BFG10K
Lifer
- Aug 14, 2000
- 22,709
- 3,003
- 126
Actually it's very much certain. See below for why.
No it couldn't.
The HL2 and CoD2 benchmarks are run on identical cards except one has 256 MB memory and the other has 512.
From those two results alone we can logically conclude the difference is coming entirely from the extra memory and therefore it does make a difference.
As for the Quake 4 and BF2 benchmarks, the performance gain is higher than the theoretical specs would allow between the cards.
Since we know for a fact there is no architectural changes to the 512 MB GTX that would allow more work per clock, we must logically conclude that at least SOME of the gain is coming from extra memory and therefore there is a difference between 256 MB and 512 MB there as well.
This is blatantly simple logic yet it seems to be escaping certain individuals who continue to post their ridiculous assumptions.
Again I'll repeat: you can't use a benchmark that shows no performance gain to infer there is no performance gain anywhere.
In any case, Anandtech's own BF2 benchmarks showed a performance gain higher than the specs would allow, just like B3D's Quake 4 benchmarks did. Therefore even Anand's benchmarks back what I'm saying.
I didn't say that at all, I merely speculated that this could be one of the reasons why he is not undertstanding the simple concepts presented in this thread. In reality there could be any number of reasons why he doesn't get it or why he refuses to get it.
Alright BFG, you've won the argument that the improvement is greater than zero from 512MB. I thought we were past that point, at least!
Some of the improvement is from the extra 256MB. Most of the improvement is from the higher GPU/RAM clocks. Furthermore, the improvements from the extra memory are only exhibited in certain games, and in those games only at very high resolutions and AA levels. Many of the tests reflect unplayably low framerates, making those results good for academic purposes, but not that useful to current gamers.
This isn't a court of law; you're not going to be convicted of any of your claims. It's merely a message board where I thought the point was to try to remain civil.
Isn't insinuation of stupidity just the intellectual's version of a schoolyard bully tactic?
Some of the improvement is from the extra 256MB. Most of the improvement is from the higher GPU/RAM clocks. Furthermore, the improvements from the extra memory are only exhibited in certain games, and in those games only at very high resolutions and AA levels. Many of the tests reflect unplayably low framerates, making those results good for academic purposes, but not that useful to current gamers.
This isn't a court of law; you're not going to be convicted of any of your claims. It's merely a message board where I thought the point was to try to remain civil.
Isn't insinuation of stupidity just the intellectual's version of a schoolyard bully tactic?
BFG10K
Lifer
- Aug 14, 2000
- 22,709
- 3,003
- 126
Not if the recording doesn't cover such a scenario. This is the fundamental point I've been making right from the start.
If your benchmark doesn't pick an area that is affected by texture swapping it can't possibly show a difference. To then make the assumption there is no difference period is a fatally flawed course of reasoning.
Honestly, I don't know how many times I need to repeat this point.
Now that we have timedemo scores showing blatant differences hitching is quite irrelevent, it's simply Dug clutching at straws and refusing to accept he's wrong.
I thought he admitted he was wrong last page when he revised/clarified his position from the extra RAM adding nothing to the extra RAM adding next to nothing.
I see what you're saying regarding the timedemos being contingent on the section benchmarked reflecting an area that is affected by texture swapping. This is a possibility that can only be verified by a thorough testing of the benchmark demos themselves, and analyzing the data.
All I can say in favour of the counterargument to your position is that timedemos are usually selected from the largest, most stressful sections of the game, meaning there is a good chance that if indeed swapping is occuring due to lack of video memory, it will be in one of the timedemos.
But I definitely can't provide you with the absolute certainty that you're looking for, so you're either going to have to dig around for the numbers you seek online, or test it yourself
.
Keysplayr
Elite Member
- Jan 16, 2003
- 21,211
- 50
- 91
But it is a pissing contest Snowman. And that's all it is when two people constanly go back and forth for eternity with no give on either side whatsoever. When I see that happening, I know that all its about is who can piss the further distance. You should know. You have been in a multitude of them as well as I have. But I choose not to anymore. Everybody knows who everybody is in here. Some people just have to be ignored if possible.
So, I won't "piss" with you on this anymore. You win.
Cheers mate.
BFG10K
Lifer
- Aug 14, 2000
- 22,709
- 3,003
- 126
Those questions have been answered multiple times. If the only difference between the cards is the VRAM and you refuse to accept this then what more do you want?
How can it be equivalent if it doesn't show a difference?
Not seeing something doesn't prove it doesn't exist.
Seeing something does prove it exists.
The hitching was brought up to illustrate why benchmarks don't always show a difference. You then asked for benchmarks to prove the hitching, when the benchmarks themselves are generally useless!
Logically you've contructed the following:
A can't prove anything.
B can prove something.
---------------------------
Therefore I want to use A to prove B.
The logic behind it is utter nonsense.
As for conjecture, like I said before if you had logged some gaming hours under your belt, had a basic understanding of 3D concepts, ran your own benchmarks & observed websites' benchmarks and listened to developers you'd know it was far more than just conjecture.
But it would still affect the minimums. And I even mentioned your point regarding averages in my post that in a longer benchmark it would get averaged out...
It may be a pissing contest, but many of us (myself included) never dropped our trousers, and I think Dug zipped up a couple pages ago. BFG seems to have an endless flow - maybe he had video card Gatorade for breakfast
.
kylebisme
Diamond Member
- Mar 25, 2000
- 9,396
- 0
- 0
Possably, but then possablity the minum might be the result of something other than the amount of ram on the cards and be the same for both, while the one with not enough ram to keep everything on the card at once will still have more frequent drops that while not lowering the minimum would defenatly make for less smooth gameplay.
If the card with less RAM has frequent drops, then it will have a lower average framerate.
If it only stutters on the 256MB card, then obvious stutters should be reflected by conspicuously low minimums. This should be easily verifiable as long as the 512MB card doesn't get single digit minimums as well, or on a graph of the fps such as what HardOCP does. A stutter should be reflected by a dip down to almost zero, or at least a drop drastically sharper than on the 512MB card.
kylebisme
Diamond Member
- Mar 25, 2000
- 9,396
- 0
- 0
IF one card gets a constant 100fps for a 1 minute long benchmark, and the other card gets 100fps 59 seconds but drops to 10fps for a half second twice during the benchmark; what is the difference in average framerate?
If it only stutters on the 256mb card then obviously it will at least a drop drastically sharper than on the 512MB card, eh? You just talked yourself in a circle there, and in doing so you totally missed my point. Or maybe you were trying to miss my point and that lead you around in a circle, either way my point still stands.
Snowman, no offense, but I have absolutely no idea where you are going with your argument, so I'll need to ask you to clarify! It is you who have talked in a circle, I think my position is pretty clear, and I will elabore it even further.
There are two cases that support your argument, and a third which does not.
1. The 256MB card does not really stutter.
2. The 256MB card stutters, meaning minimum frames will be affected. Average frames are not affected tangibly since it happens infrequently, but the stutters can be seen by observing an average framerate graph over time, such as the ones HardOCP uses.
3. The 256MB card stutters, affecting minimum frames. The timedemo is short and/or stutters relatively frequently, so average framerates are affected as well.
Your hypothetical example is totally meaningless, and does not reflect real gameplay.
Unless you are being limited by something (eg Vsynch), having a constant 100fps for 59 seconds, but two drops to 10fps does not reflect real gameplay in any way, shape or form. Take a look at HardOCP's graphs for what in game performance is like: it fluctuates tremendously, because the scene is always changing! here's one.
I have laid out a methodology for how to check for stuttering: compare the graphs of the 256MB card to the 512MB card, and see if there are any dips that drop way down to near zero for the 256MB, where the 512MB does not suffer nearly as much.
----------------
Why did you say I talked in a circle? Where and how did I do that, exactly? To give me the benefit of the doubt, re-read my post a couple of times to make sure I made a mistake; I've read it a few times and it's pretty clear to me.
But just to give you the benefit of the doubt as well, I will reword my quote so you can see my point clearer:
If only the 256MB card stutters, then you will be able to see dips into very low fps (probably under 10fps) for that card, while the 512MB exhibts a much smaller dip.
These stutters on the 256MB card will be easy to check by comparing the graph of a 256MB GTX and a 512MB GTX, as long as the 512MB card doesn't also get single digit minimums at these points as well.
An example of this methodology having problems discriminating between stuttering or just low FPS would be for example in FEAR, where both the 256MB and 512MB GTX actually dip to 0fps at points. In this case there is no way to tell by looking at the data alone whether the drops to 0fps are caused by stutters or simply by low fps (ie too much pixel shader data in the frame than can be rendered above 0fps). In this case, since both cards drop to 0fps at around the same points, it's reasonable to assume that the same factor is affecting both cards: either both the 256MB and the 512MB cards stutter in FEAR, or neither do.
Hopefully this clears it up for you. I've taken awhile to try to break it down, so please do me the courtesy of reading it
.If you're still scratching your head, then I have to say that I'm starting to empathize with BFG10K
. I'm also curious as to why you would speculate that I would try to deliberately try to decieve you (see the part I bolded of your quote). Have I ever done this in the past to you? I certainly can't think of any examples.----------------
Also please note that the HardOCP graphs are not the best possible example, since they run different settings for different cards that they judge as "playable" settings. Still, the HardOCP data is by far the best data for this purpose that I know of, since I can't even think of another site that graphs their results like this...
I'll also fix this post for clarity:
Get rid of the word more. That should make it obvious what I was trying to say:
Get rid of the word more. That should make it obvious what I was trying to say:
do any of these COD2 benchmarks that you guys are quoting say that the reviewer used "extra" for all the texture settings? if they left it at "automatic" the game will lower the settings to fit 256meg cards. someone needs to record at least a minute long demo heavy gameplay of COD2, then force the texture settings to extra, run the demo with a gtx 256 and then with a gtx 512 clocked down to the speed of the 256, while logging fps with fraps so we can see the mininmum framerates (hitching and texture swapping) and not some average number skewed by 100+ fps when staring at a wall. this will end the argument once and for all.
look what I found: a review that actually tests video memory usage (I didn't know such utilities existed).
So apparently, Riddick, HL2, Far Cry, Doom3, 3dmark05 do not exceed 256MB of video memory usage at 16X12, 4X AA 16X AF, FEAR does.
Yet, FEAR does not show any improvement from this discrepancy because at these points it is also GPU limited!
So apparently, Riddick, HL2, Far Cry, Doom3, 3dmark05 do not exceed 256MB of video memory usage at 16X12, 4X AA 16X AF, FEAR does.
Yet, FEAR does not show any improvement from this discrepancy because at these points it is also GPU limited!
Haven't read this thread, but you all might appreciate these 256 vs. 512MB benchmarks from Hardware.fr. They underclocked the GTX 512 and XT to GTX 256 and XL clocks, respectively, and then graphed the % performance gains from the extra memory w/o AA, w/AA, w/HDR, and w/HDR+AA. Surprising to see the NV cards don't gain as much as the ATI cards. That could just be b/c NV's drivers aren't optimised for 512MB framebuffers yet, as no one's used the 89.94 drivers (which are supposed to be optimized for the GTX 512).
The English version of that review should be posted at BeHardware.com in a few days or so (BeHW is HW.fr's English site), if you're interested in the rest of it.
Edit: Good catch, jiffy. Whodathunk 16x12 w/4xAA wouldn't top 240 for most games? I guess game devs budgeted for that, AA and all. I wouldn't have thought they'd account for AA's extra mem usage.
Edit Edit: It's possible, CM--there're always be more optimizations. Dunno for sure if any will come of the extra mem, though.
The English version of that review should be posted at BeHardware.com in a few days or so (BeHW is HW.fr's English site), if you're interested in the rest of it.
Edit: Good catch, jiffy. Whodathunk 16x12 w/4xAA wouldn't top 240 for most games? I guess game devs budgeted for that, AA and all. I wouldn't have thought they'd account for AA's extra mem usage.
Edit Edit: It's possible, CM--there're always be more optimizations. Dunno for sure if any will come of the extra mem, though.
Cookie Monster
Diamond Member
- May 7, 2005
- 5,161
- 32
- 86
Maybe that couldve been the reason how it performed like 1fps less then the GTX 256mb tested in firing squad.
What you are saying is that we could be seeing more perfomance out of the 512mb GTX?
this is deja*vu every damn generation of cards i swear,..lol
heres what i think(not that im claiming its important by any means,..hehe)
idsoftware engines spawn games to come ,..and thats a fact so how doom3 reacts is an indication of how quake4 acts to how prey will act to how return to castle wolfenstien 2 will act whatever... etc etc and the fact is ultra quality plays well with 512mb,.. high quality with 256mb,...doesnt take a genious and high quality in doom3 takes 246mb(i think it showed),..thats cutting it close,...but kudo's to idsoftware for the refining to max out just enough under,...i play quake4 and doom3 in high quality(though in quake4 high qualty is 4x AF instead of doom3's 8x AF,so i simply change it to 8x AF in the config in high quality),...i simply dont like how the game acts in ultra,.yeah sure its playable i guess some may think,..but my version of "playable" may be different from his or hers or the next guy,(which is why the benches showing ultra at b3d show the 512 so much faster,..because the 256 is stalling when new uncompressed textures need to be loaded,the 512mb just moves along its merry way,making the system less confused as to why they have to fill in for somethin that isnt in thier full job discription)...to me,..my playable is "flawless 100% silk!" to the point of liquid frames,..and quake4 on a 256mb card(or 2) cannot do that in ultra mode,...period,..however at high quality 2x Super Sampling/GammaAA 8xAF 1280x1024@85hz openAL eaxHD audio(for single player and multiplayer to play identically),..straight liquid!
512mb cards will eventually replace 256mb cards as developers demand larger and more uncompressed texture files,...it starts with one game,..leading to 2 games leading to 4 and more until the people with 256 cards eventually have to play in "low quality textures" when 1024mb cards are the standard,...it comes down to this,..yes the gtx 512mb is faster in every game because of its clocks and its pcb modifications nvidia has made ,..and because of the 512mb's of memory it also can handle textures twice the size of a 256mb card so it can be visually more impressive at the same time effortlessly sailing on its clock speeds through its massive 24 pixelpushing pipelines,letting the rest of the system to do its thing without having the vid card borrowing time and space from it to render a task,..so games that only use 205mb of vid mem with max settings,....yeah a 512mb card wont do anything more than a 256 would if the cards are otherwise identical
heres what i think(not that im claiming its important by any means,..hehe)
idsoftware engines spawn games to come ,..and thats a fact so how doom3 reacts is an indication of how quake4 acts to how prey will act to how return to castle wolfenstien 2 will act whatever... etc etc and the fact is ultra quality plays well with 512mb,.. high quality with 256mb,...doesnt take a genious and high quality in doom3 takes 246mb(i think it showed),..thats cutting it close,...but kudo's to idsoftware for the refining to max out just enough under,...i play quake4 and doom3 in high quality(though in quake4 high qualty is 4x AF instead of doom3's 8x AF,so i simply change it to 8x AF in the config in high quality),...i simply dont like how the game acts in ultra,.yeah sure its playable i guess some may think,..but my version of "playable" may be different from his or hers or the next guy,(which is why the benches showing ultra at b3d show the 512 so much faster,..because the 256 is stalling when new uncompressed textures need to be loaded,the 512mb just moves along its merry way,making the system less confused as to why they have to fill in for somethin that isnt in thier full job discription
)...to me,..my playable is "flawless 100% silk!" to the point of liquid frames,..and quake4 on a 256mb card(or 2) cannot do that in ultra mode,...period,..however at high quality 2x Super Sampling/GammaAA 8xAF 1280x1024@85hz openAL eaxHD audio(for single player and multiplayer to play identically),..straight liquid! 512mb cards will eventually replace 256mb cards as developers demand larger and more uncompressed texture files,...it starts with one game,..leading to 2 games leading to 4 and more until the people with 256 cards eventually have to play in "low quality textures" when 1024mb cards are the standard,...it comes down to this,..yes the gtx 512mb is faster in every game because of its clocks and its pcb modifications nvidia has made ,..and because of the 512mb's of memory it also can handle textures twice the size of a 256mb card so it can be visually more impressive at the same time effortlessly sailing on its clock speeds through its massive 24 pixelpushing pipelines,letting the rest of the system to do its thing without having the vid card borrowing time and space from it to render a task,..so games that only use 205mb of vid mem with max settings,....yeah a 512mb card wont do anything more than a 256 would if the cards are otherwise identical
You got to be careful when underclocking 1 card to compare to another. Chances are the memory on the faster card will be running at slower timings (e.g. my pc could be using DDR 400 at 2-2-2-6 or DDR 600 at 3-4-4-8), when you underclock as far as I am aware you don't change the timings (e.g. for my pc example my DDR600 underclocked to DDR400 would still be using the slower 3-4-4-8 timings). Hence when comparing say a 7800 256mb to a 7800 512mb underclocked to the 7800 256mb settings the underclocked 7800 512mb cards memory is probably running slower then the 7800 256mb one.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter