by Alexey Stepin , Yaroslav Lyssenko
05/22/2008 | 10:51 AM
The era of identical graphics cards based on the reference designs seems to have come to an end. The unification principle is still upheld by Nvidia’s flagship GeForce 9 models yet the senior GeForce 9800 GX2 is the only card to have a chance to remain unchanged because it represents a very complex design. It is indeed only a chance because we wouldn’t be really surprised to see a unique GeForce 9800 GX2 as well. Few companies would dare release it but such majors as, for example, ASUS have enough resources and experience to experiment. We can recall the triple-processor ASUS EAH3850 Trinity we’ve recently mentioned in our newswire.
Still, even though ATI’s inexpensive RV670-based solutions have become widespread and popular among gamers due to their appealing price/performance ratio, there are but few interesting versions of ATI Radeon HD 3870 and HD 3850 available. Every model we have tested in our labs so far was a copy of the reference card manufactured by the graphics card vendor or at ATI’s contracted facilities. Occasionally they were pre-overclocked as ASUS’s EAH3870 TOP and EAH3850 TOP but most often the stickers on the cooler made the only difference. Without a special feature such cards could not provoke a genuine interest among hardware testers and enthusiasts.
Gigabyte United is among the first companies to take a creative approach with respect to its versions of ATI Radeon HD 3870 and HD 3850 cards.
Gigabyte’s people were not ashamed to ask obvious questions such as why does the Radeon HD 3870 need GDDR4 memory which seems to be more expensive than the widespread GDDR3 type while being not much better in terms of frequency potential? And can an inexpensive version of the ATI Radeon HD 3870 be created that would use a new unified design without GDDR4 and would not be inferior in terms of quality and performance? These questions got a positive answer and Gigabyte has released a series of RV670-based graphics cards that are substantially different from the ordinary reference copies of Radeon HD 3870 and HD 3850.
This review is about the junior and senior models of the series, GV-RX385256H and GV-RX387512H. We’ll see what’s so special about them and how they compare with the corresponding reference samples.
Both cards are shipped in identical boxes differing with the model number, HD 3850 and HD 3870 respectively:
The boxes are not very large. Their decoration is based on the popular role-play game Neverwinter Nights 2 which is included with both products. The face side of each box shows a picture of Amon Jerro, a character from the game. The gloomy design is quite appealing in a special way.
The basic specs of the product are also shown on the box. You can learn that the cards belong to the Ultra Durable 2 series featuring low-ESR solid-state capacitors, ferrite-core inductors and power transistors with a low RDS(on) rating. The combination of these components helps lower the heat dissipation of the power circuit and increase its efficiency as well as overall reliability.
The graphics cards lie in foam-rubber trays that protect them against possible damage during transportation and storage. Both cards come with the same set of accessories:
The accessories are good for mainstream products. We can only gripe about the lack of a DVI-I → HDMI adapter and HD content player.
ATI Radeon HD 3000 series cards feature a best-in-industry hardware video processor and are often bought for use in a home multimedia system with HD video support. That’s why the lack of the mentioned accessories is a drawback really. You’ll have to spend an additional $50-100 in order to use your GV-RX385256H or GV-RX387512H for watching HD video on a large screen.
Fortunately, the lack of a HD video player is the single problem we can see here. Otherwise, the packaging and accessories of the Gigabyte GV-RX385256H and GV-RX387512H deserve our praise, particularly for the CrossFire connector and the copy of an interesting game.
The cards from Gigabyte both use a unified PCB design that is different from the reference Radeon HD 3870 and HD 3850. The azure color of the PCB is one difference that strikes the eye immediately. ATI Radeon cards are usually red. Both cards being identical, we’ll describe the senior model GV-RX387512H.
Like ATI’s reference samples, the Gigabyte cards use a two-phase GPU power circuit with four power transistors in each channel. As we’ve already noted, Gigabyte installed low-RDS(on) transistors on these cards for lower heat dissipation of the power circuit. Coupled with the ferrite-core inductors, this also increases the efficiency of the circuit. As for solid-state capacitors that feature a longer service life, they are not something extraordinary. You can often find them on top-end enthusiasts-targeted mainboards.
A Nexsem NX2420 is the controller of the GPU power circuit. It is a compact counterpart of the NX2416 PWM-controller we have seen before. A single-phase voltage regulator with a Nexsem NX2124A controller is responsible for the memory chips. This simple circuit is quite okay because the power consumption of the 55nm RV670 core is rather low. However, both cards from Gigabyte have an additional 6-pin PCI Express 1.0 connector with a load capacity of 75W. There is a seat for a standard 4-pin fan connector in the bottom right corner but the connector itself is missing because the cards are equipped with Zalman’s coolers. Instead, there is a 2-pin header located at the opposite edge of the PCB, near the bottom DVI port.
The two cards are identical on the left. Their wiring is the same as the left part of the PCB of the reference Radeon DH 3850 because both of them use GDDR3 memory. Each card has 8 chips of memory. The GV-RX387512H is equipped with Samsung’s K4J52324QE-BJ1A chips with a capacity of 512Mbit (16Mbit x 32) and capable of working at 1000 (2000) MHz. This is indeed the frequency they are clocked at on this card. For comparison, the original Radeon HD 3870 carries Samsung’s K4U52324QE-BC08 chips of GDDR4 memory clocked at 1125 (2250) MHz.
The junior model, GV-RX385256H, uses memory from Hynix. The HY5RS573225BFP-14 chips have a capacity of 256Mbit (8Mbit x 32), a voltage of 1.8V and a rated frequency of 700 (1400) MHz. The chips are clocked at their rated frequency on this card.
We don’t quite comprehend why Gigabyte equipped the GV-RX385256H with such slow memory while the reference Radeon HD 3850 has a memory frequency of 830 (1660) MHz. We doubt the Hynix chips will be any good at overclocking, which means the GV-RX385256H is going to be inferior to the Radeon HD 3850 in games. The GPU may save the day to some extent if overclocked well enough.
Both cards carry RV670 chips with hard-to-understand markings typical of ATI’s GPUs.
The only useful information you can learn is the manufacturing date. Here, it is the third week of the current year (January 13-19). The GPU has its full configuration on each card: 64 superscalar execution modules with 5 ALUs in each (a total of 320 ALUs), 4 texture processors roughly equivalent to 16 conventional TMUs, and 4 raster back-ends equivalent to 16 ROPs.
The execution modules support DirectX 10.1 extensions. Nvidia’s GPUs don’t offer this support as yet. The superscalar architecture of the RV670’s execution modules is known to be its bottleneck because it requires meticulous software optimizations for all of the ALUs in the 64 shader processors to be utilized effectively. This GPU also has low performance when filtering textures because it has only 1 filter unit per 2 address units.
The GPU clock rates coincide with those of the reference cards: 775MHz for the senior model and 670MHz for the junior model. You should be aware that despite the single declared core frequency (without the division into domains as in Nvidia’s GPUs) the RV670 actually has a number of internal frequency zones. This allows to control the power consumption of the core flexibly by means of ATI PowerPlay technology.
Both cards can be used in CrossFire configurations with two, three or four GPUs, so they have two CrossFire connectors each. A CrossFire subsystem is not necessarily symmetric. The cards can work together with a Radeon HD 3870 X2 but the amount of graphics memory available for 3D applications will equal the smallest amount of onboard memory among the graphics cards in the configuration. In other words, there is little practical sense in joining a GV-RX385256H with a Radeon HD 3870 X2 or GV-RX387512H because the amount of graphics memory will be limited to 256 megabytes which will make a bottleneck for modern games that often require more memory.
Each card from Gigabyte has two dual-link DVI-I ports and a universal 7-pin analog video output that supports RCA, S-Video and YPbPr formats. The RV670 core incorporates an audio controller and supports audio-over-HDMI but the kit doesn’t contain the adapters necessary to use this feature. Besides, the audio core driver doesn’t support Dolby TrueHD and DTS-HD formats – the PC is yet inferior to Blu-ray players in this respect. The connectors are all covered with protective gags.
Each of the described cards has a non-standard cooling system which is a simplified version of the popular VF700-AlCu from Zalman.
This system was used by Gigabyte before, e.g. on the GV-NX7795T512H-RH model. A junior model in Zalman’s GPU cooler line-up, the VF700-AlCu is itself a variation of the classic design of Zalman’s CNPS7x00 series of CPU coolers with a radial placement of the ribs. Gigabyte’s version differs from the original VF700-AlCu as it has a two-wire fan connection and a simplified fastening mechanism without a back-plate. Besides, there are no heatsinks on the memory chips. The fan’s ball bearings mean a higher level of noise but a longer service life than sleeve bearings. The heatsink is fastened to the card with two brackets screwed to the two metallic poles installed on the PCB. The fastening doesn’t look secure to us and we don’t recommend you to hold the card by the cooler to avoid damaging the GPU die.
The Zalman VF700-AlCu used to cope easily with an Nvidia GeForce 7950 GT, so we have no reason to think it won’t do its job on the Radeon HD 3850/HD 3870. The only potential drawback is the high level of noise you cannot diminish by adjusting the fan speed. We’ll check this out in the next section.
We already published information about the power consumption of ATI’s Radeon HD 3870 and 3850 cards in our reviews but we think it proper to publish it again to compare the parameters of RV670-based cards with their G94-based competitors from Nvidia. Our tests showed no difference in the power consumption of Gigabyte’s cards and the reference versions of Radeon HD 3870 and 3850.
Click to enlarge
The junior GV-RX385256H is competitive against the Nvidia GeForce 9600 GT in peak power consumption while the senior GV-RX387512H needs considerably more. But thanks to PowerPlay technology both cards are superior to Nvidia’s solution under low loads.
Both Gigabyte cards are equipped with the Zalman VF700-AlCu cooler and their noise parameters should be identical. We checked this out with a digital sound-level meter Velleman DVM1326 using A-curve weighing. The level of ambient noise in our lab was 36dBA and the level of noise at a distance of 1 meter from the working testbed with a passively cooled graphics card inside was 43dBA. Here are the results:
Alas, the cards are not noiseless. The cooler’s fan is always working at its maximum speed irrespective of the GPU load or temperature. This noise is perfectly audible among the other components of a working PC and you cannot lower it unless you use your soldering iron. The original VF700-AlCu comes with a FanMate controller, but the fan is connected right to the PCB here. On the other hand, the noise at a distance of 1 meter is not as high as to require an immediate action to reduce it. Gigabyte’s cards are about as noisy as the reference Radeon HD 3870 and GeForce 8800 GT.
The cooler is highly efficient, though. The GPU temperature is never higher than 52°C on the GV-RX385256H and 55°C on the GV-RX387512H even under continuous load. This is a better performance than what the reference GeForce 9600 GT cooler can do (60-62°C). And as for the GeForce 8800 GT, its GPU temperature can be as high as 80°C and more. Thus, we think the Zalman cooler is okay. Its highest performance makes up for its increased level of noise.
The cooler could not but affect the cards’ overclockability. We managed to raise the GPU frequency of the GV-RX387512H card by 100MHz from the default 775MHz to 875MHz. The memory chips could be overclocked to 1130 (2260) MHz, which was quite an achievement for chips with a rated frequency of 1000 (2000) MHz. The card was stable at these clock rates and easily passed the full cycle of tests.
The GV-RX385256H had more modest results at overclocking: 820MHz GPU and 800 (1600) MHz memory. That’s not a surprise really because this card uses samples of RV670 that have not passed the frequency check for the use on ATI Radeon HD 3870. Moreover, Gigabyte installs slow memory on it, with a rated frequency of only 700 (1400) MHz. As a result, we didn’t achieve the memory frequency of the reference Radeon HD 3850, which is 830 (1660) MHz. This product can hardly be interesting to overclockers.
We found no compatibility issues between the Gigabyte cards and mainboards supporting PCI Express standards earlier than 2.0 which is due to the implementation of PCI Express 2.0 support in the ATI RV670. All RV670-based cards start up in PCI Express 1.0 mode and enable the faster data-transfer mode only after checking the bus out for its compliance with PCI Express 2.0. This is a better approach than in Nvidia’s solutions. Nvidia’s early G92-based cards would often refuse to start up on some PCI Express 1.0a mainboards.
For our performance tests of Gigabyte GV-RX385256H and GV-RX387512H we put together the following testbed:
According to our testing methodology, the drivers were set up to provide the highest possible quality of texture filtering and to minimize the effect of software optimizations used by default by both: AMD/ATI and Nvidia. Also, to ensure maximum image quality, we enabled transparent texture filtering - Adaptive Anti-Aliasing/Multi-sampling for ATI Catalyst and Antialiasing – Transparency: Multisampling for Nvidia ForceWare. As a result, our ATI and Nvidia driver settings looked as follows:
For our tests we used the following games and synthetic benchmarks:
First-Person 3D Shooters
Third-Person 3D Shooters
We selected the highest possible level of detail in each game using standard tools provided by the game itself from the gaming menu. The games configuration files weren’t modified in any way. The only exception was Enemy Territory: Quake Wars game where we disabled the built-in fps rate limitation locked at 30fps. Games supporting DirectX 10 were tested in this particular mode.
Besides Gigabyte GV-RX385256H and GV-RX387512H we have also included the following graphics accelerators to participate in our test session:
The tests were performed in the following resolutions: 1280x1024/960, 1600x1200 and 1920x1200. If the game didn’t support 16:10 display format, we set the last resolution to 1920x1440. We used “eye candy” mode everywhere, where it was possible without disabling the HDR/Shader Model 3.0/Shader Model 4.0. Namely, we ran the tests with enabled anisotropic filtering 16x as well as MSAA 4x antialiasing. We enabled them from the game’s menu. If this was not possible, we forced them using the appropriate driver settings of ATI Catalyst and Nvidia ForceWare drivers.
Performance was measured with the games’ own tools and the original demos were recorded if possible. Otherwise, the performance was measured manually with Fraps utility version 2.9.1. We measured not only the average speed, but also the minimum speed of the cards where possible.
This game doesn’t support display resolutions of 16:10 format, so we use a resolution of 1920x1440 pixels (4:3 format) instead of 1920x1200 for it.
The Gigabyte cards are considerably slower than ATI’s reference samples, especially the senior GV-RX387512H. However, their performance is satisfactory in every mode, including 1920x1440, due to the rather low system requirements of the game. On the other hand, the RV670-based cards are all inferior to the GeForce 9600 GT. This must be due to their less efficient texture processor architecture.
BioShock doesn’t support FSAA when running in Windows Vista’s DirectX 10 environment. That’s why we benchmarked the cards without FSAA.
The game runs on Unreal Engine 3 using math1ematics-heavy shader effects. That’s just the right environment for ATI’s solutions although at low resolutions the Nvidia GeForce 9600 GT is only overtaken by the overclocked GV-RX387512H. The GV-RX385256H delivers acceptable performance at resolutions up to 1600x1200/1680x1050 inclusive notwithstanding the difference from the reference sample in memory frequency.
None of the cards can provide a playable frame rate in this game at our settings. We can see the senior R670-based models with 512 megabytes of local memory are in the lead at resolutions higher than 1280x1024 pixels. As opposed to them, the Nvidia GeForce 9600 GT slows down suddenly at 1600x1200 and higher. This is due to inefficient memory management which is the common problem of all Nvidia solutions. It must have something to do with the memory controller of the G8x/G9x cores.
The GeForce 9600 GT enjoys a big lead over the other cards and allows playing comfortably at resolutions up to 1600x1200 inclusive. The rest of the cards, except for the overclocked Gigabyte GV-RX387512H, can’t ensure the desired minimum of speed even at 1280x1024.
This game is tested at the High level of detail, excepting the Shaders option which is set at Very High. This way we try to achieve a compromise between image quality and speed.
Below-$200 graphics cards cannot be used for playing Crysis at the High or better settings. Still, we can note that the GeForce 9600 GT has a considerable advantage over the Radeon HD 3870 and HD 3850, especially in terms of minimum speed, and that there is almost no effect from overclocking the latter cards at 1920x1200.
The frame rate is fixed at 30fps in this game as this is the rate at which the physical model is being updated at the server. Thus, this 30fps speed is the required minimum for playing the game.
It all depends on the display resolution here. The overclocked Gigabyte GV-RX387512H is the only card that can compete with the GeForce 9600 GT at 1280x1024 but the ATI solutions with 512MB of memory go ahead at the higher resolutions while the performance of the GeForce 9600 GT lowers nearly to the level of the Radeon HD 3850 256MB. From a practical point of view, the Radeon HD 3870 allows playing at 1920x1200 whereas the Radeon HD 3850, like the GeForce 9600 GT, can be used for playing at 1600x1200/1680x1050.
The advantage of the GeForce 9600 GT over both versions of Radeon HD 3870 lowers from 40% to 30% as the display resolution grows up. The overclocked Gigabyte GV-RX387512H is 25% to 15% behind the leader. Anyway, the senior RV670-based solutions should better be used at 1600x1200 because they don’t have any reserve for 1920x1200 – the frame rate may occasionally bottom out below comfortable level in action-heavy scenes. The junior models should be limited to 1280x1024 for the same reason.
The game doesn’t support FSAA when you enable the dynamic lighting model, but loses much of its visual appeal with the static model. This is the reason why we benchmarked the cards in S.T.A.L.K.E.R. using anisotropic filtering only.
The game engine is optimized for Nvidia’s solutions and this test has always been a trial for ATI’s single-chip cards. This time the overclocked Gigabyte GV-RX387512H is only able to deliver acceptable performance at 1280x1024. The GeForce 9600 GT feels confident at 1600x1200.
Every card is very slow in this game. We can see the GeForce 9600 GT doesn’t feel a lack of memory at high resolutions as it does in some other games. On the other hand, the results of the junior Radeon HD 3000 models indicate that 256 megabytes of graphics memory is not enough, at least at the highest graphics quality settings. By the way, you can lower the level of detail to Medium to achieve a playable speed without losing much in terms of visual quality.
The senior models of Radeon HD 3870 can be used for playing this game at resolutions up to 1600x1200 inclusive while the overclocked Gigabyte GV-RX387512H delivers a playable frame rate even at 1920x1200. The Gigabyte GV-RX385256H is about as fast as the reference Radeon HD 3850, both cards providing an acceptable speed at 1280x1024 only.
The GeForce 9600 GT and the overclocked Gigabyte GV-RX387512H have rather high average results, the former being about 10% faster, but each has a very low minimum of speed. The rest of the tested cards can’t yield 30fps even.
The game loses much of its visual appeal without HDR. Although some gamers argue that point, we think TES IV looks best with enabled FP HDR and test it in this mode.
Each of the four Radeon HD 3000 series cards provides comfortable gaming conditions in every display mode including 1920x1200 although Gigabyte’s versions are somewhat slower then the reference samples. The GeForce 9600 GT looks good, too. But at high resolutions its average performance is similar to that of the ATI Radeon HD 3850 while the minimum speed is even lower.
The new add-on to Company of Heroes is tested in DirectX 10 mode only since it provides the highest quality of the visuals.
Both versions of Radeon HD 3800 from Gigabyte are a little slower than the reference cards, by less than 10%. The senior models compete with the GeForce 9600 GT but the overall performance level is not very comfortable due to the low minimum of speed. The GV-RX387512H is 20% faster when overclocked but the minimum speed doesn’t grow up much.
The game having a frame rate limiter, you should consider the minimum speed of the cards in the first place.
The reference Radeon HD 3850 is no different from Gigabyte’s versions, both providing a playable minimum speed at 1280x1024 only. Both versions of Radeon HD 3870 cope with the job in every test mode, outperforming the GeForce 9600 GT at 1920x1200. The senior model from Gigabyte doesn’t benefit much from overclocking except for a minor increase in the minimum speed.
Formally, the GeForce 9600 GT has a huge advantage over the RV670-based cards, yet this is not enough to deliver a playable frame rate even at 1280x1024, judging by the results produced by the in-game benchmarking tools.
GV-RX387512H has the highest score. The Gigabyte GV-RX385256H scores the smallest amount of points.
The individual tests produce a different picture: the GeForce 9600 GT wins the first and third tests. In the second test, which doesn’t depend much on the texture processor performance, it is about as fast as both Radeon HD 3870 including the overclocked Gigabyte GV-RX385256H. These tests are run with enabled 4x FSAA, so we can see again that FSAA is implemented more efficiently in Nvidia’s GPUs.
The GeForce 9600 GT ranks third-bottom in this test as its G94 core cannot compare with the RV670 in terms of computing resources. The other tested cards are ranked up according to their specs. Note that both versions of Radeon HD 3870 have the same result while the Gigabyte GV-RX385256H is slower than the reference Radeon HD 3850.
The performance of the GeForce 9600 GT is the main difference between the results of the SM2.0 and SM3.0/HDR tests. In the first case this card is about as fast as the ATI Radeon HD 3870 and the overclocked Gigabyte GV-RX385256H. And in the second case, it is far slower than them, being on the same level with the ATI Radeon HD 3850.
The RV670-based cards cannot match the GeForce 9600 GT in the first test. And in the second test the overclocked card from Gigabyte is the only to rival Nvidia’s solution.
It’s completely different in the SM3.0/HDR tests that demand all of the GPU’s computing resources. Both versions of Radeon HD 3870 are in the lead and the Gigabyte card is even faster when overclocked. The GeForce 9600 GT is about as fast as the Gigabyte GV-RX385256H overclocked to 820/800 (1600) MHz. The results of the individual tests agree with the results of the groups of tests.
It’s hard to make a brief and definite conclusion about Gigabyte’s GV-RX385256H and GV-RX387512H cards. On one hand, these cards feature a unique PCB design, a high-quality power circuit, and a powerful cooler. The senior model also features a good overclocking potential. On the other hand, we don’t grasp Gigabyte’s idea of lowering the frequencies of both models. As a result, their performance is lower than that of the reference Radeon HD 3870 and 3850 by 3-7%.
When it comes to the senior model this can be explained by the inexpensive GDDR3 memory instead of the rare and expensive GDDR4 – and this frequency gap can be easily closed by overclocking. But it’s just a sabotage to install chips rated for only 700 (1400) MHz on the junior model! We could not overclock them even to the level of the reference Radeon HD 3850 and we don’t think we were unlucky. Hynix memory is just no good at overclocking in general.
Thus, the Gigabyte GV-RX385256H can only be interesting for the user if its retail price is far lower than that of the GeForce 9600 GT which is superior to it in every parameter. The advanced cooler doesn’t make much sense because this model doesn’t overclock well. And even if you overclock its GPU successfully, the memory frequency will be the bottleneck. Moreover, the dimensions of the cooling system and the high level of noise from it may prevent you from installing the GV-RX385256H into your home multimedia center where the memory frequency wouldn’t be so important. In other words, the positioning of this product is not quite clear to us. The GV-RX385256H can only be viewed as an inexpensive alternative to Nvidia GeForce 9600 GT, for instance.
The Gigabyte GV-RX387512H deserves a more optimistic verdict, we think. Having a lower price thanks to its inexpensive GDDR3 memory, this card can be overclocked to deliver good performance. It belongs to the Ultra Durable 2 series, making an appealing buy for enthusiasts. And even if you don’t overclock it, it is generally but slightly slower than the reference Radeon HD 3870. So, if it comes to retail at a comparable or lower price, this card can make a good alternative to the GeForce 9600 GT.
Besides the models discussed in this review, Gigabyte’s RV670-based line-up includes the GV-RX38512H model with 512MB of fast graphics memory but with lower frequencies than those of the senior model. If priced acceptably, it can be an even more appealing variant for affordable overclocking than the GV-RX387512H.
As for the accessories to both cards, we can see one drawback only: the lack of a DVI-I → HDMI adapter and software for playing HD content. Otherwise, the accessories are good enough for the price category these products come from. You get a nice bonus, a copy of Neverwinter Nights 2. The package design is quite nice too, although not extraordinary if compared with other manufacturers’ products.