***UPDATE: changed my recommended values from 1.7x to 1.4x-1.6x... I may crunch the numbers in the comparison part again, but they shouldn't stray too far from their current values.
***UPDATE 2: check this out!: http://www.yougamers.com/hardw...ormance/?mainnavi=true
***UPDATE 3: some slight refinements and additions here and there.
The following is a guide and tool to help you determine the ideal price/performance ratio amongst video cards (although it could just as easily be used for CPU's, etc.; anything that follows a Moore's Law-esque progressive rate):
So, if card A is 50fps and card B is 75fps (a 50% increase in performance), but costs 50% more money, then it's NOT worth the added cost; in fact, I argue that you are actually paying MORE for LESS (even though they both yield the same fps/$ figure and card B performs 50% better). Thanks to progress/Moore's Law, we should get MORE performance for the SAME amount of money (well, including inflation). So, in terms of Moore's law, the performance/cost ratio between generations should always be about (going by theoretical performance) 2:1. So, in other words, if card A is 50fps and card B is 70-90fps of real-world performance or more and costs not significantly more... it's worth it! Anything less performance-wise/anything more cost-wise is probably not worth it (unless of course you need it for a specific application). So, a 1% increase in price should improve --theoretical-- performance by 2% to be worth it; 'theoretical' as in meaning an increase in say stream processors or clock speed, rather than real-world performance measured using framerate, which will always be lower... 70-90% real-world performance compared to theoretical specs is a pretty good range to expect. If assuming an 85% efficiency (which might be a little optimistic) over a 2x theoretical performance upgrade, one can expect a performance boost of 1.7x. This is the figure I chose to go with because I put a focus on value, but a more conservative estimate might be in the 1.4x-1.6x range. In fact, based on a fair bit of statistical analysis between generations, I conclude that a 1.5x average increase in real-world performance seems to occur (this is based on the averages of several benchmarks looking at 7800gtx- 8800gtx- gtx 280).
To go into more detail, one way to determine the price/performance ratio is to divide the framerate by the price-- a higher relative number is better. To compare the price/performance ratio between two cards always only compare within the same exact benchmark! So, say if card A is rated 0.5 (fps divided by price; e.g., 50fps and costs $100) and card B is rated the same (0.5, e.g., 100fps, $200), even though card B has a higher actual framerate, you'd think it would be better because they have the same performance/price ratio, but it's not because it costs more. This is because although its performance is 100% higher, its price is also 100% higher! Card B has a 1:1 ratio of increase vs Card A. Remember, we should be getting MORE (performance) for the SAME money OR, SAME performance for LESS money (not same for the same, which is a poor 1:1 ratio)-- imagine every time you walked into a computer store and said you wanted a higher performing CPU they said it would cost more (not counting inflation), like if at one point you wanted a 2GHz CPU and it cost $200, and then the next generation CPU came out (with twice the performance, say achieved with twice the clock speed) and you came in and the new 4GHz CPU cost $400...! This is a 1:1 ratio of increase. Soon, no one could (or would want to) buy anything! So, a truly good buy/deal/upgrade (of course depending on your applications/needs) compared to card A in this example would have a (real world) fps/price rating of about 0.85 (vs 0.5, or 1.7x higher) or higher for the same price, or a theoretical spec or performance/price rating of 1.0 or better (if you're too lazy to look at benchmarks ). So, card A with a lower actual framerate has a better price/performance ratio than Card B even though they both might be rated 0.5 (price divided by framerate). This kind of thinking, that double the performance is worth double the money, has led to the huge and hugely expensive monster cards of today and accordingly crappy price/performance ratios of these ultra high-end cards (*cough GTX 280 *cough-- especially at its launch price, yikes!). Interestingly, you can also relate this thinking to power consumption trends as well (like performance/watt). Although, certainly you can find poor examples of price/performance at the low-end too (or nearly anywhere in the spectrum of cards/prices). Take, for example, the current 9500 GT; with at most 1/3 the performance level of the 4670 for about the same price... that's terrible value! To make a very vague, general statement, typically the ideal price/performance ratio is in the $75-200 range.
