Hello,
In recent times, I have been involved in some major debates concerning AMD and NVIDIA and how their current generation products (i.e. Tahiti and Kepler) perform in various applications. I would like to clarify that both AMD and NVIDIA work closely with game developers and try to ensure that their products are adequately supported in games and vice versa. Though situation does sometimes end up in the favour of either RED TEAM or GREEN TEAM, depending upon how close the relationship factor is between the companies. The purpose of this thread is to educate people about the "ground realities" of the competition between AMD GCN and NVIDIA Kepler products on the basis of my personal insight.
Why focus on applications?
AMD have been very actively involved with game developers in recent times after it started its "GAMING EVOLVED" program. Here is an interesting read.
Some consumers often fail to realize the fact that both AMD GCN and NVIDIA Kepler architectures have their respective strong points (merits) and weak points (demerits) and these architectures have been designed on different principles. Game developers, working with GPU manufacturers, are able to program their games on the basis of these points.
An application which is designed to take full advantage of strong points of NVIDIA Kepler architecture will obviously (relatively) work better on NVIDIA Kepler products in comparison to their AMD Tahiti counterparts. Similarly, an application which is designed to take full advantage of strong points of AMD GCN architecture will obviously work (relatively) better on AMD Tahiti products in comparison to their NVIDIA Kepler counterparts. This is the hard truth that people need to realize and not base their judgements on plethora of simplistic reviews which fail to highlight these important details. In current times, competition is not just about technical specifications of hardware but also brand relations.
As an example:-
NVIDIA Kepler will clean sweep in Lost Planet 2 application because it makes heavy use of Kepler friendly features:
Key hint: Tesselation
Similarly, AMD Tahiti will clean sweep in Sniper Elite v2 application because it makes heavy use of Tahiti friendly features.
Key hint: Compute Shaders
However, their are applications which aren't heavily tilted towards any architecture/brand and have been designed on nuetral grounds with respect to taking advantage of technical specifications of hardware because the developers of these applications chose to work with both NVIDIA and AMD to ensure best possible performance on a wide range of products from any brand. In the nutshell, their are white shades and then their are grey shades.
So if you ask me that which graphics card is better; AMD HD 7970 GE or NVIDIA GTX 680?
On the basis of my own insight, I would say that their is no definitive answer but performance wise-both fall in the same segment/category.
-----------------
Testing:-
Test setup and drivers:
I recently came across a review which features test setup very similar to mine (NOTE: Details of my PC are in my profile):
Source: ASUS Matrix HD 7970 Platinum: L'ammiraglia ASUS è di nuovo a caccia di record
IMPORTANT: This review highlights the importance of performance improvements offered by driver related optimizations for applications with an example of a AMD Catalyst 12.11 beta driver version. In response to performance improvements offered by AMD Catalyst 12.11 driver series in many applications, NVIDIA released GeForce R310 driver series which also incorporate optimizations for many applications. I am currently using GeForce 310.90 WHQL driver release; it offers performance improvements in very few games in comparison to previous versions of R310 driver releases and these games aren't featured in this comparison. In addition, the processor in the test setup involved in the review is clocked @ 4 GHz continuous whereas mine is set at EIST; this means that my processor have been operating at relatively lower clock speeds in comparison to the processor involved in the review and is unlikely to influence performance in applications in the same manner as the processor in the review. Furthermore, my graphics card is running at stock settings. Therefore, due to all of the factors highlighted, any potential benefits on my end have been effectively minimized/eliminated to ensure that this comparison is as fair as possible. I have conducted all tests @ 1080p resolution (maximum resolution of my monitor) and matching settings in applications as in the aforementioned review.
UPDATE: Upon insistence of some members, I have incorporated some comparisons which are based on another review in which AMD Catalyst 13.1 WHQL driver have been used but the review's test setup is different from mine.
The synthetics:-
3DMark11
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 10176
AMD HD 7970 GE: 10109
AMD HD 7970 standard: 9302
GTX 680 standard: 9471
EVGE GTX 680 SC Signature: 10439
Observation: 3DMark11 is designed to make extensive use of features such as Tesselation, Compute Shaders and Multi-threading. Due to this balancing factor, performance comparisons are likely to be fair in this application. Noticeable performance gap between my GPU and GTX 680 standard in this comparison is due to an older (non-R310 series) driver version used in the review which lacks in optimization for this application.
