by Alexey Stepin , Yaroslav Lyssenko, Anton Shilov
11/03/2006 | 06:07 PM
Cut-throat competition between ATI Technologies and Nvidia encouraged them not only to introduce new high-end graphics solutions every half a year, but sometimes to even completely update their entire product line-up. Of course, this state of things is very favorable for the end-users, because it ensures pretty rapid price drop for the new products as well as quick performance improvement in all market segments.
Since it is hardly profitable to develop two-three new GPUs every half a year, the leading GPU developers try to modify their GPUs so that they could last them a little bit longer. It means that each GPU will find its way into a wide variety of graphics products. So, the flagship solutions, like ATI R520 and Nvidia G71 can be found not only in$549+ graphics cards, but also in $199 solutions. However, the latter solutions can also be based on specifically developed low-cost GPUs.
When ATI and Nvidia updated their product line-up in the fall of 2006, they pursued different strategies: the former introduced new RV570 and RV560 chips, while the latter designed new products using the already existing G71 graphics processor.
Today we are going to tell you about two new graphics cards: ATI Radeon X1950 Pro (RV570) and Nvidia GeForce 7900 GS (G71) priced at $199. In our detailed review we will try to figure out which card boasts better performance and is more promising from the future prospective, the good old buddy G71 or the specifically developed newcomer from ATI.
As a matter of fact, after the announcement of the Radeon X1000 series nearly every mainstream graphics card from ATI Technologies has been built on GPUs and PCBs that were originally meant for high-performance top-end products. For example, take the Radeon X1800 GTO or the Radeon X1900 GT that are based on the R520 and R580 chips, respectively.
Using this approach, the company avoided spending money and resources on developing new components, yet there is a downside to it, too. Based on chips and PCBs from top-end products, ATI’s mainstream graphics cards had high power consumption and heat dissipation as well as a high cost of manufacture. The company’s reserve of the out-dated R520 chip was eventually depleted while using the full-featured R580 to produce mainstream graphics cards made no economical sense. Perhaps they were using defective graphics cores to make the Radeon X1900 GT, yet this didn’t solve the problem of high power consumption and heat dissipation anyway. The need to create new mainstream GPUs became apparent. The existing RV530 chip (Radeon X1600) didn’t suit that purpose due to its rather low performance.
Surely ATI Technologies was aware of all that and new GPUs were under development, but creating a new GPU from scratch is a time- and resource-consuming task, so the above-described policy just helped the company hold its ground until the new GPUs expected to replace the R520/R580 chips in midrange products were ready. As for price segments of $259 and higher, the company still feels quite confident there. The Radeon X1950 XTX and the Radeon X1900 XT 256MB are quite competitive and will remain such until the arrival of the next generation of high-performance GPUs. As of today, ATI’s and Nvidia’s products are distributed among the different price niches like follows:
The RV570 graphics processor is a chip developed specifically for mainstream graphics cards. It incorporates about 330 million transistors and is manufactured at TSMC’s facilities on new 0.08-micron tech process. The optimal (from ATI Technologies’ point of view) number of functional subunits for this product class means there are less transistors in the chip and the die area is smaller as a consequence. The use of the thinner tech process contributes to the same purpose, too. More chips can be made out of one silicon wafer, and so the manufacturing cost of the RV570 is much lower in comparison with that of the R580. Combined with the specially developed PCB, this lowers the self-cost and, accordingly, the price of the RV570-based graphics card which is called Radeon X1950 Pro. The official price of this graphics card, the only product on the new GPU as yet, is $199. This is not much at all for a product that boasts better parameters than the new version of Radeon X1900 GT.
Architecturally, the RV570 is like other chips from the Radeon X1000 family and implements a modular design in which there are no “pipelines” because the pixel shader processors and the texture-mapping units are separated and governed by an ultra-threading dispatch processor that is trying to load them up in a most optimal way.
Following ATI’s 3-to-1 concept, the RV570 has 36 pixel processors, 12 TMUs and 12 raster operators (ROPs). In this aspect, it resembles the cut-down R580 in the Radeon X1900 GT.
It should be noted that the threading dispatch processor has been simplified in the RV570 in comparison with the one in the R580. The number of simultaneously processed threads of shader code is reduced from 512 to 384, which theoretically may have a negative affect on performance of the Radeon X1950 Pro, but this theory can be confirmed or refuted by real-life tests only. For comparison, the RV530, ATI’s first chip to use the 3-math1-operations-per-1-texture-operation concept, uses an even simpler scheduler that can only control 128 threads simultaneously.