------------------------------------
---Now, take a look at some rather limited real-world examples (dated December 2008); compare the 4670 for $61 to the more expensive, higher performing card, 4850 for $130, divide that by the price of the lower end, cheaper card, 4670 for $61= 2.13, turn that into a percentage-- which is 113% (how much more expensive it is) multiply that number by at least 1.7x (to compare real-world fps) and you should arrive at how much, at least, better (real-world) performing the 4850 would have to be to be considered a worthy upgrade; therefore, the 4850 ought to perform at least 192% better (or 2.92x better) than the 4670... but, instead, it only performs almost twice as well-- for more than twice the price, a worse than even a 1:1 ratio (36fps vs 66fps). But even if the 4850 cost only $120 and did actually perform twice as well, it would still only be a very poor 1:1 performance/price ratio.
---Let's look at the 9500gt for $57 compared to the 4670 for $61. Take 61 divided by 57= a price difference of 1.07x or 7% more; so multiply the 7 by 1.7x and you find that it should perform at least 12% better than the 9500gt to be a worthy upgrade. Take 53fps (4670) divided by 20fps (9500gt) and you see that it in fact performs 2.65x or 165% better! Obscenely good value!
---Now let's look at the 4830 for $100, vs the 4670: a price difference of 1.64x or 64% more. So multiply 1.7x this amount and we find that it should perform at least 109% better (2.09x) to justify the price increase. In fact, it performs 34.4fps vs 25.5fps for the 4670, only 1.35x better (35% better), or, put another way, the amount of increase in its performance is only a fraction of what it should at least be (it should at least be 53fps) to be a worthy upgrade over the 4670.
---Now let's check out the 3650 $55 vs the 4670 ($61): a price difference of 1.11x or 11% higher. So, 1.7x 11%= 18.7% higher performance to be a worthy upgrade. The 3650 gets 20fps vs 53fps for the 4670, it performs 2.65x better (sound familiar? yes, the 3650 was competing against the 9500gt last generation) or 165% better performance! Compare this with the fact that to be a worthy upgrade, despite the price increase, it would only have to perform 18.7% higher! Which is to say that it performs 8.8x better than what would be considered the minimally acceptable increase in performance relative to its increased price! This is a performance/price ratio of 15:1 over the 3650!!! Which is to say that for every 1% increase in price from the 3650's price, you get a 15% increase in performance with the 4670! Which is like upgrading FOUR generations at once (and for only $6!!!)-- you could perhaps argue that that's like going from a geforce 4 to a geforce 8 in terms of the theoretical value/performance ratio offered. A big part of why this is the case is because of the fierce competition between ATI and NVIDIA (with ATI being mostly responsible for the hard downward pressure on prices for the relative performance). But now you know why these companies spend SO MUCH on trying to convince people that their high-end/poor performance-price cards are worth it.
You may want to adjust the desired minimum ratio to higher or lower than 1.7x, but I don't recommend going much lower than this amount in order to offer a truly compelling value/performance for your upgrade (1.4x-1.6x or higher is the recommended minimum range-- increase according to how much value you would like to see for your dollar/upgrade depending on application).
The 9600gso has worse performance for the same or higher money, the 9600gt is a terrible buy at $100, so is the 3870 also at $100, the 9800gt is competitive with the 4830 at the same price point (~$100) and offers similar performance, which is to say that even they both have poor performance/price ratios vs the 4670. And of course all higher-end cards such as the 4870 and gtx 260 at $300+ offer far worse value.
Essentially, the 4670 is a genius buy, especially because it shifts the optimal performance/price point down to the $60 mark. Amazing! And ya gotta admit that the fact that a $60 card can play crysis on stock High and average 26fps at 1280x1024 is pretty darn impressive. FYI: 4670= slightly better than 8800gs performance for MUCH less.
