Originally posted by: Azn
incompetent like you finger nail explanation and tried to make sense that ROP is the most makes the most performance impact in games? :laugh:
No, that I shouldn't have to detail everything that "ROP" entails when cutting ROPs also results in a loss of those logical units.
Uh. Yes you did. :roll: You also said bandwidth didn't matter either. :laugh:
Keep in mind though, the Shader Core speed difference doesn't produce nearly the results in performance as a straight increase to core or memory clocks. I think AT did a comparison in that OCZ 8800GTX review and the difference wasn't linear, maybe 20-50% improvement relative to Shader Core increase.
I posted that on 2/27/2007 almost two years ago, so what were you saying again? You were probably still talking about vertex shaders being more important than pixel shaders on your crippled 7300GS back then.
WRONG!!! G92 has minimal impact when you raise core frequencies. This has already been tested on 2 G92 cards which you have been ignoring with the evidence provided. 8800gs and 8800gts. G94 is a balanced product so you would need to raise everything. Not just the core.
Minimal impact from core frequencies? Really? Is that why G92 GT, to GTS, to GTX to GTX+ and 9800GX2 compared to the SLI G92 solutions all show significant gains from core clock increases, despite similar 1000-1100MHz bandwidth? That's 600MHz to 738MHz, a 23% difference in core clock. Are you saying the performance difference between G92 is closer to the minimal differences in SP and memory frequency, or closer to the 23% differences in core clock? G92 benefits
less than G80 and GT200 from core clock frequencies, but its still has the greatest impact on performance.
ROFL... I even provided evidence back then which you ignored because it just made you look foolish with your ever retarded ROP makes the biggest performance difference. Here it is again.
http://techreport.com/r.x/rade...850/3dm-color-fill.gif
2900xt 3.8 Gpixels/second
3870 3.1 Gpixels/second
Why does it outperform in High color fill? Oh that's right ROP is tied down to the memory controller. In this case 2900xt had more bandwidth so it has more pixel performance than 3870. That 20% advantage in ROP performance equates to 1% performance difference in the real world. :roll:
Rofl, 3DMark synthetics again. ROP performance between the parts is almost the same, only 4% difference, so you can stop with whatever nonsense about ROPs performance being tied to memory controllers:
R600 and RV670 specs
But this isn't surprising from someone who claimed:
Bigger bus is just better. It's wider and able to hit some peaks a smaller bus can't sustain. when discussing 3870 and 2900XT, while linking a bunch of garbage 3DMark theoreticals (again, about as useful as counting frames on a loading screen) while ignoring
actual game benchmarks from the same site that showed minimal difference between the parts.
ROFL. You've proven to yourself that bandwidth mattered even at low resolution when you applied AA which you said it wouldn't. You even said it yourself... GTX 280 bandwidth was being wasted because it didn't have enough fillrate. When you downclocked your memory even @ 28% GTX 280 still has enough bandwidth to run efficiently. Now try downclocking it to 9800gtx+ memory bandwidth and see your card perform more like 9800gtx+ than GTX 280. :brokenheart:
Except the comparison was never between the 9800GTX+, I already know the GT200 is always faster than the 9800GTX+. The point of the exercise was to show memory bandwidth has much less impact than core clock increases to show the difference between bandwidth on GTX 295 and GTX 280 was less relevant than the loss of ROPs. My results clearly show that.
Your arguments are hypocritical and illogical at best. If someone actually read your arguments to Nvidia or AMD engineer or in a court they would surely laugh in your face. :laugh:
Like how a 27% decrease in memory bandwidth results in a 3-8% difference in performance? Anyone would clearly see that's less than the 15-25% difference between GTX 295 and GTX 280 SLI.
It was bandwidth limited @ 1680x1050. What makes you think it's not bandwidth limited @ 1280x1024? :laugh: Do you want me to test Crysis again but @ 1280x1024 with my 8800gts? Don't try to tell me about bandwidth limitation when you don't even have the slightest idea. You made a wild guess and you were wrong. Everyone can make a guess and be wrong. It doesn't mean you are stupid.
