Holiday 2006 Shopping Guide: GPUs
by Jarred Walton on December 13, 2006 5:15 AM EST- Posted in
- Guides
Performance Overview
Now that we've covered all of the various graphics cards that are currently available, you might be wondering which solution is really best for your needs. Users looking to upgrade to a faster graphics card might also be wondering how their current hardware compares to newer hardware, and how much money they might need to spend on an upgrade. One of the critical factors in determining how much graphics power you need is going to be your display. If you don't run at high resolutions with antialiasing, you really don't need a ton of horsepower for most games. Running tons of benchmarks with most of the graphics cards we've mentioned on a variety of games is beyond the scope of this article; however, we've created a rough summary of what sort of resolutions and detail settings you should be able to get from the various GPUs on the latest games. (Older titles are less demanding, so you should be able to crank up the details/resolution.)
This is of course somewhat ad hoc, and individual opinions about what is acceptable will vary, but you should be able to get some general guidelines of what to purchase - either for a new system or as an upgrade to your current hardware.
You can see in most instances that two slower GPUs will end up offering a worse price/performance ratio than a single faster GPU. That's why we've downplayed multi-GPU configurations throughout this article. Why buy two 7300 GT cards when you can get a single 7600 GT for less money? Until you max out the single GPU options, there's not much point in buying two GPUs.
In terms of the recommended settings, we generally feel that you should disable antialiasing and increase your resolution first, and only when you are running at your monitor's maximum/native resolution would we begin to worry about turning on antialiasing. That is especially true with LCDs, as running at anything less than the native resolution will tend to create a blurry effect. If you disagree and prefer lower resolutions with antialiasing, on average turning on 4xAA will have the same impact on performance as increasing the resolution one or two notches, i.e. 1024x768 4xAA will often run about as fast as 1280x1024-1600x1200 0xAA. There are of course exceptions to this rule, but it should serve as a good baseline estimate.
The recommended settings listed on the table above are basically an estimate of what settings more recent/demanding games will allow you to use - games like Prey, Oblivion, Neverwinter Nights 2, Company of Heroes, etc. There are a lot of overlapping resolutions listed, which is to be expected. For example, the difference between an X1800 XT and an X1800 XTX isn't going to be so great that the latter allows you to run higher detail settings. If you are looking at upgrading your graphics card, you'll want to get something that opens the possibility of running clearly improved resolutions/settings. As another example, we wouldn't recommend upgrading from a GeForce 6800 GT to a GeForce 7600 GT, because even though the latter is faster they are fundamentally similar in terms of performance. You would be better off spending a bit more money to get something like a Radeon X1950 Pro instead, so that you would truly notice a difference in performance.
Now that we've covered all of the various graphics cards that are currently available, you might be wondering which solution is really best for your needs. Users looking to upgrade to a faster graphics card might also be wondering how their current hardware compares to newer hardware, and how much money they might need to spend on an upgrade. One of the critical factors in determining how much graphics power you need is going to be your display. If you don't run at high resolutions with antialiasing, you really don't need a ton of horsepower for most games. Running tons of benchmarks with most of the graphics cards we've mentioned on a variety of games is beyond the scope of this article; however, we've created a rough summary of what sort of resolutions and detail settings you should be able to get from the various GPUs on the latest games. (Older titles are less demanding, so you should be able to crank up the details/resolution.)
This is of course somewhat ad hoc, and individual opinions about what is acceptable will vary, but you should be able to get some general guidelines of what to purchase - either for a new system or as an upgrade to your current hardware.
| Approximate Performance Ranking | |||
| GPU | Resolution | Detail | Price |
| Radeon X300 SE | 640x480-800x600 | Minimum-Med; 0xAA | $38 |
| GeForce 6200 TC | 640x480-800x600 | Minimum-Med; 0xAA | $38 |
| Radeon X300 LE | 640x480-800x600 | Minimum-Med; 0xAA | $70 |
| GeForce 7100 GS | 640x480-800x600 | Minimum-Med; 0xAA | $49 |
| Radeon X550 HM | 640x480-1024x768 | Low-Med; 0xAA | $47 |
| Radeon X550 | 640x480-1024x768 | Low-Med; 0xAA | $60 |
| Radeon X700 | 640x480-1024x768 | Low-Med; 0xAA | $75 |
| GeForce 7300 LE | 640x480-1024x768 | Low-Med; 0xAA | $52 |
| GeForce 7300 GS | 640x480-1024x768 | Low-Med; 0xAA | $59 |
| Radeon X1300 | 800x600-1280x1024 | Low-High; 0xAA | $52 |
| GeForce 6600 LE | 800x600-1280x1024 | Low-High; 0xAA | $66 |
| Radeon X1300 Pro | 800x600-1280x1024 | Low-High; 0xAA | $75 |
| GeForce 7300 GT | 1024x768-1280x1024 | Low-High; 0xAA | $75 |
| Radeon X1300 XT | 1024x768-1280x1024 | Low-High; 0xAA | $108 |
| Radeon X1600 Pro | 1024x768-1280x1024 | Low-High; 0xAA | $91 |
| GeForce 6600 GT | 1024x768-1280x1024 | Med-High; 0xAA | $85* |
| Radeon X1600 XT | 1024x768-1280x1024 | Med-High; 0xAA | $119 |
| Radeon X1650 Pro | 1024x768-1280x1024 | Med-High; 0xAA | $105 |
| Radeon X800 | 1024x768-1280x1024 | Med-High; 0xAA | $85* |
| GeForce 6800 | 1024x768-1280x1024 | Med-High; 0xAA | $90* |
| GeForce 7600 GS | 1024x768-1280x1024 | Med-High; 0xAA | $109 |
| Radeon X1600 Pro CF | 1024x768-1600x1200 | Med-High; 0x-2xAA | $182 |
| Radeon X800 Pro | 1024x768-1600x1200 | Med-High; 0x-2xAA | $125* |
| GeForce 6800 GT | 1024x768-1600x1200 | Med-High; 0x-2xAA | $175* |
| GeForce 6800 GS | 1024x768-1600x1200 | Med-High; 0x-2xAA | $135* |
| GeForce 6800 SLI | 1024x768-1600x1200 | Med-High; 0x-2xAA | $170* |
| GeForce 6600 GT SLI | 1024x768-1600x1200 | Med-High; 0x-2xAA | $170* |
| GeForce 7300 GT SLI | 1024x768-1600x1200 | Med-High; 0x-2xAA | $150 |
| GeForce 6800 Ultra | 1024x768-1600x1200 | Med-High; 0x-2xAA | $175* |
| Radeon X850 XT | 1024x768-1600x1200 | Med-High; 0x-2xAA | $135* |
| GeForce 7600 GT | 1024x768-1600x1200 | Med-High; 0x-2xAA | $135 |
| Radeon X1650 XT | 1024x768-1600x1200 | Med-High; 0x-2xAA | $156 |
| GeForce 7800 GS | 1024x768-1600x1200 | Med-High; 0x-2xAA | $230 |
| GeForce 7600 GS SLI | 1024x768-1600x1200 | Med-High; 0x-2xAA | $218 |
| Radeon X1600 XT CF | 1024x768-1600x1200 | Med-High; 0x-2xAA | $238 |
| Radeon X1650 Pro CF | 1024x768-1600x1200 | Med-High; 0x-2xAA | $210 |
| Radeon X1800 GTO | 1024x768-1600x1200 | Med-High; 0x-4xAA | $145 |
| GeForce 7900 GS | 1024x768-1600x1200 | Med-High; 0x-4xAA | $186 |
| Radeon X1800 XL | 1024x768-1600x1200 | Med-High; 0x-4xAA | $240* |
| Radeon X1900 AIW | 1024x768-1600x1200 | Med-High; 0x-4xAA | $220 |
| GeForce 7800 GT | 1024x768-1600x1200 | Med-High; 0x-4xAA | $185* |
| GeForce 6800 GS SLI | 1024x768-1600x1200 | Med-High; 0x-4xAA | $270* |
| GeForce 6800 GT SLI | 1024x768-1600x1200 | Med-High; 0x-4xAA | $350* |
| GeForce 6800 Ultra SLI | 1024x768-1600x1200 | Med-High; 0x-4xAA | $350* |
| GeForce 7800 GTX | 1280x1024-1920x1200 | High; 0x-4xAA | $250* |
| Radeon X1800 XT | 1280x1024-1920x1200 | High; 0x-4xAA | $300* |
| Radeon X1900 GT v2.0 | 1280x1024-1920x1200 | High; 0x-4xAA | $176 |
| GeForce 7900 GT | 1280x1024-1920x1200 | High; 0x-4xAA | $246 |
| Radeon X1900 GT | 1280x1024-1920x1200 | High; 0x-4xAA | $176 |
| Radeon X1650 XT CF | 1280x1024-1920x1200 | High; 0x-4xAA | $312 |
