GPU Cost Per FPS with Digital Foundry

[skeeder]

The last few months have been a clusker of information, driver updates, card launches, and so many other people just full of opinions it is insane.

My search for a replacement card for an aging 550Ti was starting to become this mind spinning task since my budget was between $200 and $300.

So many options, this above that, blah blah blah. I ended up using Digital Foundry's benchmark videos and snagged average FPS at the same time for each card type listed here I didn't do all the games, but wanted to grab DX12 and DX11 games. If you're unfamiliar with DF, they do great work on Youtube.
DigitalFoundry

So here is what I put together (this is not sponsored by DF at all). These are at 1080P Ultra Settings.

Note: 970 and 980 prices are what I was seeing in the forum. Truth is the 980 has held unto its pricing rather well, $250 is the lowest used price I saw, the last batch sold for $280ish it appears.

The prices were as of 9/7/16 in the morning AND IN STOCK, (eVGA...sigh...really?...)

I used brands I would actually buy so it's not a scientific fact for everyone. (eVGA, Asus, MSI in that order. Gigabyte's Windforce Model had many reviews stating it felt cheap, it wasn't worth saving the $10 over the $279 Asus model).

Dollar for frame, the 1060 3GB is actually not as terrible as the [hard] people around here would lead you to believe, DF did note some microstuttering in benches which is why I avoided it. Heat and Power requirements for one of my rigs removed the 970 and 980 and while the 470 did well in both areas it wasn't as friendly to DX11 titles that I pretty much play.

I ended up ordering a 6GB Asus 1060 for $279.99 on Amazon this morning.

I'm posting this for those still digging and looking for that same price range, you could easily reproduce the results and edit the prices to see if your fate changes, it's not a [h]ardocp article, but it might help someone out.

Cheers!

 

[BroHamBone]

Lol....and the CPU being used?

Congrats on the new card though.

 

[skeeder]

Lol....and the CPU being used?

Congrats on the new card though.

Digital foundry did all the tests with a 6700k.

I will be pairing it with a SandyBridge i5 until I upgrade.

 

[Schro]

The thing that I read into just looking at your chart, is that for the game, graphics settings and resolutions in the chart, you're not going to see a bit of difference in performance between the most expensive and least expensive cards. I would assume the data is at 1080p and is fairly maxed out - the only reason to go higher than the cheapest card in that stack would be for anticipation of future games or to support a higher resolution (like 1440p or higher).

 

[Dayaks]

Nice chart. At least you based your decision on something. The 1060 6GB should serve you well!

 

[Brian_B]

Congrats on your purchase.

 

[big_aug]

I got my used 980 for $200 shipped a couple weeks ago on the forums. Deals definitely to be had with people upgrading.

 

[skeeder]

I got my used 980 for $200 shipped a couple weeks ago on the forums. Deals definitely to be had with people upgrading.

I originally was hoping to grab one for that price, it never materialized.

 

[chenw]

The problem with religiously using $/fps to determine your GPU purchase is that it's the wrong figure to use.

The proper metric to use should be $/fps_increase, because when you buy GPU, you are not buying all of the FPS the card provides, you are paying for the fps your old GPU(s) weren't providing.

If you took that into account, the actual best GPU for the money may not have been 1060 (it could very well still be 1060, just not necessarily).

I'll give an example:

Your currently GPU gives you say 50fps in a game.

Your GPU choice #1 runs the game at 70fps, and costs $50

Your GPU choice #2 runs the game at 100 fps, and costs $100.

Using $/fps metric, your GPU choice #1 would be the better deal.

BUT, what you REALLY paid for with the $50 isn't the whole of 70fps, it's the 20fps increase over what you have.

So with #1, you paid $50 for 20fps increase
with #2, you paid $100 for 50fps increase.

The actual value you'd get depends on the fps you already have. EG if your GPU were running only 20fps, then #1 would still be better for the money (50fps for $50 vs 80fps for $100).

 

[muz_j]

I ended up ordering a 6GB Asus 1060 for $279.99 on Amazon this morning.!

