by Sergey Lepilov
06/11/2009 | 11:08 AM
The most difficult shopping decision you have to make when purchasing a new gaming computer is what graphics card to choose. The CPU’s performance ceased to be a limitation long ago. Today, a $100 CPU will do just fine for a majority of gaming applications. It is even simpler with system memory and hard disk drives: it is the amount of memory and disk space that is important for games in general. With the memory and HDD prices we have today, there is no problem in getting as much of either as you want. So, the only component left to choose is the graphics card – the main part of a gaming computer that directly determines the latter’s speed and your gaming experience.
That’s why in this review I will describe four new graphics cards (and test a total of nine graphics cards/configurations) priced at $100 to $260 and will try to figure out what performance benefits, in what games and for what money you can get by spending more and more for your graphics subsystem. To illustrate this, the next diagram shows the recommended and retail (according to Newegg.com for the 27th of May, 2009) prices on reference graphics cards selling in the mentioned price range:
The products in the diagram fall into three groups. The first group is the single $99 Radeon HD 4770. Next goes the GeForce GTS 250 which is 50% costlier. The Radeon HD 4850 512MB has the same recommended price as the GTS 250 but its retail price has declined due to the release of the Radeon HD 4770. The Radeon HD 4870 and GeForce GTX 260 (216SP) are only $20 and $30 more expensive than the GTS 250. That’s only 13.4% and 20.1% of difference, so these three cards can be viewed as belonging to the same group.
The remaining two cards stand aside from the entry-level class. The new Radeon HD 4890 and GeForce GTX 275 cost about the same money (39.1 % and 44.7 % above the GTX 260), but there is a gap of $70 between them and the GeForce GTX 260. I will fill it in my test session with two Radeon HD 4770 cards in CrossFireX mode ($198). The more expensive product costs 2.38 times as much as the cheapest one.
So, we’ve got a steady progression of price, but is there a proportional progression of performance? Are the performance benefits from purchasing a faster solution worth the money paid? I will try to answer this question in this review. Besides, I will compare two GeForce GTX 250 cards with 1 and 2GB of onboard memory and find the difference between Radeon HD 4870 and HD 4890 as well as between GeForce GTX 260 (216SP) and GeForce GTX 275. And finally, I will do a comparison of graphics cards’ performance in Windows Vista and Windows7.
First I will present the new graphics cards to you.
The first graphics card to be discussed in this review comes from Hightech Information System Limited and is a new product called HIS Radeon HD 4770 512MB. I was given two identical cards for testing. They come in small crimson-and-black boxes designed in HIS’ traditional style.
Each box contains two adapters, a disc with drivers, an installation manual, and a sticker with HIS logo:
The card has an original appearance thanks to the black-and-silver casing of the non-reference cooling system:
The card measures 208x100x38mm and fits into any ATX system case, but its cooler will block the neighboring PCI slot.
There is nothing especially interesting on the reverse side of the HIS Radeon HD 4770.
The card is equipped with two DVI outputs and an S-Video output.
There are two MIO connectors for CrossFireX configurations and a single 6-pin additional power connector.
The purpose of the power connector is unclear because the Radeon HD 4770 should not need more than 55W, which can be covered by a PCI Express slot with its load capacity of 75W. A 450W power supply is recommended for a system with one Radeon HD 4770, and a 550W power supply should be used for a CrossFireX configuration built out of two such cards.
With the cooler removed, the card looks like that:
The power circuit follows a three-phase design.
The card has a new heart indeed. Manufactured on today’s thinnest 40nm tech process, the RV470 die is only 137 sq. mm large. It is the smallest of today’s consumer GPUs.
This 826-million-transistor cutie features grownup resources including 640 unified shader processors, 32 texture-mapping units and 16 rasterization modules. Thus, the HD 4770 is functionally equivalent to the Radeon HD 4850 and HD 4830 which is currently leaving production lines. As opposed to the HD 4830’s 575MHz frequency, the HD 4770 has a GPU frequency of 750MHz! Moreover, the 40nm tech process brings a promise of high overclockability which could not be had from the HD 4830. I can also add that the card drops its GPU frequency to 250MHz in 2D mode.
The reference HIS Radeon HD 4770 is equipped with 512 megabytes of graphics memory in eight GDDR5 chips located on the face side of the PCB.
The use of GDDR5 chips in an entry-level product ($99) sets a precedent because no other maker has installed such potentially fast memory on graphics cards of this class. The memory bus is 128 bits wide. The Qimonda IDGV51-05A1F1C-40X chips are dated the 36th week of 2008 and work at a frequency of 3200MHz. The rated frequency of these chips is 4000MHz, so they may be good at overclocking. I will check this out shortly.
GPU-Z reports the full specs of this card.
