40nm Battle Heats Up

[chizow]

Originally posted by: Azn
It won't matter anyways. AMD can easily raise their performance level by adding more TMU and SP into their design without better processing technology. You do know RV770 is slightly smaller than 9800gtx and much smaller than Nvidia's behemoth GT200. Your whole logic makes very little sense if any sense at all other than you are swayed by Nvidia. It's quite amusing.

That'd be the expectation, as stated in my OP with news of a 40nm RV790.....

Also, G92b on the same 55nm process is smaller than RV770 at roughly 231mm^2 to 250mm^2. so I'm not quite sure how my logic is swayed by Nvidia.....

Originally posted by: Azn
It's quite proportionate considering it has more vram, whole lot of bandwidth, more rop, SP. the fact that you think ROP is the driving force is laughable.

Let's not forget GT 295 doesn't have full 240SP either nor does it have 80TMU which Anandtech forgot to mention.

But GTX 260 doesn't have proportionately more SP capability, I've already clearly shown the two factors you claimed had the biggest impact on performance, SP and TMU clearly do not. I showed this with G80 to G92 and now G92 to GT200.

Also, GT295 does have the full 240SP and 80TMUs....which again brings us back to ROPs, bandwidth and VRAM.....

 

[nosfe]

yes, the G92b is smaller than the RV770 but you're forgetting that ati uses a different method for increasing yields, those chips have 900shaders of which 100 are eaten up to make the yields high while on the G92b they have to sell crippled versions of those cards when some of the chip goes bad. which of the methods is superior? i don't know and i don't care, but it's worth noting that when talking about die size

 

[AzN]

Originally posted by: chizow

That'd be the expectation, as stated in my OP with news of a 40nm RV790.....

Also, G92b on the same 55nm process is smaller than RV770 at roughly 231mm^2 to 250mm^2. so I'm not quite sure how my logic is swayed by Nvidia.....

What? Your whole AMD doesn't have an answer post has Nvidia persuasion all over it.

But GTX 260 doesn't have proportionately more SP capability, I've already clearly shown the two factors you claimed had the biggest impact on performance, SP and TMU clearly do not. I showed this with G80 to G92 and now G92 to GT200.

Also, GT295 does have the full 240SP and 80TMUs....which again brings us back to ROPs, bandwidth and VRAM.....

It's quite proportionate considering it has more vram, whole lot of bandwidth, more rop, SP where 9800gtx is bandwidth starved however you look at it. the fact that you think ROP is the driving force is laughable

Let's not forget GT 295 is clocked lower than GTX 280 in SLI which Anandtech forgot to mention and mostly tested bandwidth limited situations.

 

[chizow]

Originally posted by: nosfe
yes, the G92b is smaller than the RV770 but you're forgetting that ati uses a different method for increasing yields, those chips have 900shaders of which 100 are eaten up to make the yields high while on the G92b they have to sell crippled versions of those cards when some of the chip goes bad. which of the methods is superior? i don't know and i don't care, but it's worth noting that when talking about die size

Yep and G92b also has the integrated NVIO chip that is estimated at something like ~100-150m transistors. The point was someone was trying to make a comparison to die size and number of transistors and missed the boat completely.

Originally posted by: Azn
What? Your whole AMD doesn't have an answer post has Nvidia persuasion all over it.

Uh, no I said AMD still has not responded to GT200, and now GT200b. And they haven't.

Originally posted by: Azn
I had edited my post before you replied.

It's quite proportionate considering it has more vram, whole lot of bandwidth, more rop, SP where 9800gtx is bandwidth starved however you look at it. the fact that you think ROP is the driving force is laughable

Let's not forget GT 295 is clocked lower than GTX 280 in SLI which Anandtech forgot to mention and mostly tested bandwidth limited situations

Maybe you should edit again, because you're just digging that hole deeper and deeper now. If GTX 280 SLI comparisons aren't good enough because of a 25MHz difference in core clock and difference in bandwidth, just compare it to the GTX 260 Core 216 SLI, which is clocked identically to the GTX 295 at 576/1242/1000. The only differences between the GTX 295 and GTX 260 core 216 GPUs are 24SP and 8TMUs.

So once again, does the GTX 295 perform more like GTX 260 SLI? Or GTX 280 SLI. Differences are evident even in non-bandwidth/VRAM limited resolutions and settings. Don't bother to reply, I already know the results.

 

[chizow]

Originally posted by: qbfx
That's not true, the HD4870X2 dominates:

1280x1024, 8xAA/16xAF
1680x1050, 8xAA/16xAF
2560x1600, 4xAA/16xAF
2560x1600, 8xAA/16xAF
and logically derived from the above 1920x1200 which they skip in this test.

4870X2 wins 4 games at 1280/1680 8xAA
GTX 295 wins 3 games at 1280/1680 8xAA

GTX 295 also wins in 4 other titles with 4xAA, the highest allowed/tested. Results are similar with 2560, as already discussed.

I was looking at the 2560x1600, 8xAA/16xAF benches where the HD4870X2 clearly runs circles around the GTX295. If we consider all of the 2560x1600 tests I'd put FarCry 2 (the HD4870X2 beats the GTX295 in 2560x1600, 1xAA/1xAF) and Lost Planet (the HD4870X2 beats the GTX295 in 2560x1600, 8xAA/16xAF) in splits too.