There are more transistors in the RV570 than we had expected. More than in the R520 chip. As we found out in our Radeon X1900 XTX review, pixel processors account for but a small percent of transistors in the Radeon X1000 architecture, so the transistor count hasn’t been reduced greatly. The reduction is noticeable, though, and we shouldn’t forget about the new tech process, too!
The Radeon X1950 Pro specifications are almost identical to the original Radeon X1900 GT’s specs, but the new card has one important difference. Its GPU incorporates a Compositing Engine. Until today, CrossFire technology was implemented either with the help of a special Master card that carried a separate compositing engine on board (based on a Xilinx Spartan-3 FPGA) or the data was transferred via the PCI Express bus that had a negative effect on the efficiency of the multi-GPU configuration. By integrating the compositing engine right into the GPU, ATI kills two birds with one stone: they maintain the same high efficiency of hardware CrossFire but do not have to release yet another Master card. The user doesn’t have to bother about purchasing a Master, either.
In its traditional form, CrossFire made use of an ordinary DVI link to transfer data from the Slave card to the Master. This required the Compositing Engine to incorporate two TMDS receivers and two TMDS transmitters, and an external RAMDAC. With this engine now in the GPU, the additional links in the compositing engine design are removed and data flows from one card to another directly, without conversions into DVI format and back again. Otherwise, the CrossFire concept has remained unchanged.
The graphics cards are connected with two flexible cables. As opposed to Nvidia SLI, the cables will be enclosed with the cards, most likely one cable per each sample of Radeon X1950 Pro. You anyway need two such cards to build a CrossFire subsystem. In the new version of CrossFire the cards are both identical. The Master is appointed on the software level. This depends on what mainboard slot the card is plugged into. The two CrossFire connectors may mean bi-directional data transfers or a dual-channel interface or the opportunity to build graphics subsystems with three, four, five or more GPUs. According to ATI Technologies, the bandwidth of the new CrossFire interface allows using graphics modes up to 2560x2048@60Hz which should be more than enough for resolutions like 2560x1600 without dropped frames. By the way, monitors with a resolution of 2560x2048 are as yet a prerogative of health and other specialized industries and are not produced for home or office use.
So, the technical characteristics of the new Radeon X1950 Pro are very optimistic, especially in contrast with the new version of Radeon X1900 GT that has a greatly reduced core frequency. Practice shows that the original version of Radeon X1900 GT is quite capable of challenging not only the GeForce 7900 GS but also the GeForce 7900 GT card that has more pixel processors and TMUs. We can expect the same from the Radeon X1950 Pro, but as always, we can only be sure about that after we’ve done our tests. Let’s take a closer look at the new graphics card now.
The Radeon X1950 Pro uses a specially developed PCB with a unique design that differs greatly from the PCB design used earlier in the R580-based products. Before the announcement of the Radeon X1950 Pro, the new PCB debuted on the market in the new version of Radeon X1900 GT with a reduced core frequency but an increased memory clock rate relative to the original.
The new PCB has the same dimensions as the old one, but the differences are remarkable, particularly in the right part of the card where the power circuitry is located. As opposed to the Radeon X1900 GT, the Radeon X1950 Pro’s power circuit is digitally controlled and is simplified to the utmost. It incorporates fewer elements, so there is quite a lot of free space in this part of the PCB. Using less transistors and a thinner tech process, the RV570 should boast considerably low power consumption – we’ll check out later on how exactly low it is. The additional power connector has moved from the top right corner down to the middle of the card, near its edge. This may turn to be inconvenient as you are plugging the power cable into an installed Radeon X1950 Pro. Some elements of the power circuit are installed on the reverse side of the PCB, but there are few of them there, too. The power circuit is governed by a Volterra VT1165MF controller.
In its main part, the PCB resembles the Radeon X1900 GT, but only at first sight. There are indeed no fewer differences there than in the right part with the GPU and memory voltage regulators. The memory chips are located at the same spots, but are turned around by 90 degrees. They are not screened as on the previous PCB design. The new PCB must not have been intended for memory frequencies higher than 1.4GHz. In the left top corner you can see two CrossFire connectors. They are not taller than the MIO connectors on Nvidia’s cards, but are wider and have more contacts.
The VIVO-supporting Rage Theater chip still resides in its place. It may not be very useful nowadays but there’s no harm from it, either. It can be viewed as a nice and free bonus Nvidia’s recent cards lack. The Radeon X1950 Pro has a traditional configuration of main connectors: two DVI-I and one S-Video/VIVO. The reference cooler occupies just one slot, and the card’s mounting bracket has an appropriate height.