Looking at dual card solutions, if we take a 2x 4850 (crossfire) solution, it will offer better performance/price ratio than a single gtx 280, 260, or even 4870 (although a practical consequence of this, of course, means that if your needs change, you cannot add another card). However, from a purely performance/price ratio, ALL crossfire/SLI solutions are poor as they all offer about 0.7:1 ratios (not even the already poor 1:1) at best compared to the original first card (i.e., they can only boost performance by 70% at best, usually, and obviously always less than 100%, which is 1:1). However, if you upgrade and add a second card later when prices have dramatically fallen, it may very well make good price/performance sense.
-------------------------------------
Another good reason to not exceed the curve of the rate of progress (performance/price over time) is that if you reach ahead of the curve now (say going for a 4830 or 9800gt), you are not only getting truly lousy value today, but when the next generation comes out, you will have to at least stay at the same point ahead of the curve to experience that nice up to ~2x performance increase-- thus starting a trend of continuously bad performance for your dollar every time you upgrade. To figure out if it's worth upgrading from your current card, you simply use the price you originally paid for it and compare it with the price/performance of the new card. If it isn't around the 1.4:1 or higher range, then don't upgrade (though depending upon your needs). And of course, ideally you would compare performance using fully identical benchmarks and hardware/setup.
Limitations of this guide include that it assumes an identical feature set between cards; this is usually not the case. Another is that it fails to take into account the relevant applications that you intend to use your card for. If, for example, you need game x to run at an average smooth framerate of 35fps+ and that is all you intend to use it for, then anything significantly above this performance level is irrelevant (as is that which is significantly lower); and, certainly, try to look at benchmarks with similar settings and resolutions with what you'll be using.
Finally, it is worth noting that upgrading less than once every generation (defined roughly as a doubling in theoretical performance) is a good way to improve your performance/price ratio even further and to even get MORE for LESS (rather than just more for the same amount of money)! That is to say that by waiting another generation, one can improve the ratio to 2x-2.56x from 1.4x-1.6x for the same amount of money. To put it into perspective, upgrading once every generation (which seems to yield an average of a 1.5x increase in real-world performance) is like going from 30fps to 45fps (1.5x), so it doesn't make a particularly compelling argument (unless of course your favourite game is stuttering and you don't mind shelling out); however, upgrading once every other generation is like going from 30fps to 67.5fps (1.5x1.5= 2.25x). So, if this is your strategy shoot for at least 2.2x (slightly below average to give you a guaranteed minimum performance increase). I should also note, however, that all games are designed to be run on several different performance levels, so a 2.2x difference may not necessarily change what games you can and cannot play, therefore, you may want to upgrade every third generation (yielding 101fps; 67.5x1.5). Upgrading ever third generation is probably the limit at which you can maintain your ability to play all or most games between each upgrade (e.g., 3 years), depending upon other hardware too, of course.
PLEASE NOTE: Even if your "needs" fall at a lower-value performance/price ratio (i.e., you need higher-performing cards), then please still at least take away from this guide the underlying principle that 1:1 is bad value. Even at your higher performance level, you can still use the principles/methods here to help determine the best performance/price ratio FOR YOU.
PLEASE READ my first post on page 5 of the thread before posting, thanks. When the formatting is ambiguous, my responses are listed in bold. If your question/criticism is the same as has already been raised and responded to, I will not bother responding. Thanks
----
We're done here, folks. And with that, allow me to say a few things:
djayjp: Amber asked you to stop namecalling, only for you to call garritynet "pathetic". We have rules for good reason, and no one is to be above that (not even the mods). Regardless of what other people do, it is never appropriate for you to break the rules; "but he started it" has no standing here, you should always be on your best behavior. For future reference please follow instructions from the moderators, I do not expect we'll give you a warning in the future.
Furthermore in accordance with the needs of community, I must ask you to deeply rethink your guide. While everyone is entitled to their opinion, opinions become problems when they are consistently bad advice. As others have correctly pointed out, this guide is bad advice because it takes only in to consideration a few factors, fewer than need to be considered, resulting in a bad recommendation. What you want to do is noble and I completely understand where you want to go, but the product does not match the intentions. Please do not repost this guide.
The rest of you: I'm writing down names and you know who you are. Just because the op is giving bad advice, does not give you the right to treat him poorly - these are things I should not need to tell you.