So it's more like:

HD4870X2: Jericho, GRID, RS, CoJ, Crysis
GTX295: COD5, Bioshock, WIC
Splits: Assasin's Creed, STALKER, FarCry 2, Lost Planet

Why would you count LP and FC2 as splits when the GTX 295 wins the majority? I guess you'd also have to count Jericho as a split as well?

2560x1600, 8xAA/16xAF IS compared under Performance Rating, there's a little + on the bottom of the page.

Ah yep, I see it now. Again it proves my point the Performance Rating can be misleading and meaningless.

%-based results are meaningful when you have multiple benches and you want to summarize the results, as simply there is no other/better way to do that.

Sure there is, you show difference in FPS and % difference on a per title and resolution basis. Not only is it easier to read, its actually meaningful as well. For example, I can look at a 65% difference and 5.2 FPS at 2560 and dismiss the result as meaningless.

This review includes 12 titles that are not handpicked like nV does when testing their new drivers. Lost Planet and FarCry 2 both favor nV, yet you see the HD4870X2 is on par with the GTX295.

If by hand-picked you mean Top 10 titles for the last 2-3 months at any given time, I'd be glad to have hand-picked titles for every review. Certainly more relevant than old mainstays for certain GPU vendors. Jericho, CoJ, RS: Vegas hmm.... lol.

"Here we want our results to benchmark the Radeon HD 4870 and expand the driver, presumably as a Catalyst 9.1 will be released against his predecessor Catalyst 8:10, 8:11 and Catalyst Catalyst 8:12"

Ya, it means its the same driver, which means AT used the same driver in their testing to come to the conclusion the GTX 295 was the faster card.

 

[AzN]

Originally posted by: chizow

Uh, no I said AMD still has not responded to GT200, and now GT200b. And they haven't.

These kind of posts just make you look silly. Not that you your posts aren't silly to begin with.

Maybe you should edit again, because you're just digging that hole deeper and deeper now. If GTX 280 SLI comparisons aren't good enough because of a 25MHz difference in core clock and difference in bandwidth, just compare it to the GTX 260 Core 216 SLI, which is clocked identically to the GTX 295 at 576/1242/1000. The only differences between the GTX 295 and GTX 260 core 216 GPUs are 24SP and 8TMUs.

So once again, does the GTX 295 perform more like GTX 260 SLI? Or GTX 280 SLI. Differences are evident even in non-bandwidth/VRAM limited resolutions and settings. Don't bother to reply, I already know the results.

You've already dug a hole bigger than grand canyon which you can't even get out of at this point. :laugh: Why do you keep insisting looking at core clocks alone when anandtech tested mostly bandwidth starved situations? If you knew the results than why ask? GTX perform more like 260SLi because of bandwidth at 2560x1600 4xAA. Why even ask stupid questions?

 

[chizow]

Originally posted by: Azn
These kind of posts just make you look silly. Not that you your posts aren't silly to begin with.

No, my replies just show you lack basic reading comprehension skills and frequently post misinformation as if it were fact.

You've already dug a hole bigger than grand canyon which you can't even get out of at this point. :laugh: Why do you keep insisting looking at core clocks alone when anandtech tested mostly bandwidth starved situations? If you knew the results than why ask? GTX perform more like 260SLi because of bandwidth at 2560x1600 4xAA. Why even ask stupid questions?

Rofl figures, once again you've been proven wrong so you just start typing nonsensical replies. Once again, the differences between the parts are still clearly evident even at lower resolutions that would not be VRAM/bandwidth limited.

Here's some 1680 examples from the review:

Game - 295...280SLI...260SLI

COD5 - 119...133...115
AOC - 52...62...42
CRY- 41...46...36
FC2- 77...88...75
Grid- 109...138...108
L4D- 124...121...118

There's certainly some instances where the GTX 295 performs very similarly to the 280 but its obvious that its overall performance is much closer to GTX 260 SLI rather than GTX 280 SLI, a difference that 4% core clock wouldn't be able to overcome.

 

[AzN]

Originally posted by: chizow

No, my replies just show you lack basic reading comprehension skills and frequently post misinformation as if it were fact.

No your posts are silly swayed by Nvidia marketing gimmicks and whole lot of BS. :laugh:

Rofl figures, once again you've been proven wrong so you just start typing nonsensical replies. Once again, the differences between the parts are still clearly evident even at lower resolutions that would not be VRAM/bandwidth limited.

Here's some 1680 examples from the review:

Game - 295...280SLI...260SLI

COD5 - 119...133...115
AOC - 52...62...42
CRY- 41...46...36
FC2- 77...88...75
Grid- 109...138...108
L4D- 124...121...118

There's certainly some instances where the GTX 295 performs very similarly to the 280 but its obvious that its overall performance is much closer to GTX 260 SLI rather than GTX 280 SLI, a difference that 4% core clock wouldn't be able to overcome.

I knew you were going to bring up lower resolution. Your post wouldn't be silly other wise. If you look at 1680x1050 they were using AA. AA uses bandwidth whether it be lower resolution or higher resolution. Yeah really.. :roll:

Try Crysis warhead bench where no AA was used.

295...280SLI...260SLI @ 2560x1600

31.5....35.4.....28.3

Again 280SLI is clocked higher and bandwidth also does play some role here but not as much as with AA.

 

[OCGuy]

Originally posted by: Azn

Originally posted by: chizow

No, my replies just show you lack basic reading comprehension skills and frequently post misinformation as if it were fact.

