by Sergey Lepilov
11/21/2008 | 11:38 AM
Since the release of the Radeon HD 48xx series of graphics cards based on AMD/ATI’s Radeon processor, competition between the GPU developers has become tougher in every market sector. It would be even more correct to say that Nvidia, after having ruled unchallenged in nearly every price category for a while, has now got problems fighting for customers. According to an analytical report from Jon Peddie Research, AMD won 5% of the graphics card market from Nvidia in Q3. The company owes this success to the triumphant introduction of the Radeon HD 48xx series in the first place. Now AMD has 40% of the market while Nvidia keeps 60%, but the red team is going to push its advances further.
Spurred by the financial crisis and lowering demand, Nvidia has prepared itself to a radical cut in the pricing of its GeForce 9800 GT and 9600 GT series as well as of the GeForce GTX 2x0. The former cards do not provide a comfortable frame rate at high-quality settings and high resolutions in modern games but can suit a low-end computer. As opposed to them, the GeForce GTX 280 and 260 are the best option for overclockers among Nvidia’s cards. This option will become even more appealing after the expected price cut. The GTX 280 may get as cheap as $375 and the GTX 260 may drop to $235 in price, according some sources. We just have to wait and see.
Nvidia doesn’t limit itself to price cuts only, though. The GeForce GTX 260 has a dangerous opponent, the Radeon HD 4870, which is generally somewhat faster at high-quality display modes and is cheaper in retail (although the recommended prices of both cards are identical at $299 at the time of my writing this). Therefore, besides the prospective price reduction, Nvidia has released a reinforced version of GeForce GTX 260 that features not 192 but 216 unified shader processors. This is only 24 processors less than in the top-end GeForce GTX 280. There are also rumors on the Web about new drivers from Nvidia that are going to be released soon to push the performance of these cards up to a new level.
And what about AMD/ATI? Judging by what information you can get from the available sources, the company is not much concerned about Nvidia’s planned attack. The RV790 processor is looming in the horizon but we can’t be sure we’ll see RV790-based cards by the end of this year. Well, why should AMD/ATI worry at all? The Radeon HD 4670, 4830, 4850 and 4870 models are actually faster (or at least no slower) than their opponents and cover the entire price range from $80 to $300. And surely AMD has some reserves to lower the recommended prices when necessary.
As opposed to Nvidia’s G200 processor, the RV770 chip, the heart of Radeon HD 4850 and 4870 cards, always came with all of its shader processors active and available. The graphics card models were differentiated by frequencies and memory type. The amount of graphics memory is 512 megabytes on every reference card whereas the GeForce GTX 260, the market opponent to the Radeon HD 4870, has 896 megabytes. Many graphics card vendors corrected this deficiency with their own hands by installing 1024MB of memory on their versions of Radeon HD 4870. One such card from Palit is going to be tested in this review in comparison with the new GeForce GTX 260 card (with 216 stream processors) from Leadtek.
The graphics card from Palit Microsystems whose products are known for moderate pricing but not-always-immaculate quality comes in a pretty cardboard box. The face side of the box shows a dangerous-looking armed and armored frog and has two stickers that inform you about the card’s dual fan and dual BIOS.
You can also find information about the amount of graphics memory and the two frequencies it can be clocked at. The key features of the card are listed at the back of the box.
The accessories to the card are scanty including one HDMI→DVI adapter, a power cable, a CD with Catalyst, and an installation guide.
You may be surprised that this rather expensive product comes without a copy of a game or some original accessory, but that’s quite typical of Palit’s graphics cards.
The Palit Radeon HD 4870 Sonic 1024MB is designed in a unique way. Let’s take a look at it.
So, this is surely not the reference cooling system. I will discuss it in detail shortly. Note that this is a dual-slot cooler, too. It has become quite normal already for a modern graphics card to occupy two expansion slots.
The PCB has become 14mm shorter and 4mm thicker than the reference Radeon HD 4870 and now measures 230x99x39mm. The reduced length of the card should be appreciated by its users because many top-end cards are just too long to fit into small system cases.
The card’s mounting bracket has an original design, too. It carries analog and DVI video outputs, HDMI and DisplayPort interfaces, and a small switch for changing the card’s operation mode from Normal to Turbo and back again.
