Anandtech GPU Benchmarking

[maniacalpha1-1]

I have noticed that on Anandtech benchmarks, the FPS on GPUs generally seem higher than on other reviews, at the same resolution and same or close to same AAing(is there really that big a difference between no AAing or 2 or even 4 AA?). I should mention the only benchmark I am comparing here is Bad Company 2. Below is an example of a GTS 450:

http://www.anandtech.com/show/3909/nvidias-geforce-gts-450-pushing-fermi-in-to-the-mainstream/10

It shows the worst GTS 450 getting 55.1 FPS at 1680X1050 on Bad Company 2, yet other reviewer sites show as low at 32 FPS. There are similar differences between Anandtech reviews and other reviews for other cards I've noticed, such as Radeon 6850/70 etc.

So what I wonder is, are these Anandtech numbers trustworthy in a sense of I would probably get those with the same card? Or do they disable/turn down other things that have nothing to do with the resolution/AAing that push the FPS up? (Like shadows, and other details)

I mean, if a GTS 450 can really get 55 FPS at 1680 and not 32 like the other reviews suggest, that might be good enough for me....and certainly a Radeon 6850 would be good enough at Anandtech's 88 FPS(whereas other reviews show it as low as 50 FPS at 1680 in Bad Company 2).

Here's the link for reference that shows 6850 getting 88 FPS:
http://www.anandtech.com/show/3987/...enewing-competition-in-the-midrange-market/14

 

[Genx87]

How about you link the review that shows 32FPS so we can compare?

 

[maniacalpha1-1]

Techpowerup GTS 450:
http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTS_450_SLI/4.html

Techpowerup 6850:
http://www.techpowerup.com/reviews/HIS/Radeon_HD_6850/7.html

Guru 3D 450
http://www.guru3d.com/article/geforce-gts-450-sli-review/6

Guru 3D 6850:
http://www.guru3d.com/article/radeon-hd-6850-6870-review/14

All of them show lower FPS for the 450 and 6850.

 

[Termie]

Based on my testing, I would say the Anandtech review is the most accurate. In regard to your link to Guru3d, which is testing at 8AA, the comparison is not valid because going to 8xAA makes a huge difference in performance (which answers your question on AA levels). As for Techpowerup, their numbers are about 10% lower, and while I tend to agree with Anandtech's numbers more, the two tests aren't that far off and you're probably safe guessing that you'll get somewhere in that range. Much of the performance in a game like BC2 depends on the level you're playing on and what is happening on screen during the test.

To answer your other implied question as to whether a 6850 is enough for BC2, the answer is that it's definitely enough for 4xAA settings at 1680 or 1920, possibly 8xAA at 1680. The catch is that you really need a fast CPU, preferably a quad-core, for the graphics card to perform its best. I promise that you will not get the numbers in the Anandtech test with a dual-core processor. The GTS450 is not sufficient for running 4xAA at 1920 and is barely sufficient for running 4xAA at 1680, but if you're running an older dual-core, it's a better match and will work well at the moderate level of image quality that a dual-core can support.

 

[maniacalpha1-1]

My processor is an AMD Phenom II X4 925, and my current GPU is a Radeon 4850. I was thinking that if the Anandtech number of 55 FPS(and it gave 65 FPS for some Palit Platinum GTS 450 that, according to Newegg anyway, is not significantly more expensive) was accurate, then it might be good enough. Currently there's an HIS 6850 that's only $180 though.

I should mention my monitor only goes to 1680, so anything higher than that is blocked from me. With that said, I have another question...performance aside, just imagine that you had the most powerful processor and GPU setup possible; is it possible to compare resolutions from an older game to a newer game? I realized that, even though a 4850 can easily run Battlefield 2 at any resolution well, I find I often play it at 1024 X 768 and I like it just fine. Does that mean, of necessity, that 1024 X 768 in a newer game like Bad Company 2 would be comparably enjoyable, or does it not necessarily work that way?

One thing I might mention with regard to my questions is that I actually have Bad Company 2, and it's playable with my 4850, but the reality is I despise BC2(it's not currently installed or I'd play it quick at 1024 X768 to answer the above question, lol) - my purpose in this inquiry is actually preparation for Battlefield 3, and BC2 is probably the most valid thing to compare since it's not out yet. Although system requirements may turn out differently as we get closer to release, I am for the moment proceeding under the assumption that BF3 will be within a margin of 20% tougher on GPUs than BC2, thus if 60 FPS at 1680 is the sweet spot minimum for ensuring smoothness, than a GPU needs to be capable of at least 72 FPS in BC2 at 1680 in order for me to be confident in it for Battlefield 3. Thus, if it's accurate that a 6850 can get 88 FPS as per Anandtech and not 51 as shown in Techpowerup, then I would consider a 6850 to be perfect and just save up for that. As far as I can tell, both Anandtech and Techpowerup are using 4x AA at 1680 in their reviews...