No your posts are silly swayed by Nvidia marketing gimmicks and whole lot of BS. :laugh:

Comments like this are getting old and should be grounds for vacation.

Everytime some says something + for nV, we shouldnt have to hear:


"Zomg focus group marketing lulz AMDzone sent me lulz"


Im getting alot of good information and it is fun to read, but that is unnecessary.

 

[toslat]

Recipe almost complete, all we need is to add a dash of larabee!

 

[AzN]

Originally posted by: Ocguy31

Originally posted by: Azn

Originally posted by: chizow

No, my replies just show you lack basic reading comprehension skills and frequently post misinformation as if it were fact.

No your posts are silly swayed by Nvidia marketing gimmicks and whole lot of BS. :laugh:

Comments like this are getting old and should be grounds for vacation.

Everytime some says something + for nV, we shouldnt have to hear:


"Zomg focus group marketing lulz AMDzone sent me lulz"


Im getting alot of good information and it is fun to read, but that is unnecessary.

Telling people they lack basic reading comprehension is okay? That's quite humorous considering coming from a guy who has been cheerleading.

 

[chizow]

Originally posted by: Azn
No your posts are silly swayed by Nvidia marketing gimmicks and whole lot of BS. :laugh:

Ya, whole lot of gimmicks and BS. So if someone claimed G92b was larger than RV770, when its not, and consistently posts similar misinformation despite clear evidence to the contrary, what would their posts be swayed by? Incompetence. And that's not a personal attack, its just the truth.

I knew you were going to bring up lower resolution. Your post wouldn't be silly other wise. If you look at 1680x1050 they were using AA. AA uses bandwidth whether it be lower resolution or higher resolution. Yeah really.. :roll:

Try Crysis warhead bench where no AA was used.

295...280SLI...260SLI @ 2560x1600

31.5....35.4.....28.3

Again 280SLI is clocked higher and bandwidth also does play some role here but not as much as with AA.

Yes, of course I'm going to bring up lower resolutions if you're going to incorrectly claim performance differences are a result of bandwidth. If bandwidth was an issue at 1680 with 4xAA, what exactly do you think would happen at even more bandwidth intensive resolutions and settings? You'd expect larger differences in performance, yet the GTX 260 SLI and 295 scale as expected until 2560 with AA at which point some titles begin to show bandwidth/VRAM as a significant factor and performance drops off precipitously. I could link to a variety of other reviews showing 1680, even 1280 with no AA but there really is no point with you as you'll just come up with some other nonsensical reply.

As for Crysis, why would you go back to 2560 if you think bandwidth is an issue? Not to mention Crysis is actually one of the few titles that benefits as much or more from shader performance compared to ROP/core clocks. Which isn't something I ever disputed. Its clear however that ROPs are more important than SP/TMUs, which I've shown now over 3 different architectures.

 

[AzN]

Originally posted by: chizow
Ya, whole lot of gimmicks and BS. So if someone claimed G92b was larger than RV770, when its not, and consistently posts similar misinformation despite clear evidence to the contrary, what would their posts be swayed by? Incompetence. And that's not a personal attack, its just the truth.

Now you are being hilarious. :laugh: Quote on me where I said G92b was smaller than RV770? From a guy who says I have basic reading comprehension you seem to be the one lacking in that dept. When did 9800gtx become a 92b chip? Oh that's right they aren't the new 55nm. They are built with 65nm. the new 9800gtx+ is g92b chip. Makes you look real smart there with the whole incompetence and personal attack speech. :laugh:

Yes, of course I'm going to bring up lower resolutions if you're going to incorrectly claim performance differences are a result of bandwidth. If bandwidth was an issue at 1680 with 4xAA, what exactly do you think would happen at even more bandwidth intensive resolutions and settings? You'd expect larger differences in performance, yet the GTX 260 SLI and 295 scale as expected until 2560 with AA at which point some titles begin to show bandwidth/VRAM as a significant factor and performance drops off precipitously. I could link to a variety of other reviews showing 1680, even 1280 with no AA but there really is no point with you as you'll just come up with some other nonsensical reply.

Your whole lower resolution needing less bandwidth than higher resolution is quite wrong. At 4xAA it uses a certain bandwidth for not to bottleneck. Just because it uses higher resolution and same old 4xAA doesn't mean it's going to have bigger performance drop. That only occurs when you raise to 8xAA yet it still wouldn't matter what resolution which is determined by GPU core. You should get a clue before trying to to tell people not to comeback with nonsensical reply or what not.

As for Crysis, why would you go back to 2560 if you think bandwidth is an issue? Not to mention Crysis is actually one of the few titles that benefits as much or more from shader performance compared to ROP/core clocks. Which isn't something I ever disputed. Its clear however that ROPs are more important than SP/TMUs, which I've shown now over 3 different architectures.

Because bandwidth still isn't an issue @ 2560x1600. Raw performance is mostly determined by the core not bandwidth.

 

[chizow]

Originally posted by: Azn
Now you are being hilarious. :laugh: Quote on me where I said G92b was smaller than RV770? From a guy who says I have basic reading comprehension you seem to be the one lacking in that dept. When did 9800gtx become a 92b chip? Oh that's right they aren't the new 55nm. They are built with 65nm. the new 9800gtx+ is g92b chip. Makes you look real smart there with the whole incompetence and personal attack speech. :laugh:

You said:

It won't matter anyways. AMD can easily raise their performance level by adding more TMU and SP into their design without better processing technology. You do know RV770 is slightly smaller than 9800gtx and much smaller than Nvidia's behemoth GT200. Your whole logic makes very little sense if any sense at all other than you are swayed by Nvidia. It's quite amusing.