There is also a vent grid for exhausting the hot air out of the system case even though the cooler’s fans are oriented towards the mainboard rather than in parallel to it as in the reference cooler.
The cooler’s plastic casing can be removed easily. Now we can take a look at the cooler’s heatsink and the face side of the card’s PCB.
The heatsink is fastened with four screws to an X-shaped backplate you can remove easily.
The four-phase voltage regulator looks like this:
The power circuit differs considerably from the reference Radeon HD 4870 although the two 6-pin power connectors have been left in their places (but they are oriented upward here). The Radeon HD 4870 specifications declare a peak power draw of 170 watts, so a 500W power supply is recommended for a computer with such a card. A 600W or higher PSU is recommended for a CrossFire configuration.
A small aluminum heatsink is installed via a thin thermal pad on the card’s power components.
There was a surprisingly thin layer of gray and thick thermal grease on the GPU. I removed it to examine the latter.
So, this sample of Palit Radeon HD 4870 Sonic employs a 55nm RV770 processor manufactured in Taiwan on the 32nd week of this year. The GPU has 800 unified shader processors, 40 texture processors and 16 render back-ends. Now, when the switch on the card’s mounting bracket is set at Normal, the GPU frequency is 750MHz in 3D mode, which matches the Radeon HD 4870 specs. But when you set the switch at Turbo and reboot your computer, the GPU frequency grows up to 775MHz. That’s a small but nice bonus. The GPU frequency is reduced to 500MHz in 2D mode, irrespective of the position of the switch, to save power and lower the temperature of the graphics core.
The key difference of the Palit Radeon HD 4870 Sonic from the reference card and competitor’s versions is that it has twice the standard amount of memory. Its 1024 gigabytes of memory are provided by eight GDDR5 chips from Qimonda manufactured on the 34th week of 2008.
These Revision A1 chips are marked as IDGV1G-05A1F1C-40X. The specified access time is 1 nanosecond. The theoretical effective clock rate is 4000MHz. As opposed to the reference Radeon HD 4870, the memory frequency is 3800MHz in the Normal mode (200MHz higher than on the reference card) and grows up to 4000MHz in the Turbo mode (+400MHz). Again, you have to reboot your system after you change the operation mode (this is necessary for the card to switch to another BIOS).
The following screenshot shows the card’s characteristics in the Turbo mode:
Now let’s check out the cooling system. It is based on an aluminum heatsink consisting of thin ribs that hang on three 6mm nickel-plated copper heat pipes. The pipes are soldered to the copper plate of the cooler’s sole which is about 2 millimeters thick.
This heatsink contacts with the GPU only and is cooled with two fans installed in a plastic frame.
Note that the fans are different. The fan that cools the heatsinks right above the GPU has a diameter of 80 millimeters and nine blades. The second fan has a diameter of 70 millimeters and 11 slim blades. It cools the heatsink above the card’s power circuit and the power components themselves. Both fans are made by EverFlow and have the same speed (PWM-regulated).
The card’s temperature was measured in a simple test. I loaded it by running the Firefly Forest test from 3DMark06 at 1920x1200 with 16x anisotropic filtering for 10 times. I didn’t enable FSAA as the GPU load and temperature would have been lower then. The test was performed in a closed ASUS Ascot 6AR2-B system case (its fan configuration is described below in the Testbed and Methods section). The ambient temperature was 23.5°C. The card’s frequencies and temperature were monitored with RivaTuner 2.11. As I had dismantled the card before testing it, I replaced the thermal interface of the GPU with a thin layer of high-efficiency Gelid GCI thermal grease.
So, here are the results of the test with the card working in the Normal mode with automatic fan speed management.
As you can see, the GPU temperature is low in comparison with the reference cooler working in automatic mode. Take note that the fan speed is never higher than 1240rpm during the test! Of course, the card remains very quiet. You just can’t hear it against the noise from a quiet system case. Unfortunately, the sensors of the card’s power elements do not work (or are not installed) on the Palit Radeon HD 4870 Sonic. It would be interesting to know their temperature, too.
And now let’s see how effective the cooler is at the highest speed of the fans.
