Over the last few years the market has been inundated with more and more GPU releases. In a ridiculous effort to keep to an arbitrary two year refresh cycle both nVidia and ATi (AMD) have been pushing out one refresh after the other. These newer GPUs really offer very little other than the ability to empty your wallet. Recently ATi decided to stop competing release for release with nVidia and try to concentrate on releasing GPUs that offer performance and value. This “best-bang-for-the-buck” strategy is what originally got AMD its name and place in the enthusiast heart. But can they compete with the massive cards coming out from the big green one? We decided to take a look at an assortment of GPUs both from nVidia and from ATi to see what differences we can find.
Product: Multiple Products
Author: Sean Kalinich
Category: Gamer
Reviewed on: July 20th 2008
Spelling and Grammatical Editors: Planetx64 Staff
{mospagebreak title=The Market}
What is the point?
This is a question that many people are asking today. Why do I need to replace my existing GPU for a new one? “The games I like to play run great.” In the eyes of the mainstream gamer the top of the heap does not hold much more than a resigned sigh or a small shake of the head. To the ultra elite gamer or enthusiast that just HAS to have the highest 3DMark number they want at least one and are willing o pay for it. The question many wonder is why is there such a big range in thought?
Psychology of gaming:
As the title implies I am going to embark on a brief explanation as to why I think our market is driven the way it is. To do this I will have to cover some background facts and principles behind what makes a game “fluid” and playable.
The first concept we cover is: full fluid motion
Full Fluid Motion is described as the speed (in FPS) where video (and games) appear to the human eye as matching what we expect from real life motion. Now there are many schools of thought as to what this magic number is but most put it at 24(1) FPS for film (traditional movies) and 32-36 for digital. But there are variances in this, these are all centered on another item; Refresh Rate.
Refresh Rate is the speed (measured in Hz) that a monitor or light refreshes or strobes, for most new movies (in theaters) and TVs it is 60Hz (for US and 50Hz for most of Europe) for analog and roughly 72 Hz for Digital (Not all HD sets will run at 72Hz some still use the 60Hz standard). For Monitors (including LCD) this is usually set to 72Hz with the ability to alter it within a set range depending on resolution. Older monitors had a set refresh at 62-65Hz. This causes the strobe or pulsing effect you see on many old monitors and TVs.
The Full Fluid Motion rating is affected by Refresh Rate in that a higher Refresh Rate can fool the eye into perceiving something as fluid that might fall below the technical level, while a lower Refresh Rate might make a high frame rate appear to stutter. This is why you will see an option in many games to sync your vertical refresh. It eliminates the refresh rate getting out of sync with the rendering of frames and causing texture tearing. So now we see that in technical terms there are two very important numbers Full Fluid Motion (32-36) and refresh (commonly 72Hz) This covers the basics now let’s talk about how this affects gaming specifically.
The next item is more subjective, in fact it is entirely subjective. That is perceived motion. In other words what do you as the game player/ video watcher see? To one person 32FPS might be very fast while to someone else it might be so slow as to be unplayable. This is based on perception and our notions of what makes a good gaming experience.
Here is where it gets tricky, we know that Full Fluid Motion is roughly 36FPS and that refresh should be at 72Hz to maintain this, but what do we know about each person’s experience? If you become used to one thing would you know to be frustrated with it?
For example If I limit you to running all of your games at 15 FPS and then expose you to 30FPS this new experience will dazzle you. Your gaming experience and perception does not even match what we technically consider Full Fluid Motion, yet you are happy with this amazing leap in performance.
At the upper end it is a little different. Here we base our experience on perceived performance. Your eyes cannot tell the difference between 50 FPS and 70 FPS but in the days of the bench mark we know that one is faster than the other. Higher FPS= better performance right? Well yes and No, to be blunt the human eye is not capable of registering much over 32 FPS in video or gaming. Yet we perceive that if I can get one or two more FPS I am doing better! This is a great marketing strategy that has worked for a long time and has us by the short and curlies. We happily shell out more and more money for new GPUs in the hope for another 500 points in 3DMark or an extra 10 FPS that our eyes cannot register.
To me this is simply wrong. So how does this play into what I am talking about?