-ViRGE
***UPDATE 2: check this out!: http://www.yougamers.com/hardw...ormance/?mainnavi=true
***UPDATE 3: some slight refinements and additions here and there.
The following is a guide and tool to help you determine the ideal price/performance ratio amongst video cards (although it could just as easily be used for CPU's, etc.; anything that follows a Moore's Law-esque progressive rate):
So, if card A is 50fps and card B is 75fps (a 50% increase in performance), but costs 50% more money, then it's NOT worth the added cost; in fact, I argue that you are actually paying MORE for LESS (even though they both yield the same fps/$ figure and card B performs 50% better). Thanks to progress/Moore's Law, we should get MORE performance for the SAME amount of money (well, including inflation). So, in terms of Moore's law, the performance/cost ratio between generations should always be about (going by theoretical performance) 2:1. So, in other words, if card A is 50fps and card B is 70-90fps of real-world performance or more and costs not significantly more... it's worth it! Anything less performance-wise/anything more cost-wise is probably not worth it (unless of course you need it for a specific application). So, a 1% increase in price should improve --theoretical-- performance by 2% to be worth it; 'theoretical' as in meaning an increase in say stream processors or clock speed, rather than real-world performance measured using framerate, which will always be lower... 70-90% real-world performance compared to theoretical specs is a pretty good range to expect. If assuming an 85% efficiency (which might be a little optimistic) over a 2x theoretical performance upgrade, one can expect a performance boost of 1.7x. This is the figure I chose to go with because I put a focus on value, but a more conservative estimate might be in the 1.4x-1.6x range. In fact, based on a fair bit of statistical analysis between generations, I conclude that a 1.5x average increase in real-world performance seems to occur (this is based on the averages of several benchmarks looking at 7800gtx- 8800gtx- gtx 280).
To go into more detail, one way to determine the price/performance ratio is to divide the framerate by the price-- a higher relative number is better. To compare the price/performance ratio between two cards always only compare within the same exact benchmark! So, say if card A is rated 0.5 (fps divided by price; e.g., 50fps and costs $100) and card B is rated the same (0.5, e.g., 100fps, $200), even though card B has a higher actual framerate, you'd think it would be better because they have the same performance/price ratio, but it's not because it costs more. This is because although its performance is 100% higher, its price is also 100% higher! Card B has a 1:1 ratio of increase vs Card A. Remember, we should be getting MORE (performance) for the SAME money OR, SAME performance for LESS money (not same for the same, which is a poor 1:1 ratio)-- imagine every time you walked into a computer store and said you wanted a higher performing CPU they said it would cost more (not counting inflation), like if at one point you wanted a 2GHz CPU and it cost $200, and then the next generation CPU came out (with twice the performance, say achieved with twice the clock speed) and you came in and the new 4GHz CPU cost $400...! This is a 1:1 ratio of increase. Soon, no one could (or would want to) buy anything! So, a truly good buy/deal/upgrade (of course depending on your applications/needs) compared to card A in this example would have a (real world) fps/price rating of about 0.85 (vs 0.5, or 1.7x higher) or higher for the same price, or a theoretical spec or performance/price rating of 1.0 or better (if you're too lazy to look at benchmarks ). So, card A with a lower actual framerate has a better price/performance ratio than Card B even though they both might be rated 0.5 (price divided by framerate). This kind of thinking, that double the performance is worth double the money, has led to the huge and hugely expensive monster cards of today and accordingly crappy price/performance ratios of these ultra high-end cards (*cough GTX 280 *cough-- especially at its launch price, yikes!). Interestingly, you can also relate this thinking to power consumption trends as well (like performance/watt). Although, certainly you can find poor examples of price/performance at the low-end too (or nearly anywhere in the spectrum of cards/prices). Take, for example, the current 9500 GT; with at most 1/3 the performance level of the 4670 for about the same price... that's terrible value! To make a very vague, general statement, typically the ideal price/performance ratio is in the $75-200 range.