The coolers of the graphics cards covered in this review were tested in a closed system case at an ambient temperature of 24°. The cards’ temperatures and frequencies were monitored with RivaTuner 2.24 excepting the Radeon HD 4770 which is not yet supported by RivaTuner and was monitored with GPU-Z. Each card was loaded by FurMark version 1.6.5 running in stability mode at 1280x960. This test loads the GPU very heavily. It ran for about 15 minutes until the GPU temperature fully stabilized. I used Arctic Silver 5 thermal grease as I had dismantled the cards before the tests.
So, the cooler of the HIS Radeon HD 4770 is an aluminum fan-like thing painted the color of copper. The heatsink is cooled with an 80mm fan and has a plastic cap. The cooler was quiet most of the time but would give out an occasional wail for a couple of seconds. Its performance was good enough, though:
Unfortunately, the Arctic Cooling Accelero Twin Turbo could not be installed on the Radeon HD 4770 as it did not fit the latter’s size so I instead installed a Thermaltake S-Orb which helped reduce the GPU temperature by as much as 23°C.
As you can see, the GPU temperature of the Radeon HD 4770 512MB (at 750/3200MHz) is no higher than 47°C under FurMark! This is an impressive result, so it’s time to check the cards out at overclocking.
I overclocked them with Catalyst Control Panel and GPUTool.
Alas, the results are not very good. The GPUs could only be overclocked to 850 and 865MHz. The graphics memory could be overclocked to 3340MHz on one card and could not be increased at all on the other.
Of course, that’s not what you can expect from a 40nm GPU and GDDR5 memory. The memory chips behaved oddly: they could work at a high frequency but then would hang the system at a lower one. After I finished my tests, ATI released a new BIOS for the reference Radeon HD 4770. Users report that this BIOS helps overclock the card’s memory better. Perhaps we will check it out in a future review.
The Radeon HD 4770 has a recommended price of only $99.
The GeForce GTS 250 is the new reincarnation of GeForce 8800 GTS, 9800 GTX and 9800 GTX+. It will be represented by two cards from Palit Microsystems Ltd. in this review. Both cards come in large yellow boxes with a picture of an armored frog on the face side and a brief specification on the back.
The cards differ with the red stickers in the bottom right corner: the first sticker shows the access time and amount of graphics memory while the other indicates that the card has a double amount of memory.
I will discuss these differences shortly. Let’s first take a look at the cards’ accessories.
The accessories are rather scanty. There is only one power cable and no S-Video cable. No games are included, either. On the other hand, these are entry-level products, so you should not expect them to have gorgeous accessories.
The cards are identical and look like shown in the photo:
Each card measures 230x100x39mm. The cooler has a dual-slot design and blocks the neighboring expansion slot on the mainboard.
Each card is equipped with an analog, DVI and HDMI connector. There is a vent grid in the mounting bracket.
There are MIO connectors at the top of the PCB for building SLI and 3-way SLI configurations. There are also two 6-pin power connectors there.
The reference GeForce GTS 250 should not consume more than 150 watts. A 450W power supply with a load capacity of 24A on the 12V rail is recommended for it.
Now let’s have a look at the graphics cards with their coolers and memory heatsinks removed:
As you can see, there are memory chips on both sides of the PCB. The power circuit uses a 4-phase design.
The G92b chips were manufactured on 55nm tech process in Taiwan. They are revision B1 and differ with the week of manufacture and the third line of the marking.
Each GPU incorporates 128 unified shader processors, 64 texture-mapping units, and 16 raster operators. The ROP/TMU frequencies are 745/1848MHz which is just a tiny bit higher that the reference GeForce GTS 250’s frequencies (738/1826MHz). These GPUs do not have a power-saving mode.
As these two cards have different amounts of GDDR3 memory, you can guess that they have different memory chips. Indeed, the 1GB version comes with Samsung K4J52324QH-HJ08 chips (0.83ns, a rated frequency of 2400MHz) whereas the 2GB version comes with Hynix H5RS1H23MFR N2C chips (0.8ns, a rated frequency of 2400MHz).
|
|
Oddly enough, the 1GB version has a memory frequency of only 2000MHz, which is 200MHz lower than the reference card’s. The 2GB version has a memory frequency of 2200MHz. You should keep this fact in mind if you are choosing between these two products.
The card’s characteristics are overall as follows:
|
|
The cooling system is downright primitive. Its top part is a plastic casing with an 80x15mm fan.
The 9-blade fan was manufactured by Everflow (the R128015BH model). Its speed is varied by the card in an unknown range (the fan performance is reported in percent rather than rotations per minute).
This fan cools a rather small heatsink with a copper base and two copper nickel-plated 6mm heat pipes that carry thin aluminum ribs soldered to the base.
The memory chips on both sides of the PCB are cooled with aluminum plates that have thermal pads.