Well it is very simple to the mainstream gamer who only wants to play his games getting 32-36 FPS in any game is simply good enough. They question become which GPU will give me this for the best price?
We can add to this short and uncomplicated list Power Efficiency and Heat. The Guy/Girl that HAS to have the highest numbers will always shell out more and more for the next fastest product but to the majority (the casual gamer and prosumer) Those cards are almost laughable in their cost, power and size.
Now that I have bored you with my over simplified version of how people view games (and I know this does not cover all aspects and is by no means complete) I will take a look at four GPUs that are still considered current. The ATi 3850 and 4850 and the nVidia 8800GTX and 9800GTX, I will be testing these based on their ability to meet the 32-26 FPS mark using the least amount of power and producing the least amount of heat.
(1)This is the speed that most films are recorded at, however with the move to more theatrical movies being shot on full digital media (HDD) there is some controversy about this number.
With current 24 FPS films to be converted to “video” a 3:2 pull down is used to correct issues between the speed the movie was filmed at and the refresh the TV will display (60 or 50Hz) if your refresh rate is a multiple of the film speed (i.e 48Hz or 72Hz) you do not need this Pull Down to correct discrepancies in playback. These topics are beyond the scope of this article and will not be discussed in detail.
{mospagebreak title=Testing and Test Systems I}
The Subjects:
The test systems are going to be static. I will have one Intel System to use for this testing. In order to rule out possible system issues I am going to use the fastest CPU that I have on hand, the QX9770. System specs are shown below.
Asus EAH3850 Smart OC:
This card, is based off of the ATi 3850 GPU but instead of the 512MB of GDDR3 on most models Asus have chosen to drop a hefty 1GB of Fast DDR2 memory as well as an oversized GPU cooler (making this card a double height card) and a large heatsink for the power regulators. This card features PCI-e 2.0 support, dual dual-link DVI ports, CrossfireX support, an A/V out port and comes complete with a DVI to HDMI converter (the EAH3850 Smart OC supports HDCP). The software bundle is also very nice and includes Asus’ own splendid software for color correction and control, Game OSD for real-time overclocking and in game benchmarking and capture, Video Security a software for home security (requires a web or IP camera), and SmartDoctor which is described as an overclocking and hardware protection tool But these features are not what makes the EAH3850 Smart OC stand out. What makes the EAH3850 Smart OC stand out is the Smart OC button, this small button on the I/O plate (or tang) instantly sends more voltage to the card for overclocking it does not actually change the clocks itself. With this extra voltage you can get these cards quite high on Core and Memory Clocks. The Smart OC button also turns on a red LED that shines through the DVI ports adding to the visual appeal of these cards (if you are looking at the back of the case).
|
|
|
|
|
|
|
|
|
|
GIGABYTE GV-R485-512H-B ATi 4850 512MB
Our second entrant from ATi/AMD is the 4850. This pair of cards hails from GIGABYTE and is a purple and red explosion of color and graphical power. The 4850 arrives with support for PCI-e 2.0, DX 10.1, OpenGL 2.1, HDCP. The GV-R485-512H-B features dual dual-link DVI-I, HDMI support (with an adapter included), a 55nm die, 512MB of GDDR3, with a 256bit memory interface. Full shadermodel 4.1 support, CrossfireX support, is ROHS compliant and is just plain fast. The 4850 is also a single slot card allowing for the very real possibility of Quad Crossfire using these cards on the right system. The software bundle is not as good as on the Asus and the fan can get very loud under heavy gaming but in all this offering from GIGABYTE is a nice one.
|
|
|
|
|
|
|
|
{mospagebreak title=Testing and Test Systems II}
Foxconn FV-N88XMAD2-ON 8800GTX 768MB
Since we have already review the Foxconn FV-N88XMAD2-ON 8800GTX 768MB in the past I will not go into detail but will link the review if you are interested in the cards features.
Foxconn 9800GTX.
Our second nVidian entry is also from Foxconn (being one of the few vendors that will actually send nVidia products our way, Thank you Foxconn!)