The new graphics processor from ATI has a rather small die area. It is about 230 sq. mm. Compare this to the R580’s 350 sq. mm and to the G71’s 196 sq. mm. The RV570 still has quite a lot of transistors after all.
The die has a mirror-like surface. The marking tells that we are dealing with a sample of the RV570 chip manufactured on the 34th week of the current year, i.e. at the end of August. The die package is equipped with a metal frame that protects it from damage during installation. A distinguishing feature of the RV570 is that there are very few miniature components on the GPU package: 13 as compared with the R580’s 90.
There are eight chips of GDDR3 memory (Samsung K4J55323QG-BC14) on board, 256Mb each (8Mx32). This yields a total of 256MB of graphics memory. The chips are rated to work at 700 (1400) MHz, but they are clocked at 690 (1380) MHz on this graphics card. One BC14 chip has a max current drain of 935mA and works at 1.8V voltage, so it is easy to calculate that the memory subsystem of the Radeon X1950 Pro card has a peak consumption of only 13.5W.
Stock coolers of Radeon X1800 and X1900 graphics cards used to come under deserved criticism, from us too, due to their insufficient efficiency and high noise level. ATI Technologies has listened to the users’ complaints and equipped its new products with absolutely new coolers. The cooling system of the Radeon X1950 XTX can be called blameless, for example. The Radeon X1950 Pro comes with a new cooler, too.
Trying to make the Radeon X1950 Pro a compact device, ATI didn’t use a dual-slot cooler design. Its design has even been simplified over the earlier reference cooler. The new cooler doesn’t have any technical advances like heat pipes or something. It consists of a large copper base and a heatsink whose tubular cells spread out from the fan axis. The base has two juts with milled ribs at the rear part. These are intended for better cooling of the power elements of the power circuit. The rather small fan is connected to the PCB with a 3-pin connector. The whole design is covered with a red plastic casing to match the style of the Radeon X1950 XTX’s cooler.
Dark-gray thermal grease is used as the thermal interface between the GPU die and the heatsink base. We have no complaints about its quality, it does its job well. The memory chips contact the heatsink through elastic white pads made from some porous material. This should be sufficient considering the low level of heat dissipation of today’s GDDR3 chips.
The new cooler is unlikely to set any performance records, yet it should be somewhat more efficient than the older model due to the larger heat-spreader surface. Moreover, the RV570 should prove a much cooler chip than the R580, so there is no reason why we should be alarmed. We’ll tell you later on how the new tech process has affected the heat dissipation of the new chip. Right now we’ll introduce to you another contestant, the Nvidia GeForce 7900 GS graphics card.
When replacing the GeForce 7800 series, based on the then revolutionary G70 processor, with the GeForce 7900 family that uses the more progressive 0.09-micron G71 chip, Nvidia discovered a gap in its product line-up right in the very popular price category of $199. The gap stretched from the GeForce 7600 GT to the GeForce 7900 GT that belonged to price ranges of $149-199 and $249-299, respectively. There is indeed an abyss between these two graphics cards in terms of price as well as of technical characteristics. The former has 12 pixel shader processors, 12 TMUs and a 128-bit memory bus whereas the latter uses a 256-bit memory bus, has 24 TMUs and 24 pixel processors. A $199 solution must be somewhere in between.
In the ATI camp, this niche had been filled with the Radeon X1800 GTO which was subsequently replaced with the Radeon X1900 GT. And now the latter is going to leave room for the Radeon X1950 Pro. The GeForce 7600 GT was competitive against the first solution from the three, but it looks downright weak against the Radeon X1900 GT or Radeon X1950 Pro. After all, it belongs to a lower price niche. So, Nvidia wanted a graphics card that would be able to challenge ATI Technologies’ solutions right in the $199 sector. How did they do that?
In our review of the XFX GeForce 7600 GT XXX Edition card we found a disabled quad of pixel processors in the G73 chip. By one supposition, the disabled quad was a kind of reserve Nvidia might use to quickly respond to the rival’s releasing a new mainstream graphics card. We won’t probably know if it was so or not because Nvidia took another approach. Instead of improving upon the G73, they cut down the G71. In other words, the GeForce 7900 GS resembles the GeForce 7800 GT which had also been made out of the G70 chip with some of the pixel and vertex processors disabled.