Well, I can just applaud Palit for equipping its graphics card with such an effective cooler. I should confess, however, that you may only want to run the card at the highest fan speed during a short benchmarking session or something because the level of noise is rather high then.
As opposed to the reference cooler which exhausts the hot air out of the system case, the cooler of the Palit card leaves most of the air inside the system case. Considering the high heat dissipation of the Radeon HD 4870, this will increase the overall temperature inside your computer.
So, I decided to set a subjectively comfortable speed of the fan, 1770rpm, and performed my overclockability test then. I found my sample of Palit Radeon HD 4870 Sonic to be stable and free from visual artifacts at a GPU frequency of 820MHz and an effective memory frequency of 4152MHz.
Frankly speaking, the core proved to have low overclocking potential. It didn’t reach 840-850MHz I had achieved with two Radeon HD 4870 cards from HIS. The memory chips overclocked better, but I think the memory frequency has but a small effect on the performance of a Radeon HD 4870.
Here is the card’s temperature at the overclocked frequencies:
After I had found the maximum frequencies of the card, I replaced its cooler with an Arctic Cooling Accelero Twin Turbo. And I found that under load the default cooler of the Palit Radeon HD 4870 Sonic was only 3°C inferior to one of the best coolers available (the fan speed of the Twin Turbo was set at 1770rpm, too). This test is yet another proof of the highest efficiency of Palit’s cooler.
One of the card’s BIOSes can be downloaded here (a 44.8KB WinRAR file). This card costs somewhat higher in retail than the recommended price of the Radeon HD 4870 512MB ($299), but I guess the extra 512 megabytes of graphics memory, the efficient cooler, the increased frequencies and dual BIOS are worth the difference.
The newest card from Leadtek Research is shipped in a small cardboard box covered with a layer of original lacquer that shimmers in many colors at different angles of view.
Besides the name of the card, Leadtek WinFast GeForce GTX 260 Extreme+, the front of the box reports key features of the product such as bus type, amount and type of graphics memory, support for DirectX 10 and HDCP. It also tells you that a copy of Neverwinter Nights 2 is included into the box. The card is claimed to have a quiet cooling system and this is not just a marketing trick as I’ll explain below.
The back of the box provides a lot of information about the graphics card, the GPU and technologies implemented in it. System requirements and included accessories are also listed there.
The accessories are (from left to right and from top to bottom):
The new card looks exactly like earlier GeForce GTX 260 because nearly all the changes are limited to the GPU.
To remind you, the PCB of GeForce GTX 260 and 280 series cards is encased within a dual-slot cooler:
The new GeForce GTX 260 measures 270x100x32mm, just as the old one. The graphics card has a PCI Express x16 interface, two dual-link DVI ports and an S-Video output. The card’s mounting bracket also has a vent grid to exhaust the hot air out of the system case.
In the top part of the PCB there is a connector for an S/PDIF cable, two 6-pin connectors for additional power, and two MIO connectors for building SLI and 3-way SLI configurations out of two or three identical graphics cards with Nvidia’s GPUs.
Notwithstanding the less cut-down GPU, the peak power draw of the new GeForce GTX 260 has not changed. It is declared to be 182W. Like before, a 500W power supply is recommended for a computer with such a card installed.
Without the cooling system the card looks like this:
I scrutinized the PCBs of an old and the new GeForce GTX 260 but could not spot any difference, particularly in the power sections.
The GPU of the new GeForce GTX 260 can be identified by the second number in the last line of the marking. It used to read “G200-100-А2,” but now it reads “G200-103-A2.”
Manufactured on the 29th week of this year on Taiwan, the GPU is revision 2. Compared with the GPU of the older GeForce GTX 260, the GPU of the new GTX 260 has 216 rather than 192 shader processors and 72 rather than 64 texture processors. The number of render back-ends has remained the same at 28. The specified GPU frequencies of the new GeForce GTX 260 are 575/1242MHz like those of the older version. However, Leadtek pre-overclocked its version of the card, adding “Extreme+” to the product name. Thus, the default frequencies of the WinFast GeForce GTX 260 Extreme+ are 602/1296MHz. In other words, the card’s GPU is somewhat lacking in shader and texture processors in comparison with the GeForce GTX 280, but has the same frequencies. The card’s GPU frequencies are automatically reduced to 300/600MHz in 2D applications to save power and to lower the GPU’s heat dissipation.
The graphics card is equipped with 896 megabytes of GDDR3 memory. The memory bus is 448 bits wide. These parameters have not changed in comparison with the older version of GTX 260. Leadtek’s card carries memory manufactured by Hynix.
Marked as H5RS5223CFR NOC, the chips have an access time of 1 nanosecond which corresponds to an effective frequency of 2000MHz. The GeForce GTX 260 specs have a memory frequency of 1998MHz but it is increased to 2214MHz on the WinFast GeForce GTX 260 Extreme+. This is not the highest frequency among pre-overclocked GeForce GTX 260, yet quite high anyway.
Thus, the new graphics card has the following parameters:
The cooling system is the same as the reference cooler of GeForce GTX 280/260 cards.
It consists of a copper base, five copper heat pipes (6mm in diameter), thin-ribbed aluminum heatsink, aluminum frame and a blower that creates airflow through the ribbing. An aluminum cap contacts with the memory chips on the reverse side of the PCB via soaked thermal pads. The GPU’s heat-spreader contacts with the cooler’s copper base via a thick layer of gray thermal grease.
If the cooling system has remained unchanged, how can Leadtek claim that it is exceptionally quiet on its new card? Well, let’s check out the card’s temperature at the frequencies which correspond to the clock rates of the reference GeForce GTX 260:
Take note that the cooler’s fan did not accelerate to its maximum 3300rpm but was working at a rather quiet 1700rpm. The temperatures of the GPU, power elements and PCB were not higher than 76°C, 79°C and 63°C, respectively, which was quite low for a graphics card of this class.
I then decided to compare the temperature and the fan speed of the Leadtek card with an ordinary GeForce GTX 260 (192 stream processors) from BFG and was surprised to find that the latter was as hot as 84°C, 66°C and 58°C, respectively, under the same conditions and with the same thermal interface (Gelid GC1).
That is, the GPU of the BFG card was 8°C hotter while the power circuit and the PCB at large were 13°C and 5°C cooler. It is easy to explain the difference in the latter two parameters. The BFG card’s cooler worked at 3200rpm in automatic mode, which was far noisier than the Leadtek’s 1700rpm. But why was the GPU of the older GeForce GTX 260 hotter than the GPU of the newer one?
I found the answer when I analyzed the BIOSes of the cards. The BFG card proved to have a core voltage of 1.12V under load whereas the Leadtek had a core voltage of only 1.06V. This seemingly insignificant difference affected the results. I want to also note that at the same frequencies the peak current is not higher than 53A on the Leadtek card but as high as 62A on the BFG card. I am not absolutely sure, but considering the above-mentioned facts and the new marking of the chip, it is possible that the new card’s GPU is manufactured on 55nm tech process. I will return to this topic again when I will be measuring power consumption of the cards.
And now you can see how effectively cooled the Leadtek WinFast GeForce GTX 260 Extreme+ is at the maximum speed of the cooler’s fan.
As you can see, the graphics card is very cool, especially for its class, at the full speed of the fan. The level of noise is high, though. You may want to run the card in the automatic fan management mode instead as its temperature is rather low then.
Considering the high efficiency of the reference cooler and the low level of noise, I checked the card’s overclockability without installing an alternative cooler. Notwithstanding the pre-overclocked frequencies, the card managed to speed up to 663/1404/2448MHz. It was stable at these frequencies and showed no visual artifacts.
That’s not a record-breaking result, but the graphics card’s temperature proved to be rather low in the overclocked mode:
The fan speed of the reference cooler is set manually at 2160rpm. It is the upper limit of the comfortable noise range.
The recommended price of the Leadtek WinFast GeForce GTX 260 Extreme+ is rather high at $390. You can download the card’s BIOS here (a 41.3KB WinRAR).
The following table lists the specifications of the reviewed graphics cards in comparison with the GeForce GTX 280.
The power consumption of the graphics cards was measured with a multifunctional panel from Zalman called ZM-MFC2. This panel measures the overall consumption of the system (without the monitor) rather than of an individual component. The configuration of my testbed is listed in the next section. The measurement was performed in 2D (Word and Web surfing) as well as in 3D mode (a recorded demo from S.T.A.L.K.E.R.: Clear Sky running at 1920x1200 without FSAA but with 16x anisotropic filtering).
I want to note one thing about Zalman’s monitoring panel. I had used to power my testbed by a 1000W Enermax Galaxy DXX power supply (EGA1000EWL). After I replaced it with a 1500W Thermaltake Toughpower (W0218), Zalman’s panel began to report a 25W lower consumption for the system than before. I don’t have more accurate tools for measuring power consumption and I can’t tell you how exact the numbers are (the measurement conditions were identical, of course). Anyway, this panel is quite good for the purpose of comparison.
To differentiate between graphics cards that have the identical name of GeForce GTX 260 and to avoid adding more symbols into the overpopulated diagrams, I will be referring to the new GeForce GTX 260 as GeForce GTX 266 because it has 216 shader processors. So, the power consumption results are listed in the next diagram:
It is interesting to compare the new GeForce GTX 260 with the older one. I have already written about the reduced heat dissipation of the newer version. Here, we can also see that at identical frequencies the system with the newer GeForce GTX 260 requires almost 30W less power than the system with the older GeForce GTX 260. Moreover, the system with the new GTX 260 that has 216 shader processors and is clocked at its default frequencies now consumes less power than the same system with a Radeon HD 4870. The same goes for 2D mode where the graphics cards are loaded less and work at reduced frequencies.
So, this comparison is in favor of the GeForce GTX 260. I can only add that there is but a negligible difference in power consumption of the system with the Radeon HD 4870 cards that have 512 and 1024 megabytes of graphics memory.
The graphics cards were benchmarked in a closed system case with the following configuration:
To minimize the CPU’s influence on the graphics cards’ performance I overclocked the CPU to 4.00GHz at 1.575V voltage during the tests.
The system memory worked at a frequency of 1066MHz with 5-4-4-12 timings (Performance Level = 6) and 2.30V voltage.
The tests were run under Windows Vista Ultimate Edition x86 SP1. I used the latest drivers available at the moment of my tests (some screenshots were made before or after the tests and may show different driver versions):
The graphics card drivers were set at High Quality. Thus, all the optimizations available in the ForceWare and Catalyst drivers were disabled, save for the Catalyst A.I. option that was left at its default value (Standard). I turned full-screen antialiasing and anisotropic filtering on from the menu of each game. If the game didn’t provide such options, I enabled FSAA and AF from the control panels of ForceWare and Catalyst. The Transparency Antialiasing (multisampling) option was turned on in ForceWare and Adaptive Antialiasing (multisampling) was turned on in ATI Catalyst.
The graphics cards were tested at two resolutions: 1280x1024 and 1920x1200. I didn’t test them at 1680x1050 to save time.
The cards were benchmarked in the following games and applications:
I added the results of three more cards into this review for the sake of comparison. First, it is the HIS Radeon HD 4870 512MB at the default (750/3600MHz) and overclocked (845/3920MHz) frequencies. Comparing this card to the Palit Radeon HD 4870 1024MB at the same frequencies will show us if the Radeon HD 4870 can get any benefits from having twice the standard amount of graphics memory. Second, the pre-overclocked Leadtek GeForce GTX 260 will be compared with a BFG GeForce GTX 260 896MB so that we could see the difference in performance of the two versions of the same card at the same frequencies. Third, an XFX GeForce GTX 280 1024MB will help us see if the new GeForce GTX 260 is much slower than Nvidia’s flagship product. By the way, the XFX card comes with a full version of Far Cry 2. I mention this fact because you usually get obsolete games together with graphics cards.
The graphics cards are listed in the diagrams in the order of ascending price. The results of Radeon cards are marked with red while the results of GeForce cards are marked with teal. Thus, the graphics cards are listed like follows in the tests:
First go the two synthetic benchmarks.
There are no surprises in 3DMark06. There is no difference between the two Radeon HD 4870 when no image-enhancing methods are used. But when full-screen antialiasing and anisotropic filtering are turned on, the card with the double amount of memory goes ahead. It doesn’t overtake the GeForce GTX 260/216SP, though. The latter is also 3-4% faster than the ordinary GeForce GTX 260. So, there is nothing particularly interesting about these results.
The newer version of 3DMark agrees with the previous one. The Radeon HD 4870 1024MB is but slightly faster than the reference Radeon HD 4870 512MB. Both cards are somewhat slower than their opponents from Nvidia. The diagram indicates that Nvidia’s solutions show better scalability with the growth of the frequencies than AMD’s cards.
This technical demo suggests that the Radeon HD 4870 are faster than their opponents. The top-end GeForce GTX 280 is somewhat competitive to the leaders at 1280x1024 without FSAA and AF, but loses in the higher-quality mode. Both versions of GeForce GTX 260 are on the losing side here. Note that there is no difference between the two versions of Radeon HD 4870 with different amount of graphics memory. The new GeForce 260 (with 216 streamed processors) enjoys a solid advantage (up to 20%) over the older GeForce GTX 260 (with 192 streamed processors).
The benchmark integrated into World in Conflict shows that the GeForce GTX 260/280 cards are somewhat faster than the Radeon HD 4870 cards at low resolutions and easy display modes. As for the two versions of Radeon HD 4870, the card with the double amount of graphics memory enjoys certain advantage. Moreover, it overtakes the GeForce GTX 260/216SP at 1920x1200 with FSAA and AF when working at the default frequencies. The latter is in its turn ahead of the ordinary GeForce GTX 260 by 4-8%.
You can’t feel the difference between the two versions of Radeon HD 4870 at such a high frame rate, yet this difference is present. The same goes for the two versions of GeForce GTX 260. Then, you can see that the results of the cards in the low-quality mode (i.e. without FSAA and AF) are similar to those we have seen in the previous benchmarks. But as soon as 8xFSAA and anisotropic filtering are turned on, the Radeon HD 4870 leave no chance to their opponents, including the top-end GeForce GTX 280. That’s interesting because I didn’t see such an advantage before when I used the older drivers and version 1.4 of the game.
This game draws a picture of performance that is quite different from what we’ve seen in Enemy Territory: Quake Wars. Here, the two Radeon HD 4870 cards are good in the low-quality display mode, differing but little from each other. But the GeForce GTX 260/280 cards are ahead in the high-quality mode. The new GeForce GTX 260 is 4-5% faster than the previous version of the same card.
The still rather popular shooter Unreal Tournament 3 shows a difference between the 512MB and 1024MB versions of Radeon HD 4870 in those modes where the requirements to the amount of memory are high. For example, the 1024MB version is 4% faster in the low-quality mode and at 1024x768, but as much as 13% faster at 1920x1024.
Nvidia’s solutions are somewhat faster than the Radeons in this game. The new GeForce GTX 260/216 is 3-5% ahead of the older version and 6-15% slower than the GeForce GTX 280 depending on quality mode and resolution.
The new GeForce GTX 260 enjoys a record advantage over the older version of the card in this test (in one quality mode only, though). Otherwise, the results do not differ much from what we’ve seen in the previous tests.
The first impression from benchmarking graphics cards in S.T.A.L.K.E.R.: Clear Sky is that they deliver a very low frame-rate at the maximum graphics quality settings. Alas, none of the tested cards can provide a comfortable speed. Anyway, we can see a difference between the two Radeon HD 4870 (the 1024MB version is faster, of course) and ATI’s victory over Nvidia. The GeForce GTX 280 is only competitive to the cheaper Radeon HD 4870. The GeForce GTX 260 cards are slower and compete between each other: the newer version with 216 unified shader processors is 7-8% faster than the older version.
This is yet another shooter none of the tested cards can provide a playable frame rate in. As opposed to S.T.A.L.K.E.R.: Clear Sky, Nvidia’s solutions are superior in Crysis Warhead. The 1024MB version of Radeon HD 4870 is occasionally as fast as the older GeForce GTX 260 but the latter is the overall winner, anyway. The two Radeon HD 4870 cards differ by 9-17% in this test, which is quite a lot. As for the two versions of GeForce GTX 260, the newer card is 2-13% faster depending on quality mode and resolution.
It is good that the graphics cards deliver higher frame rates in this new game than in the previous two tests. Then, the bottom speed is rather high in every mode and at every resolution, which means that the gameplay is going to be smooth. The Radeon HD 4870 1024MB is up to 11% faster than the reference version of the card. And the GeForce GTX 260/216SP is ahead of the GeForce GTX 260 with 192 streamed processors by 3-5%. The gaps between the competing cards are rather small in Far Cry 2.
First of all, I want to show you the comparison of the Radeon HD 4870 with 1024 megabytes of graphics memory with the reference Radeon HD 4870 512MB at the same frequencies of 750/3600MHz. The reference card is taken as the baseline in the diagram. An increase or reduction of performance is shown relative to it.
As you can see, the extra 512 megabytes of graphics memory are not always a good thing. The 1024MB version shows best results in such resource-consuming games as World in Conflict, S.T.A.L.K.E.R.: Clear Sky, Crysis Warhead, Far Cry 2 and in the graphics tests from 3DMark 2006. 3DMark Vantage, Unigine Tropics, Call of Duty 4 and Devil May Cry are largely indifferent to the increased amount of memory. The 1024MB card is 3.6% faster on average across all the tests (the geometric mean was calculated) at 1280x1024 and 5.1% faster at 1920x1200. When FSAA was in use, the 1024MB version was 7% and 5.8% faster, respectively, than the 512MB version.
The next summary diagram shows the advantage of the new GeForce GTX 260 with 216 unified shader processors over the older version of the same card that had 192 such processors. The cards were compared at the default frequencies of 575/1242/1998MHz.
The performance growth is inexplicably high in Unigine Tropics Demo and the simple mode of Devil May Cry although I verified the results twice. Overall, the new GeForce GTX 260/216Sp is 4.6-6.9% faster than the older one depending on resolution and display quality mode.
The third diagram will show you how slower the GeForce GTX 260/216SP is in comparison with the top-end GeForce GTX 280.
So, the GeForce GTX 260/216SP is slower than the GeForce GTX 280 by 10-13%. To remind you, the GeForce GTX 260 with 192 unified shader processors used to be 15-18% slower than the GeForce GTX 280. Thus, the gap from the top-end card has diminished.
And finally, let’s compare the performance of the GeForce GTX 260/216SP 896MB at the default frequencies of 575/1242/1998MHz with the Radeon HD 4870 1024MB clocked at 750/3600MHz. The RV770-based card is the baseline in the diagram.
It is hard to prefer one of these cards as they win the tests alternately. I wouldn’t choose any of them without specifying the game, quality, resolution you want to play at and other factors.
Summing up this review, it must be noted that the Radeon HD 4870 and GeForce GTX 260 have both become stronger indeed. The increased amount of memory of the former card has a positive effect on its performance in resource-consuming games, at high resolutions and in high display quality modes. When the conditions are the opposite, the extra 512 megabytes of memory do not provide substantial benefits. As for the latter card, its GPU has got more advanced in comparison with the first version of GeForce GTX 260 and brings about higher performance, lower power consumption and heat dissipation, and a lower level of noise.
Choosing between these two cards, you should keep it in mind that GeForce GTX 260 series cards now cost $30 more than Radeon HD 4870 series products even though the GeForces have declined in price at a more rapid rate than the Radeons. The arrival of the improved version of the GTX 260 will be impeded by the older version as Nvidia did not differentiate the new and old versions in terms of model name and recommended price, confusing the customer. At the same time, Radeon HD 4870 with different amounts of memory live quite well on the market together and differ in price by $15-50 – there is no confusion for the potential customer. I hope you will be able to make an informed and unconfused choice after having read this review.
As for the particular graphics cards reviewed today, the Palit Radeon HD 4870 Sonic 1024 is good in every respect. The double amount of memory, competitive price, efficient and quiet cooler, dual BIOS and two operation modes of the card can’t leave you indifferent. Its GPU didn’t do well in the overclockability test but that’s just a problem of the particular sample. Our recent experience shows that the quality and service life of this brand’s products have most likely improved.
The Leadtek WinFast GeForce GTX 260 Extreme+ is appealing with its pretty packaging as well as pre-overclocked frequencies and reinforced core. It would make a bestseller if it price were $50-70 lower. I hope the price cut predictions about Nvidia’s products will come true indeed.