The Evolution of GPU Prices and Performance Scaling

In recent years, the prices of graphics processing units (GPUs) have seen significant increases, sparking debate among gamers and tech enthusiasts. The price inflation can be attributed to various factors, including the advancement of technology, market competition, and the shift in demand for lower-end GPUs.

Crossfiring with a friend's rig was an experiment that revealed interesting insights into GPU performance scaling. The friend's high-end system, powered by a Devil's Canyon i7 processor, was equipped with two cards: one with an r9290 and another with an r9390. To observe the effects of crossfiring, they ran the GPUs in tandem, noticing that the second card, which was supposedly at full potential, was running at X4 or BY4 (four tap PCIe). This anomaly was attributed to the motherboard's PCIE slot configuration, which may not have been electrically wired for a bus16 setup. However, if it were a full bus6 slot, further investigation would be necessary to confirm this.

Motherboard configurations and their impact on GPU performance cannot be overstated. Some motherboards may only support certain PCIe slots, even those with four slots, due to electrical wiring limitations. This can lead to reduced performance or even prevent the use of certain GPUs in crossfiring setups. It is essential to consult the motherboard manual to ensure compatibility and optimal performance.

The price inflation of GPUs over the past five years is staggering. In 2011, a GTX 560 TI was sold for $250, which doubled to $500 in 2012, with an increase of $50 for the GK 104 GTX 680, reaching $550. The Nvidia GeForce Founders Edition GM 204 followed suit, increasing prices by another $150, reaching $700 for the GP 104 equivalent in the 1080. This significant price hike raises concerns about market competition and the justification for such increases.

One possible explanation for the price inflation is the reduced demand for lower-end GPUs due to advancements in technology and the proliferation of more powerful cards that can handle demanding games at lower resolutions. The performance gap between previous generations' lower-end GPUs and their modern counterparts has narrowed, making it easier for gamers to get by with lower-end or mid-range cards.

Another factor contributing to the price inflation is the increased competition between Intel and AMD in the CPU market. As both companies push for dominance in the low-end market, they are driving up prices due to increased demand and reduced supply. This shift in the gaming PC market has created a gap in the middle range, where consumers can no longer afford traditional lower-end GPUs.

The impact of this price inflation is far-reaching, affecting not only GPU enthusiasts but also gamers who rely on mid-range cards for smooth performance. As prices continue to rise, it becomes increasingly difficult for companies to justify these increases without alienating their customers. The situation is further complicated by the fact that some AMD partners have not yet reached the $600 price point for their cards, which would be seen as a competitive benchmark.

The author of the article seems uncertain about the exact causes of the price inflation and invites discussion among readers. However, they do express concern about the price increase, citing the $700 Founders Edition 1080 as an example of this issue.

As we wait for the Polaris 10 series to be released, which will undoubtedly bring new challenges and opportunities in the GPU market, it is essential to remain vigilant and monitor developments closely. The article concludes by inviting readers to check back for further updates on upcoming events, including factory tours, headquarters visits, and participation in the CompY Tech show floor.

Patron Link: If you like this content, we'll be visiting Taiwan, China, and Maau next week, where we'll be attending factory tours, headquarters tours, and exploring the Compy Tech Show floor. We'll also be participating in some random walking around an abandoned mall, which promises to be a fun experience. Be sure to check back for more updates on our upcoming adventures.