The graphics cards’ coolers were tested according to the same method as was used for the above-discussed Radeon HD 4770. They proved to have identical temperatures so I will only show you the results of the 2048MB version.
|
|
Unfortunately, the cards’ power components and PCBs are not equipped with thermal sensors but you can see that the GPU is not very hot. The coolers are loud, but I will discuss the noise factor later on. When I installed an Accelero Twin Turbo, the GPU was no hotter than 57°C although the temperature reduction did not affect the cards’ overclockability. By the way, neither sample of Palit GeForce GTS 250 did particularly well at overclocking.
|
|
The 2GB version of the card is $15-20 more expensive than the 1GB one. You can email me for the cards’ BIOSes.
The above-discussed GeForce GTS 250 cards cost about the same money as the Radeon HD 4850. The Radeon HD 4870 and GeForce GTX 260 are more expensive. These three graphics card models have been discussed in our earlier reviews, so I will proceed right to the next step: Radeon HD 4890 1GB.
This model is represented by two products in this review: one more card from HIS and a card from Diamond. Both come in small but nice-looking boxes.
Inside the cardboard wrapper there are smaller boxes that contain the cards and accessories.
HIS Radeon HD 4890 1 GB
| Diamond Radeon HD 4890 1 GB XOC
|
The HIS card comes without the exclusive flashlight-screwdriver the manufacturer used to include with its products before. There are no games enclosed with either card.
These products are exact copies of the reference card and only differ from it with the stickers on the coolers.
|
|
Each card measures 244x100x35mm and has a dual-slot cooling system.
These Radeon HD 4890 from HIS and Diamond are equipped with a PCI Express x16 version 2.0 interface, two dual-link DVI-I ports with support for high resolutions, and an S-Video output. There is a grid in the mounting bracket for the hot air to leave the system case.
The connectors for CrossFireX mode are still at the top of the PCB (it would be hard to place them anywhere else, though).
Let’s take a look at a Radeon HD 4890 with its cooler removed:
The power circuit is considerably revised in comparison with the Radeon HD 4870’s and now uses a 7-phase design (refer to our article called ATI Radeon HD 4890: 1GHz Conquered! for details).
The Radeon HD 4890 has a peak power consumption of 190 watts which is 20 watts higher than that of the Radeon HD 4870. A 550W or higher PSU is recommended for a single Radeon HD 4890. A CrossFireX configuration may require a 700W or higher PSU. Our recent review showed, however, that the power consumption estimates provided by the manufacturers are often overstated.
The key difference of the Radeon HD 4890 over its predecessor Radeon HD 4870 is the new RV790 graphics processor. The GPUs installed on my samples of the cards were manufactured in Taiwan on the 8th and 10th weeks of this year.
|
|
The GPU has not changed functionally. It still incorporates 800 unified shader processors, 40 texture modules, and 16 raster back-ends. The transistor count has grown up by 3 million to 959 million while the die has grown from 256 to 282 sq. mm. The GPU frequency has bee increased from 750 to 850MHz and even to 900MHz for the OC version. Moreover, some brands even turn out Radeon HD 4890 with a core frequency of 950 and even 1000MHz! That’s an impressive progress (a 25% frequency growth) within the same tech process although such cards are going to be rather rare. My cards differ in GPU frequencies: 850MHz for the HIS card and 925MHz for the Diamond (which is 75MHz or 9% above the reference frequency). Each card drops its GPU frequency to 240MHz and voltage from 1.31 to 1.05V in 2D mode.
The cards come with 1024MB of memory in eight GDDR5 chips from Qimonda manufactured on the 46th week of 2008.
The chips are revision A1 and are marked as IDGV1G-05A1F1C-40X. They have an access time of 1.0 nanosecond and a rated frequency of 4000MHz, but the reference memory frequency of the Radeon HD 4890 is 3900MHz (300MHz higher than that of the Radeon HD 4870). The memory bus is 256 bits wide. The HIS card has the same memory frequency as the reference card while the Diamond’s memory frequency is overclocked to 4200MHz (300MHz or 7.7% above the reference card’s).
Here are the full specs of these two products:
|
|
The cooling system of the Radeon HD 4890 has not changed much since the Radeon HD 470 and still consists of an aluminum plate that touches the power components and memory chips, a heatsink with copper base, three heat pipes and aluminum ribs, and a plastic casing with blower.
The only differences in the new card’s cooler is that it now has one 8mm and two 6mm heat pipes (instead of two 8mm pipes) pumping the heat from the copper base to the heatsink ribbing.
I don’t think much about this change because the heatsink, which is the key component of every cooler, is still too small to cool the GPU effectively. On the other hand, you can see in the photo above that the current cooler design does not allow to make the heatsink larger.
The blower has not changed, either. It is the CF1275-B30H-C004 model from NTK Technologies Inc.
Besides checking the efficiency of the Radeon HD 4890’s cooler, I will compare it with the reference Radeon HD 4870 that has the following specs:
The Radeon HD 4890 is represented by the reference card from HIS. First let’s see what temperatures the cards have at the default frequencies and in the automatic fan control mode.
|
|
Notwithstanding the higher frequencies, the Radeon HD 4890 is cooler than the Radeon HD 4870 but you should note that the HD 4890’s fan worked at 2800rpm whereas the HD 4870’s at 2200rpm only. Next, let’s see what temperatures the cards have at the maximum speed of their fans:
|
|
Now the Radeon HD 4870 is cooler, just as you can expect. The difference in the previous test must have been due to the difference in the fan speeds. The overclocked Diamond is just as hot as the HIS card under load but its fan is rotating 240-300rpm faster.