So did I - and I'm really happy with it as an upgrade from a 660 Ti for 1080P gaming.
I'd add two comments - the GPU generation is relevant in terms of features and support - which is worth keeping in mind when comparing cards from different generations.
And secondly, I tend to spend a couple of dollars extra and get a card with a good aftermarket cooling solution. The benefits being more efficient cooling, which often means a lower noise level for a given card - and some of the stock coolers aren't really that good on things like the RX 480's.
But all in all, that's a useful reference chart as a rough guide

The problem with religiously using $/fps to determine your GPU purchase is that it's the wrong figure to use.
The proper metric to use should be $/fps_increase, because when you buy GPU, you are not buying all of the FPS the card provides, you are paying for the fps your old GPU(s) weren't providing.

Good points and very worth while to anyone looking at a new card.
You've always got to compare any new hardware versus your current stuff. My own personal target for a new GPU was something with good bang for the buck and I wanted to basically double my 660 Ti's performance and at least double it's on board memory.
So having a clear idea of your end goal and your budget is all part of it for me.

 

[skeeder]

Good points and very worth while to anyone looking at a new card.
You've always got to compare any new hardware versus your current stuff. My own personal target for a new GPU was something with good bang for the buck and I wanted to basically double my 660 Ti's performance and at least double it's on board memory.
So having a clear idea of your end goal and your budget is all part of it for me.

I think chenw's point is more relevant for people upgrading from the 900 series. Since I'm coming from a 550Ti, virtually everything was upwards from there.

I didn't want the stock cooler either, never been a fan of blower style cards since my X1900XTX.

 

[mogar]

Just upgraded to a gtx1060 6gb as well from a 1gb HG6950, couldn't really lose with any of my choices either .

 

[evilwon]

I think chenw's point is more relevant for people upgrading from the 900 series. Since I'm coming from a 550Ti, virtually everything was upwards from there.

I didn't want the stock cooler either, never been a fan of blower style cards since my X1900XTX.

I concur...that or the AMD 3xx series or an AMD 29x series card.

I only purchased a 1070 (may upgrade to a 1080) because I want to game at 4k. Otherwise the 970 would do for almost everything.

 

[skeeder]

I concur...that or the AMD 3xx series or an AMD 29x series card.

I only purchased a 1070 (may upgrade to a 1080) because I want to game at 4k. Otherwise the 970 would do for almost everything.

The end destination for the 1060 is my htpc in about 1 year. I almost was able to save myself the cash and get one of those but the heat and power requirements didn't match my needs. For those doing an several generation upgrade, you'd be a fool to not consider a 970.

 

[SvenBent]

This way of evaluation price/performance of an upgrade is horrible wrong.

You try to think NewFPS/Cost = Money Efficiency, but that not right either because a lot of those FPS you already have. If you go buy 12 more glasses to your already set of 12 you don't say hmm 24glasses with this price on a 12pack glass is so and so. No you say I'm getting 12 MORE glasses with this cost show me my money efficiency.
So you should establish a baseline for 0 cost. aka you current FPS. Lets say 40 and then you look at what you are getting with the new card lets say 60. Then you need to take the DELTA and realize that is what you are actually getting from the upgrade. You delta is 60-40=20, so its 20/Cost = cost per upgraded FPS. Its not 60/cost.

That however leaves out what you might sell the old for so if you want to count that in as well it shouls be ( NewFPS - OldFPS ) / (NewCardCost - OldCardSellValue) and then you see exactly what you get of more FPS per spend buck.

You will notice that this weight heavily more to get cards with higher FPS even if they're are a bit more expensive per total FPS.


Please don't keep doing this stupid way of optimizing performance it directly hurts your money efficiency.

 

[skeeder]

This way of evaluation price/performance of an upgrade is horrible wrong.

You try to think NewFPS/Cost = Money Efficiency, but that not right either because a lot of those FPS you already have. If you go buy 12 more glasses to your already set of 12 you don't say hmm 24glasses with this price on a 12pack glass is so and so. No you say I'm getting 12 MORE glasses with this cost show me my money efficiency.
So you should establish a baseline for 0 cost. aka you current FPS. Lets say 40 and then you look at what you are getting with the new card lets say 60. Then you need to take the DELTA and realize that is what you are actually getting from the upgrade. You delta is 60-40=20, so its 20/Cost = cost per upgraded FPS. Its not 60/cost.