The solution looks logical to us. The G71 is simple even in comparison with the G70, let alone the ATI R580, and is, accordingly, cheap, even though it doesn’t look too advantageous in comparison with the RV570. Moreover, Nvidia didn’t have to design a new PCB because they already had the simple and cheap PCB from the GeForce 7900 GT. That PCB suited fine for the GeForce 7900 GS with its fewer active GPU subunits. So, the new graphics card has 20 pixel pipelines and 7 vertex pipelines. That is, it was made out of the GeForce 7900 GT by disabling one quad of pixel processors and one block of vertex processors. You can learn more about the GeForce 7900 GS specs from the table at the beginning of this review.
The GPU and memory frequencies were left intact. Having such specs, the GeForce 7900 GS looks a rather fearsome rival to both versions of Radeon X1900 GT as well as to the new Radeon X1950 Pro. Let’s scrutinize this card closely now.
The photographs show quite clearly that the etalon sample of the GeForce 7900 GS card doesn’t differ visually from the GeForce 7900 GT:
The cards use identical PCBs and coolers, but the GeForce 7900 GS is a mainstream product and is not covered by Nvidia’s strict policy concerning the purchase of ready-made cards as is the case with the higher-performing GeForce 7900 GTX and GeForce 7950 GX2 models. So, we can expect to see original models of GeForce 7900 GS built on non-reference PCBs designed by the graphics card manufacturers on their own.
The main frequency of the GeForce 7900 GS GPU is 450MHz, but its vertex processors work at 470MHz. In these parameters the card doesn’t differ from its ancestor. As we said above, there are 20 active pixel processors (5 quads) and 7 active vertex processors in the GeForce 7900 GS. The other processors are disabled on the hardware level by cutting up the bridges on the die package with a laser. It means it is impossible to unblock those subunits and transform the GeForce 7900 GS into a GeForce 7900 GT. But considering the polished-off tech process, most of the chips that go into GeForce 7900 GS production are likely to be fully functional and without any defects. So, it is probable that there will appear “overclocker-friendly” versions of this graphics card in which some of the subunits are disabled on the software level and can be unblocked by using an appropriate BIOS or with RivaTuner.
An overwhelming majority of graphics cards of a high enough class tested in our labs have been equipped with Samsung’s memory, but the GeForce 7900 GS is an exception. This card carries eight Hynix HY5RS573225A GDDR3 chips, 256Mb each. The chips’ FP-14 suffix denotes the appropriate access time and a rating frequency of 700 (1400) MHz. The real frequency the memory works at on this card is the same as on the GeForce 7900 GT, namely 660 (1320) MHz, and provides a bandwidth of 42.2GB/s.
The GeForce 7900 GS is considerably better than both versions of Radeon X1900 GT in this respect, but not than the Radeon X1950 Pro. We should also keep in mind the more efficient ring-bus memory controller of ATI’s graphics cards.
The GeForce 7900 GS comes with the same cooler as the GeForce 7900 GS is equipped with. It consists of a simple copper base to which a folded sheet of copper is glued. This heatsink is topped with a small blower, about 45mm in diameter. The whole arrangement is covered with a plastic casing and is fastened to the PCB with four spring-loaded screws. The solution is not very efficient, but does its job well even on the GeForce 7900 GT considering the low heat dissipation of the G71 chip. So, the cooler won’t have any problems on the 7900 GS which has some of the subunits disabled. The memory chips are not cooled, just like on the GeForce 7900 GT.
We measured the level of noise produced by the graphics cards’ coolers with a digital sound-level meter Velleman DVM1326 (0.1dB resolution) using A-curve weighing. At the time of our tests the level of ambient noise in our lab was 36dBA and the level of noise at a distance of 1 meter from a working testbed with a passively cooled graphics card inside was 40dBA. We got the following results:
The ATI Radeon X1950 Pro boasts excellent noise characteristics. Its cooler’s fan is rather noisy by itself, but you can only hear it for the first few seconds after you turn your PC on. After that, the level of noise goes down to the measured value, 40.1dBA. This is very low, considering the background noise and the noises from the other components of the working testbed. Subjectively, the graphics card is not heard at all against the background of a working PC. We are not sure if the speed management controller on the Radeon X1950 Pro supports higher speeds because we couldn’t manage to make the card increase its fan speed even when testing it under maximum load.
The Nvidia GeForce 7900 GS behaves in a similar manner. As opposed to the GeForce 7900 GT, this card is equipped with a full-featured speed controller, so there is no problem with high noise level as before. The fan is working at its highest speed for only 1-2 seconds on your turning the PC on. Then it slows down to such a speed that the cooler becomes absolutely quiet. This is confirmed by the results of our measurements. The speed control system is set up in such a way that it has two modes: a quiet 2D and a noisier 3D. But even in the latter mode the GeForce 7900 GS has somewhat better results than the recognized leader in quietness, the GeForce 7900 GTX card.