You claimed ATI could increase performance without upgrading their process and attempted to draw parallels to die sizes comparing Nvidia chips of a different process. In effect you were attempting to compare apples to oranges, ie. 55nm to 65nm to make a point when that difference disappears once you compare apples to apples, ie. 55nm to 55nm.

The lack of reading comprehension skills and incompetence comments have been demonstrated repeatedly, even through all of your edits. I've again shown 9800GTX+ is always slower than GTX 260 despite similar theoreticals of factors you considered the most important and I've also shown the GTX 295 performs closer to GTX 260 SLI despite having the same number of TMUs and SPs as the GTX 280.

Your whole lower resolution needing less bandwidth than higher resolution is quite wrong. At 4xAA it uses a certain bandwidth for not to bottleneck. Just because it uses higher resolution and same old 4xAA doesn't mean it's going to have bigger performance drop. That only occurs when you raise to 8xAA. You should get a clue before trying to to tell people not to comeback with nonsensical reply or what not.

Rofl what? No its really simple. If a GTX 295 and GTX 260 SLI are bandwidth limited at 1680 with 4xAA and you're claiming bandwidth is the reason for the 10-15% difference in performance from GTX 280 SLI, then you would expect greater drops than 10-15% at higher resolutions with AA. But instead, we see similar scaling and differences in the same 10-15% range, which tells us bandwidth is *NOT* an issue. Only at 2560/AA do you start seeing greater drops in performance than the 10-15% differences at 1680.

Because bandwidth still isn't an issue. Raw performance is mostly determined by the core not bandwidth.

ROFL. You're claiming bandwidth is an issue at 1680 with 4xAA, but not at 2560 which by default is 2.5x as many pixels? Also I don't disagree raw performance is mostly determined by the core, just not the parts of the core that you think, as I've proven time and time again. :laugh:

 

[AzN]

Originally posted by: chizow
You said:

It won't matter anyways. AMD can easily raise their performance level by adding more TMU and SP into their design without better processing technology. You do know RV770 is slightly smaller than 9800gtx and much smaller than Nvidia's behemoth GT200. Your whole logic makes very little sense if any sense at all other than you are swayed by Nvidia. It's quite amusing.

You claimed ATI could increase performance without upgrading their process and attempted to draw parallels to die sizes comparing Nvidia chips of a different process. In effect you were attempting to compare apples to oranges, ie. 55nm to 65nm to make a point when that difference disappears once you compare apples to apples, ie. 55nm to 55nm.

And your point? GT200 on 55nm is what? 470mm which is nothing but clock increase? and what RV770 die size? 260mm? I don't see why ATI can't fit more SP and TMU into their chip to compete with GTX 285. Of course they would have to raise die size but it still wouldn't be big as GT200 on 55nm.

The lack of reading comprehension skills and incompetence comments have been demonstrated repeatedly, even through all of your edits. I've again shown 9800GTX+ is always slower than GTX 260 despite similar theoreticals of factors you considered the most important and I've also shown the GTX 295 performs closer to GTX 260 SLI despite having the same number of TMUs and SPs as the GTX 280.

You haven't done a thing except personal attack. :laugh: Never denied 9800gtx+ wasn't slower but in non bandwidth limited situations it's only 10% faster if that. As for your GTX 295 I had to correct you to determine bandwidth was the limitations in most of those tests followed by GTX 280 SLI having faster clocks.

Rofl what? No its really simple. If a GTX 295 and GTX 260 SLI are bandwidth limited at 1680 with 4xAA and you're claiming bandwidth is the reason for the 10-15% difference in performance from GTX 280 SLI, then you would expect greater drops than 10-15% at higher resolutions with AA. But instead, we see similar scaling and differences in the same 10-15% range, which tells us bandwidth is *NOT* an issue. Only at 2560/AA do you start seeing greater drops in performance than the 10-15% differences at 1680.

Why even try when you have no idea what you are talking about?

It doesn't matter what resolution. that is determined by the core. Same reason why 9800gtx is only 10% slower with no AA even at 2560x1600 vs GTX260. At 4xAA it will be about 30% slower across resolution again because of bandwidth limitations long as it's there's no vram limitations.

GTX 295 is working with same bandwidth whether it be higher resolution or lower resolution as same for GTX 280 SLI. Performance difference is going to mostly same throughout at 4xAA because of bandwidth limitations.

ROFL. You're claiming bandwidth is an issue at 1680 with 4xAA, but not at 2560 which by default is 2.5x as many pixels? Also I don't disagree raw performance is mostly determined by the core, just not the parts of the core that you think, as I've proven time and time again. :laugh:

That's right. Fillrate. not bandwidth. What part of the GPU makes fillrate? Don't tell me bandwidth. :laugh: This just shows how clueless you really are.

 

[wlee15]

Originally posted by: chizow


Also, G92b on the same 55nm process is smaller than RV770 at roughly 231mm^2 to 250mm^2. so I'm not quite sure how my logic is swayed by Nvidia.....