That however leaves out what you might sell the old for so if you want to count that in as well it shouls be ( NewFPS - OldFPS ) / (NewCardCost - OldCardSellValue) and then you see exactly what you get of more FPS per spend buck.

You will notice that this weight heavily more to get cards with higher FPS even if they're are a bit more expensive per total FPS.


Please don't keep doing this stupid way of optimizing performance it directly hurts your money efficiency.

That is only applicable to those selling their old cards. Last time I priced out a used 550Ti, someone was asking $40...and with skipping 5 generations of GPU, I'm not sure it would be worth while to bench 1080p ultra settings to truely do an apples to apples comparison.

You're welcome to.

 

[SvenBent]

That is only applicable to those selling their old cards. Last time I priced out a used 550Ti, someone was asking $40...and with skipping 5 generations of GPU, I'm not sure it would be worth while to bench 1080p ultra settings to truely do an apples to apples comparison.

You're welcome to.

If you cant sell the old card you just set the value of the old card tor 0. The math still pans out perfectly and gives you the price you are paying for the new FPS. The important pant is to realize that you are paying for an UPGRADE and not a new product. you already have an amount of FPS so its only the extra FPS that are of any value.

Lets say you have Card A and it gives you 40 fps. and you can buy card Card B for 60 bucks and it gives you 60fps and Card C gives you 90FPS but it cost 100 bucks
byt the OP FPS/cost method you would say oh card B is a better buy because you get 1FPS per buck and card C only gives me 0.9FPS per buck. But in reality you just paid 60dollars to get 20 more fps and you could have paid 100 buck to get 50 more fps
aka
Card B: 20newFPS/60bucks = 0.33newFPS per buck spend
Card C: 50newFPS/100bucks = 0.5newFPS per buck spend

Card C gives you more new FPS per buck spend than card B even though its looks expensive for it total FPS.



Worst case example if we have to go with the "Stupidway" you could pay for no gains at all
Card A 40fps you currently have. Card B gives you also 40fps for40 bucks. Card C gives 60FPS for 70bucks.
By the old way of just saying TotalFPS/Cost= moeny efficienct, you would say getting a card that perform the same as your current is a better buy, cause its cheaper per FPS. So now you are paying the 40 bucks to get no improvements... doesn't sound sane to me really


so again you have to realize its an upgrade and you have to realize what the upgrade is actually improving. Because THATS what you are paying for.



Another thing to realize is that you have pretty big pockets and must live a weird place if you keep paying with animals

 

[Dayaks]

That is only applicable to those selling their old cards. Last time I priced out a used 550Ti, someone was asking $40...and with skipping 5 generations of GPU, I'm not sure it would be worth while to bench 1080p ultra settings to truely do an apples to apples comparison.

You're welcome to.

Yeah a card that old I usually just keep around as a backup or gift away. I don't factor it into the new card equation.

 

[horrorshow]

http://gpu.userbenchmark.com/Compare/Nvidia-GTX-550-Ti-vs-Nvidia-GTX-1060/3161vs3639

This outta give you a baseline idea of what to expect

 

[cageymaru]

I think you need to have 2 charts; one with MSRP and one with current pricing. I got my RX 480 for $249 a few weeks ago by waiting a until I saw one for that price on Amazon. They do come into stock for MSRP; it's just that they are sold and relisted even faster.

 

[skeeder]

I think you need to have 2 charts; one with MSRP and one with current pricing. I got my RX 480 for $249 a few weeks ago by waiting a until I saw one for that price on Amazon. They do come into stock for MSRP; it's just that they are sold and relisted even faster.

Thus the excel document.

I was able to adjust prices on the fly and have the formulas do the work. The original price of the 1060 6GB I had in there was $300.

As I dug around I would update the prices. I was hoping to jump to a 1070, but I needed the extra cash for other parts.

 

[SpongeBob]

has both 1060's but not both 470s or 480s bummer. These aren't using Vulkan are they?