"WEBVTTKind: captionsLanguage: enhey everyone we're back for another episode of ask GN and this is our last one before flying out to computex or type for computex we might do one there but I'm not sure it's going to be pretty crazy the first question for this episode I can't remember which video I saw it on but I do remember it being asked by one of the regulars so maybe Leo di or Street Guru but whoever it was asked how much variance there is in testing data and so the the question was basically uh you guys run three tests for parody on all the FPS benchmarks is there that much variance the answer is yes so uh a lot of these other outlets will present data that's maybe 22 games tested or 30 games benchmarked or something like that and that's just an absurd amount of benchmarks to run unless they have multiple systems running it's normally not really feasible in the review time that we have uh so if you run I I could certainly run more benchmarks if we only did one pass per game but the thing is with these even the automated benchmarks there's still a variance of sometimes 1 to 3 FPS and that's a big deal when you might be comparing things that are one to 3 FPS different so it just it depends on the game and the throttling where the game throttles if it's more intensive on one component or the other if you're testing gpus and it's more intensive on uh the CPU then you're going to hit CPU throttles and so if the CPU boosts or does any um energy control which we turn all that stuff off then you'll see fluctuations there but normally it's only a couple FPS so not enough to be an invalid tool for the test Suite but enough where you want to run it multiple times and uh that's just because if maybe the first test it card a tests at 76 FPS and card B tests at 73 FPS that looks like a pretty reasonable difference but uh over multiple test passes they might average out and be the same or very close to the same so yes there is actually difference and it is important to do multiple passes especially and this is the important part with the frame time so if anyone who's doing frame time testing should do multi-pass testing because uh especially with the 1% 0.1% lows that we use 0.1% might only be a couple frames that might be like five frames for the entire test sequence so that's a pretty low samp small sample size uh doing multiple tests will smooth out that data and make sure that from one to the next it looks consistent and if it's inconsistent in one of the passes we know we need to run it a few more times just to try and understand what's going on if we can't repr reproduce that inconsistency then we know that it's basically an outlier and we discard that particular piece of data so that hopefully answers that question next question is uh Mark Rau who says dude to benefit from the new GTX 1080 am I going to need a new CPU and motherboard my current CPU is an i5 4670 4 67 nonk I've been saving up for a rift uh my current GPU is a 780 and then some other words I'm just making sure there's no other questions in here do I need to save for a whole new machine if I want to see benefits from the 1080 uh so I haven't tested this yet but I can tell you from really brief testing I just did on Total War Warhammer which our benchmarks aren't up for that yet really brief testing there using a 593k which is more powerful than the 4670 uh the 59 30k at 4.2 GHz versus 36 or 8 GHz at a 15 or 20 FPS swing and that was with a GTX 1080 so what that tells us is that in the very least frequency in that particular game has a huge impact and will throttle the GPU uh in if you sort of extrapolate that any other games that act similarly that are maybe CPU bound a lot of racing games are CPU bound like grid and dirt those are instances where you will be bottlenecking the GTX 1080 on a 4670 because we were bottlenecking it on a 5930 so that would be certainly a consideration the 4670 is still a good CPU uh but I I need to test it more personally we are working on that but it'll be at least a few weeks till it gets up because of this compy Tech trip it is something that you may end up replacing though so just be aware of that the next question zero Sami Sami Sammy who says hi recently I switched to a mini ITX lowend motherboard but the former overclock I sustained on my R9 290 GPU couldn't be achieved on this new board what could cause this I only switched the motherboard and nothing else so I know for sure it's the motherboard but why it's possible depending on what you had for your previous board it's possible that maybe there were additional phases or something the better power cleaning going on that was stabilizing your clock so that would work on stabilizing voltage Supply to the GPU in that case maybe that's what it was but that's not super common uh for a change like that to to destroy an overclock it could also be maybe the I don't know maybe the pcie lanes are handled differently or something like that but I that doesn't really doesn't make a whole lot of sense to me so I would probably suspect two things I would look at bios or UEFI and see if there's anything in there that handles pcie timings or latency or pcie bus clock or something like that bus speed um that would be one place to look the next place to look would be uh the the power delivery design on the new board versus the old one and see if the old one is better and on Mini ITX that may be the case because they're really limited on on space so uh those are the two places I would look of course also look at your CPU clock because maybe well depends on what you mean by you can't achieve the same clock uh it may be the case though for gaming FPS that you were overclocked on one board and you're not on the new one for the CPU that would impact it too but if you're just straight not seeing the new clock be achieved on the GPU then it's one of the first two issues I thought of uh the next part of this question was also I crossfir r9290 with an r9390 in my friend's high in machine for fun to see performance scaling I noticed in the GPU panel that the second card ran at X4 or by4 or four tap pcie noting that my friend's rig has Devil's Canyon i7 uh that sounds like a motherboard potentially motherboard thing so I don't again don't know what motherboard this is it's not listed but um boards will even though it's there's a f length pcie slot that would normally be called x16 even on newag or Amazon a lot of those slots even if there's four on a board a lot of those slots will only be electrically wired for byy four or by eight and that's basically so they can put a full slot to fit the cards but with Crossfire you may end up running like a x X8 X4 X4 or something like that um so it's possible that the slot used