To compare the reference Radeon HD 4890 cooler with an alternative, I took one of the best options in this class – an Arctic Cooler Accelero Twin Turbo. I installed it on the GPU while the power components and the memory chips were cooled by the plate from the reference cooler.
My attempt failed, however, Even at the max speed of the Twin Turbo fans the card got alarmingly hot 3 minutes into FurMark and I had to stop the test.
Therefore I checked the cards’ overclockability with the reference cooler working at its max speed. Alas, the HIS card could only be accelerated to 910/4400MHz. The pre-overclocked card from Diamond did not show a good result, either.
|
|
The memory frequency of the overclocked HIS is good but its GPU is a disappointment. There was no talking about 1000MHz with that sample. I hope you’ll be luckier with yours. If you don’t want to rely on luck, you may want to consider the HIS Radeon HD 4890 TURBO 1GB that has default frequencies of 900/4000MHz. The GPU of the Diamond card could not be overclocked at all while its memory worked at 4560MHz.
The Radeon HD 4890 has a recommended price of $249. You can email me for the BIOSes of the two described models.
The last new card to be presented in this review is manufactured by Zotac International (MCO) Limited. It is a new GeForce GTX 275 896MB that comes in a large and eye-catching box with a cutout in the face panel for you to see some of the cooler’s casing:
The text on the face panel of the box informs you about the type of memory, the width of the memory bus, video interfaces, etc. Besides, the box has a GRID sticker meaning that the namesake game is included into the box.
You get the following stuff together with a copy of Zotac GeForce GTX 275:
I should confess Zotac’s accessories look the most appealing among the products presented in this review. This might have been expected, though, as this is one of the most expensive cards here, too. The new graphics card follows the reference design and is no different from the GeForce GTX 260 but, like on the GeForce GTX 285, there is no aluminum cap on the reverse side of the PCB:
The GeForce GTX 260 is cooler without the back cap, which must be the reason why this cap was abandoned in the GeForce GTX 285 (and in the GTX 275, too).
The graphics card is equipped with a version 2.0 PCI Express interface, one S-Video connector and two DVI-I ports. There are slits in the card’s mounting bracket for exhausting the hot air.
There is a MIO connector at the top part of the PCB. At the opposite side of the PCB there are two 6-pin power connectors and an SPDIF header.
According to the specs, the GeForce GTX 275 has a peak power draw of 219W, so a 550W or higher PSU is recommended for it.
Here is this graphics card with the cooler removed:
And this is its power circuit:
Covered by a heat-spreader, the GT200b processor was manufactured on 55nm tech process on the second week of 2009. It is revision B3.
The GPU incorporates 240 unified shader processors, 80 texture-mapping units and 28 raster operators. Thus, the GeForce TX 275 nearly equals the GeForce GTX 285 functionally but its frequencies are lower: 633/1404MHz as opposed to 648/1476MHz. In fact, the difference is negligible in terms of resulting performance. The GPU frequencies are dropped to 300/600MHz when the card works in 2D mode.
The card is equipped with 896MB of GDDR3 memory in 0.83ns Samsung K4J52324QH-HJ08 chips.
The rated frequency of the memory chips is 2400MHz but the GeForce GTX 275 has a memory frequency of 2268MHz in 3D mode, which is 216MHz lower than that of the GeForce GTX 285. Besides, the GeForce GTX 275 has a 448-bit memory bus whereas the GTX 285 has a 512-bit bus. The memory frequency is dropped to 200MHz in 2D applications.
Here is a summary of the graphics card’s characteristics:
The GeForce GTX 275 is equipped with the same cooler as the GeForce GTX 285. You can learn about it in our earlier review, so I will only show you a photo of it:
Now let’s check out the graphics card’s temperature:
|
|
So, the GeForce GTX 275 is a very hot graphics card and is no cooler than the Radeon HD 4890. However, its blower is much quieter than the blower installed on Radeons. I will discuss the noise factor later in this review, though.
I checked the card’s overclockability using its reference cooler working at 100% speed. As a result, I increased the GPU frequencies to 702/1557MHz (+10.9%) and the graphics memory frequency to 2522MHz (+11.2%).
That’s not much of a frequency gain, I should say. The good news is that the card’s GPU grew only 1°C hotter when overclocked.
The Zotac GeForce GTX 275 comes at a recommended price of $259.