Actually analysis by the folks at beyond3d have found out that the G92b is actually is slightly larger than the RV770.

http://forum.beyond3d.com/show...t=G92b+die+size&page=2

 

[chizow]

Originally posted by: Azn
And your point? GT200 on 55nm is what? 470mm which is nothing but clock increase? and what RV770 die size? 260mm? I don't see why ATI can't fit more SP and TMU into their chip to compete with GTX 285. Of course they would have to raise die size but it still wouldn't be big as GT200 on 55nm.

Its not just a clock increase, its a clock increase with a reduction in power draw, ~40W, enough to reduce its power consumption below 65nm GT200 and HD 4870. It was also enough to reduce power consumption on the GTX 295 so that it draws less power than HD4870X2.

As for adding transistors to RV770, I'm sure they could but considering RV770's power draw and TDP already match or exceed Nvidia's 55nm GT200, they'd most likely need a process shrink or a reduction in clock speeds to keep thermals in check. That's where the news of a 40nm RV790 comes in, as AMD is either going to add transistors with a shrink or try and improve yields enough to take advantage of the 2 additional shader clusters RV770 is rumored to have (12x80).

You haven't done a thing except personal attack. :laugh: Never denied 9800gtx+ wasn't slower but in non bandwidth limited situations it's only 10% faster if that. As for your GTX 295 I had to correct you to determine bandwidth was the limitations in most of those tests followed by GTX 280 SLI having faster clocks.

Heh, you keep saying 10%, but that doesn't make it true. 1920 Comparison I put that together at launch to compare performance differences and found 25-100% performance increases between GT200 and G80, even without AA. I'm sure I'd see similar differences today.

As for correcting me with regards to GTX 295...you still haven't acknowledged how it performs more like GTX 260 core 216 SLI than GTX 280SLI, and in that case there is no bandwidth or VRAM disadvantage.

Bit-Tech GTX 295 Review

Feel free to look through that review, they have plenty of results without AA and in cases where all of the multi-GPU solutions aren't completely CPU bottlenecked, you'll once again see GTX 295 performs much more similarly to GTX 260 SLI than GTX 280 SLI, even in non-bandwidth intensive resolutions and settings.

Why even try when you have no idea what you are talking about?

It doesn't matter what resolution. that is determined by the core. Same reason why 9800gtx is only 10% slower with no AA even at 2560x1600 vs GTX260. At 4xAA it will be about 30% slower across resolution again because of bandwidth limitations long as it's there's no vram limitations.

GTX 295 is working with same bandwidth whether it be higher resolution or lower resolution as same for GTX 280 SLI. Performance difference is going to mostly same throughout at 4xAA because of bandwidth limitations.

ROFL. So now resolution and bandwidth aren't important? You're arguing one side of the coin with one breath and contradicting yourself in the next. You don't need to enable AA to increase bandwidth requirements, simply increasing resolution will as well. If you're going to claim a low resolution with AA is bandwidth limited then you must also acknowledge higher resolutions with or without AA are also going to be bandwidth limited. If that were the case, the drop in performance between GTX 295 and GTX 280 SLI would be similar to GTX 260 and 9800GTX+ in bandwidth limited situations, except its not.

Bandwidth is not an issue until 2560 and usually requires AA before significant bandwidth/VRAM limitations become apparent, as every benchmark has shown time and again with these high-end parts. Until bandwidth and/or VRAM become an issue, then yes, core architecures and differences are going to be most responsible for diffferences in performance, which is how we can conclude ROPs are more important than TMUs and SPs given the difference in core/speeds between GTX 260, GTX 295 and GTX 280.

That's right. Fillrate. not bandwidth. What part of the GPU makes fillrate? Don't tell me bandwidth. :laugh: This just shows how clueless you really are.

ROFL. When I have I claimed bandwidth was an issue with GTX 295 other than extremely high resolutions with AA in some titles? I haven't, so why would I claim bandwidth would have a greater impact on performance, especially without AA? I've already stated numerous times that fillrate is more important than bandwidth in most situations, hence my emphasis on ROPs, as they're directly responsible for both fillrate and peformance with or without AA.

Anyways, thanks again for reminding me why its completely unproductive arguing with you. Half the time you don't even know what you're arguing. :laugh:

 

[AzN]

Originally posted by: chizow
Its not just a clock increase, its a clock increase with a reduction in power draw, ~40W, enough to reduce its power consumption below 65nm GT200 and HD 4870. It was also enough to reduce power consumption on the GTX 295 so that it draws less power than HD4870X2.

As for adding transistors to RV770, I'm sure they could but considering RV770's power draw and TDP already match or exceed Nvidia's 55nm GT200, they'd most likely need a process shrink or a reduction in clock speeds to keep thermals in check. That's where the news of a 40nm RV790 comes in, as AMD is either going to add transistors with a shrink or try and improve yields enough to take advantage of the 2 additional shader clusters RV770 is rumored to have (12x80).

Wow more Nvidia marketing lingo.

Heh, you keep saying 10%, but that doesn't make it true. 1920 Comparison I put that together at launch to compare performance differences and found 25-100% performance increases between GT200 and G80, even without AA. I'm sure I'd see similar differences today.

Was I even talking about 8800gtx vs GTX 280? Of course not. You thought it actually had something to do with 9800gtx+ vs 260 raw performance.

Try something like this... http://www.anandtech.com/GalleryImage.aspx?id=3152

As for correcting me with regards to GTX 295...you still haven't acknowledged how it performs more like GTX 260 core 216 SLI than GTX 280SLI, and in that case there is no bandwidth or VRAM disadvantage.