The 9800GTX is a monster of a card, it is even longer than the 8800GTX. This makes for some interesting installation situations if you do not measure your case properly. There are also issues on many mainboards as the length of the card interferes with PATA and SATA ports on a few.
The plastic fan shroud runs the full length of the card and to be perfectly honest makes the 9800GTX look very cheesy, and cheap.
Despite the length and looks of the card the 9800GTX boast some eye turning specs. The 9800GTX has support for DX 10, Shadermodel 4.0, OpenGL 2, HDCP support (through DVI-I), comes with some very fast (2200MHz) GDDR3, and many others. The bundle that comes with the Foxconn 9800GTX is nothing to be amazed at and simply includes 2x DVI-I to VGA adapters, 2 power adapters (4-pin Molex to 6-Pin PCI-e), Drivers and utilities CD-ROMs. One thing to note is that the 9800GTX like the 8800 requires two 6-pin PCI-e power connections. The ATi 3850 and 4850 require only one per card.
|
|
|
|
|
|
|
|
|
|
|
|
{mospagebreak title=The Tests}
The tests:
For testing a picked out 6 full games and one demo this was in addition to 3DMark Vantage (for the numbers people) Each game was run through the selected level 3 times and the average of the minimum, Maximum and Average frame rate were used for final scoring. The game settings and level used will be listed with each game. All screen shots and frame rate recordings were done using FRAPS.
{mospagebreak title=Power and Heat}
Power consumption was tested using a P9 Kill A Watt this was attached to the AC cord then plugged into the wall. Power consumption testing was done for the whole system; no locks were used to determine GPU power consumption alone. The reasoning behind this is that when you drop in a new GPU your overall system power consumption is going to change. We looked at the idle power usage in terms of wattage draw from the wall then we ran 3DMark Vantage and observed the power draw for the GPU tests only. This was then averaged into a single wattage draw number listed for each GPU. For gaming we tested in each game after 30 minutes of gaming. These scores were also averaged for a single wattage draw result. We then looked at the delta between the two to determine the power efficiency. This was not meant to be a complete method for testing power draw but is s fairly effective way to determine what a GPUs effect on power draw will be on a given system. As with all reviews remember that these results are not all encompassing, they show power draw as it occurred in my lab and with my test system setup, your results may be different.
|
|
Heat:
Heat or more to the point heat dissipation plays a big role on efficient performance. If your system produces more heat than it can dissipate then you may begin to get sluggish performance. One of the leading heat producers in any system is the GPU. This is why so many companies have begun to put double height coolers massive shrouds, etc on their cards. It is with the intent of getting that heat out of the system (as if you did not already know that). But even with this you still get ambient heat from the card, this is what helped me develop our testing method. For idle temperatures I used a Fluke Laser thermometer. I read temps from the center of the heatsink as well as the back of the card. Measurements were taken while idle, during gaming and while running Cinebench’s OpenGL test. This produced a good range for Idle and load temperatures. For Crossfire setups I took a reading of the ambient temperature between the two cards in addition to the heatsink and back readings.
A foreword on these readings for ease of testing I used an open stand (HSPC test station) this means there was constant air movement around the cards, your temperatures may vary with your cases and their air flow patterns.
|
|
We saw some interesting numbers here. The double height coolers on the nVidia cards tend to pay off, as the higher clocked GTX parts still managed to run cooler under load than the ATi cards. Even the older 3850 has trouble keeping cool during gaming. When the cards were idling the winner here was the 3850 Smart OC with an average temp of 34c. For cooling efficiency the 8800GTX wins as the delta between idle and load was 9 degrees.
Ambient room temp was between 24-27c during power and heat testing
{mospagebreak title=Performance I}
The Games and Applications:
3DMark Vantage:
I threw this one in as a baseline test, it is a synthetic and not representative of real world gaming (which has to deal with sound, AI and other calculations) it is a good test of system potential.
|
|
Although I am sure it is no surprise that the 4850 won out for single card performance, what I find shocking is that the 8800GTX actually beat the 9800GTX by over 350 points. The only reason I can think of for this is the extra memory in the 8800GTX. As expected Crossfire shows a healthy boost to the scores for both the 3850 and the 4850 while the 3850’s magic SmartOC button showed a large gain in only single card mode.