 

[bryanW1995]

at 1680x1050 a 6850 will last you for years to come.

have you overclocked your cpu? you should shoot for the 3.5-3.8 range if you plan to game with it for the next few years.

 

[maniacalpha1-1]

I haven't tried overclocking at all, I have the default AMD cooler and I know it could still probably be overclocked slightly, but I haven't messed with it.

Do you think I'll be able to handle BF3 without overclocking? If so, I can probably survive for a couple of years until it's CPU upgrade time.

 

[Termie]

My processor is an AMD Phenom II X4 925, and my current GPU is a Radeon 4850.

. . .

I should mention my monitor only goes to 1680, so anything higher than that is blocked from me. With that said, I have another question...performance aside, just imagine that you had the most powerful processor and GPU setup possible; is it possible to compare resolutions from an older game to a newer game? I realized that, even though a 4850 can easily run Battlefield 2 at any resolution well, I find I often play it at 1024 X 768 and I like it just fine. Does that mean, of necessity, that 1024 X 768 in a newer game like Bad Company 2 would be comparably enjoyable, or does it not necessarily work that way?

One thing I might mention with regard to my questions is that I actually have Bad Company 2, and it's playable with my 4850, but the reality is I despise BC2(it's not currently installed or I'd play it quick at 1024 X768 to answer the above question, lol) - my purpose in this inquiry is actually preparation for Battlefield 3, and BC2 is probably the most valid thing to compare since it's not out yet. Although system requirements may turn out differently as we get closer to release, I am for the moment proceeding under the assumption that BF3 will be within a margin of 20% tougher on GPUs than BC2, thus if 60 FPS at 1680 is the sweet spot minimum for ensuring smoothness, than a GPU needs to be capable of at least 72 FPS in BC2 at 1680 in order for me to be confident in it for Battlefield 3. Thus, if it's accurate that a 6850 can get 88 FPS as per Anandtech and not 51 as shown in Techpowerup, then I would consider a 6850 to be perfect and just save up for that. As far as I can tell, both Anandtech and Techpowerup are using 4x AA at 1680 in their reviews...

You have a good processor, so overclocking it should allow a modern graphics card to work well. Comparing image quality in a game like BF2 to BC2 is difficult. Will BC2 at 1024x768 look as good as BF2? Sure. Will it look good enough to play at? Depends what your standards are, but in my opinion it will look pretty bad, as you'll lose much of the quality and detail that sets it apart from BF2 in regard to realism. I don't think that's how you'll want to run BC2, or BF3 for that matter. Running BC2 at such a low resolution would compromise far more than in BF2 because there is more detail to lose.

All that being said, I think your graphics card search might be a bit premature. You've said that you don't like BC2, so you don't need a card to play that game. And we'll all be lucky if BF3 is here in less than 12 months, so you might as well hold off of serious analysis of graphics cards until more news of BF3 comes out. If your budget is ~$150, any card you get today will be inferior to what you'll be able to buy in 2H'11. Plus the hardware demands of BF3 are really just speculation at this point.

 

[bryanW1995]

I haven't tried overclocking at all, I have the default AMD cooler and I know it could still probably be overclocked slightly, but I haven't messed with it.

Do you think I'll be able to handle BF3 without overclocking? If so, I can probably survive for a couple of years until it's CPU upgrade time.

try it out and see. worst case scenario you decide to OC the cpu after playing the game, best case it's good enough for you and you leave it alone. the stock amd cooler will allow you to get a decent OC if you need to do it btw.

 

[maniacalpha1-1]

thanks....Well, I do currently play Civ 5, not sure if a better GPU would help there, and at a bare minimum I'll also be getting Dragon Age II in March. Back when I got Dragon Age I in 2009 I only had a 9500GT, I'd hate to repeat that debacle, lol.

 

[bryanW1995]

you should DEFINITELY oc your cpu if you're playing civ5 or DAO. I saw a huge improvement going from 3.5 to 4.0 on my i7 rig in DAO, and civ5 is similarly cpu-limited. any of the better dx11 gpus (say gtx 460 and up) will massively improve your gaming/resolution in civ5. It's funny, on the civfanatics forums I see lots of people who run min specs on civ5 then complain that civ4 had similar/better graphics. with a 4850 @ 1680x1050 you are probably stuck on medium graphics with no tesselation and probably low textures. turn up the graphics on civ5 and it's a huge upgrade however, and with a good video card you'll get better performance and much better quality.

 

[maniacalpha1-1]

To be honest, in Civ 5 I don't care too much about graphics, I get annoyed by the inter-turn calculation times which get bad after a couple hundred turns, given that I like playing the 12 civilization/24 city state maps. Turning down graphics didn't help turn loading times, so to keep it fast I had to drop it to 4 civs/4city states. Would overclocking the CPU affect turn calculation? I saw on some forums that the load times were bad for everyone so I didn't give a lot of thought to it.

 

[maniac5999]

Yes, overclocking will help the turn processing times in civ5. That is massively CPU and I/O limited. Remember to OC the HTT/CPU-NB links as well, as that will help with the I/O bottleneck.