| GeForce 7600 GT SLI | 1280x1024-1920x1200 | High; 0x-4xAA | $270 |
| GeForce 7950 GT | 1280x1024-1920x1200 | High; 2x-4xAA | $249 |
| GeForce 7800 GTX 512 | 1280x1024-1920x1200 | High; 2x-4xAA | $460* |
| Radeon X1950 Pro | 1280x1024-1920x1200 | High; 2x-4xAA | $206 |
| GeForce 7900 GTO | 1280x1024-1920x1200 | High; 2x-4xAA | $310 |
| Radeon X1900 XT 256MB | 1280x1024-1920x1200 | High; 2x-4xAA | $275 |
| GeForce 7800 GS SLI | 1280x1024-1920x1200 | High; 2x-4xAA | $460 |
| GeForce 7800 GT SLI | 1280x1024-1920x1200 | High; 2x-4xAA | $370* |
| GeForce 7900 GS SLI | 1280x1024-1920x1200 | High; 2x-4xAA | $372 |
| GeForce 7900 GTX | 1280x1024-1920x1200 | High; 2x-4xAA | $430 |
| Radeon X1900 XT | 1280x1024-1920x1200 | High; 2x-4xAA | $335 |
| Radeon X1900 XTX | 1280x1024-1920x1200 | High; 2x-4xAA | $400 |
| Radeon X1950 XT 256 | 1280x1024-1920x1200 | High; 2x-4xAA | $259 |
| Radeon X1950 XTX | 1280x1024-1920x1200 | High; 2x-4xAA | $378 |
| Radeon X1900 GT CF | 1280x1024-1920x1200 | High; 2x-4xAA | $352 |
| Radeon X1800 XT CF | 1280x1024-1920x1200 | High; 2x-4xAA | $600* |
| Radeon X1950 Pro CF | 1280x1024-1920x1200 | High; 2x-4xAA | $412 |
| GeForce 7800 GTX SLI | 1280x1024-1920x1200 | High; 2x-4xAA | $500* |
| GeForce 7900 GT SLI | 1600x1200-1920x1200 | High; 2x-4xAA | $492 |
| GeForce 7950 GX2 | 1600x1200-2560x1600 | High-Max; 0x-4xAA | $465 |
| GeForce 8800 GTS | 1600x1200-2560x1600 | High-Max; 0x-4xAA | $455 |
| GeForce 7950 GT SLI | 1600x1200-2560x1600 | High-Max; 0x-4xAA | $498 |
| GeForce 7800 GTX 512 SLI | 1600x1200-2560x1600 | High-Max; 0x-4xAA | $920* |
| GeForce 7900 GTO SLI | 1600x1200-2560x1600 | High-Max; 0x-4xAA | $620 |
| GeForce 7950 GX2 QSLI | 1600x1200-2560x1600 | High-Max; 2x-4xAA | $930 |
| GeForce 7900 GTX SLI | 1600x1200-2560x1600 | High-Max; 2x-4xAA | $860 |
| Radeon X1900 XT CF | 1600x1200-2560x1600 | High-Max; 2x-4xAA | $770 |
| Radeon X1950 XTX CF | 1600x1200-2560x1600 | High-Max; 2x-4xAA | $774 |
| GeForce 8800 GTX | 1920x1200-2560x1600 | High-Max; 2x-4xAA | $610 |
| GeForce 8800 GTS SLI | 1920x1200-2560x1600 | Maximum; 2x-4xAA | $910 |
| GeForce 8800 GTX SLI | 1920x1200-2560x1600 | Maximum; 4xAA | $1220 |
* - Prices for these parts are prone to fluctuation, as these are discontinued products.
You can see in most instances that two slower GPUs will end up offering a worse price/performance ratio than a single faster GPU. That's why we've downplayed multi-GPU configurations throughout this article. Why buy two 7300 GT cards when you can get a single 7600 GT for less money? Until you max out the single GPU options, there's not much point in buying two GPUs.
In terms of the recommended settings, we generally feel that you should disable antialiasing and increase your resolution first, and only when you are running at your monitor's maximum/native resolution would we begin to worry about turning on antialiasing. That is especially true with LCDs, as running at anything less than the native resolution will tend to create a blurry effect. If you disagree and prefer lower resolutions with antialiasing, on average turning on 4xAA will have the same impact on performance as increasing the resolution one or two notches, i.e. 1024x768 4xAA will often run about as fast as 1280x1024-1600x1200 0xAA. There are of course exceptions to this rule, but it should serve as a good baseline estimate.
The recommended settings listed on the table above are basically an estimate of what settings more recent/demanding games will allow you to use - games like Prey, Oblivion, Neverwinter Nights 2, Company of Heroes, etc. There are a lot of overlapping resolutions listed, which is to be expected. For example, the difference between an X1800 XT and an X1800 XTX isn't going to be so great that the latter allows you to run higher detail settings. If you are looking at upgrading your graphics card, you'll want to get something that opens the possibility of running clearly improved resolutions/settings. As another example, we wouldn't recommend upgrading from a GeForce 6800 GT to a GeForce 7600 GT, because even though the latter is faster they are fundamentally similar in terms of performance. You would be better off spending a bit more money to get something like a Radeon X1950 Pro instead, so that you would truly notice a difference in performance.
Facebook
Twitter
RSS
51 Comments
View All Comments
justly - Thursday, December 14, 2006 - link
I apoligize for straying from the video topic, but I really get annoyed at the all to often trash talk about VIA and SiS chipsets.I understand that this is a GPU article so I can see Anandtech not recommending SiS or VIA integrated graphics based on their lack luster video capabilities. My question (or maybe I should call it a complaint) is how can Anandtech claim SiS and VIA boards are not stable or reliable? The last reviw (that I can remember) of a SiS based board was over a year ago, even then I dont think it was a production board. Coverage of VIA based boards isn't much better but at least Anandtech does give VIA some budget coverage.
I can fully understand if Anandtech doesn't want to recommend VIA or SiS to their enthusiast crowd due to poor overclocking, or being "a bit more quirky" as your article states.
I'm not going to read all the way through old articles just to try and figure out what these stability and reliability issues mighy be (mainly because most of the articles are so old that a BIOS update could easily have made any stability issues invalid). Well I lied a little, I did briefly look through the VIA board articles within the last year and found no stability issues at stock settings. In fact, the only stability issues I saw mentioned in an article happened when "we tried to exceed the SPD settings of our DDR memory modules" but the next line reads "We did not experience these same issues with our DDR2 memory modules" (and that article is 1 week shy of 9 months old).
I hope Anandtech decides to either stop repeating these claims of unstable, unreliable and quirky boards based on VIA & SiS or start reviewing these boards and show its readers why they deserve these remarks.
Then again if the only thing we as readers get from reviews of these chipsets/boards is complaints about how budget boards are not able to overclock, or the lack of a tweakable BIOS in a sub $60 board then blame the board not the chipset as most people are already aware that budget boards are like this reguardless of what chipset they use.
JarredWalton - Thursday, December 14, 2006 - link
I know of at least one attempted SiS board review in the past year that was canned because our reviewer could not get the board to function properly (after several BIOS updates and two boards, IIRC). Motherboards (and chipsets) are such an integral part of any computer that I would never skimp in that area. Then again, maybe I'm just too demanding of my computers?If you read user reviews of VIA/SiS boards you typically see a pattern that indicates the boards are overall "less reliable" - periodic instabilities and far higher failure rates. Some people report no problems and love the low prices, while others try to do a bit more with their systems and encounter difficulties.
If you just want to use a computer for office tasks, just about any system will be fine... but then again, if you're doing office work and your computer crashes, you probably won't be too happy. Anyone planning on running a higher-spec GPU should avoid cheaper motherboards IMO, as running a $300+ GPU in a <$75 board is just asking for problems. (For the same reason, I recommend $75+ PSUs for anyone running a CPU+GPU that cost more than $400 combined.)
Basically, I just can't recommend a questionable motherboard that saves a person $10-$20. The fact that the companies aren't out there promoting their products says something. If they're not proud enough of their work to try hard to get reviews at reputable sites, perhaps it's because they know their boards won't pass muster.
I actually had a company representative complain to me once about my stress tests being "unrealistic". He asked, "How many people actually try to run Folding@Home and a bunch of gaming benchmarks in sequence?" Basically, the system would crash if I used my script to benchmark games at various resolutions without rebooting in between each run. It's true that a lot of people might never stress a system to that level, but when I've looked at dozens of computers that handle that workload without problems, a system that crashes/locks in the same situation is clearly not as "stable or reliable" as competing solutions. All things being equal, I would recommend a different PC at the same price.
That's basically how I see the VIA/SiS situation. $10 is about 100 miles of driving, a trip to most restaurants, a two hour movie.... It's not worth the risk just to save $10. If it is, maybe a new computer isn't what you really need; a used PC would probably be just as good and likely a lot cheaper (and possibly faster as well).
justly - Thursday, December 14, 2006 - link
I agree with most of what you say, no one wants a system that crashs.One thing I do notice though, is that most of your arguments can be atributed to low priced boards, yet the comments I find annoying are the generalizations about chipsets. Do you actually believe a $50 nvidia based board is significantly more stable or reliable than any other chipset? and if you do, couldn't this just be a side effect of being a more popular chipset thus less work programming a bios? I'm sure this isn't what you meant, but going by your comments about motherboard pricing, if I found a $100 SiS based board it should be more stable and reliable than a $50 nvidia board.
You also want me to read "user reviews"? this doesn't sound like a good way to judge reliability to me. Most user reviews are either in enthusiast fourms like the ones you have here, these usually only rewiew overclocking abilities, or on retail sites like Newegg, and to be honest most of the bad reviews I see there look more like PEBKAC.
You really haven't cleared up why VIA or SiS chipsets should be considered unreliable or unstable, although your dislike of budget boards is quite evedent.
I'm not trying to deny you your opinion, I'm just asking that you refrain from singling out specific chipsets if what you are really having a problem with is all budget boards, if there actually is a chipset specific problem please try to get a review published indicating what the problem is.
BTW if the board that wouldn't function, and had the review canned was a production board I feel sorry for the person that bought it without a proper warning from a review site that knew it was flawed (you don't want to know what I think of the review site that would let this happen).
Knowing what to expect from a product can help a budget builder as much as it can help an overclocker.
Sunrise089 - Thursday, December 14, 2006 - link
I tend to view guids like these through the eyes of my own system, and having a 7900GT at 500/1500, there is little reason to upgrade if I'm going to continue to play games at 1280x1024. However, 22" (widescreen) LCDs have also become a lot cheaper, and with my poor eyes, the 1650x1050 or so resolution will probably work pretty well. That leads me to the great situation I'm apparently in - it looks like my card will fetch around $200 if I sell it, and I have the option of either a perhaps slightly faster X1950pro for $199, basically making it a free change but only slightly faster, or a X1950XT 256meg for only $249. That's a lot of additional card for only $50, and pretty tempting. I cannot see why the $249 part doesn't get the nod for your pick over the 7950GT though.JarredWalton - Thursday, December 14, 2006 - link
Despite the fact that they are separated by quite a few cards in the table, the X1950 XT 256MB and the 7950 GT give relatively similar performance. The XT is probably 10-15% faster depending on game, but that's not really enough to mean the difference between one resolution and another in my opinion. You also get 512MB of RAM with the 7950GT, and it tends to overclock better than the XT resulting in performance that is basically equal.However, you're right that it is still worth considering, and so I added it to the final table. This is particularly true for people that don't like NVIDIA hardware for whatever reason - just as the 7950GT is worth considering for people that don't like ATI's drivers. Honestly, I'm still unhappy with ATI's drivers overall; they NEED TO DITCH .NET! What's next, writing low level drivers in C# or Jaba (that's big, fat, slow Java for the uninformed)? I know the .NET stuff is just for the UI, but it still blows, and I get about a 45 second delay after Windows loads while the ATI driver starts up. If I weren't running CrossFire, I might not have as many issues with ATI's drivers, though.
JarredWalton - Thursday, December 14, 2006 - link
As a side note, Neverwinter Nights 2 appears to require/use .NET 2.0, and for those who have played the game that probably explains a lot of the performance issues. I'm not sure if CrossFire/SLI support is working yet, but I do know that my CrossFire X1900 XT config can't handle running with antialiasing, and/or water reflections/refractions at resolutions above 1280x1024. Seems decent without the AA and water stuff at 1920x1200 with the latest drivers and patch, though.PrinceGaz - Thursday, December 14, 2006 - link
Something seems to be missing from this part of the last paragraph on page 8.JarredWalton - Thursday, December 14, 2006 - link
Weird speech recognition there, I guess. I'm pretty sure it was supposed to be "they are" instead of "fair so"... but I can't honestly remember if that's what I said or not. LOLgerf - Thursday, December 14, 2006 - link
On the second page, were should be "where."BTW, good article. Laptop integrated's good enough for me though (ex-gamer).
Noya - Thursday, December 14, 2006 - link
Chart of best values jumps from about $100 w/rebate to $200+, while a highly overclockable 7900gs can be had for $145 after rebate (about $35 over a 7600GT).