------
3DMark Vantage (GPU Scores only)
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 37338
AMD HD 7970 GE: 37081
AMD HD 7970 standard: 34889
GTX 680 standard: 34367
EVGA GTX 680 SC Signature: 36625
Observation: 3DMark Vantage is sensitive to characteristics such as clock speeds, bandwidth and/or ROPs. AMD Tahiti products have advantage in these aspects.
Technologies:-
Unigine Heaven 3.0
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 57 FPS (AVG)
AMD HD 7970 GE: 55 FPS (AVG)
AMD HD 7970 standard: 53 FPS (AVG)
GTX 680 standard: 58 FPS (AVG)
EVGA GTX 680 SC Signature: 63 FPS (AVG)
Observation: This program is designed to take advantage of nearly all aspects of GPU capabilities. Heavy use of tesselation changed the game in favour of NVIDIA Kepler comfortably in this comparison.
------
ComputeMark
ASUS HD 7970 Matrix Platinum: 3756
AMD HD 7970 GE: 3574
AMD HD 7970 standard: 3305
GTX 680 standard: 2939
EVGA GTX 680 SC Signature: 3129
Observation: Compute shaders is undoubtedly strong point of AMD Tahiti architecture.
------
TessMark
ASUS HD 7970 Matrix Platinum: 78 FPS (AVG)
AMD HD 7970 GE: 75 FPS (AVG)
AMD HD 7970 standard: 69 FPS (AVG)
GTX 680 standard: 119 FPS (AVG)
EVGA GTX 680 SC Signature: 130 FPS (AVG)
Observation: Tesselation is undoubtedly strong point of NVIDIA Kepler architecture.
The games:-
PLA
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 76 FPS (AVG)
AMD HD 7970 GE: 74 FPS (AVG)
AMD HD 7970 standard: 68 FPS (AVG)
GTX 680 standard: 75 FPS (AVG)
EVGA GTX 680 SC Signature: 87 FPS (AVG)
Observation: Seems balanced in the context of hardware utilization. This application is sensitive to nearly all aspects of hardware.
------
Crysis 2 (Comparison 1)
Time Square Map:
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 81 FPS (AVG)
AMD HD 7970 GE: 78 FPS (AVG)
AMD HD 7970 standard: 72 FPS (AVG)
GTX 680 standard: 70 FPS (AVG)
EVGA GTX 680 SC Signature: 79 FPS (AVG)
Central Park Map:
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 67 FPS (AVG)
AMD HD 7970 GE: 65 FPS (AVG)
AMD HD 7970 standard: 59 FPS (AVG)
GTX 680 standard: 68 FPS (AVG)
EVGA GTX 680 SC Signature: 76 FPS (AVG)
Observation: Seems balanced in the context of hardware utilization. This application is sensitive to nearly all aspects of hardware. However, these comparisons exclude the use of tesselation setting which may tilt the performance in favour of NVIDIA Kepler if enabled.
-------
Crysis 2 (Comparison 2)
Mine:
FPS (MIN): 32
FPS (AVG): 83
FPS (MIN): 24
FPS (AVG): 79
Observation: This application is sensitive to nearly all aspects of hardware. However, these comparisons exclude the use of tesselation setting which may tilt the performance in favour of NVIDIA Kepler if enabled. My GPU dominates in this application, regardless of settings.
DiRT 3
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 140 FPS (AVG)
AMD HD 7970 GE: 138 FPS (AVG)
AMD HD 7970 standard: 124 FPS (AVG)
GTX 680 standard: 134 FPS (AVG)
EVGA GTX 680 SC Signature: 137 FPS (AVG)
---
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 126 FPS (AVG)
AMD HD 7970 GE: 124 FPS (AVG)
AMD HD 7970 standard: 111 FPS (AVG)
GTX 680 standard: 126 FPS (AVG)
EVGA GTX 680 SC Signature: 126 FPS (AVG)
---
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 111 FPS (AVG)
AMD HD 7970 GE: 108 FPS (AVG)
AMD HD 7970 standard: 98 FPS (AVG)
GTX 680 standard: 109 FPS (AVG)
EVGA GTX 680 SC Signature: 111 FPS (AVG)
Observation: This game is AMD oriented in marketing context; the Colin McRae series generally is. However, this particular game may not be much AMD oriented in technical context as the performance numbers seem to suggest.