LOL? Uh, maybe because reducing resolution means a reduction in bandwidth requirements? There's no wild guessing here other than what gibberish will come out of your mouth next.
Bandwidth plays a role whether it be low or high resolution. What you haven't figured out is that more fillrate you have more bandwidth you need to run efficiently. Just because you have certain amount of bandwidth doesn't mean you aren't bandwidth limited.
Yes it plays a role but it clearly plays less of a role at lower resolutions as there's simply fewer pixels per frame, which reduces how much data passes to/from the frame buffer. I'm not arguing efficiency, I'm arguing which factor has the larger impact and clearly its not bandwidth. Bandwidth is only an issue if you clearly don't have enough and its completely crippling performance so that gains in other areas show no gain.
Your GTX 280 or GTX 260 isn't bandwidth starved. You and I both can agree to that but you keep implying g92 isn't bandwidth starved when it has texture fillrate closer to GTX 280 is ludicrous..
Its not ludicrous when G92 always benefits more from core/shader clock increases than increases to memory bandwidth. If G92 was so bandwidth starved as you claim, simply increasing memory bandwidth by itself would yield a bigger gain than core/shader increases, but it does not. There's at least 5 G92 parts that show this to be the case, scaling from 600 to 750MHz with memory clocks locked at 1000-1100MHz.
That's all you can come up with (BS) when evidence is provided for you. :disgust:
I pretty much explained this in my previous post in this thread and 9600gt thread with Keys. Here I will explain it again since you seem to think I couldn't explain it.
G92 is bandwidth starved and I've pretty much said this from the very start while G94 is a balanced product. With AA 9600gt comes within 10% of 9800gt because of bandwidth limitations of the 9800gt but when you compare raw performance without AA 9800gt is roughly 20-35% faster than 9600gt when bandwidth restrictions were less critical. Now if both cards has all the bandwidth it needs 9800gt would be much faster with or without AA.
Yep, it is BS because you haven't and still can't explain away the G94. But this should help straighten things out for you. You should be familiar with it, as you've referenced it in the past:
Expreview G94 to G92 GS
Please explain how a G92 card with more pixel/texture fillrate and shader performance is able to perform within 5% of G94, despite a 33% reduction in bandwidth? You said G92 didn't have enough bandwidth to satisfy its texture fillrate, yet here's a G92 part that shows no adverse effects from less bandwidth.
Also please explain to us how G94 with 33% fewer SP and TMUs is able to stay competitive with G92 GS if SP and TMU are the most important aspects of performance. Yes it has more bandwidth, but that shouldn't matter since there's less texture fillrate to begin with.
ROFL. you are the one here spreading Nvidia marketing jargon.
I've proven almost everything you claimed my benchmarks and your benchmarks. :laugh: If that makes me a troll to you I guess I am.
Explaining the differences in architecture and how they corrolate to real-world performance between the parts is marketing jargon? More like squashing misinformation from someone who has repeatedly demonstrated incompetence and the inability to absorb readily available information.
Because you need bandwidth to take full advantage of that texture fillrate not to mention AA needs more bandwidth.
But the 4870 has less texture fillrate than 9800GTX+ and beats the 9800GTX+ even without AA, so additional bandwidth shouldn't be an issue, yet the 4870, like the 260 runs circles around the 9800GTX+. Weird.
I wouldn't say circles now. It's roughly 20% faster.
Yep, its always faster despite lower texture fillrate theoreticals. But is this surprising given the GTX 260 follows the same pattern as well?
Gotta love how you throw up all these theoretical numbers which never bear out in real world applications. Just shows theoreticals are just that, theoretical and ultimately useless.
Exactly my point. 9800gtx+ is bandwidth limited to a point increasing core had very little impact on performance as shown on my Crysis benchmark. Did I say shader didn't impact performance? Of course I didn't I said texture and shader makes bigger impact than ROP. :roll:
Except that's clearly not true as G92 has gone through 25% increases to core from 8800GT to 9800GTX+ with a minimal 10%-15% increase in memory speed, yet it still scaled signficiantly with core clock increases.
ROFL. What hint? That you are making shit up? RV770 can only do 16 pixels per clock at 32bit color and GTX 280 can do 32 pixels per clock. Only time it can write/blend 2 pixels per clock is with MSAA or 64bit color consisting of HDR scenes. :laugh:
http://techreport.com/r.x/rade...ender-backend-ppcs.gif
Ah yep, I hadn't referenced that graph in a while and was still thinking of parts that could only write/blend at half speed. However it still shows 4870 is still faster than 9800GTX+ even without AA, even though it can write similar pixels per clock.
Grasp straws is right. You were trying to compare L2Cache to GPU logical units which I corrected your flawed argument. :gift: Depending on what you were doing those units can have bigger performance impact. That's the point. But in a game ROP doesn't pull nearly as much frames as texture or SP would long as you weren't limited anyway.
LOL, BS, you linked a die shot to GT200 and claimed a ROPs that only appeared to be 25% of the die couldn't have the biggest impact on performance because they were only 25%, at which point I clearly illustrated that equating die size to performance was clearly a flawed analysis. See, the difference is I was using L2 Cache to clearly show die size is not proportionate to performance, where you tried to show that it did with GT200. So do you think ROP performance and size is still a relevant comparison, or not?
Why don't you test your GTX 280 while you are at it. Try to see if you get the same results as I did. Of course not because GTX 280 is core hungry while my GTS is bandwidth hungry. Now if your GTX 280 showed same results as I did than I would be wrong and you would be right but then again you are too chicken shit to test your GTX 280 because you already know what the outcome will be. I would test more games but majority of games out there don't have a built in game benchmark. I don't have WIC and demo doesn't work with the latest drivers. GTA4 shouldn't be a factor because that game is CPU dependent.
Blah blah blah. I did test Crysis, it showed an 8% difference from 27% reduction in memory clocks. I also tested 4 other titles that showed 0-5% difference. Its obvious you're too "chicken shit" to stray away from Crysis as its one of the few titles that is bandwidth intensive enough to show a significant decrease in performance from a reduction in bandwidth at a lower resoluton like 1680, yet its still only 8%. Run a straight line and use FRAPs for all I care, just don't hide behind a lame excuse like "I don't have enough games with a built-in benchmark so I can only use Crysis".
Look at your minimum frame rate in Crysis and WIC at 21% and 14%. I don't know how much more relevance you need. :roll: Now clock your GTX 280 memory clocks to 550mhz and do the benches again and compare it with 9800gtx+. I dare you to post your results.
I'll only do it if you double dog dare me! :laugh: What about minimum frame rates? You claimed the difference between GTX 295 and GTX 280 SLI was due to bandwidth, they used averages so minimums were never in question. My benchmarks clearly showed a 3-8% difference from a 27% reduction in memory, proving your theory wrong while showing memory bandwidth was indeed less significant at lower resolutions, just as I stated.
ROFL. you never made any claims? WTF??? This post alone you were taking jabs about GTX 260 vs 9800gtx+ being not bandwidth starved for performance and you made no claims? W@W!! I'm speechless. Pathetic. High resolutions? I tested 1680x1050 with my 8800gts and 1440x900 with my 8800gs. :roll:
What do your results with 8800GTS and GS have to do with my claims about GTX 260 and 9800GTX+? If you're going to try and illustrate they're bandwidth limited, then show how much of an
increase in performance is gained from an
increase from memory clocks. Increasing bandwidth requirements and then reducing bandwidth doesn't prove your point about being bandwidth limited. That'd be like saying 9800GTX+ is bandwidth limited, so to prove this, I'm going to clock memory to 128-bit like an 8600GT.
That's because you don't have the slightest clue when it comes to GPU. I think you are no better than any of the new forum members asking which card is faster when it comes to GPU except those newbies are willing to learn but you. :laugh: Nazi's used to believe something too. That didn't make them right.
Ya, you're an imbecile for equating anything on a tech forum to Nazism. But I'm sure all those new forum members are tickled pink to hear garbage like "GTX 260 isn't much better than 9800GTX+" or "9800GTX+ would stomp GT200 with more memory bandwidth" or "9800GTX+ is actually faster than GT200, but not really" And you still can't explain away the G94. :laugh:
Employee discount I bet for spreading Nvidia marketing jargon. :laugh: There's no way I'm paying more than $200 for GTX 280. The card performs fine for older games but new games like Crysis it still chokes not to mention ridiculous power requirements for such a slow ass card.
Yeah, employee discount from one of my other employers, Microsoft. Cash back employee benefit program was great, you didn't even have to be an employee to get in on it.
As for marketing jargon, slow ass card. LMAO. Funny coming from someone who has a long history of only using "slow ass cards". But keep pouring it on my GTX 280, it can take the criticism, really. LOL.
I bet you downclocked your core and SP clocks by now only to find out I was right all along. That 10% turned into 15-25% with AA. ROFL... No wonder you don't want to show results. Such a silly boy.
I bet....you have no clue what you're talking about. So you're saying GT200 gets more than a linear increase from core clock increases? I'm a big fan of overclocking Nvidia parts and the performance gain they give, but even I can't make that claim. A 4% increase/decrease is going to yield a 4% increase/decrease at best, which isn't enough to make up for the 15-25% difference between GTX 295 and GTX 280 SLI.
Knowledgeable observers like who? Yourself? :laugh: You've also linked bit-tech benchmarks with GTX 260 SLI beaing GTX 295 too when it's not theoretically possible. :laugh:
Like Derek and Bit-Tech who both stated ROPs or bandwidth were bottlenecking GTX 295, and I'm sure most other review sites as well. And it is theoretically possible that GTX 260 SLI beats GTX 295, which just shows TMU/SPs aren't the most significant factor when it comes to performance.
Exactly no proof. When you lowered your bandwidth your minimum frames were all over the place dropping as much as 21%.
It showed a 27% reduction in bandwidth resulted in 3-8% difference in FPS and proved my point that bandwidth alone wasn't enough to explain away 15-25% differences between GTX 295 and GTX 280 SLI.
Only if it worked that way. Poor Chizow. :frown: When you have combination for both core, SP, bandwidth you get much greater drops than just clocking each clocks separately. Of course you would have to be knowledgeable with GPU to actually know this but then again you don't know anything about GPU so.
Yes, you get bigger decreases when you lower all simultaneously but if you decrease all equally, the drop in performance will not be greater than the % that you decreased them, the difference should be very linear. When you change them individually, you can then come to the conclusion which factor has the greatest impact on performance by comparing actual clock decreases to % drop in performance. And this is clearly illustrated when I decrease memory bandwidth 27% and only see a 3-8% drop in performance.
This is really elementary stuff when it comes to overclocking, enthusiasts have been making these comparisons for years and have generally found memory clocks historically have much less impact on performance than core clock. I guess introducing that third shader clock threw you off somehow lol.
Of course I did but it's okay. I can easily clock my 8800gts to stock clocks and only raise memory clocks and core clocks separately to show you same kind of results but then again it would make you look arguments look silly. You wouldn't want that. Of course not. You would will just say it doesn't prove anything when the evidence is before you.
Just as long as you understand why your initial benchmark was flawed, I don't really care if you run the benchmark again or not because I already know you're going to see less return on the increase to bandwidth.
Benchmarks talk chizow barks.
Yep and 27% talks, 3-8% barks, really softly.
Ruff ruff. errr.... :laugh: Useless without benches.
There's about 20 GTX 285 benchmarks that show a 10% increase in clocks result in <=10% performance. No need for me to prove a 4% difference in clocks is going to result in a <=4% difference in performance when that information is readily available and verifiable fact for anyone who has actual experience with the parts (and isn't completely incompetent, like you).
Make some sense. Why would I bench a 3dmark fillrate test and measure fps? In 3dmark fillrate test the screen isn't black with a logo. It's testing peak texures with multiple layers of textures.
Because it shows GPU workload and not theoreticals ultimately determine fillrate and FPS. It also shows pixels are still being drawn regardless how blank/empty you think they are.
What does downclocking the core of your GTX 280 have anything to do with what was stated? You said memory clocks wouldn't matter but it showed huge drops in minimum frame rates and in average frame rates by 3-8%. As for core I've already mentioned this numerous times in this thread. Your GTX 280 is core hungry and my 8800gts is bandwidth hungry. :roll:
No, I said bandwidth was less significant at lower resolutions even with AA, and it clearly is. Now going back to my original point about ROPs being more important than SP and TMU and bandwidth, do you think downclocking the core clock by 27% would result in a miniscule 3-8% drop in performance? Think about this a second before you reply, given you've already questioned how much a 4% core clock difference would make.
In lower resolutions you need less vram. Come on chizow you show know this. I thought you were brighter than this but apparently you are not. I only brought it up because GTX 260 having more VRAM for higher resolutions and or with AA. When those vram abnormally isn't effected I implied it's not much faster.
I'm well aware VRAM is less significant at lower resolutions, which is why I asked why you brought it up when we were specifically discussing lower resolutions where VRAM would be less of an issue. Oh right, its because you're not smart enough to focus on what you're arguing.
So you are implying all cards are built like your GTX 280 so all cards behave the same to core frquency like your GTX 280. How retarded of you.
Yep, all Nvidia parts since G80 that I've owned have behaved similarly where they benefit the most from increases to core clock over shader or memory. This is obvious to anyone who has used these parts, the fact you haven't come to the same conclusion, despite experience with only older/crippled parts would indicate you're incompetent.
Significantly? :laugh: I wouldn't say all that when it only performs 5-10% faster.
This is where the extra shader clocks and core clocks come in. Although it's bandwidth limited it still able to pull more fillrate with high core clocks but not as much as a card that has combination of bandwidth and core efficiency.
LMAO finally, progress. Its more than 5-10% from 8800GT or GTS to 9800GTX+, its closer to the 15-25% core/shader difference between the parts. Its obvious core/shader still has a greater impact than bandwidth despite your claims G92 was bandwidth limited to begin with.
If I prove it will you acknowledge that g92 is bandwidth starved? Of course not. You will say something as stupid as Crysis is bandwdith starved and we are back to square 1. It wouldn't matter anyway because I already have the results ready if you are willing to acknowledge you were wrong.
Sure you do, post them. I already know a memory bandwidth increase will result in less of an increase than than core/shader increases. Its really simple, which of the 3 yields the bigger increase? That means increasing each individually, not keep 2 factors as high as they can go, then dropping the one you claim will gain from an increase by 25%.
Did you even understand what I had originally posted? You are disagreeing with something you didn't even understand? :roll:
I mentioned that if 8800gs showed 10% dropped by downclocking memory, 8800gts should show bigger drops at the same resolution. In this case my 8800gts dropped 16% while 8800gs 10%. Why you ask? Because 8800gs was bandwidth starved in the first place. If you add 25% more of everything in a bandwidth limited card you would get amplified results.
That might be true, unfortunately you didn't test them at the same resolution, which was my point to begin with. Do you think the 8800GTS would've showed a bigger % drop at 1440 than the 8800GS? Maybe, but not for any certainty given you've decreased resolution 36% and decreased bandwidth 33% as well.
ROFL. You said you would show me benchmarks of exact same card performing same at lower resolution at lower bandwidth in fact there are no benchmarks available on the web which you have lied about. Then say go research? Pathetic feeble attempts. :laugh: I can easily do this with my 8800gts. Cut my bandwidth to half and make it craw but what's the point. :roll:
Yep pretty sure I came across a benchmark with it, of course you claimed it was a BS part when it clearly isn't. And yes cutting your bandwidth in half would prove the point that insufficient bandwidth at higher resolutions can cripple performance.
You cool you drop lot of money on gaming hardware. I'm 33 years old. I game on the side. Nothing major but I do it as a hobby but that's me. I'm content with what I have because that's all I need.
Oh wow you said that as if I cared. It just makes your comments and comparisons about GTX 280 that much funnier.
ROFL... Now you are denying... Seriously Lame. :thumbsdown:
What am I denying? One of the first things I said was not to bother pulling Crysis benches as I'm well aware its responsive to core/shader/memory clocks individually, and much moreso than other games.
Facebook
Twitter