Bit-Tech GTX 295 Review

Feel free to look through that review, they have plenty of results without AA and in cases where all of the multi-GPU solutions aren't completely CPU bottlenecked, you'll once again see GTX 295 performs much more similarly to GTX 260 SLI than GTX 280 SLI, even in non-bandwidth intensive resolutions and settings.

There's obviously something wrong with bit-tech's benches. A 216SLI is actually faster than GTX 295 in some of those benches when GTX295 more SP and TMU. It's more of driver issues than anything else at this point. It could also be GTX 295 processing issue, user error or what not.

ROFL. So now resolution and bandwidth aren't important? You're arguing one side of the coin with one breath and contradicting yourself in the next. You don't need to enable AA to increase bandwidth requirements, simply increasing resolution will as well. If you're going to claim a low resolution with AA is bandwidth limited then you must also acknowledge higher resolutions with or without AA are also going to be bandwidth limited. If that were the case, the drop in performance between GTX 295 and GTX 280 SLI would be similar to GTX 260 and 9800GTX+ in bandwidth limited situations, except its not.

Bandwidth is not an issue until 2560 and usually requires AA before significant bandwidth/VRAM limitations become apparent, as every benchmark has shown time and again with these high-end parts. Until bandwidth and/or VRAM become an issue, then yes, core architecures and differences are going to be most responsible for diffferences in performance, which is how we can conclude ROPs are more important than TMUs and SPs given the difference in core/speeds between GTX 260, GTX 295 and GTX 280.

You are quite wrong again. Raw performance has very little to do with bandwidth whether you'd be raising higher resolution or not long as you aren't on 128bit bus or something of that nature. There is memory footprint no doubt but when that memory footprint is available it doesn't matter if you are pulling 2560x1600 resolution. More bandwidth isn't the driving force if any at all. What matters is fillrate.

ROFL. When I have I claimed bandwidth was an issue with GTX 295 other than extremely high resolutions with AA in some titles? I haven't, so why would I claim bandwidth would have a greater impact on performance, especially without AA? I've already stated numerous times that fillrate is more important than bandwidth in most situations, hence my emphasis on ROPs, as they're directly responsible for both fillrate and peformance with or without AA.

Anyways, thanks again for reminding me why its completely unproductive arguing with you. Half the time you don't even know what you're arguing. :laugh:

LOL Of course you claimed bandwidth was an issue @ 2560x1600 no AA because it's pushing 2.5x as more pixels than 1680x1050 4xAA.

Here you go. :laugh:

ROFL. You're claiming bandwidth is an issue at 1680 with 4xAA, but not at 2560 which by default is 2.5x as many pixels?

 

[Tempered81]

quick snapshot of Extreme IQ high-res gaming 4870x2 vs. gtx295:

http://www.hardwarecanucks.com...eo-card-review-16.html

 

[chizow]

Originally posted by: Azn
Wow more Nvidia marketing lingo.

And more incompetence in reply.

Was I even talking about 8800gtx vs GTX 280? Of course not. You thought it actually had something to do with 9800gtx+ vs 260 raw performance.

Try something like this... http://www.anandtech.com/GalleryImage.aspx?id=3152

Actually there were some comparisons to the 260 as well, and given the 8800GTX and 9800GTX+ perform nearly identically the comparisons would be relevant. Your linked galleries also show this. AC doesn't show much advantage, but this isn't a huge surprise given Derek acknowledged it was either CPU limited or frame rate capped. Other titles do show much more than 10% difference though, which again backs my point that the GTX 260 is always faster than the 9800GTX+, despite similar theoreticals which you claim have the greatest impact on performance.

There's obviously something wrong with bit-tech's benches. A 216SLI is actually faster than GTX 295 in some of those benches when GTX295 more SP and TMU. It's more of driver issues than anything else at this point. It could also be GTX 295 processing issue, user error or what not.

Sure it could be driver issues, or PCIE bandwidth issues, scheduling overhead, user error or whatever else. That doesn't change the fact the majority of reviews show GTX 295 performing more like GTX 260 SLI than GTX 280 SLI even at resolutions that are not bandwidth limited.

You are quite wrong again. Raw performance has very little to do with bandwidth whether you'd be raising higher resolution or not long as you aren't on 128bit bus or something of that nature. There is memory footprint no doubt but when that memory footprint is available it doesn't matter if you are pulling 2560x1600 resolution. More bandwidth isn't the driving force if any at all. What matters is fillrate.

ROFLMAO! If bandwidth doesn't matter, why are you claiming GTX 280 SLI can't be compared to GTX 295 at 1680 with AA because the 280 has more bandwidth? LOLOL. Seriously, just stop, its painful watching you contradict yourself with every other sentence.

Also, fillrate requirements are going to be less at lower resolutions than higher resolutions regardless of AA or bandwidth, so again, if you think fillrate and raw performance alone determines performance you shouldn't have any issues comparing results at lower resulolutions with AA. Here's a hint, ROPs are responsible for both AA and pixel fillrate, which is largely why the GTX 260 always outperforms the 9800GTX+ and moreso when AA is enabled, even at lower resolutions where bandwidth isn't an issue.

LOL Of course you claimed bandwidth was an issue @ 2560x1600 no AA because it's pushing 2.5x as more pixels than 1680x1050 4xAA.

Here you go. :laugh:

ROFL. You're claiming bandwidth is an issue at 1680 with 4xAA, but not at 2560 which by default is 2.5x as many pixels?