What about FXAA? Does it makes difference?
Answer is YES.
Consider the example of PLA application for this discussion.
Test run without AA:-
Test run with FXAA 8x:-
Test run with MSAA 8x:-
NOTES:
I may include more comparisons in this thread in the future to further solidify my personal insight about this subject. In my debates on this subject, I have always stressed on the fact that comparisons between hardware are not just technical specifications dependent but also application/programming dependent. In addition, this thread also discloses some merits of both NVIDIA Kepler and AMD Tahiti architectures and how these merits can make difference in applications designed to support them. Competition between AMD Tahiti and NVIDIA Kepler is not so black and white as some reviews would want you to believe.
Now some may argue that situation may change in higher resolutions but I disagree because this is also a theoretical assumption. Even in higher resolutions, picture will remain the same as is the case in this comparison with minor tilts here and there.
Here is an example:
In addition, examples of Lost Planet 2 and Sniper Elite v2 are also in front of you.
It may be possible that raw power oriented technical characteristics such as; greater memory interface; larger VRAM size; and significantly greater bandwidth output; of AMD HD 7970 series may permit it to; close the gap; or help it equalize performance; or grant it advantage; in some applications against NVIDIA GTX 680 series at extreme resolutions but NVIDIA also offers GTX 680 with 4 GB of VRAM as an alternative. Once again, performances in applications at larger resolutions will be application/programming dependent and not just technical specifications dependent.
CONCLUSION: AMD GCN and NVIDIA Kepler architectures have their respective strong points (merits) and weak points (demerits) and these architectures have been designed on different principles. Game developers, working with GPU manufacturers, are able to program their games on the basis of these points. Therefore, performance of hardware (regardless of brand) in applications will be dependent upon their programming nature; the applications may be designed to take advantage of merits of AMD Tahiti architecture and/or they may be designed to take advantage of merits of NVIDIA Kepler architecture.
DISCLAIMER (IMPORTANT):
In recent times, I have been involved in some major debates concerning AMD and NVIDIA and how their current generation products (i.e. Tahiti and Kepler) perform in various applications. I would like to clarify that both AMD and NVIDIA work closely with game developers and try to ensure that their products are adequately supported in games and vice versa. Though situation does sometimes end up in the favour of either RED TEAM or GREEN TEAM, depending upon how close the relationship factor is between the companies. The purpose of this thread is to educate people about the "ground realities" of the competition between AMD GCN and NVIDIA Kepler products on the basis of my personal insight.
Why focus on applications?
AMD have been very actively involved with game developers in recent times after it started its "GAMING EVOLVED" program. Here is an interesting read.
Some consumers often fail to realize the fact that both AMD GCN and NVIDIA Kepler architectures have their respective strong points (merits) and weak points (demerits) and these architectures have been designed on different principles. Game developers, working with GPU manufacturers, are able to program their games on the basis of these points.
An application which is designed to take full advantage of strong points of NVIDIA Kepler architecture will obviously (relatively) work better on NVIDIA Kepler products in comparison to their AMD Tahiti counterparts. Similarly, an application which is designed to take full advantage of strong points of AMD GCN architecture will obviously work (relatively) better on AMD Tahiti products in comparison to their NVIDIA Kepler counterparts. This is the hard truth that people need to realize and not base their judgements on plethora of simplistic reviews which fail to highlight these important details. In current times, competition is not just about technical specifications of hardware but also brand relations.
As an example:-
NVIDIA Kepler will clean sweep in Lost Planet 2 application because it makes heavy use of Kepler friendly features:
Spoiler: show
Key hint: Tesselation
Similarly, AMD Tahiti will clean sweep in Sniper Elite v2 application because it makes heavy use of Tahiti friendly features.
Spoiler: show
Key hint: Compute Shaders
However, their are applications which aren't heavily tilted towards any architecture/brand and have been designed on nuetral grounds with respect to taking advantage of technical specifications of hardware because the developers of these applications chose to work with both NVIDIA and AMD to ensure best possible performance on a wide range of products from any brand. In the nutshell, their are white shades and then their are grey shades.
So if you ask me that which graphics card is better; AMD HD 7970 GE or NVIDIA GTX 680?