First, let me say a few words about the hardware that we used during this test session. All graphics cards were benchmarked in a closed system case with the following configuration:
To minimize the platform influence on the performance of the tested graphics cards I overclocked our quad-core CPU to 4.01GHz at 1.36875V voltage:
The system memory worked at 1527MHz frequency with 7-7-7-14_1T timings and 1.64V voltage:
We tested the platforms stability with these settings during two consecutive runs of LinX (Linpack 32-bit) with 15 cycles in each, which is more than enough to confirm systems stability within our today’s test session.
Besides six graphics accelerators we have just talked about above, we have also added the results for the following cards:
|
|
|
|
|
|
Now let’s move on to the software components. The main part of our test session took place in Windows Vista Ultimate Edition x86 SP1 (plus all critical updates as of 04.16.2009). We started testing on 04.16.2009, so these are the driver versions we used, that were available at the time:
The only exception was Radeon HD 4770. For some reason I couldn’t install the Catalyst 9.4 driver for it. I could do Catalyst 9.3 or 9.5 just fine, but not 9.4. Therefore, I decided to go with version 9.5, as the latest available at that point. Of course, it is incorrect to compare the performance of graphics cards tested with different driver versions, but on the other hand, there is no mention in the Catalyst 9.5 release notes about any performance differences compared with version 9.4 in the games and benchmarks that we will be using today. That is why we decided not to rerun the tests for Radeon HD 4870, HD 4890 and HD 4850 with the newer Catalyst 9.5.
Once the drivers had been installed, I made the following changes in their control panels: image quality set from “Quality” to “High Quality”; Adaptive Antialiasing set to “Quality”; vertical synchronization set to “Always Off”. There were no other changes. I turned on full-screen antialiasing and anisotropic filtering in the menu of each game. If the game didn’t provide such options, I enabled FSAA and AF from the GeForce and Catalyst control panels.
The graphics cards were tested in two resolutions: 1280x960 and widescreen 1920x1200. Just like before, we decided to go with a not very common first resolution instead of the popular 1280x1024, because GeForce GTX 260 doesn’t work in 1280x1024 (we have already discussed this issue in our article called “Evolution of Nvidia GeForce Driver: GeForce GTX 260 (216SP) under Investigation” and the problem hasn’t been solved since then). The tests were performed in two image quality modes: “High Quality” without any image quality enhancements and “HQ+ AF16x+AA4/8x” with enabled 16x anisotropic filtering and 4x full screen anti-aliasing (or 8x FSAA if the average framerate was high enough for comfortable gaming experience).
We have modified the list of games for our today’s test session. We no longer have Call of Duty 4: Modern Warfare that has finally been replaced with Call of Duty 5: World at War (patch version 1.3 allows recording demos). Besides, we also added such gaming titles as Stormrise, Tom Clancy's H.A.W.X. and The Chronicles Of Riddick: Assault On Dark Athena; and World in Conflict is being used with the new patch version 1.0.1.0, which provides a better benchmark. So, the complete list of test applications includes the two popular synthetic benchmarks and 15 games of various genres:
Here I’d like to add that if the game allowed recording the minimal fps readings, they were also added to the charts. We didn’t use FRAPS utility during our tests because it produced high measuring error.
The graphics cards are ranked in the diagrams in the order of ascending recommended price. Here is the full list:
As you can note, I brought the cards’ frequencies in compliance with the respective reference cards’ frequencies for this test session. I will not test them at overclocked frequencies because there are too many models some of which are quite poor in terms of overclockability.
Synthetic benchmarks come first.
Obviously, there is no strict correlation between the pricing and performance of the cards in 3DMark 2006 although the Radeon HD 4770, the cheapest product in this test session, has the lowest performance indeed. Next go the two GeForce GTS 250. By the way, the two GeForce GTS 250 with different amounts of onboard memory are absolutely identical in terms of performance in this benchmark as well as in the other tests. Therefore I will not comment upon this issue in this section anymore.
The recommended price of the Radeon HD 4850 512MB is $10 higher than that of the GeForce GTS 250 but its 3DMark06 results are lower – it only has a small advantage over the Radeon HD 4770. The Radeon HD 4870 1GB and GeForce GTX 260 896MB are similar to each other and somewhat faster than the GTX 260. Then, the CrossFireX configuration built out of two Radeon HD 4770 cards proves to be the fastest in 3DMark 2006, outperforming the more expensive Radeon HD 4890 1GB and GeForce GTX 275 896MB. This is a nice start for CrossFireX technology, but we are yet to see how it will do in real-life games. Let’s check out the second synthetic benchmark now.
The diagram below shows not the overall 3DMark Vantage scores but the results of the cards in the graphics tests.
According to 3DMark Vantage, there is no point in overpaying for the graphics cards from Radeon HD 4770 to HD 4850. The four products in that range score the same amount of points in the GPU test. Then, there is a 25% performance jump as we move from the Radeon HD 4870 to the GeForce GTX 260 while the CrossFireX HD 4770 configuration turns in an impressive result, winning the test once again. We can also note that the GeForce GTX 275 enjoys a small advantage over the Radeon HD 4890.