Uh, no I was pointing out another of your contradictions where you claimed bandwidth was an issue at 1680 with 4xAA and therefore, invalid, but then turned around and pointed to results at 2560 when raising resolution naturally increases bandwidth requirements as well.

 

[BFG10K]

Originally posted by: chizow

If by hand-picked you mean Top 10 titles for the last 2-3 months at any given time, I'd be glad to have hand-picked titles for every review. Certainly more relevant than old mainstays for certain GPU vendors. Jericho, CoJ, RS: Vegas hmm.... lol.

Who gives a shit what games you want to see? I don't and neither does anyone else that plays games you don't.

I play Call of Juarez and Jericho so I want to see those figures, as does anyone else who plays those games. That you don't play said games doesn't somehow invalidate the scores just because you say so.

The games are very relevant because they?re extremely taxing even compared to current titles, plus they aren?t cherry-picked for marketing purposes.

And remember, these are 2007 titles (i.e. less than three years old) so according to your past statements they fall under your umbrella of expecting IHVs to support them and deliver performance gains when required. Again those are your words.

Ya, it means its the same driver, which means AT used the same driver in their testing to come to the conclusion the GTX 295 was the faster card.

Ah yes, the old "quote Anandtech whenever it suits me". Anandtech also claimed Big Bang does nothing for performance and you were running around screaming their tests were an outlier. You?ve also criticized their testing methodology whenever ATi is shown in a positive light, even going so far as to ask Derek to step down as a reviewer.

But whenever Anandtech posts positive nVidia reviews you don?t bat an eyelid and are more than happy to link to them. Interesting, that. :roll:

Again your agenda and bias are as clear as day.

Nobody is claiming the 4870X2 is the faster card, just that ComputerBase's results clearly show massive performance gains from the 9.1 drivers, more than enough to beat the GTX295 in some cases. That you employ hand-having and try to dismiss the scores is irrelevant because people that actually play those games don?t give a shit about your comments.

 

[MegaWorks]

Don't worry BFG10K, I really don't give a shit about his comments.

 

[AzN]

Originally posted by: chizow

And more incompetence in reply.

No needed reply to marketing jargon by a guy who thinks ROP was the biggest factor when it comes to performance. :laugh: You even had a thread about it. ROFL...

Actually there were some comparisons to the 260 as well, and given the 8800GTX and 9800GTX+ perform nearly identically the comparisons would be relevant. Your linked galleries also show this. AC doesn't show much advantage, but this isn't a huge surprise given Derek acknowledged it was either CPU limited or frame rate capped. Other titles do show much more than 10% difference though, which again backs my point that the GTX 260 is always faster than the 9800GTX+, despite similar theoreticals which you claim have the greatest impact on performance.

Don't give me that shit about 8800gtx is same as 9800gtx+. That's full of shit and you know it. Even in my link 8800gtx is in the benchmark. 12fps slower 9800gtx+ or 28% faster than 8800gtx far as raw performance goes.

Sure it could be driver issues, or PCIE bandwidth issues, scheduling overhead, user error or whatever else. That doesn't change the fact the majority of reviews show GTX 295 performing more like GTX 260 SLI than GTX 280 SLI even at resolutions that are not bandwidth limited.

There's something wrong with your logic. You are basically agreeing it could be all those issues you mentioned but then you go ahead emphasis GTX 295 performs more like GTX 260? :laugh:

ROFLMAO! If bandwidth doesn't matter, why are you claiming GTX 280 SLI can't be compared to GTX 295 at 1680 with AA because the 280 has more bandwidth? LOLOL. Seriously, just stop, its painful watching you contradict yourself with every other sentence.

Wow.. Reading comprehension much? I was talking about RAW performance but then you go ahead emphasis on 1680x1050 with AA and says bandwidth doesn't matter and you chuckle by yourself like a guy on drugs. :Q

Also, fillrate requirements are going to be less at lower resolutions than higher resolutions regardless of AA or bandwidth, so again, if you think fillrate and raw performance alone determines performance you shouldn't have any issues comparing results at lower resulolutions with AA. Here's a hint, ROPs are responsible for both AA and pixel fillrate, which is largely why the GTX 260 always outperforms the 9800GTX+ and moreso when AA is enabled, even at lower resolutions where bandwidth isn't an issue.

Why would fillrate requirements become less long as it's not being bottlecked by CPU? More fillrate the merrier.

Hint? :laugh: Bandwidth not being an issue @ lower resolutions is quite funny. with AA it's quite the issue no matter what resolution. just let me know I could downclock my g92 memory and give you whole lot of benchmarks which would prove your incompetence.

Uh, no I was pointing out another of your contradictions where you claimed bandwidth was an issue at 1680 with 4xAA and therefore, invalid, but then turned around and pointed to results at 2560 when raising resolution naturally increases bandwidth requirements as well.

Talk about contradictions.

 

[chizow]

Originally posted by: BFG10K
Who gives a shit what games you want to see? I don't and neither does anyone else that plays games you don't.

I play Call of Juarez and Jericho so I want to see those figures, as does anyone else who plays those games. That you don't play said games doesn't somehow invalidate the scores just because you say so.

The games are very relevant because they?re extremely taxing even compared to current titles, plus they aren?t cherry-picked for marketing purposes.

And remember, these are 2007 titles (i.e. less than three years old) so according to your past statements they fall under your umbrella of expecting IHVs to support them and deliver performance gains when required. Again those are your words.