------------------------------------
---Now, take a look at some rather limited real-world examples (dated December 2008); compare the 4670 for $61 to the more expensive, higher performing card, 4850 for $130, divide that by the price of the lower end, cheaper card, 4670 for $61= 2.13, turn that into a percentage-- which is 113% (how much more expensive it is) multiply that number by at least 1.7x (to compare real-world fps) and you should arrive at how much, at least, better (real-world) performing the 4850 would have to be to be considered a worthy upgrade; therefore, the 4850 ought to perform at least 192% better (or 2.92x better) than the 4670... but, instead, it only performs almost twice as well-- for more than twice the price, a worse than even a 1:1 ratio (36fps vs 66fps). But even if the 4850 cost only $120 and did actually perform twice as well, it would still only be a very poor 1:1 performance/price ratio.
---Let's look at the 9500gt for $57 compared to the 4670 for $61. Take 61 divided by 57= a price difference of 1.07x or 7% more; so multiply the 7 by 1.7x and you find that it should perform at least 12% better than the 9500gt to be a worthy upgrade. Take 53fps (4670) divided by 20fps (9500gt) and you see that it in fact performs 2.65x or 165% better! Obscenely good value!
---Now let's look at the 4830 for $100, vs the 4670: a price difference of 1.64x or 64% more. So multiply 1.7x this amount and we find that it should perform at least 109% better (2.09x) to justify the price increase. In fact, it performs 34.4fps vs 25.5fps for the 4670, only 1.35x better (35% better), or, put another way, the amount of increase in its performance is only a fraction of what it should at least be (it should at least be 53fps) to be a worthy upgrade over the 4670.
---Now let's check out the 3650 $55 vs the 4670 ($61): a price difference of 1.11x or 11% higher. So, 1.7x 11%= 18.7% higher performance to be a worthy upgrade. The 3650 gets 20fps vs 53fps for the 4670, it performs 2.65x better (sound familiar? yes, the 3650 was competing against the 9500gt last generation) or 165% better performance! Compare this with the fact that to be a worthy upgrade, despite the price increase, it would only have to perform 18.7% higher! Which is to say that it performs 8.8x better than what would be considered the minimally acceptable increase in performance relative to its increased price! This is a performance/price ratio of 15:1 over the 3650!!! Which is to say that for every 1% increase in price from the 3650's price, you get a 15% increase in performance with the 4670! Which is like upgrading FOUR generations at once (and for only $6!!!)-- you could perhaps argue that that's like going from a geforce 4 to a geforce 8 in terms of the theoretical value/performance ratio offered. A big part of why this is the case is because of the fierce competition between ATI and NVIDIA (with ATI being mostly responsible for the hard downward pressure on prices for the relative performance). But now you know why these companies spend SO MUCH on trying to convince people that their high-end/poor performance-price cards are worth it.
You may want to adjust the desired minimum ratio to higher or lower than 1.7x, but I don't recommend going much lower than this amount in order to offer a truly compelling value/performance for your upgrade (1.4x-1.6x or higher is the recommended minimum range-- increase according to how much value you would like to see for your dollar/upgrade depending on application).
The 9600gso has worse performance for the same or higher money, the 9600gt is a terrible buy at $100, so is the 3870 also at $100, the 9800gt is competitive with the 4830 at the same price point (~$100) and offers similar performance, which is to say that even they both have poor performance/price ratios vs the 4670. And of course all higher-end cards such as the 4870 and gtx 260 at $300+ offer far worse value.
Essentially, the 4670 is a genius buy, especially because it shifts the optimal performance/price point down to the $60 mark. Amazing! And ya gotta admit that the fact that a $60 card can play crysis on stock High and average 26fps at 1280x1024 is pretty darn impressive. FYI: 4670= slightly better than 8800gs performance for MUCH less.