There’s no sense in comparing the GeForce 7900 GS with the Radeon X1950 Pro because the difference between the noise characteristics of the two graphics cards is so small that it can only be caught by a measuring instrument, not by ear. So, we can say the cards are equally quiet at work.
The power consumption of the ATI Radeon X1950 Pro and the Nvidia GeForce 7900 GS was measured according to our traditional method on the following testbed:
The mainboard in this testbed is modified: measuring shunts are connected inline into the power lines of the PCI Express x16 slot. The shunts have connectors for measuring instruments. The 2xMolex → PCI Express power adapter is also equipped with such a shunt. We perform our measurements with a Velleman DVM850BL multimeter that has a measurement accuracy of 0.5%.
To create a 3D load on the graphics card, we ran the first SM3.0/HDR test from Futuremark 3DMark06 in a loop at 1600x1200 resolution with 16x anisotropic filtering. A Peak 2D load was created by launching the 2D Transparent Windows test from Futuremark PCMark05. Here are the results:
The new ATI Radeon X1950 Pro is no record-breaker in terms of power saving. At least its power consumption proved to be higher than we had anticipated. The new card consumes less than the Radeon X1900 GT but only by 9.3W under max load. The integration of the compositing engine is unlikely to have affected that. The problem must be in the rather large number of transistors (more than in the R520) and the higher, in comparison with the opponent, core frequency. Considering these factors, the result is good enough, yet we would want to see more from the 0.08-micron tech process.
The power consumption of the GeForce 7900 GS is a little lower than that of the GeForce 7900 GT, quite in accordance with the disabled processors (four pixel and one vertex). Of course, the card can’t show as superb results as the G73-based solutions do, but it is more economical than the Radeon X1800 GTO and much more so than the Radeon X1900 GT or the X1950 Pro.
The detailed picture has nothing new to tell us. As expected, the GeForce 7900 GS behaves like its elder brother, having a near zero consumption on the +3.3V line. In 3D mode, however, the external +12V line bears a somewhat higher load than the internal one, but the difference is less than 5 watts.
The Radeon X1950 Pro behaves differently. Like all modern graphics solutions from ATI Technologies, it loads the +3.3V line heavily. It also puts a higher load on the internal +12V line in 3D mode than on the external one.
So, our tests show that the GeForce 7900 GS is a more economical solution than the Radeon X1950 Pro, but the G71 GPU, although manufactured on a “thicker” tech process, is less complex and works at a lower clock rate than the ATI RV570. It is thanks to this moderate appetite that the GeForce 7900 GS can do without a massive cooling system and is quite satisfied with the compact reference cooler from Nvidia without any danger of overheat whereas the Radeon X1950 Pro has to use a rather massive cooler with well-developed ribbing.
Unfortunately, we couldn’t check out the overclocking potential of the Radeon X1950 Pro because there were no overclocking utilities that would support the RV570 chip at the time of our writing this review. ATI’s Overdrive feature in the Catalyst Control Center worked incorrectly, at least with our sample of the Radeon X1950 Pro. When we tried to increase the GPU clock rate, the frequency was stepped up so high that we got visual artifacts right away. We will try to overclock the Radeon X1950 Pro in the future as soon as we’ve got a retail version of the card and appropriate utilities.
The GeForce 7900 GS met our expectations in the overclockability tests. The reason of its success is obvious – a solid reserve of core frequency because the same G71 chip works at 650MHz on the GeForce 7900 GTX. We only set an additional 120mm fan to blow at the card’s PCB and achieved stability at 580/850 (1700) MHz frequencies. This amounts to a 30% core frequency growth and a 20% memory frequency growth. Our overclocked GeForce 7900 GS surpassed the GeForce 7900 GTX in terms of memory frequency and equals the GeForce 7800 GTX 512 whose memory was clocked at the highest frequency among all of Nvidia’s cards, 850 (1700) MHz.
As for the graphics core frequency, we didn’t modify the frequency delta of our G71 chip, so the pixel and vertex processors and the TMUs worked at 580MHz and the ROPs were clocked at 600MHz. This good overclocking achievement provides an appropriate performance growth as you will see right now in the gaming tests.
We compared ATI Radeon X1950 Pro and Nvidia GeForce 7900 GS graphics cards on the following platforms:
The drivers were set up in such a way as to provide a comparable, yet highest possible texture filtering quality.