Think about it this way. The CPU does the thinking, the GPU puts pretty pictures on the screen. i\If you put enough pretty pictures on the screen in rapid succession it turns into video. If the GPU isn't powerful enough, it puts out too few pictures, and you get a slideshow instead of video. (or you turn down the image quality, and get more, but less detailed pictures)

Now, if the CPU isn't powerful enough it won't have updated info for the GPU to draw, or as in your case it'll spend time thinking about what each of the computers is going to do, and nothing is going to seem to happen.

 

[Termie]

This thread is overrun by maniacs!

 

[bryanW1995]

yeah, my i7 @4.0 still struggles on huge maps. with my top rig in sig I keep to either large maps (default 10 civs/20 CS's) on standard speed or standard (8/16) maps on epic speed. with your rig not-oc'd you will probably experience lategame slowdowns even on small maps. depending on your case, you might even spend $30-40 to get a good cpu cooler and max out the oc. you'll be glad you did, especially on those larger maps.

an easy way to check cpu impact is to load up a large map lategame turn. time the inter turn with a stopwatch. drop cpu speed and do it again, measuring the difference. if you're patient do it in 100mhz intervals, if you're like me just change it 500 mhz or so see what happens. easiest way now is to underclock it first b/c you don't know how high you can oc the cpu yet. I don't know what change you'll see, but I know from my personal experience that you'll be glad you did it.

 

[jacktesterson]

The Palit GTX cards are clocked much higher then reference cards in a lot if examples.

For example, a normal 1GB GTX 460 is 675/1400 reference (correct if wrong) but my Palit 460 is 800/1600, for the same price on newegg.

Little things like that, other system components, settings, etc can swing things 10% easily

 

[maniacalpha1-1]

Got another question about Anandtech benchmarking -
From March 26, 2010, the GTX 470 and 480 review:
http://www.anandtech.com/show/2977/...x-470-6-months-late-was-it-worth-the-wait-/14

From October, the 6850 Review:
http://www.anandtech.com/show/3987/...enewing-competition-in-the-midrange-market/14

The March review shows the GTX 470 getting 67.3 FPS at 1680, whereas the listing for the 470 in the 6850's October review shows GTX 470 getting 103.9, and that's with 4x AA(the March review did not specify, so I presume no AA was used).

That's a huge increase, is there something I'm not seeing or does 6 months of driver improvements mean that much?

 

[Childs]

Got another question about Anandtech benchmarking -
From March 26, 2010, the GTX 470 and 480 review:
http://www.anandtech.com/show/2977/...x-470-6-months-late-was-it-worth-the-wait-/14

From October, the 6850 Review:
http://www.anandtech.com/show/3987/...enewing-competition-in-the-midrange-market/14

The March review shows the GTX 470 getting 67.3 FPS at 1680, whereas the listing for the 470 in the 6850's October review shows GTX 470 getting 103.9, and that's with 4x AA(the March review did not specify, so I presume no AA was used).

That's a huge increase, is there something I'm not seeing or does 6 months of driver improvements mean that much?

That is odd. All the Nvidia cards show significant improvements on the same hardware and games, not just the 470, and not just in any one game.

 

[happy medium]

That is odd. All the Nvidia cards show significant improvements on the same hardware and games, not just the 470, and not just in any one game.

Drivers my friends.

 

[Termie]

That is odd. All the Nvidia cards show significant improvements on the same hardware and games, not just the 470, and not just in any one game.

Just ignore the original Fermi results from March. They're dead wrong. The drivers were terrible. No one would argue that the HD5830 should ever have matched the GTX480 in BC2. I'm surprised Nvidia was so sloppy with its rollout of Fermi - I think the cards would have been considered a success had it not been for the terrible drivers. High heat and power use were not unprecedented - it was the poor performance in comparison to the heat/noise that was the problem.

Look at the HAWX benchmarks - almost no change at all. Nvidia should have focused on games that people actually play, rather than optimizing its release drivers for games like HAWX. Clearly, there was nothing wrong with Anandtech's benching setup.

 

[maniacalpha1-1]

I was thinking drivers of course, but I thought it was AMD, not Nvidia, that was known to be the one that started out with whacky drivers and cleaned up as they went along. 67.3 to 103.9 FPS is a lot...

Anandtech does their GPU reviews before cards are even released for sale, right? Maybe Anandtech should put dates on their benchmarks, right on the graph image itself, so that they can go back 6 months after release and benchmark cards again and merge the new benchmark graphs into the old review documents side-by-side(or uh, right below, anyway), this would serve an excellent method of letting people who like to wait til prices drop keep tabs on the performance of a card after there's been time for the company to clean up drivers.

As it stands right now, the only way to gauge 6 month after performance is to ask a ton of people who may or may not be using settings you will be, read through hundreds of newegg reviews hoping to spot FPS numbers, etc. Or, in some cases you do find updated benchmarks like in this case but nothing organized for that purpose.