Looking at dual card solutions, if we take a 2x 4850 (crossfire) solution, it will offer better performance/price ratio than a single gtx 280, 260, or even 4870 (although a practical consequence of this, of course, means that if your needs change, you cannot add another card). However, from a purely performance/price ratio, ALL crossfire/SLI solutions are poor as they all offer about 0.7:1 ratios (not even the already poor 1:1) at best compared to the original first card (i.e., they can only boost performance by 70% at best, usually, and obviously always less than 100%, which is 1:1). However, if you upgrade and add a second card later when prices have dramatically fallen, it may very well make good price/performance sense.
-------------------------------------
Another good reason to not exceed the curve of the rate of progress (performance/price over time) is that if you reach ahead of the curve now (say going for a 4830 or 9800gt), you are not only getting truly lousy value today, but when the next generation comes out, you will have to at least stay at the same point ahead of the curve to experience that nice up to ~2x performance increase-- thus starting a trend of continuously bad performance for your dollar every time you upgrade. To figure out if it's worth upgrading from your current card, you simply use the price you originally paid for it and compare it with the price/performance of the new card. If it isn't around the 1.4:1 or higher range, then don't upgrade (though depending upon your needs). And of course, ideally you would compare performance using fully identical benchmarks and hardware/setup.
Limitations of this guide include that it assumes an identical feature set between cards; this is usually not the case. Another is that it fails to take into account the relevant applications that you intend to use your card for. If, for example, you need game x to run at an average smooth framerate of 35fps+ and that is all you intend to use it for, then anything significantly above this performance level is irrelevant (as is that which is significantly lower); and, certainly, try to look at benchmarks with similar settings and resolutions with what you'll be using.
Finally, it is worth noting that upgrading less than once every generation (defined roughly as a doubling in theoretical performance) is a good way to improve your performance/price ratio even further and to even get MORE for LESS (rather than just more for the same amount of money)! That is to say that by waiting another generation, one can improve the ratio to 2x-2.56x from 1.4x-1.6x for the same amount of money. To put it into perspective, upgrading once every generation (which seems to yield an average of a 1.5x increase in real-world performance) is like going from 30fps to 45fps (1.5x), so it doesn't make a particularly compelling argument (unless of course your favourite game is stuttering and you don't mind shelling out); however, upgrading once every other generation is like going from 30fps to 67.5fps (1.5x1.5= 2.25x). So, if this is your strategy shoot for at least 2.2x (slightly below average to give you a guaranteed minimum performance increase). I should also note, however, that all games are designed to be run on several different performance levels, so a 2.2x difference may not necessarily change what games you can and cannot play, therefore, you may want to upgrade every third generation (yielding 101fps; 67.5x1.5). Upgrading ever third generation is probably the limit at which you can maintain your ability to play all or most games between each upgrade (e.g., 3 years), depending upon other hardware too, of course.
PLEASE NOTE: Even if your "needs" fall at a lower-value performance/price ratio (i.e., you need higher-performing cards), then please still at least take away from this guide the underlying principle that 1:1 is bad value. Even at your higher performance level, you can still use the principles/methods here to help determine the best performance/price ratio FOR YOU.
PLEASE READ my first post on page 5 of the thread before posting, thanks. When the formatting is ambiguous, my responses are listed in bold. If your question/criticism is the same as has already been raised and responded to, I will not bother responding. Thanks
----
We're done here, folks. And with that, allow me to say a few things:
djayjp: Amber asked you to stop namecalling, only for you to call garritynet "pathetic". We have rules for good reason, and no one is to be above that (not even the mods). Regardless of what other people do, it is never appropriate for you to break the rules; "but he started it" has no standing here, you should always be on your best behavior. For future reference please follow instructions from the moderators, I do not expect we'll give you a warning in the future.
Furthermore in accordance with the needs of community, I must ask you to deeply rethink your guide. While everyone is entitled to their opinion, opinions become problems when they are consistently bad advice. As others have correctly pointed out, this guide is bad advice because it takes only in to consideration a few factors, fewer than need to be considered, resulting in a bad recommendation. What you want to do is noble and I completely understand where you want to go, but the product does not match the intentions. Please do not repost this guide.
The rest of you: I'm writing down names and you know who you are. Just because the op is giving bad advice, does not give you the right to treat him poorly - these are things I should not need to tell you.
-ViRGE