ATI Catalyst:
Nvidia ForceWare:
We selected the highest possible graphics quality level in each game. We didn’t edit the games’ configuration files. The speed was measured using the game’s integrated tools or, if not available, by means of the Fraps program. We also measured the minimum speed of the cards where possible.
Besides the two standard resolutions of 1280x1024 and 1600x1200 pixels, we also used a resolution of 1920x1200 pixels with an aspect ratio of 16:10 in games supporting widescreen modes. We enabled FSAA and anisotropic filtering from the game’s menu. If this was not possible, we forced them using the appropriate driver settings. We decided not to run any tests with disabled FSAA, because with a $199 graphics card you can hope to get relatively high performance even with FSAA 4x enabled.
We ran the tests with disabled FSAA only for those games that do not support FSAA due to the specifics of their engine or make use of HDR (FP16). The thing is that the GeForce 7 family cannot perform FSAA along with floating-point HDR.
Besides ATI Radeon X1950 Pro and Nvidia GeForce 7900 GS we also tested the following graphics cards:
The following games and applications were used in this test session:
First-person 3D shooters:
Third-person 3D shooters:
RPG
Simulators:
Strategies:
Synthetic benchmarks:
The new member of the GeForce 7 family enters the battlefield with much confidence, providing performance comparable to that of the Radeon X1800 GTO and delivering a comfortable frame rate in all resolutions up to 1600x1200 inclusive. On the other hand, the GeForce 7900 GS is slower than the Radeon X1950 Pro, which costs the same money, at the default frequencies.
In the previous test the GeForce 7900 GS was slower than the Radeon X1950 Pro, but it easily provides the same speed in this OpenGL application. However, both cards are so slow here that you can’t really turn 4x FSAA on. The overclocking doesn’t help, either.
The Nvidia GeForce 7900 GS is ahead of the Radeon X1800 GTO in every resolution except for 1920x1200 where they are equals. When overclocked to 580/850 (1700) MHz frequencies, the GeForce 7900 GS leaves the Radeon X1950 Pro behind in the two lowest resolutions. But in the widescreen 1920x1200 resolution the ATI card delivers the same performance at its default frequencies as the overclocked GeForce 7900 GS does.
The Radeon X1950 Pro feels even better on the Research map where the overclocked GeForce 7900 GS can outperform it in 1280x1024 resolution only and by a small margin, too. The fact is the pixel shader performance of the GPU is more important here than anything else, and the Radeon X1950 Pro with its 36 pixel processors can deliver it.
The difference between the cut-down GeForce 7900 GS and the GeForce 7900 GT is not felt strong in the HDR mode. The memory subsystem performance the two cards are equals in seems to be the limiting factor here. The Radeon X1950 Pro easily beats the non-overclocked GeForce 7900 GS despite the imperfect implementation of FP16 HDR for ATI’s cards in the current version of Far Cry.
The Radeon X1800 GTO improves its position on the Research map, but otherwise there are no changes. The GeForce 7900 GS looks better than that old solution from ATI, but can compete with the Radeon X1950 Pro only at the overclocked frequencies.
Nvidia’s ForceWare driver has been recently improved for this game and this helps the GeForce 7900 GS almost match the Radeon X1950 Pro. If successfully overclocked, the 7900 GS can be used to play this game with comfort at 4x full-screen antialiasing.
The deferred rendering technique employed in this game is incompatible with full-screen antialiasing, so we can only benchmark the cards in the anisotropic filtering mode.
The GeForce 7900 GS turns in a good result (for its product class) in Ghost Recon, too. It is 50% faster than the Radeon X1800 GTO and equals the performance of the Radeon X1950 Pro. The average performance of 37-39 may seem rather low to the fans of this game, though.
It is only in the resolution of 1920x1200 pixels that the GeForce 7900 GS is really slower than the Radeon X1950 Pro, but it enjoys a 35% advantage over the Radeon X1800 GTO and GeForce 7600 GT in this case. However, this resolution is unavailable for practical purposes. 1600x1200 is the limit.
The GeForce 7900 GS does well again, but it must be due to the good OpenGL driver and the game’s using OpenGL and UltraShadow technology that this card manages to match the Radeon X1950 Pro. There’s no talking about using full-screen antialiasing on mainstream cards in this game.
The results of this test are similar to those we’ve seen in Prey. That’s expectable since Quake 4 runs on a modified version of the Doom 3 engine, too. But Quake 4 is older and graphically simpler, so the speeds are higher here, and the mainstream cards allow using full-screen antialiasing in the resolution of 1280x1024.