Mass Effect:
Mass Effect is one of two Third Person games that I used for this evaluation. It is a game very much like the Knights of the Old Republic series in terms of look and game play. For testing I ran the Eden level from you drop off onto the planet until your return to the ship. Settings and image quality comparison are shown below.
|
|
|
4850 |
3850 SMOC |
8800GTX |
9800GTX |
|
|
|
|
|
|
|
In Mass Effect, for single card average performance the 8800GTX is the winner by a little over a frame per second. The 4850 followed close behind by roughly 3 FPS. Now for minimum frame rate (single card) the 4850 was the winner by a single frame per second. For multi-card performance the 4850 in Crossfire is the winner by a mile out performing everything by over 10 FPS in average performance and 17 FPS in minimum performance. Interestingly the 3850 in Crossfire was only able to squeak by the 4850 in single card mode. For our first run we find that only the 3850 at stock speed is unable to meet our full fluid motion standard of 32 FPS.
Bioshock.
The first of our DX10 games is Bioshock. This game is from Looking Glass Studios and in terms of game play there is very little like it. The game offers the player a wide range of choices. Graphically the game developers used some DX10 textures for water effects and lighting but on the whole you are seeing mostly DX9. For testing I used the welcome to rapture level up until you enter the 2nd bathosphere.
Settings and Image quality shots are below.
|
|
|
4850 |
3850 SMOC |
8800GTX |
9800GTX |
|
|
|
|
|
|
|
Score another win for the GIGABYTE 4850 in single card mode for average frame rate and a tie for minimum frame rate. For multi-card performance I was shocked, the 3850 Smart OC won out easily over everything else with an average frame rate of 72 although minimum frame rate was taken by the 4850s in Crossfire. The 3850 in single card, stock mode was not able to meet our full fluid motion standard but was not too far under it.
Call of Duty 4 Modern Warfare:
The Call of Duty series have always been fun games that challenge players in a number of ways. Call of Duty 4 is a no exception, with good character models, particle effects and some interesting lighting Call of Duty Modern Warfare can slow down your GPU. For testing I ran the “Crew Expendable” level from beginning to completion. Image quality comparison and settings are shown below.
|
|
|
|
4850 |
3850 SMOC |
8800GTX |
9800GTX |
|
|
|
|
|
|
|
The 4850 manages to eek out a victory by less than 1 FPS, however for minimum frame rate there is no question, the 4850 comes in at 62 FPS while the 9800GTX shows 59. For Multi-card performance the 4850 pair takes this one with no effort. For Call of Duty 4 only the stock 3850 did not make the cut for full fluid motion.
{mospagebreak title=Performance II}
Assassins Creed (DX10)
This title from Ubisoft has managed to grab more headlines for what has been done to it than what it does. I ran Assassin’s creed in its original installed form for testing here, no patches were applied. The level I used was the beginning level up to your execution. Image quality comparison and settings are shown below.
|
|
|
4850 |
3850 SMOC |
8800GTX |
9800GTX |
|
|
|
|
|
|
|
I am sure it came as no surprise to see the 4850 win again in both single and multi-card modes. This was for average and minimum frame rates. Full fluid motion was achieved by all of the cards but the 3850 in single card mode. Perhaps with a little overclock this card would be fitting for your Assassins Creed fix.
GRID (Demo)
From Codemasters GRiD is our only other third person game and is a pretty graphically intense game.
The smoke and dust effects as well as the in game physics can seriously slow you down. Image quality comparison and settings are shown below.
|
|
|
|
|
4850 |
3850 SMOC |
8800GTX |
9800GTX |
|
|
|
|
|
|
|
Here is another big win for the 4850. The 3850, while it did not do bad, was not able to meet our full fluid motion standard of 32 FPS. I do think that with a little overclocking (made easier by the Smart OC button) it would be able to both in single card mode and in Crossfire.
Crysis- x64:
Although I am not a fan of this game I will not deny the impact it has on a system and a GPU. For testing I used the Deactivate GPS Jammer level and ran until after the fire fight on the 2nd beach. Image quality comparison and setting are below.
|
|
|
|
4850 |
3850 SMOC |
8800GTX |
9800GTX |
|
|
|
|
|
|
|
I was shocked that the 9800GTX came out on top here. It even outperformed the 4850s in Crossfire.