was just not wired electrically for a bu16 setup but if it's a full bu6 slot you also need to check the manual and make sure that it's not suggesting that you run a different slot for SL Crossfire because that is also something that happens uh next question final question Homer Thompson says Steve what do you make of the price inflation we've seen in gpus The Last 5 Years in 2011 a GTX 560 TI was sold for 250 in 2012 doubled for 500 for a full size the GK 104 gtx680 $50 price increase to 550 for the Nvidia uncut GM 204 and then another increase to $700 for the equivalent uncut GP 104 in the 1080 how much until we're paying $350 for the small GM 206 equivalent in gtx960 uh yeah that's a real concern so the price gains are there's a few things I'm not 100% sure what I think of this just yet I'm still thinking about it but one thing is these cards like the the 560ti GPU equivalent of today they're a lot more powerful and the games aren't necessarily growing in their demands a lot of the games anyway so um I guess the you can get away with a lowend or mid-range card a lot better now than you used to be able to so that's one thing that might help the companies justify the price increases the next thing is uh just a general competition um I guess misalignment so because Nvidia is not super threatened right now in their market share they have a ton of market share and just strictly speaking to gaming PCs we're not counting consoles here gaming PCs nvidia's aib presence is 70 something perc so they're certainly not feeling the the threat to lower their prices AMD is putting pressure On The Low End Market but they've always been there now another thing that I think is more relevant than that is uh Intel and AMD with their CPUs they're both really pushing this low-end market so the the demand for maybe a gt740 or something that demand has kind of evaporated because the reason you would buy one of those is if you didn't have on board video or you didn't have one that was good enough to play some simple games but you also didn't want a full end full mid-range or low-end GPU uh you would buy something like a gt740 or whatever now you don't need to do that because the CPUs have these igps apus have their uh their GPU components so those kind of invalidate the need for a lot of the really cheap gpus and I think that probably creates a weird Gap in the market where now the low n gpus are priced a little higher than they used to be and that of course raises the price of everything else but um of course there's also just general inflation but I'm not 100% sure what I think of that just yet I do I will say that $700 for Founders Edition 1080 is too much uh but the $600 price point seems good if any of the aib partners actually reach that price when they make their cards so that seems good 700 is certainly cheap uh we don't know what AMD stuff will cost yet but we'll see and and analyze that once we're reviewing the Polaris stuff Polaris 10 so that's all I got for you this week as always patreon link the post video If you like this content we'll be in Taiwan China and maau next week and we're going to be doing uh factory tours headquarters tours and the compy tech show floor and we'll probably do some kind of random walking around the giant empty abandoned mall and Shen so that'll be fun but check back for that thanks for watching I'll see you all next timehey everyone we're back for another episode of ask GN and this is our last one before flying out to computex or type for computex we might do one there but I'm not sure it's going to be pretty crazy the first question for this episode I can't remember which video I saw it on but I do remember it being asked by one of the regulars so maybe Leo di or Street Guru but whoever it was asked how much variance there is in testing data and so the the question was basically uh you guys run three tests for parody on all the FPS benchmarks is there that much variance the answer is yes so uh a lot of these other outlets will present data that's maybe 22 games tested or 30 games benchmarked or something like that and that's just an absurd amount of benchmarks to run unless they have multiple systems running it's normally not really feasible in the review time that we have uh so if you run I I could certainly run more benchmarks if we only did one pass per game but the thing is with these even the automated benchmarks there's still a variance of sometimes 1 to 3 FPS and that's a big deal when you might be comparing things that are one to 3 FPS different so it just it depends on the game and the throttling where the game throttles if it's more intensive on one component or the other if you're testing gpus and it's more intensive on uh the CPU then you're going to hit CPU throttles and so if the CPU boosts or does any um energy control which we turn all that stuff off then you'll see fluctuations there but normally it's only a couple FPS so not enough to be an invalid tool for the test Suite but enough where you want to run it multiple times and uh that's just because if maybe the first test it card a tests at 76 FPS and card B tests at 73 FPS that looks like a pretty reasonable difference but uh over multiple test passes they might average out and be the same or very close to the same so yes there is actually difference and it is important to do multiple passes especially and this is the important part with the frame time so if anyone who's doing frame time testing should do multi-pass testing because uh especially with the 1% 0.1% lows that we use 0.1% might only be a couple frames that might be like five frames for the entire test sequence so that's a pretty low samp small sample size uh doing multiple tests will smooth out that data and make sure that from one to the next it looks consistent and if it's inconsistent in one of the passes we know we need to run it a few more times just to try and understand what's going on if we can't repr reproduce that inconsistency then we know that it's basically an outlier and we discard that particular piece of data so that hopefully answers that question next question is uh Mark Rau who says dude to benefit from the new GTX 1080 am I going to need a new CPU and motherboard my current CPU is an i5 4670 4 67 nonk I've been saving up for a rift uh my current GPU is a 780 and then some other words I'm just making sure there's no other questions in here do I need to save for a whole new machine if I want to see benefits from the 1080 uh so I haven't tested this yet but I can tell you from really brief testing I just did on Total War Warhammer which our benchmarks aren't up for that yet really brief testing there using a 593k which is more powerful than the 4670 uh the 59 30k at 4.2 GHz