The Radeon X1950 Pro and the GeForce 7900 GS deliver the same performance in 1280x1024, but the ATI solution provides a somewhat higher speed in the higher display modes.
Overclocking gives an edge to the Nvidia card, but the game is so heavy that you get a mere 35fps at best, which is not sufficient for a first-person shooter, especially as action-rich as Serious Sam.
The lack of difference between the results of the GeForce 7900 GT and the GeForce 7900 GS points at the fact that their performance is limited either by the number of ROPs (each card has 16 ROPs on board) or by the memory bandwidth (both cards use 1320MHz GDDR3 across a 256-bit bus).
There’s no talking about playing Hitman: Blood Money on these graphics cards with enabled 4x FSAA, even though you only need a frame rate of 40-45fps to play a third-person shooter comfortably. Here, it is only in 1920x1200 resolution that the GeForce 7900 GS is much slower than the Radeon X1950 Pro. In the lower resolutions the two cards are contending as equals. Note also the higher minimum of speed the Nvidia card provides.
The GeForce 7900 GS isn’t as much better than the Radeon X1800 GTO as in the previous tests, yet its advantage amounts to about 10% here. It’s only at the overclocked frequencies that the Nvidia card can match the performance of the Radeon X1950 Pro.
The game’s Next Generation Content mode is too heavy for graphics cards like GeForce 7900 GT/GS or Radeon X1800 GTO/X1900 GT/ X1950 Pro. They are too slow in this mode with enabled full-screen antialiasing. However, the close collaboration of Nvidia with Eidos helps the GeForce 7900 GS perform in Tomb Raider: Legend almost as well as the new Radeon X1950 Pro does.
The Nvidia GeForce 7 architecture doesn’t permit to use FP16 HDR along with full-screen antialiasing, but without HDR the game loses much of its visual appeal. That’s why we test the cards here with anisotropic filtering only. The game doesn’t offer integrated benchmarking tools, so we measured the frame rate manually with the Fraps utility. The numbers may be somewhat inaccurate as a consequence.
The GeForce 7900 GS looks preferable to the Radeon X1800 GTO and GeForce 7600 GT in closed game scenes because it has 20 pixel processors against the opponents’ 12. Moreover, the new product from Nvidia successfully rivals the Radeon X1900 GT and X1950 Pro that are built according to ATI’s 3-to-1 concept and have 36 pixel processors, but only 12 TMUs and 12 ROPs.
As opposed to closed environments, the Radeon X1800 GTO feels much more at ease in the open and overtakes the GeForce 7900 GS. This must be due to its more efficient memory controller. The cards have low speeds in general, so you won’t have much comfort when playing this game on them, but note that the Radeon X1950 Pro provides the best performance among non-overclocked cards.
The results produced by this test are quite a surprise. The GeForce 7900 GS is much slower not only than the Radeon X1950 Pro but also than the Radeon X1800 GTO. We hadn’t seen Nvidia’s cards perform like than in Titan Quest, so this must be some problem with the ForceWare driver. Anyway, the results suggest that 4x full-screen antialiasing is not recommended on mainstream $199 graphics cards in Titan Quest due to very low performance in that mode.
Radeon X1000 series cards do not support vertex texturing, so they can’t use the appropriate technology to render the water surface with the highest quality in this game. This is only available on Nvidia GeForce 7 GPUs.
Thanks to its efficient OpenGL driver and the overall Nvidia-directed orientation of the game engine, the GeForce 7900 GS shows a very good result even in 1600x1200 with enabled 4x full-screen antialiasing. It’s quite possible to play the game with comfort in that mode.
The X3 engine is known to prefer ATI Technologies’ hardware. This case is not an exception. Notwithstanding its better specs, the GeForce 7900 GS is a little slower than the less advanced Radeon X1800 GTO in every resolution and is much slower than its immediate market rival.
The GeForce 7900 GS is slightly slower than the Radeon X1950 Pro in two resolutions out of the three tested but beats it in 1920x1200 for some reason. But like in many other modern games, it is possible to enable FSAA on graphics cards of that class only in resolutions not higher than 1280x1024.
Contrary to what we’ve seen in Age of Empires, the GeForce 7900 GS cannot compete with the Radeon X1900 GT and the Radeon X1950 Pro anymore, but it is always ahead of the Radeon X1800 GTO. The speed of the cards is even lower here, so using FSAA on the mainstream cards makes no sense even in low resolutions in this game.