On the basis of my own insight, I would say that their is no definitive answer but performance wise-both fall in the same segment/category.
-----------------
Testing:-
Test setup and drivers:
I recently came across a review which features test setup very similar to mine (NOTE: Details of my PC are in my profile):
Spoiler: show
Source: ASUS Matrix HD 7970 Platinum: L'ammiraglia ASUS è di nuovo a caccia di record
IMPORTANT: This review highlights the importance of performance improvements offered by driver related optimizations for applications with an example of a AMD Catalyst 12.11 beta driver version. In response to performance improvements offered by AMD Catalyst 12.11 driver series in many applications, NVIDIA released GeForce R310 driver series which also incorporate optimizations for many applications. I am currently using GeForce 310.90 WHQL driver release; it offers performance improvements in very few games in comparison to previous versions of R310 driver releases and these games aren't featured in this comparison. In addition, the processor in the test setup involved in the review is clocked @ 4 GHz continuous whereas mine is set at EIST; this means that my processor have been operating at relatively lower clock speeds in comparison to the processor involved in the review and is unlikely to influence performance in applications in the same manner as the processor in the review. Furthermore, my graphics card is running at stock settings. Therefore, due to all of the factors highlighted, any potential benefits on my end have been effectively minimized/eliminated to ensure that this comparison is as fair as possible. I have conducted all tests @ 1080p resolution (maximum resolution of my monitor) and matching settings in applications as in the aforementioned review.
UPDATE: Upon insistence of some members, I have incorporated some comparisons which are based on another review in which AMD Catalyst 13.1 WHQL driver have been used but the review's test setup is different from mine.
The synthetics:-
3DMark11
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 10176
AMD HD 7970 GE: 10109
AMD HD 7970 standard: 9302
GTX 680 standard: 9471
EVGE GTX 680 SC Signature: 10439
Observation: 3DMark11 is designed to make extensive use of features such as Tesselation, Compute Shaders and Multi-threading. Due to this balancing factor, performance comparisons are likely to be fair in this application. Noticeable performance gap between my GPU and GTX 680 standard in this comparison is due to an older (non-R310 series) driver version used in the review which lacks in optimization for this application.
------
3DMark Vantage (GPU Scores only)
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 37338
AMD HD 7970 GE: 37081
AMD HD 7970 standard: 34889
GTX 680 standard: 34367
EVGA GTX 680 SC Signature: 36625
Observation: 3DMark Vantage is sensitive to characteristics such as clock speeds, bandwidth and/or ROPs. AMD Tahiti products have advantage in these aspects.
Technologies:-
Unigine Heaven 3.0
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 57 FPS (AVG)
AMD HD 7970 GE: 55 FPS (AVG)
AMD HD 7970 standard: 53 FPS (AVG)
GTX 680 standard: 58 FPS (AVG)
EVGA GTX 680 SC Signature: 63 FPS (AVG)
Observation: This program is designed to take advantage of nearly all aspects of GPU capabilities. Heavy use of tesselation changed the game in favour of NVIDIA Kepler comfortably in this comparison.
------
ComputeMark
ASUS HD 7970 Matrix Platinum: 3756
AMD HD 7970 GE: 3574
AMD HD 7970 standard: 3305
GTX 680 standard: 2939
EVGA GTX 680 SC Signature: 3129
Observation: Compute shaders is undoubtedly strong point of AMD Tahiti architecture.
------
TessMark
ASUS HD 7970 Matrix Platinum: 78 FPS (AVG)
AMD HD 7970 GE: 75 FPS (AVG)
AMD HD 7970 standard: 69 FPS (AVG)
GTX 680 standard: 119 FPS (AVG)
EVGA GTX 680 SC Signature: 130 FPS (AVG)
Observation: Tesselation is undoubtedly strong point of NVIDIA Kepler architecture.
The games:-
PLA
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 76 FPS (AVG)
AMD HD 7970 GE: 74 FPS (AVG)
AMD HD 7970 standard: 68 FPS (AVG)
GTX 680 standard: 75 FPS (AVG)
EVGA GTX 680 SC Signature: 87 FPS (AVG)
Observation: Seems balanced in the context of hardware utilization. This application is sensitive to nearly all aspects of hardware.
------
Crysis 2 (Comparison 1)
Time Square Map:
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 81 FPS (AVG)
AMD HD 7970 GE: 78 FPS (AVG)
AMD HD 7970 standard: 72 FPS (AVG)
GTX 680 standard: 70 FPS (AVG)
EVGA GTX 680 SC Signature: 79 FPS (AVG)
Central Park Map:
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 67 FPS (AVG)
AMD HD 7970 GE: 65 FPS (AVG)
AMD HD 7970 standard: 59 FPS (AVG)
GTX 680 standard: 68 FPS (AVG)
EVGA GTX 680 SC Signature: 76 FPS (AVG)
Observation: Seems balanced in the context of hardware utilization. This application is sensitive to nearly all aspects of hardware. However, these comparisons exclude the use of tesselation setting which may tilt the performance in favour of NVIDIA Kepler if enabled.
-------
Crysis 2 (Comparison 2)
Mine:
FPS (MIN): 32
FPS (AVG): 83
FPS (MIN): 24
FPS (AVG): 79
Observation: This application is sensitive to nearly all aspects of hardware. However, these comparisons exclude the use of tesselation setting which may tilt the performance in favour of NVIDIA Kepler if enabled. My GPU dominates in this application, regardless of settings.
DiRT 3
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 140 FPS (AVG)
AMD HD 7970 GE: 138 FPS (AVG)
AMD HD 7970 standard: 124 FPS (AVG)
GTX 680 standard: 134 FPS (AVG)
EVGA GTX 680 SC Signature: 137 FPS (AVG)
---
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 126 FPS (AVG)
AMD HD 7970 GE: 124 FPS (AVG)
AMD HD 7970 standard: 111 FPS (AVG)
GTX 680 standard: 126 FPS (AVG)
EVGA GTX 680 SC Signature: 126 FPS (AVG)
---
NOTE: Focus on RED Bars only
ASUS HD 7970 Matrix Platinum: 111 FPS (AVG)
AMD HD 7970 GE: 108 FPS (AVG)
AMD HD 7970 standard: 98 FPS (AVG)
GTX 680 standard: 109 FPS (AVG)
EVGA GTX 680 SC Signature: 111 FPS (AVG)
Observation: This game is AMD oriented in marketing context; the Colin McRae series generally is. However, this particular game may not be much AMD oriented in technical context as the performance numbers seem to suggest.
What about FXAA? Does it makes difference?
Answer is YES.
Consider the example of PLA application for this discussion.
Test run without AA:-
Test run with FXAA 8x:-
Test run with MSAA 8x:-
NOTES:
I may include more comparisons in this thread in the future to further solidify my personal insight about this subject. In my debates on this subject, I have always stressed on the fact that comparisons between hardware are not just technical specifications dependent but also application/programming dependent. In addition, this thread also discloses some merits of both NVIDIA Kepler and AMD Tahiti architectures and how these merits can make difference in applications designed to support them. Competition between AMD Tahiti and NVIDIA Kepler is not so black and white as some reviews would want you to believe.
Now some may argue that situation may change in higher resolutions but I disagree because this is also a theoretical assumption. Even in higher resolutions, picture will remain the same as is the case in this comparison with minor tilts here and there.
Here is an example:
Spoiler: show
In addition, examples of Lost Planet 2 and Sniper Elite v2 are also in front of you.
It may be possible that raw power oriented technical characteristics such as; greater memory interface; larger VRAM size; and significantly greater bandwidth output; of AMD HD 7970 series may permit it to; close the gap; or help it equalize performance; or grant it advantage; in some applications against NVIDIA GTX 680 series at extreme resolutions but NVIDIA also offers GTX 680 with 4 GB of VRAM as an alternative. Once again, performances in applications at larger resolutions will be application/programming dependent and not just technical specifications dependent.
CONCLUSION: AMD GCN and NVIDIA Kepler architectures have their respective strong points (merits) and weak points (demerits) and these architectures have been designed on different principles. Game developers, working with GPU manufacturers, are able to program their games on the basis of these points. Therefore, performance of hardware (regardless of brand) in applications will be dependent upon their programming nature; the applications may be designed to take advantage of merits of AMD Tahiti architecture and/or they may be designed to take advantage of merits of NVIDIA Kepler architecture.
DISCLAIMER (IMPORTANT):
Spoiler: show
Last edited:
