Test Setup
We felt it best to include as many games as possible for our performance tests in order to get the best view of how these cards perform on both newer games and ones that have been around for a while. The games we used for our tests are Oblivion, Battlefield 2, Half-Life 2 Episode 1, Quake 4, Rise of Legends, Splinter Cell Chaos Theory, and F.E.A.R. When possible, we ran the benchmarks at four resolutions, 800x600, 1024x768, 1280x1024, and 1600x1200. F.E.A.R doesn't support 1280x1024, so we used 1280x960 instead, and Rise of Legends doesn't support resolutions lower than 1024x768.
Because some of these cards have only 128 MBs of memory, we ran tests at lower quality settings in each of the games for these cards, but we'll deal with these cards in the following "Lower Range Performance" section. We generally test with higher quality settings enabled because we find many gamers prefer to play games with graphics quality settings on high with "bells-and-whistles" like HDR effects, sacrificing their screen resolution to make up for the performance hit (especially given that a large number of gamers still use older monitors that can only display resolutions of up to 1280x1024 or 1600x1200). This isn't the case for all gamers, and many prefer higher resolutions and lower quality graphical settings; but for the purposes of this article we find this to be the best choice for testing.
Something to note is that many of these cards are clocked at the same reference speeds, so we will see several cards with identical numbers in our tests. Because we are dealing with so many cards, comparing performance between them all is somewhat daunting, but we hope to give the reader a good idea of how these cards rank in performance relative to each other for each of the games in the next sections.
This is the system we used for testing:
*Note that sound was disabled for testing. Testing was also performed in an open air environment rather than inside a case. This eliminated the need for a case fan, but it is not fully representative of how the various cards will run once placed inside a typical case. We will have a future article examining how the various silent cards perform inside a case, and depending on the case and cooling used overall temperatures may be lower or higher. As we stated in the introduction, we would advice some caution for those looking to build a fully silent computer. It is likely easier, cheaper, and more reliable to build a nearly-silent system by using a single low-RPM 120mm fan (perhaps with a voltage mod to further reduce fan speed) than to have an absolutely silent system that overheats under load.
We felt it best to include as many games as possible for our performance tests in order to get the best view of how these cards perform on both newer games and ones that have been around for a while. The games we used for our tests are Oblivion, Battlefield 2, Half-Life 2 Episode 1, Quake 4, Rise of Legends, Splinter Cell Chaos Theory, and F.E.A.R. When possible, we ran the benchmarks at four resolutions, 800x600, 1024x768, 1280x1024, and 1600x1200. F.E.A.R doesn't support 1280x1024, so we used 1280x960 instead, and Rise of Legends doesn't support resolutions lower than 1024x768.
Because some of these cards have only 128 MBs of memory, we ran tests at lower quality settings in each of the games for these cards, but we'll deal with these cards in the following "Lower Range Performance" section. We generally test with higher quality settings enabled because we find many gamers prefer to play games with graphics quality settings on high with "bells-and-whistles" like HDR effects, sacrificing their screen resolution to make up for the performance hit (especially given that a large number of gamers still use older monitors that can only display resolutions of up to 1280x1024 or 1600x1200). This isn't the case for all gamers, and many prefer higher resolutions and lower quality graphical settings; but for the purposes of this article we find this to be the best choice for testing.
Something to note is that many of these cards are clocked at the same reference speeds, so we will see several cards with identical numbers in our tests. Because we are dealing with so many cards, comparing performance between them all is somewhat daunting, but we hope to give the reader a good idea of how these cards rank in performance relative to each other for each of the games in the next sections.
This is the system we used for testing:
Test Bed Components | |
Motherboard: | NVIDIA nForce4 |
Processor: | AMD Athlon 64 FX-55 2.6 GHz |
Memory: | 2x1024MB OCZ 2-2-2-6 DDR-400 |
Hard Drive: | Seagate 7200.7 120GB |
Power Supply: | OCZ 600W PowerStream |
*Note that sound was disabled for testing. Testing was also performed in an open air environment rather than inside a case. This eliminated the need for a case fan, but it is not fully representative of how the various cards will run once placed inside a typical case. We will have a future article examining how the various silent cards perform inside a case, and depending on the case and cooling used overall temperatures may be lower or higher. As we stated in the introduction, we would advice some caution for those looking to build a fully silent computer. It is likely easier, cheaper, and more reliable to build a nearly-silent system by using a single low-RPM 120mm fan (perhaps with a voltage mod to further reduce fan speed) than to have an absolutely silent system that overheats under load.
49 Comments
View All Comments
Josh Venning - Thursday, August 31, 2006 - link
I also forgot to mention that some people use their pcs in home theater systems as well. This would be another case when you want as little noise from your computer as possible.imaheadcase - Thursday, August 31, 2006 - link
That was not always the case, my 9700 Pro i still use when fan went out a year ago, works like a charm without it on. It was in its time the high end card, lets hope those days come buy again :Deckre - Thursday, August 31, 2006 - link
What a great review, when tom did their silent VC review, they included a grand total of three cards...pfft. nice job anand.I have the 7600GT, very sweet and 0dB is oh so nice.
Josh Venning - Thursday, August 31, 2006 - link
We just wanted to say thanks all for your comments and we are still trying to make sure we've caught any errors. (there are actually only 20 cards in the roundup and not 21) As Derek said, these cards were included in the article because we requested any and all silent cards that any of the manufacturers were willing to give us to review. That's also why we have more cards from ASUS and Gigabyte than the others.Olaf van der Spek - Thursday, August 31, 2006 - link
Because the videocard industry hasn't introduced such a bad design as the netburst architecture.
epsilonparadox - Thursday, August 31, 2006 - link
No they've introduced worse. When they recommend a second PS just for grafx or even a 1Kw single PS, they've taken intel's lack of thermal control to a whole new level.DerekWilson - Thursday, August 31, 2006 - link
graphics cards use much much less power in 2d mode than in 3d mode -- and even their 3d power saving capabilities are really good.this is especially true when you consider the ammount of processing power a GPU delivers compared to a CPU.
Theoretical peak performance of a current desktop CPU is in the 10-15 GFLOPS range at best. For a GPU, theoretical peak performance is at least one order of magnitude larger reaching up over 200 GFLOPS in high end cases.
I'm not saying we can reach these theoretical peak rates on either a CPU or a GPU, but a GPU is doing much much more work under load than a CPU possibly could.
Keep in mind we aren't even up to GHz on GPU cores. On the CPU front, Intel just shortened the pipeline and decreased clock speeds to save power -- doing more work in one cycle. This is absolutely what a GPU does.
And the icing on the cake is the sheer options on the silent GPU front. Neither AMD nor Intel make a fast desktop CPU that can be (easily) passively cooled. These parts are a testiment to the efficiency of the GPU.
On the flip side, ATI and NVIDIA push their high end parts way up in clock speed and power consumption trying as hard as possible to gain the performance crown.
There are plenty of reasons GPUs draw more power than a CPU under load, but a lack of thermal control or inefficient desing is not one of them. It's about die size, transistor count, and total ammount of work being done.
JarredWalton - Saturday, September 2, 2006 - link
I disagree with Derek, at least in some regards. The budget and midrange GPUs generally do a good job at throttling down power requirements in 2D mode. The high-end parts fail miserably in my experience. Sure, they consume a lot less power than they do in 3D mode, but all you have to do is look at the difference between using a Radeon Mobility X1400 and a GeForce Go 7800 in the Dell laptops to http://www.anandtech.com/mobile/showdoc.aspx?i=276...">see the difference in battery life.In 2D mode, graphics chips still consume a ton of power relatively speaking -- probably a lot of that going to the memory as well. A lot of this can be blamed on transistor counts and die size, but I certainly think that NVIDIA and ATI could reduce power more. The problem right now is that power use is a secondary consideration, and ATI and NVIDIA both need to have a paradigm shift similar to what Intel had with the Pentium M. If they could put a lot of resources into designing a fast but much less power-hungry GPU, I'm sure they could cut power draw quite a bit in both idle and load situations.
That's really the crux of the problem though: resources. Neither company has anywhere near the resources that AMD has, let alone the resources that Intel has. Process technology is at least a year behind Intel if not more, chip layouts are mostly computer generated as opposed to being tweaked manually (I think), and none of the companies have really started at square one trying to create a power efficient design; that always seems to be tacked on after-the-fact.
GPUs definitely do a lot of work, although GFLOPS is a terrible measure performance. The highly parallel nature of 3D rendering does allow you to scale performance very easily, but power requirements also scale almost linearly with performance when using the same architecture. It would be nice to see some balance between performance scaling and power requirements... I am gravely concerned about what Windows Vista is going to do for battery life on laptops, at least if you enable the Aero Glass interface. Faster switching to low-power states (for both memory and GPU) ought to be high on the list for next-generation GPUs.
DaveLessnau - Thursday, August 31, 2006 - link
I'm wondering why Anandtech tested Asus' EN7800 GT card instead of their EN7600 GT. That card would be more in line with Gigabyte's 7600 GT version and, I believe, is more available than the 7800 version. In the near future, I'd like to buy one of these silent 7600GTs and was hoping this review would help. Oh, well.DerekWilson - Thursday, August 31, 2006 - link
you can get a really good idea of how it would perform by looking at Gigabyte's card.as I mentioned elsewhere in the comments, we requested all the silent cards manufacturers could provide. if we don't have it, it is likely because they were unable to get us the card in time for inclusion in this review.