3DMark05 confirms the GeForce 7900 GS’s superiority over the Radeon X1800 GTO, although the difference of 550 points is not impressive at all. At the default frequencies the Nvidia card scores 2277 points less than the ATI Radeon X1950 Pro. That’s bad for the 7900 GS considering the equal price of the two solutions. As always, we will publish the results of each of the gaming tests of this benchmark.
The first and third tests depict large scenes and thus require a high fill rate and a high speed of raster operations. The GeForce 7900 GS does much better than the Radeon X1800 GTO, but, quite expectably, worse than the Radeon X1950 Pro. The new graphics card from ATI is indeed an unrivalled leader in the $199 category.
In the second test with a smaller scene the GeForce 7900 GS equals the performance of the Radeon X1800 GTO. The ATI solution is even somewhat better in high resolutions, probably due to an extra vertex processor (the importance of vertex processors in this test is indicated by the results of the Radeon X1800 GTO in comparison with the GeForce 7900 GT).
The difference is less obvious in 3DMark06. The ATI Radeon X1950 Pro scores 491 points more. The overall scores of the cards are lower here than in 3DMark05.
The Radeon X1950 Pro enjoys a much higher advantage over the GeForce 7900 GS in the SM3.0/HDR than in the SM2.0 tests because the former group of tests requires processing very complex pixel shaders and working with FP16 HDR. Moreover, with its only 12 TMUs and 12 ROPs, the Radeon X1950 Pro easily competes with the more expensive GeForce 7900 GT in the first group of tests and leaves the Nvidia solution behind in the second group of tests.
So, while the GeForce 7900 GS is just a little slower than the Radeon X1900 GT and the Radeon X1950 Pro in the first test, it looks humble in the second, which is a modernized version of the second game test from 3DMark05. Note that the results of the GeForce 7900 GS and the 7900 GT are very close in the last case, so the disabled pixel and vertex processors of the latter card have nothing to do with it. Perhaps the ROPs are the bottleneck, but we don’t see that in the first SM2.0 test which contains a larger scene and thus puts a higher load on the raster operators.
The new graphics processor ATI RV570 has showed itself a well-made solution. Manufactured on a new tech process, the GPU has a smaller die area in spite of a large number of transistors it incorporates and thus has a rather low manufacturing cost. As a result, the recommended price of the Radeon X1950 Pro is set at $199 although it surpasses its predecessor in functionality. This is not a limit. We guess ATI has some elbowroom for price maneuvering in the future because the new product is a simple and cutting-edge solution. The Radeon X1950 Pro provides support for multi-GPU configurations by incorporating a full-featured compositing engine, which is integrated into the GPU itself.
Launching successfully its GeForce 7 graphics architecture, Nvidia could refrain from developing absolutely new GPUs for mainstream graphics cards for about a year and a half already. The small G71 chip not only helps the company keep its leadership in the premium class with its GeForce 7950 GX2, but also offer small and inexpensive products like the GeForce 7900 GS at a profit. The GeForce 7900 GS is a good graphics card that delivers higher performance than the GeForce 7600 GT. Considering the small chip size, we guess the price of the 7900 GS may be soon lowered towards $150.
The Radeon X1950 Pro has had a small advantage over the original Radeon X1900 GT in most of the tests, which is natural considering the similar characteristics of the two cards.
There are few tests in which the Radeon X1950 Pro is slower than its main opponent GeForce 7900 GS. And the difference in speed is usually very small and often can be seen in certain resolutions only. These tests are Ghost Recon Advanced Warfighter, Prey, Quake 4, Pacific Fighters and TES IV: Oblivion. These are mostly OpenGL games, some of them are also optimized for the specific features of the Nvidia GeForce 7 architecture.
If we were to compare the available $199 graphics cards in terms of performance, the new ATI Radeon X1950 Pro is not just better than the Nvidia GeForce 7900 GS but can successfully compete with the more expensive Nvidia GeForce 7900 GT that belongs to the $259 category. Also important is that the Radeon X1950 Pro offers you the option of enabling high-quality anisotropic filtering that surpasses everything Nvidia’s solutions can offer as of today. In other words, the ATI Radeon X1950 Pro is overall the best graphics card in the $199 price category. It is possible, however, that the price of products like GeForce 7900 GS, 7900 GT and 7950 GT will be going down at a fast rate and the Radeon X1950 Pro may find itself between a cheaper 7900 GS and a tremendous 7950 GT.
Highs:
Lows:
Highs:
Lows: