260 216 576/1350/629
23.39 fps


1680x1050
Crysis 4xAF high settings

8800gts @ 756/1836/1026
34.43 fps

CORE REDUCTION 590/1836/1026
32.13 fps -7.2% difference

BANDWIDTH REDUCTION 756/1836/800
29.72 fps -15.8% difference

 

[AzN]

Originally posted by: BFG10K
Actually it doesn?t because I?ve demonstrated how a 8800 GT is faster than a 8800 GTS 640 despite having less bandwidth. Therefore the G92 is not primarily limited by bandwidth. Do you understand that if I reduce bandwidth but performance still goes up, it means bandwidth isn?t the limitation?

For that matter why do you constantly ignore the plethora of evidence that proves you wrong?

It's pointless to argue with you when I already gave you an explanation why the 8800gt beats the 8800gts 640mb. Although the 8800gt is bandwidth starved it's able to beat 8800gts 640 mb because it has more SP and TMU. With more bandwidth it would easily beat 8800gts 640mb by 2 folds and even further. While 8800gts 640mb has been saturated by bandwidth and more bandwidth would not help this card much like 2900xt.

Again a total skirt of the issue. If the 4770 isn?t bandwidth happy then by your admission bandwidth isn?t the primary limitation because it has less of it than the 4830, but it?s still faster.

Again do you understand the concept that if I reduce bandwidth but performance still goes up, it means bandwidth isn?t responsible for that increase?

Do you understand what bandwidth limitation is? It's when a card can't fully stretch it's performance because of bandwidth. Not when the card has more bandwidth so it should be able to beat the card. That's where you base your entire argument on and have a hard time understanding this concept.

My benchmarks largely show the same thing as Derek?s where the core clock showed the biggest difference. So in one corner we have mine, Derek?s figures, and a plethora of DX10 benchmarks that mostly show memory bandwidth isn?t the primary limiting factor.

In another corner we have Azn and his results; your results are a clear outlier.

Notice Derek's benchmarks. That's a GTX275. Your card you did the testing was on a Ultra. These 2 cards are both bandwidth saturated cards. A G92 is not which I gave you results on.

1680x1050
Crysis 4xAF high settings

8800gts @ 756/1836/1026
34.43 fps

CORE REDUCTION 590/1836/1026
32.13 fps -7.2% difference

BANDWIDTH REDUCTION 756/1836/800
29.72 fps -15.8% difference

So you were saying? I suggest you go pick up a G92 and go study up on it. This card severely bottlenecked by bandwidth.

Why? You?re the one claiming the G92 is primarily limited by bandwidth so I demonstrated how the 8800 GT (with less bandwidth) is faster than the 8800 GTS 640, even with AA. Therefore your claims are nonsensical given real-world results contradict them.

Again you have a hard time understanding what bandwidth limitation means. I tried to explain to you in many occasions and gave you plenty of examples. Not my fault you can't seem to understand it spit out rhetoric. All I can hope is that you try.

So in other words you?re claiming the G92 is bottlenecked primarily by bandwidth, but it doesn?t matter because the core is faster? Again how can that be if bandwidth is the major component holding back performance like you claim, but the part has even less of it of the 8800 GTS 640, yet is still faster?

The core isn't just faster it has more TMU and SP. I gave you plenty of examples to show you how 8800gt is able to outperform 8800gts 640mb.

Think of bandwidth like stretchable hose but it has a limitation how far it stretches. On the 8800gt it has been stretched to it's limitation while 8800gts 640mb hasn't been stretched at all. Since 8800gt has 3x texture fillrate it can fit more on that hose (memory bandwidth) compared to 8800gts 640mb that has 1/3 texture fill. Not to mention 8800gt has 35% more processing power than 8800gts 640mb.

Again I?ll ask for the third time whether you understand the concept of reducing bandwidth means the performance increase can?t possible be coming from it?

I think my explanation right above show you you can get more performance even if you lower bandwidth but this is besides the point. We are trying to find out if G92 is hindered by bandwidth. After all that is the topic on hand.

It?s like Keys says, you can?t provide a straight answer to anything that disputes your theories.

Keys says a lot of things. Primarily tell other people what to do even though he's persuaded by Nvidia and other forum members. I'm trying to make you understand by giving you examples. You ask questions because that's just how you post to win an argument. I did however gave you a straight answer after I gave you an explanation and examples.

The 4850 isn?t primarily limited by bandwidth either. Again this is trivial to prove by simply moving the sliders one at a time and seeing which has the biggest impact on performance (hint: it?s not memory).

4850 is sure limited by bandwidth. You move the memory slider up and you get better average and minimum frame rates. Not to mention 4850 core clocks are locked with their SP clocks. Now what you are describing is what gets more gains from the GPU. SP/core vs memory bandwidth. In this case combination of SP and core wins over bandwidth. Like I've already mentioned 4850 is pretty much a balanced card.