As opposed to the two synthetic benchmarks, the real-life game shows no surprises. The Radeon HD 4770 is the slowest card in the total of test modes and resolutions but not much slower than the GeForce GTS 250. This card is even ahead of Nvidia’s products and of the low-frequency Radeon HD 4850 in the hardest visual mode. Thus, the four cards priced below $150 are actually similar in World of Conflict. And then, when we switch to the Radeon HD 4870 and GeForce GTX 260, we can see the frame rates grow up. The Radeon HD 4770 CrossFireX subsystem is faster yet and beats the more expensive Radeon HD 4890 and GeForce GTX 275 in the simpler test mode. The CrossFireX configuration is as fast as the GeForce GTX 275 in the AF+AA test mode, although the latter card is preferable in terms of bottom speed.
We have interesting results in this game. First, the Radeon HD 4770 performs very well, beating both GeForce GTS 250 (which are again equal to each other) as well as the Radeon HD 4850! It looks like AMD should stop producing not only Radeon HD 4830 but also HD 4850 now that there is the Radeon HD 4770. Second, AMD might think about introducing a dual-chip Radeon HD 4770 X2 with a recommended price of $199 or something because the CrossFireX configuration with two such cards is very impressive. Well, I have to confess that a rather annoying flickering of textures could be seen on the Radeon HD 4770 CrossFireX tandem in the FSAA mode, so CrossFireX technology does not work well in this game. But perhaps the driver for the new card is not yet polished off. And third, all the cards with Nvidia chips slow down surprisingly with 8x FSAA and fall behind their Radeon opponents.
The average frame rate grows up in nearly equal steps from one graphics card to another, more expensive one in Unreal Tournament 3. There are three exceptions, though. First, the two GeForce GTS 250 are not only faster than the Radeon HD 4850 but equal the Radeon HD 4870 (in the AF+AA mode). Second, the Radeon HD 4770 CrossFireX outperforms the two more expensive single cards. And third, the Radeons are inferior to the GeForces at the high-quality settings. It is just the opposite to what we’ve seen in the previous game.
Besides the almost 100% performance scalability delivered by the Radeon HD 4770 in CrossFireX mode, we can note the impressive results of the GeForce GTS 250 in the easy test mode. They beat not only their closest opponents Radeon HD 4770 and HD 4850 but also the Radeon HD 4870 and even GeForce GTX 260! I guess they do so because of their very high GPU frequencies. This supposition is confirmed by the fact that the cards take their habitual standings when full-screen antialiasing and anisotropic filtering are turned on. Here, the GeForce GTX 260 and GTX 275 are inferior to the Radeon HD 4870 and HD 4890, respectively, at 1920x1200. CrossFireX technology works blamelessly in Devil May Cry 4.
S.T.A.L.K.E.R. is a playground for Radeon cards as you can see in the diagrams. The CrossFireX configuration with two Radeon HD 4770 is outstanding!
It is clear that the cards with a small amount of graphics memory (Radeon HD 4770 in single and CrossFireX mode, and Radeon HD 4850) are inferior to their opponents with twice the amount of memory. Even the GeForce GTS 250 doesn’t slow down as much as those cards. There is no smooth growth of performance depending on price here. The GeForce GTX 260 is almost equal to the Radeon HD 4870 and the GeForce GTS 275, to the Radeon HD 4890. The Radeons are occasionally better in terms of average frame rate.
Far Cry 2 produces the same picture as Crysis: the graphics cards with 512MB of onboard memory slow down in the difficult test modes. As a result, the Radeon HD 4770 CrossFireX tandem wins the easy modes but can’t cope with 8x full-screen antialiasing and 16x anisotropic filtering. The same goes for the Radeon HD 4770 and HD 4850. The two GeForce GTS 250 with 1 and 2GB of memory don’t suffer such a performance hit and do not fall far behind the senior models. The GeForce series wins in the top price segment.
The Radeons beat the GeForces in X3: Terran Conflict. CrossFireX technology works well, and the Radeon HD 4770 tandem wins this test. Take note that every graphics solution ensures a comfortable frame rate in every mode and at any resolution in this game.
The Half-Life 2 engine this highly popular game runs on has always been favorable towards Radeon GPUs. This is true this time around, too. The red team wins in every segment although the difference in the bottom segment is negligible: the Radeon HD 4770, two GeForce GTS 250 and Radeon HD 4850 are close to each other, and even the slowest of them can ensure a comfortable frame rate at every resolution. Therefore, the HD 4770 is the optimal choice for playing Left 4 Dead and there is no point in paying extra for a faster card.
As opposed to Left 4 Dead, this game runs faster on GeForce GPUs which win in every price segment. They are only challenged by the CrossFireX configuration with two Radeon HD 4770 cards that offers a 90% performance scalability relative to the single card. A stable performance growth can be observed within both AMD’s and Nvidia’s product ranges but we can note that the Radeon HD 4770 is again as fast as the Radeon HD 4850 that is up to 50% more expensive.