versus 36 or 8 GHz at a 15 or 20 FPS swing and that was with a GTX 1080 so what that tells us is that in the very least frequency in that particular game has a huge impact and will throttle the GPU uh in if you sort of extrapolate that any other games that act similarly that are maybe CPU bound a lot of racing games are CPU bound like grid and dirt those are instances where you will be bottlenecking the GTX 1080 on a 4670 because we were bottlenecking it on a 5930 so that would be certainly a consideration the 4670 is still a good CPU uh but I I need to test it more personally we are working on that but it'll be at least a few weeks till it gets up because of this compy Tech trip it is something that you may end up replacing though so just be aware of that the next question zero Sami Sami Sammy who says hi recently I switched to a mini ITX lowend motherboard but the former overclock I sustained on my R9 290 GPU couldn't be achieved on this new board what could cause this I only switched the motherboard and nothing else so I know for sure it's the motherboard but why it's possible depending on what you had for your previous board it's possible that maybe there were additional phases or something the better power cleaning going on that was stabilizing your clock so that would work on stabilizing voltage Supply to the GPU in that case maybe that's what it was but that's not super common uh for a change like that to to destroy an overclock it could also be maybe the I don't know maybe the pcie lanes are handled differently or something like that but I that doesn't really doesn't make a whole lot of sense to me so I would probably suspect two things I would look at bios or UEFI and see if there's anything in there that handles pcie timings or latency or pcie bus clock or something like that bus speed um that would be one place to look the next place to look would be uh the the power delivery design on the new board versus the old one and see if the old one is better and on Mini ITX that may be the case because they're really limited on on space so uh those are the two places I would look of course also look at your CPU clock because maybe well depends on what you mean by you can't achieve the same clock uh it may be the case though for gaming FPS that you were overclocked on one board and you're not on the new one for the CPU that would impact it too but if you're just straight not seeing the new clock be achieved on the GPU then it's one of the first two issues I thought of uh the next part of this question was also I crossfir r9290 with an r9390 in my friend's high in machine for fun to see performance scaling I noticed in the GPU panel that the second card ran at X4 or by4 or four tap pcie noting that my friend's rig has Devil's Canyon i7 uh that sounds like a motherboard potentially motherboard thing so I don't again don't know what motherboard this is it's not listed but um boards will even though it's there's a f length pcie slot that would normally be called x16 even on newag or Amazon a lot of those slots even if there's four on a board a lot of those slots will only be electrically wired for byy four or by eight and that's basically so they can put a full slot to fit the cards but with Crossfire you may end up running like a x X8 X4 X4 or something like that um so it's possible that the slot used was just not wired electrically for a bu16 setup but if it's a full bu6 slot you also need to check the manual and make sure that it's not suggesting that you run a different slot for SL Crossfire because that is also something that happens uh next question final question Homer Thompson says Steve what do you make of the price inflation we've seen in gpus The Last 5 Years in 2011 a GTX 560 TI was sold for 250 in 2012 doubled for 500 for a full size the GK 104 gtx680 $50 price increase to 550 for the Nvidia uncut GM 204 and then another increase to $700 for the equivalent uncut GP 104 in the 1080 how much until we're paying $350 for the small GM 206 equivalent in gtx960 uh yeah that's a real concern so the price gains are there's a few things I'm not 100% sure what I think of this just yet I'm still thinking about it but one thing is these cards like the the 560ti GPU equivalent of today they're a lot more powerful and the games aren't necessarily growing in their demands a lot of the games anyway so um I guess the you can get away with a lowend or mid-range card a lot better now than you used to be able to so that's one thing that might help the companies justify the price increases the next thing is uh just a general competition um I guess misalignment so because Nvidia is not super threatened right now in their market share they have a ton of market share and just strictly speaking to gaming PCs we're not counting consoles here gaming PCs nvidia's aib presence is 70 something perc so they're certainly not feeling the the threat to lower their prices AMD is putting pressure On The Low End Market but they've always been there now another thing that I think is more relevant than that is uh Intel and AMD with their CPUs they're both really pushing this low-end market so the the demand for maybe a gt740 or something that demand has kind of evaporated because the reason you would buy one of those is if you didn't have on board video or you didn't have one that was good enough to play some simple games but you also didn't want a full end full mid-range or low-end GPU uh you would buy something like a gt740 or whatever now you don't need to do that because the CPUs have these igps apus have their uh their GPU components so those kind of invalidate the need for a lot of the really cheap gpus and I think that probably creates a weird Gap in the market where now the low n gpus are priced a little higher than they used to be and that of course raises the price of everything else but um of course there's also just general inflation but I'm not 100% sure what I think of that just yet I do I will say that $700 for Founders Edition 1080 is too much uh but the $600 price point seems good if any of the aib partners actually reach that price when they make their cards so that seems good 700 is certainly cheap uh we don't know what AMD stuff will cost yet but we'll see and and analyze that once we're reviewing the Polaris stuff Polaris 10 so that's all I got for you this week as always patreon link the post video If you like this content we'll be in Taiwan China and maau next week and we're going to be doing uh factory tours headquarters tours and the compy tech show floor and we'll probably do some kind of random walking around the giant empty abandoned mall and Shen so that'll be fun but check back for that thanks for watching I'll see you all next time\n"