Like I said, if you and others are going to claim Top 10 titles are "cherry-picked for marketing purposes", I have absolutely no problems whatsoever with cherry-picked titles at any given time. He claimed the Computerbase selections were objective based on game popularity, when in reality there's always going to be subjective influence with a limited testing suite. Like I said, inserting say, CoH and Dead Space, two popular titles that clearly favor Nvidia parts would significantly change the results.

Ah yes, the old "quote Anandtech whenever it suits me". Anandtech also claimed Big Bang does nothing for performance and you were running around screaming their tests were an outlier. You?ve also criticized their testing methodology whenever ATi is shown in a positive light, even going so far as to ask Derek to step down as a reviewer.

Of course I'm going to point out inconsistencies and problems with Derek's reviews, that doesn't invalidate all of his testing, research, insights and opinions. His Big Bang results were clearly an outlier and I pointed that out, especially given Nvidia did not list improvements in the titles AT tested. If Nvidia claims driver improvements in 10 recent games and AT doesn't test any of them, then someone says "well AT's results didn't show improvement" while every other review site does, AT would clearly be the outlier.

And of course I'm going to criticize his testing methodology when he flat out admits to fabricating results and ignoring significant stuttering and performance problems with ATI parts just to get some numbers up. I would absolutely point out the same if he said or did similarly with Nvidia. In fact I did point out the fact Derek uses archived drivers and results between reviews, which is a poor habit and something I don't agree with. If he doesn't feel up to producing quality reviews he should absolutely step down as a video reviewer, but that doesn't mean his opinion and editorials wouldn't be valuable as an editor.

But whenever Anandtech posts positive nVidia reviews you don?t bat an eyelid and are more than happy to link to them. Interesting, that. :roll:

Actually given the recent problems with their reviewing methodology I prefer not to link to AT reviews, but in this case they specifically mentioned updated results with updated drivers. And specifically, the 8.12 hot fix/9.1 beta driver in question. Given reviews based on this driver are few and far between its obviously going to provide more insight to driver improvements over a single review.

Again your agenda and bias are as clear as day.

LOL, looks like someone's still salty about my accurate analysis of your buying habits.

Nobody is claiming the 4870X2 is the faster card, just that ComputerBase's results clearly show massive performance gains from the 9.1 drivers, more than enough to beat the GTX295 in some cases. That you employ hand-having and try to dismiss the scores is irrelevant because people that actually play those games don?t give a shit about your comments.

Where am I dismissing the scores? I'm pointing out their performance rating aggregates aren't an accurate gauge of actual performance. Like I said earlier, I certainly enjoy seeing results from a variety of games but to say these results are completely objective or vendor agnostic is laughable.

 

[chizow]

Originally posted by: Azn
No needed reply to marketing jargon by a guy who thinks ROP was the biggest factor when it comes to performance. :laugh: You even had a thread about it. ROFL...

Yep, and I was right there as well. G80 to G80 GTS, G80 to G92, G92 to G94, G92 to GT200, GT200 to GT200b. They ALL show TMU and SP are less significant than ROP when it comes to performance.

Don't give me that shit about 8800gtx is same as 9800gtx+. That's full of shit and you know it. Even in my link 8800gtx is in the benchmark. 12fps slower 9800gtx+ or 28% faster than 8800gtx far as raw performance goes.

Based on AT's review that's right, however that wasn't the case with the original 9800GTX and GTS 512MB. Still doesn't change the fact the GTX 260 always outperforms the 9800GTX+.

There's something wrong with your logic. You are basically agreeing it could be all those issues you mentioned but then you go ahead emphasis GTX 295 performs more like GTX 260? :laugh:

Ya, it could be anything but at the end of the day GTX 280 SLI distances itself from both and GTX 295 ends up looking like GTX 260 SLI. That's the point.

Wow.. Reading comprehension much? I was talking about RAW performance but then you go ahead emphasis on 1680x1050 with AA and says bandwidth doesn't matter and you chuckle by yourself like a guy on drugs. :Q

Yep and raw performance would be just as important at 1680 as 2560, so if bandwidth isn't an issue as you've repeatedly claimed, you'd certainly have to acknowledge the results at 1680 with 4xAA are completely relevent in proving the point ROP are more significant than SP and TMU given the performance differences between GTX 280 SLI, 260 SLI and 295.

Why would fillrate requirements become less long as it's not being bottlecked by CPU? More fillrate the merrier.

Because lower resolutions have fewer pixels to draw per frame, regardless of AA.

Hint? :laugh: Bandwidth not being an issue @ lower resolutions is quite funny. with AA it's quite the issue no matter what resolution. just let me know I could downclock my g92 memory and give you whole lot of benchmarks which would prove your incompetence.

So is bandwidth an issue at lower resolutions or not? You keep arguing both, either bandwidth is important or only "RAW performance", which is it?

I'd love for you to run some benchmarks at 1680 and compare performance differences at 1680 and see which yields a bigger increase, core, shader or memory. I already know the difference but now that you have a somewhat relevant part you might find out more on your own so you can stop posting nonsense about SPs, TMUs, and bandwidth.

Talk about contradictions.

Yep, you've flip-flopped on the bandwidth issue 3-4x at least. You claim bandwidth isn't important and only "RAW performance" matter, but then claim 1680 with 4xAA results aren't valid because of bandwidth.