As far as our Full Fluid motion standard only the 9800GTX, the 4850 single and the 4850 in Crossfire was able to meet or exceed the 32 FPS mark. Meaning if Crysis is your thing these would be the cards for you to look at.
HalfLife 2 Episode 2
The HalfLife Series is still one of the best games ever written; graphically they might not be at the same level as Assassin’s Creed or Crysis (although their engine is more efficient than either). HalfLife still requires some power to run at very high resolutions and with all settings maxed out. For testing I used the “This Vortal Coil” level and played through all of the Ant Lion attacks. Image quality comparison and settings are shown below.
|
|
|
|
4850 |
3850 SMOC |
8800GTX |
9800GTX |
|
|
|
|
|
|
|
The 4850 from GIGABYTE wins out again, although, the Asus 3850 performs quite well for an older generation card. All of the cards we tested met the full fluid motion standard we set so any of these would be perfect for HalfLife2 gaming.
{mospagebreak title=Value}
Price/Warranty:
The Asus EAH3850 Smart OC will set you back about $150 making a pair of these a very reasonable $300. Warranty is a standard three year warranty on defects in manufacture.
The GIGABYTE GV-R485-512H-B runs around $179.99 although you can find them for as low as $150. This makes about $360 for a pair of these. That is not bad at all considering the performance they exhibited.
Warranty information can be found on GIGABYTE’s web site
Foxconn 9800GTX this card can be found around the Internet for about $210.00
Foxconn FV-N88XMAD2-ON 8800GTX 768MB, although we have reviewed this card before the prices have changed so we will list the current pricing here, it looks like the Foxconn 8800GTX can be picked up for about $250.
Foxconn warranty information can be found on Foxconn’s site.
{mospagebreak title=Conclusions}
Conclusion:
Performance
Looking at the results it is easy to declare a winner for fastest card. The 4850 from GIGABYTE is simply the fastest single card solution and a pair of them is amazingly quick. But does everyone want or need that high of a frame rate? For the most part all of the cards (with the exception of the Stock EAH3850 Smart OC and the game Crysis) were able to surpass the 32 FPS we were looking for. This means that anyone of these cards would work for most people. If all you want s good looking full motion gaming at a reasonable price, the Asus EAH3850 Smart OC is probably the best card for you. With the Asus 3850 Smart OC you can use the smart OC utility to gain a few extra FPS in the few games where stock speed just won’t cut it. If, on the other hand, you are looking for the fastest gaming experience you can get then the GIGABYTE GV-R485-512H-B in crossfire is what you are looking for. If Crysis is your bag, the 9800GTX is the right card for you.
If you are looking for the middle of the road then either the Asus 3850 or the single 4850 will fit the bill.
Heat and Power: None of these cards sip power or are what you would call “cool”. But there are some that stand out while others fail. If you are power conscious and looking for the best performance you might consider than the Asus EAH3850 Smart OC running in Crossfire (at stock) uses less power than a single 4850 while generating the same amount of heat. These same two cards put up some impressive frame rates, but they do cost more than a single 4850.
Final Thoughts:
Although this is setup as a competitive evaluation, this was not really the intent. What I wanted to show is that there are compromises with each choice we make in terms of computer components. I also wanted to demonstrate that the market image that once a new product comes out, the old one is obsolete is not true. The Asus 3850 did not perform faster than the other cards but in most cases it performed more than well enough to work for the mainstream gamer. We also saw that in many cases the older 8800GTX outperformed the newer 9800GTX, in the end my hope is that this article will give you the information to make the best choice for your needs if any of these cards are on your list.
I know that many of you will be asking me "Why didn’t you have the 200 series cards from nVidia?" well the simple truth is that the site has been unable to get one from ANY manufacturer. It would seem that nVidia’s threat to "cut me off" has worked with regards to their current products. As I write for this site on a voluntary basis and will not spend my money on a company that acts like that the current gen nVidia products will remain abesnt until this changes.