 

[skeeder]

has both 1060's but not both 470s or 480s bummer. These aren't using Vulkan are they?

Unfortunately, the only full Vulkan game I'm seeing benchmarked is Doom. That wasn't one of the benchs from DF.

It would of definitely thrown a bone for ATI--er...AMD.

 

[Palladium@SG]

I too have issues with the cost/average FPS metric: Avg fps doesn't take frame times into account, plus the obvious fact that target FPS differs from user to user.

 

[Ruoh]

I got my ASUS strix oc 8gb 480 for like 265 shipped. You could have done better than the 1060.

 

[rudy]

This is very complicated we need to add in the time benefit you would save by buying a better card and not needing to upgrade as often. Then we have to deal with electricity usage and costs. We could tack on a ton of other metrics as well. There is a reason we don't see reviewers going down this path much.

Because the reality is most people just don't make decisions like this. Most people just have a budget in mind and they are trying to stick with it. A $700 card is just outside of that budget for a lot of people.

I try to predict technology as part of my decision-making process. And I think many other people do the same and that is why the nVidia cards are selling so hot. People are probably thinking this is the big jump and next we will just see incremental changes for a while. So if I buy a 1080 now I can get a couple years of high performance out of it, and I might as well buy it now because even if I wait 6 or even 9 months I will very likely end up with the same choice card.

 

[wigglywaffles]

I try to predict technology as part of my decision-making process. And I think many other people do the same and that is why the nVidia cards are selling so hot. People are probably thinking this is the big jump and next we will just see incremental changes for a while. So if I buy a 1080 now I can get a couple years of high performance out of it, and I might as well buy it now because even if I wait 6 or even 9 months I will very likely end up with the same choice card.

Well people are thinking wrong then because you can't even get a year's worth of high performance out of a 1080.

Until single GPUs start having enough power to drive 4K @ 60 FPS with some power left over, we're not at the point where cards are 'future-proofed' for years.

Example would be for the peasants still gaming at 1080p resolution -- Pascal is overkill, because cards years old can still max games at 1080p60. Kepler still does a pretty good job at 1080p. So the GTX 700 series was an example of a future-proofed investment if you wanted to stay at 1080p.

GTX 1080 sits in that awkward territory where it's overkill for resolutions lower than 4K but not quite enough for 4K resolution by itself. It's a pretty terrible value whether you're gaming at 4K or lower; for 4K you still need multiple GPUs and for lower resolutions a GTX 1070 or something cheaper will suffice.

 

[rudy]

Well people are thinking wrong then because you can't even get a year's worth of high performance out of a 1080.

Until single GPUs start having enough power to drive 4K @ 60 FPS with some power left over, we're not at the point where cards are 'future-proofed' for years.

Example would be for the peasants still gaming at 1080p resolution -- Pascal is overkill, because cards years old can still max games at 1080p60. Kepler still does a pretty good job at 1080p. So the GTX 700 series was an example of a future-proofed investment if you wanted to stay at 1080p.

GTX 1080 sits in that awkward territory where it's overkill for resolutions lower than 4K but not quite enough for 4K resolution by itself. It's a pretty terrible value whether you're gaming at 4K or lower; for 4K you still need multiple GPUs and for lower resolutions a GTX 1070 or something cheaper will suffice.

In one years time there is very unlikely to be a card that is a game changer or significantly more powerful than a 1080. You buy a 1080 or titan now and you are going to be sitting in the high performance bracket for a while.

The beauty of PC gaming is you cannot pin yourself to the standards you are trying to, you can always shift graphics up or down to make more or less use of what you have. But at the top end of performance you can only buy the top end cards available. Pitan / 1080 is the target for a while. Sure they might release 1080ti which will likely at most be equivalent to a pitan. From the AMD front they have been botching things so much I highly doubt they are going to come rushing to our rescue with some pitan killer. The argument you are making is not one that you make about GPUs its one you make when deciding which monitor to buy.

People are not thinking wrong they are thinking exactly right. This is a great time to buy a GPU and be able to sit on it for a while.

 

[Roberto Lequeux]

Nice post (op)