But you still haven?t addressed my question:

How is it that the 4850 competes with the 8800 Ultra with 4xAA and beats it with 8xAA when it?s bandwidth limited like you claim, but has around half the bandwidth of the 8800 Ultra?

How is it that the ROP improvements increase AA performance if you claim memory is holding things back?

How is it that the 4850 shows bigger increases from core clocks when you claim memory is the primary limiting factor?

LOL... Again with that same rhetoric. I already posted the answers already multiple times.

I don?t have a problem with Tech-Report, I have a problem with you putting 3DMark?s theoretical figures and your guesses ahead of real benchmark results that directly disprove your claims.

As for my ?partner?, he?s free to use any testing tool he pleases, as am I. I personally don?t use 3DMark because a tool that claims a 2900XT is faster than a 8800 GTX simply can?t be trusted.

But it was techreport who posted 3dmark theoretical figures and even posted a conclusion why bandwidth limits fillrate. I'm just citing what they posted and giving you examples.

Then you should do something about your partner since you can't trust 2900xt is faster than 8800gtx in 3dmark. If you don't know the reason why 2900xt beats 8800gtx in 3dmark 2k6 what makes you so qualified to do any kind of articles and make conclusions?

Wonderful, so we?ve established you don?t consider either the 4850 or the 4770 to be bandwidth happy, and you also think they?re bandwidth limited. Then again for the fourth time I?ll ask how you can claim this when a reduction of bandwidth still shows performance gains with AA?

If these parts are primarily bandwidth limited like you claim then performance should be going down with reduced bandwidth, but it isn?t. That?s the whole point; we?re reducing bandwidth, but performance is still going up. Therefore the primary performance bottleneck is elsewhere.

Already answered your question multiple times already. I feel like a broken record for crying out loud.

I don?t have a problem accepting other peoples? findings. On the contrary; other peoples? findings show me the 8800 GT is faster than the 8800 GTS 640 despite having less bandwidth, thereby telling me bandwidth isn?t the primary limiting factor as it currently ships on the G92.

Oh right but my benchmarks are outlier and your benches are legit. :laugh:

Here it is again...

1680x1050
Crysis 4xAF high settings

8800gts @ 756/1836/1026
34.43 fps

CORE REDUCTION 590/1836/1026
32.13 fps -7.2% difference

BANDWIDTH REDUCTION 756/1836/800
29.72 fps -15.8% difference

Complete opposite of your ultra results and Derek's GTX275 results that's been saturated by bandwidth. To answer OP's question. Yes G92 is hindered by memory bandwidth.

I?m not saying memory makes zero difference, only that it isn?t the primary limiting factor i.e. other clocks make much more difference.

And actually in many most of those titles the card is either tied with the 4870, or the 4870 is less than 10% faster, with many of gaps narrowing even further when the resolution is increased, despite having much less bandwidth courtesy of GDDR3.

Now we are getting somewhere...

In case of g92 memory bandwidth makes the most difference as shown by Crysis benchmark. Something you haven't tested although you tested this on your ultra and citing Derek's GTX275 that follows Ultra's tradition which show different results because it's core starved. After all those 2 cards are nothing like G92 that is starved for bandwidth.

Look at 4870 minimum frame rates compared to gainward 4850 that's clocked to 4870 core and maxed out memory clocks. You see 25fps on the Gainward @ 1920x1200 while 4870 minimum frame rate is 33fps. that's quite a bit of a jump than 10-15% average frame rate gains. That's 25% better minimum frame rates.

Again I?m not saying memory bandwidth makes zero difference, only that it?s not the primary limiting factor with most DX10 parts. We can see this with the 4770 reducing bandwidth over the 4830, yet still being faster with AA.

So you agree bandwidth and fillrate go hand in hand. Without the bandwidth fillrate would just sit there doing nothing. I've never said bandwidth is the primary factor with most dx10 parts. What you did say however is that bandwidth is a non-issue among dx10 parts although it is very much an issue.

 

[AzN]

Originally posted by: apoppin

Originally posted by: Azn

Originally posted by: apoppin
i am testing a GTS 250 right now

anything you want to see while i have it ?
- i am comparing it to 9800GT

Can you compare with a GTX260? Get to the bottom of this thread.

i don't have one and i am still waiting on GTX 275

i will temporarily trade someone for it; maybe for a 4870/1GB
-BFG10K has one and and a 4850 but you don't like his results

I wouldn't care if BFG did the benchmarks. This wouldn't be the first time where he does the benchmarks only to find out I was right all along. His ultra core/sp/bandwidth reduction tests were done because of our arguments. He said SP makes the most difference while I said it was fillrate and right combination of bandwidth would win over SP long as the game wasn't SP limited back in late 2007.

He only cites a card and then think all cards behave this way when in fact this isn't the case. Different cards have strengths and weaknesses and behave differently much like anything else.