It’s the same as in Lost Planet: Colonies except that the Radeon HD 4770 CrossFireX cannot beat the single Radeon HD 4890. The game engine is obviously optimized for Nvidia’s GPU architecture.
We can again observe a steady performance growth within each GPU developer’s product range. Nvidia’s products are generally faster than AMD’s in Call of Duty 5.
Stormrise, a beautiful DirectX 10 based real-time strategy, shows quite a different picture. AMD is beyond competition as the Radeon HD 4770 beats the GeForce GTX 275 that costs 2.5 times as much as it. Nvidia’s programming team must have yet been unable to optimize the driver for the new game. Hopefully, they will do that soon.
The leaders change again. Nvidia wins this test and only meets some competition from the Radeon HD 4770 CrossFireX. It is only in the high quality mode at 1920x1200 that we see the performance increase in neat steps.
The last game in today’s test session shows an increase of average frame rate depending on the price of the graphics cards but the Radeons are generally faster. CrossFireX works correctly in the new game in terms of both scalability and performance.
A comparative diagram with the results of the two GeForce GTS 250 cards with different amounts of memory was meant to be the first in this section, but the test session showed no difference between the 1GB and 2GB versions of GeForce GTS 250. Therefore there is no need to build such a diagram. So, the first diagram will show how faster the new Radeon HD 4890 is in comparison with the Radeon HD 4870 whose performance is considered the baseline.
In the low-quality mode, the Radeon HD 4890 has an average advantage of 6.4% and 10.1% over the Radeon HD 4870 at 1280x960 and 1920x1200, respectively. When FSAA and anisotropic filtering are turned on, the advantage is 9.7% and 10.2%, respectively.
The next diagram compares the GeForce GTX 260 and GeForce GTX 275 with the former card as the baseline.
These two cards differ more than the Radeon HD 4870 and HD 4890. The GeForce GTX 275 is an average 13.3% and 15.1% faster at 1280x960 and 1920x1200, respectively, in the low-quality mode and 15.3% and 14.5% in the high-quality mode.
Then, let’s check out the efficiency of the CrossFireX tandem based on two Radeon HD 4770 512MB cards as compared to a single Radeon HD 4770.
CrossFireX technology works in every game now but the performance growth seems to be limited by the amount of graphics memory which is not enough for modern games at high graphics quality settings.
Additionally, you can view a diagram comparing the Radeon HD 4770 CrossFireX with the Radeon HD 4890 and a diagram comparing the GeForce GTX 275 with the Radeon HD 4890.
I performed the power consumption test with a multifunctional Zalman ZM-MFC2 panel. This panel measures the overall power draw of the computer (without the monitor) from the wall outlet. There were two test modes: 2D (Word and Web surfing) and 3D (FurMark 1.6.5 running in Burn mode in a 1280x960 window). FurMark seems to be the heaviest load possible for a graphics card, so I chose it for the testing purposes.
Here are the results:
It is clear that the system with one Radeon HD 4770 is the most economical as its overall power draw is no higher than 260 watts. The Radeon HD 4850 512MB goes next, and the GeForce GTX 260 896MB is third in terms of power saving. Interestingly, each GeForce GTS 250 consumes more than the GeForce GTX 260, which must be due to their high frequencies and GPU tech process (65nm for the GTS 250 and 55nm for the GTX 260). The system with two Radeon HD 4770 in CrossFireX mode has about the same power consumption. The Radeon HD 4870, GeForce GTX 275 and Radeon HD 4890 are the most uneconomical. Note also that Nvidia’s products consume less power in 2D mode.
The amount of noise produced by the graphics cards’ coolers was measured with a digital noise-level meter Center-321 at night in a totally quiet apartment from a distance of 1 meter from the closed system case. The speed of the five 120mm system fans was reduced to 660rpm for the test. The background noise of the system case was as low as 32.9dBA. When the PC was turned off, the sound-level meter reported 28.8dBA. Subjectively, the noise is comfortable at 34.5dBA, which is marked with a red line in the diagram. The test method is simple: the speed of the cooler’s fan is being changed with RivaTuner from 100 to 20% (or 25%) stepping 5% and the amount of noise was measured at each step.
The diagram shows the results of four graphics cards for easier reading. The Radeon HD 4770 is always working at 35.6dBA but can give an occasional wail up to 50.2dBA for a couple of seconds. Both GeForce GTS 250 are as noisy as 37.2dBA in the automatic fan control mode and 46.7dBA at full speed. The diagram shows the results of the most interesting products (in my opinion).
Easy to see, both Radeon HD 48xx cards are louder than their opponents. The bad news is that their noise is growing up faster than that of the GeForce cards (you can see a steep rise of the graphs in the diagram). The GeForces differ in terms of noise and fan speed control. For example, the GeForce GTX 275’s cooler works at the same speed (of about 1400rpm) and produces 36.0dBA of noise until 40% capacity. The GeForce GTX 260’s cooler works at 900rpm and is quieter. The fan speed control can be adjusted in the graphics card’s BIOS (this works for both AMD and Nvidia products). So, you can achieve an acceptable ratio of noise and temperature by doing such adjustment. The more radical way is to install an alternative cooling system.
The final diagram in this section shows the temperatures of the tested cards’ GPUs.
The more advanced graphics cards have higher temperatures. The only exception is the Radeon HD 4770 and HD 4850: these cards were roughly equal in the performance tests but the Radeon HD 4770 is preferable in terms of heat dissipation (and power consumption). On the other hand, the HIS Radeon HD 4770 comes with an alternative cooler (although it is not as effective as the Thermaltake S-Orb, for example) whereas the HIS Radeon HD 4850 has the reference cooler with a shrieking and inefficient blower.
To wind up this review I will test two graphics cards in Windows Vista x86 SP1 and Windows7 x64 (build 7100) to compare their performance under the different versions of Windows and see if AMD’s and Nvidia’s drivers are ready for the new OS. I chose the 64-bit version of Windows 7 due to the low price of system memory (32-bit OSes do not support more than 3.7GB). As the memory requirements of modern applications and games are growing steadily, most users will switch from Windows Vista x86 (the most popular Vista) to a 64-bit version of Windows7. I guess this comparison should be interesting.
The two fastest graphics cards were tested: Diamond Radeon HD 4890 1GB XOC and ZOTAC GeForce GTX 275 896MB. Both were overclocked to the following frequencies:
This test was performed after the gaming tests and I could use the latest versions of the drivers: Catalyst 9.5 from May 15, and GeForce 185.85 (Windows Vista x86 and Windows7 x64) from May 6. The drivers were set exactly as for the gaming tests. There were no changes in the list of benchmarks.
The results of this test are listed in the following table (the percentage shows the advantage of Windows7 x64 relative to Windows Vista x86 SP1):
Click to enlarge
The results indicate that both GPU developers are ready for Windows7, their new drivers delivering at least the same performance under the new OS as under Windows Vista SP1. Moreover, we can see a performance growth under the new OS in such games as Enemy Territory: Quake Wars, Unreal Tournament 3, S.T.A.L.K.E.R.: Clear Sky and The Chronicles of Riddick: Assault on Dark Athena as well as in both synthetic benchmarks. It is only in the high-quality modes of Far Cry 2 that the GeForce GTX 275 slows down under the new OS. There is still half a year till the official release of Windows7, so we can expect the new platform and graphics card drivers to get even more optimized.
Getting back to the topic of this review, I will again show you a diagram with the recommended and retail prices of the graphics cards. According to the results of the gaming benchmarks (we didn’t take the results of synthetic tests into account here) we calculated the average performance gain for all graphics cards in relation to the lowest-end and cheapest solution of our today’s roundup – Radeon HD 4770. We take its performance as 100% and the speed of all other solutions is shown in percentage to that of Radeon HD 4770 and added to the pricing chart. Here is what we got:
The smaller the difference between the price and performance columns, the better price/performance ratio the graphics card has. Thus, purchasing a GeForce GTS 250 or a Radeon HD 4850 instead of a Radeon HD 4770 is not wise. Switching to a Radeon HD 4870 1GB or a GeForce GTX 260 896MB is more reasonable. The HD 4870 actually seems to be the most optimal buy. The CrossFireX tandem with two Radeon HD 4770 cards is the best of the three top-performance solutions of this test session, but the previous two cards seem to be a better buy than it or than the Radeon HD 4890 1GB or the GeForce GTX 275 896MB. Unfortunately, the noise and power consumption data could not be fitted into the diagram although these factors often determine users’ shopping decisions. Anyway, you can find such data in the body of the review and make up your own mind.
I won’t repeat my summary about the comparisons made in this review (between the Radeon HD 4870 and HD 4890, GeForce GTX 260 and GTX 275) as the diagrams have showed everything clearly. As for the specific cards tested for this review, I can’t single out any of them. The reference Radeon HD 4890 come with standard accessories and could not reach a 1000MHz core frequency at overclocking but the Diamond card is more appealing due to factory overclocking. The HIS Radeon HD 4770 is a cheap and economical product. It would make an ideal low-end card if it had a quieter cooler. The Palit GeForce GTS 250 cards also come with noisy coolers. Moreover, the engineers have implemented queer solutions, equipping one of them with absolutely useless 2GB of memory and reducing the memory frequency of the other. The reference Zotac GeForce GTX 285 can be commended for good accessories and lower noise in comparison with the Radeon HD 4890.
I personally would want to see a dual-processor Radeon HD 4770 X2 with two RV740 chips and with a total of 2GB of memory (1GB per each GPU) and perhaps with increased GPU and memory frequencies (800MHz should not be a problem for this GPU as I have made sure). Equipped with a quiet cooler and priced at $200-220, such a card would make a bestseller.
