by Alexey Stepin
08/03/2005 | 10:12 PM
Back in 2004 things didn’t immediately go well for NVIDIA’s newly-announced GeForce 6800 Ultra, 6800 GT and 6800 graphics cards. The problems involved in the manufacture of an ultra-complex GPU that consisted of 222 million transistors resulted in a significant delay between the date of the announcement and the date the users could actually buy the devices. It was a real problem to find a GeForce 6800 Ultra in shops, while ATI Technologies was quite successfully supplying the RADEON X800 PRO to market.
NVIDIA Corporation learned that lesson well and got all prepared to the release of their new G70 graphics processor (for details see our article called NVIDIA GeForce 7800 GTX: New Architecture Exposed). The GeForce 7800 GTX graphics card, based on the G70 core, came to retail shops almost immediately after the announcement (for details see our article called NVIDIA GeForce 7800 GTX: Monstrous Gaming Performance Unleashed). NVIDIA showed proper promptness, and that’s good, but it also lead to all GeForce 7800 GTX graphics cards being identical irrespective of the manufacturer. The only thing those cards differed in was the design of the plastic casing on the cooler. Even ASUSTeK Computer, a major manufacturer always into offering original solutions, came up with its Extreme N7800GTX that was an exact copy of the reference GeForce 7800 GTX.
So, it seems we should wait for a while for G70-based cards of an original design to appear. We have no doubt many manufacturers will try to experiment with quieter or more efficient coolers and to attract the customer with a non-standard color of the PCB, cooler highlighting, etc. But again, all GeForce 7800 GTX graphics cards are absolutely identical as yet. And this poises a problem for us, at X-bit labs, too. Having posted our comprehensive review of the original GeForce 7800 GTX, we can’t now just repeat ourselves talking about this same card as it comes from different brands.
Fortunately for us, NVIDIA’s new graphics processor has an interesting feature we didn’t cover in our analysis of the G70 architecture. This feature is important enough to be described separately and we are going to deal with it in the Overclocking section of this review. Ah yes… this review is about the e-GeForce 7800 GTX graphics card from EVGA!
Like a majority of serial graphics cards we have tested, the EVGA e-GeForce 7800 GTX came to our test lab in its full rig – package, accessories and all. The box looks rather plain and unimposing. There are no fantastic creatures, space ships or robots here. Silver or gold inserts don’t catch your eyes. All we have here is just a modest mix of light-green and dark-blue, the name of the product and a list of the product’s parameters like the amount of graphics memory on board and the interface.
This simplicity has its own appeal amongst gaudy packages from other manufacturers. It just looks like there’s a serious product inside. The text on the box reads “NVIDIA authorized solution provider”. EVGA, along with XFX and BFG Technologies, is really manufacturing graphics cards only on chips from NVIDIA.
Take a look at the back side of the box for detailed info on the technical characteristics of the e-GeForce 7800 GTX. There’s also a small window here through which one can see the graphics card lying on a special plastic tray. The box has two EVGA seals on both sides. It’s impossible to open it without breaking a seal, so check them out to make sure you have an original EVGA product. The folded plastic tray inside holds the graphics card (in an antistatic bag) and various accessories:
The user’s manual isn’t very detailed, but it does describe the basic steps you should take to install the graphics card and the ForceWare driver in the system and adds some troubleshooting info. The language of the manual is English since the product is mainly intended for the U.S. market. The manual, CDs and stickers are packed into a small cardboard parcel with an EVGA logo. The text on the parcel says the card may also come with optional CDs with a software DVD player and demo versions of various games.
The VIVO switch unit connects to a 9-pin mini-DIN connector via a well-screened 115cm cable. The unit has 6 connectors, four of which are video outputs (three RCA outputs for devices with a component YPbPr input and a four-pin S-Video output). The blue connector that outputs the Pb component can also work as a composite video output. The remaining two connectors (composite and S-Video) are located on the opposite side of the unit and connect to video sources. There’s a place on the case of the VIVO unit for a sticker, but the sticker is missing among the accessories, so you don’t have much freedom in placing the unit. The S-Video cable included with the e-GeForce 7800 GTX is well-screened and flexible. The power splitter and the DVI-I to D-Sub adapters are quite ordinary.
So, the accessories to the EVGA e-GeForce 7800 GTX aren’t really gorgeous, but you do get everything necessary to use the card.
There’s no originality or creativeness here – this graphics card is a copy of the reference GeForce 7800 GTX, only differing from it in the design of the cooler’s plastic cover.
This cover is painted black and carries an EVGA logo where the reference card has an NVIDIA logotype. The cooler proper is, however, the same aluminum thing as on NVIDIA’s reference sample.
A SAA7115HL chip from Philips is responsible for digitizing the analog video signal. For some reason this chip was rarely installed on GeForce 6800 Ultra and GT cards, although they had an appropriate bonding area on their PCBs. But now the top-end graphics card with an NVIDIA processor boasts an extended functionality: the SAA7115HL chip roughly corresponds to ATI’s Rage Theater in capabilities, but not to the more advanced Theater 200 (this Theater has 12-bit ADCs, while the Philips chip only has 9-bit ones). Anyway, GeForce 7800 GTX chips can now match the functionality of the RADEON X850/X800 family and be used to digitize video from analog sources.
The special feature of the G70 processor we mentioned at the beginning of the review is the frequencies this chip is clocked at. It’s not a mistake to use the plural form of the word frequency here: different subunits of this GPU do work at different frequencies! The main frequency of 430MHz is for the pixel processors and raster operators (ROPs) whereas the vertex processors work at 470MHz (we’re talking about the 3D mode; in 2D applications all subunits of the processor are clocked at 275MHz).
We must make a small digression from the topic here. As our theoretical investigation revealed (for details see our article called NVIDIA GeForce 7800 GTX: New Architecture Exposed), the geometrical performance of the G70 GPU roughly equals that of the RADEON X850 XT Platinum Edition. The latter works at 540MHz, but has only six vertex processors against the G70’s eight. So it seems like the efficiency of the vertex processors in the G70 chip isn’t as high as with the RADEON X850 and X800, and NVIDIA had to increase not only their number but also their frequency to get to the same level of performance.
As we already know, the higher frequencies and the increase in the complexity of the logics didn’t affect the power consumption and heat dissipation properties of the G70 much if we compare it with the NV40, so the increase of the frequency of some subunits of the chip is a clever solution. Besides that, this separate-clocking feature can theoretically be used to control the power consumption more flexibly. But let’s get back to the EVGA card.
The GPU of the e-GeForce 7800 GTX works at 450MHz which is higher than the reference card’s 430MHz. The memory works at the same frequency as on the reference sample, i.e. at 1200MHz. But what subunit of the chip does this 450MHz frequency refer to?
NVIDIA’s representatives officially confirmed the fact that different units of the G70 work at different frequencies, yet we decided to check this out using the RivaTuner utility. The current version of this program (version 15.6) proved to keep track of the frequency of the vertex processors. When the GeForce 7800 GTX graphics card was working at its default frequencies (430/1200MHz), RivaTuner reported 470MHz in 3D applications and the difference between the two frequencies of the GPU remained constant (40MHz) whatever we did to them. The e-GeForce 7800 GTX thus has higher GPU frequencies (450 and 490MHz).
There is an opinion that the G70 can only change the frequency of the pixel processors and the raster operators with a step of 27MHz (this opinion is based on the fact that the GeForce 7800 GTX uses 27MHz quartz to issue the clock signal, but the same quartz is used in the GeForce 6800 series, by the way) and the frequency of the vertex processors with a step of 13MHz. We decided to check this out in 3DMark05 since this application always yields repeatable benchmarking results. For more precision we performed each test two times.
First we tested the EVGA card at its default frequencies, i.e. 450/490/1200MHz. Then we increased the frequencies of the GPU by 9MHz and repeated the tests once again. We selected 9MHz on purpose: if the G70 could only change its frequencies with a discrete 13/27MHz step, then a frequency growth of 9MHz wouldn’t reflect on the performance of the graphics card. Despite the above-mentioned theory, RivaTuner reported the increase of the frequency of the vertex processors to 497.25MHz when we overclocked the EVGA card to 459MHz. We also performed this test on a reference sample of the GeForce 7800 GTX, overclocking it from 430 to 439MHz. And here are the numbers we got:
So in both cases we had a minor performance improvement (about 40-50 points irrespective of the number of times we ran 3DMark) after we had increased the GPU frequency by 9MHz. So, the supposition that NVIDIA’s G70 controls the frequency with a discrete step of 13/27MHz doesn’t come true. If it were true, we wouldn’t have any performance gain at the 9MHz step. But since we do have that gain, it means the new graphics processor from NVIDIA overclocks just like older solutions, except for the fact that the frequency of its vertex processors is always 40MHz higher than that of its pixel pipelines and raster operators.
We also found that the 450 megahertz is the frequency of the pixel processors on the e-GeForce 7800 GTX, so this graphics card should have a small advantage over the regular GeForce 7800 GTX in gaming benchmarks. Unlike some other manufacturers who declare the frequency of the vertex processors (470MHz) as the main one, EVGA honestly specifies 450MHz which is the frequency of the pixel pipelines and ROPs, i.e. the frequency of the units the overall performance of a GPU depends the most on.
As for our overclocking achievements, we got 490MHz on the GPU (530MHz on the vertex processors) and 1380MHz on the memory. We think it is a nice gain, especially for GDDR3 chips that already worked near their limit. We set an additional 120mm fan from A.C.Ryan to blow at the card when we were testing it in the overclocked mode.
The acoustic parameters of the EVGA e-GeForce 7800 GTX are the same as those of the reference GeForce 7800 GTX card (for details see our article called NVIDIA GeForce 7800 GTX: New Architecture Exposed). With rather small dimensions, the new cooling system from NVIDIA is much quieter than the one they used to install on the GeForce 6800 Ultra. The fan is only loud for the first few seconds after you turn the system on. The clever design helps the cooler handle such a complex chip as the G70 even in the summer heat. As we said above, we only used an additional fan when overclocking the card – and we did so “just in case” rather than out of real necessity since the EVGA card worked well at the overclocked frequencies even without the additional cooling.
EVGA doesn’t want to limit itself with the reference cooler, by the way. We have information that e-GeForce 7800 GTX graphics cards will be equipped with the exclusive cooling system called ACS, and the company will pay special attention to cooling the components on the reverse side of the PCB (for details see our news story called Graphics Cards Maker Revamps NVIDIA GeForce 7800 Cooling System. EVGA Returns Asymmetric Cooling System).
We checked the quality of the 2D image produced by the card on Dell P1110 and Dell P1130 monitors (which are actually the same model, differing in the design of the case only). The maximum display mode these monitors support is 1800x1440@75Hz (the unofficially supported 2048x1536@60Hz mode isn’t of much use due to the low refresh rate), and the EVGA e-GeForce 7800 GTX yielded a crystal-sharp image without any ghosting or blur, just as expected from a top-end graphics card.
We performed our tests on the following testbed:
Graphics cards:
Drivers:
Following our methodology, we enabled the ForceWare optimizations, except the Anisotropic mip filter optimization. The Image settings slider was set to the Quality position, and we also left the Gamma correct anti-aliasing option enabled. In ATI’s Catalyst we enabled Catalyst A.I. using the Standard mode. The Mipmap Detail Level option was set to Quality. We disabled the VSync option in both drivers.
If possible, we control FSAA and anisotropic filtering from the application. Otherwise, we force the necessary mode from the driver. We don’t edit any configuration files. The graphics quality settings in the games were set to the maximum level, the same for graphics cards from NVIDIA and ATI Technologies. The following games and applications were used:
First Person 3D Shooters:
Third Person 3D Shooters:
Simulators:
Strategies:
Semi-Synthetic Benchmarks:
Synthetic Benchmarks:
Today’s graphics hardware hits the performance limit even in this rather recently released 3D shooter, at least in low resolutions. As for higher display modes, the GeForce 7800 GTX is obviously the best. You can tell the e-GeForce 7800 GTX’s default frequencies are higher than the reference card’s ones by the small gap between them.
NVIDIA’s new-generation solution is so fast that it is the CPU rather than the graphics subsystem that is the system bottleneck even in Doom 3 . At least, this is true for low resolutions. High resolutions and the “eye candy” mode clearly show that EVGA’s version of the GeForce 7800 GTX is really a little bit faster than the reference sample, having a 20MHz higher GPU frequency. The winner is the SLI configuration of two GeForce 6800 Ultra cards, however. They are ahead even of the overclocked e-GeForce 7800 GTX.
The performance ceiling goes somewhat higher on the d3dm4 map since this multiplayer map lacks monsters. The general picture is the same as in the previous case, but the gap between the GeForce 6800 Ultra SLI and the GeForce 7800 GTX is bigger.
The open-environment scene recorded on the Pier map depends heavily on the speed of the CPU, especially when it comes to top-end graphics cards. We encountered a strange problem with ForceWare 77.72: the performance of the GeForce 6800 Ultra was higher than that of the GeForce 7800 GTX, while the two SLI platforms performed identically. We repeated the tests but to the same outcome. The reason for the strange behavior of the GeForce 7800 GTX cards must lie somewhere in the driver, and we hope it will be solved in next versions of ForceWare.
We have the same problem on the Research map, too. We don’t think the numbers to be correct and can’t make any comparisons for Far Cry .
The e-GeForce 7800 GTX doesn’t have any advantage over the ordinary GeForce 7800 GTX or, rather, you can’t see it in the diagram since even overclocking gives the EVGA card a paltry speed gain of only 2fps. The two GeForce 6800 Ultra cards in the SLI configuration can’t keep up with the single GeForce 7800 GTX. As for full-screen anti-aliasing modes, only SLI platforms can provide a barely acceptable speed – the power of a single GeForce 7800 GTX isn’t enough.
The Canals levels in Half-Life 2 are open environments with a water surface, technically speaking. That’s why the speed of the computer there is limited by the CPU, like with Far Cry ’s Pier record. It’s only in the “eye candy” mode that you can see some difference between the participating devices. Particularly, the EVGA e-GeForce 7800 GTX is again a couple of frames per second faster than the reference sample from NVIDIA (and that’s another confirmation that the GeForce 7800 GTX does not change its frequencies with a discrete 27MHz step).
The EVGA card is either a little slower than or equal to the reference sample of the GeForce 7800 GTX in the street fighting scene we recorded on the d3_c17_02 map. This may be due to measurement errors. Anyway, the cards easily reach the performance ceiling imposed by the central processor in all cases, except 1600x1200 with enabled full-screen anti-aliasing and anisotropic filtering where only the SLI configuration of two GeForce 7800 GTX cards hits it.
The results in this game are of no use for us: the participating cards, except the single GeForce 6800 Ultra and the RADEON X850 XT Platinum Edition, have the same frame rate of 180fps in all the modes and resolutions. We think that only the extremely resource-consuming 8xS anti-aliasing mode would be a heavy enough load for the GeForce 7800 GTX here, but even in this case the card would most probably ensure a playable frame rate.
Unreal Tournament 2004 isn’t a good benchmarking tool anymore. The game is too simple for today’s ultra high-end and high-end products. Note the problems with the SLI configuration of two GeForce 6800 Ultra with ForceWare 77.72: the SLI platform is slower than the single graphics card.
The demo recorded on the Metallurgy map is less CPU-dependent, but G70-based graphics cards still easily reach the speed ceiling even here. Again, this shooter isn’t appropriate for benchmarking the recently released graphics solutions. We should wait for a next game in the Unreal Tournament series – people from Epic Games promise us a stunning spectacle.
The EVGA card is on top, again. When overclocked, it even leaves the SLI pair of two GeForce 7800 GTX behind! The effect from the 20MHz higher GPU frequency gain is clear in this game: the e-GeForce 7800 GTX enjoys an advantage of 8-12fps over the reference GeForce 7800 GTX.
We chose the same level of detail for the GeForce 7800 GTX and 6800 Ultra and for the RADEON X850 Platinum Edition. The latter doesn’t support Shader Model 3.0 and was working in the SM 1.1 mode.
The e-GeForce 7800 GTX again shows it works at higher frequencies than the ordinary GeForce 7800 GTX. The RADEON X850 XT Platinum Edition is slightly slower than the G70-based cards but it worked in the easier Shader Model 1.1 mode.
The EVGA e-GeForce 7800 GTX wins the “pure speed” mode, but falls behind the RADEON X850 XT Platinum Edition in the “eye candy” mode, even though the RADEON is formally a solution of the previous generation. We think it’s all because of the not very efficient memory controller which also prevented GeForce 6800 cards from performing well in full-screen anti-aliasing modes. As for the SLI platform with two GeForce 7800 GTX, it is simply beyond competition.
The multi-GPU technology doesn’t work right in IL-2 : the SLI platform do worse than the single card, at least with ForceWare 77.72. The G70-based solutions allow playing this flight sim with comfort even in hardest modes.
There seems to be a speed barrier lying at about 28-30fps when you use the maximum graphics quality settings in Lock On. Even the SLI platform on two GeForce 7800 GTX can’t overcome it.
The participating cards all have almost the same speed here, with a few exceptions. It seems that it is not the GPU performance that is the limiting factor at the max settings of this game, but you couldn’t see that until the arrival of such a high-performing solution as the G70 processor.
All the G70-based graphics cards easily get to the speed barrier set by the system’s central processor, but we do get some valuable info in the “eye candy” mode. Particularly, we can see that the GeForce 6800 Ultra SLI platform is faster not only than the single GeForce 7800 GTX but also than the overclocked e-GeForce 7800 GTX. NVIDIA’s solutions ensure a playable frame rate in all resolutions even with enabled full-screen anti-aliasing.
The RADEON X850 XT Platinum Edition is a little faster than the GeForce 6800 Ultra in the “pure speed” mode and much faster in the “eye candy” mode, but this is the only success of ATI Technologies here. The G70-based solutions are superior, and the e-GeForce 7800 GTX again enjoys a 1-2fps advantage over the standard GeForce 7800 GTX.
This semi-synthetic benchmark isn’t new and doesn’t use all the capabilities of modern GPUs. It prefers graphics cards with an efficient memory controller and a high geometrical performance. It’s not all perfect with the GeForce 7800 GTX’s memory controller if we compare it with ATI Technologies’ solutions, but the performance of the vertex processors is high. Eight such processors clocked at 470MHz (490MHz with the EVGA card) help the G70-based solutions outperform the RADEON X850 XT Platinum Edition, even though the gap is very small. The fact that the e-GeForce 7800 GTX enjoys a small advantage over the standard GeForce 7800 GTX is indicative of the higher operational frequencies of the EVGA card and of the lack of the discrete 27MHz step at overclocking.
The difference between the GeForce 6800 Ultra and the RADEON X850 XT Platinum Edition seems negligible against the speeds of the GeForce 7800 GTX SLI, GeForce 7800 GTX and GeForce 6800 Ultra SLI platforms.
The highest possible speed is reached in the first two resolutions. In 1600x1200 the GeForces leave no chance to the RADEON X850 XT Platinum Edition. The latter, however, beats the GeForce 6800 Ultra in the “eye candy” mode.
NVIDIA’s cards have been good in the second test since the times of the GeForce FX architecture. The GeForce 6800 Ultra is similar to the RADEON X850 XT Platinum Edition here, while the more advanced solutions and the SLI platforms have much higher speeds.
The third test is technically similar to the second one, except that it features a more complex geometry. That’s why the overall situation among the participating cards is the same as above.
The fourth test uses Shader Model 2.0 and is the only test from 3DMark03 where the RADEON X850 XT Platinum Edition is really better than the GeForce 6800 Ultra. But that’s the only real success of ATI Technologies today. As for the EVGA card, it is again a little faster than the reference GeForce 7800 GTX.
Thus, the GeForce 7800 GTX have proved their belonging to a new generation of graphics hardware in 3DMark03’s tests.
The G70-based solutions are supreme and the SLI configuration of two GeForce 7800 GTX scores over 10,000 points. The RADEON X850 XT Platinum Edition is rather far ahead of the GeForce 6800 Ultra even though the latter supports Shader Model 3.0.
The e-GeForce 7800 GTX performs confidently in the first game test which is a typical 3D shooter scene. When overclocked, the EVGA card isn’t much slower than the SLI platform with two GeForce 6800 Ultra cards and the gap diminishes in the ”eye candy” mode. Note also that the performance ceiling of the first test is at 42fps (the performance of the two SLI platforms is almost the same in the lowest resolution). Factors other than the GPU’s capabilities limit the speed from that point on.
The second test loads the vertex processors much, so the GeForce 6800 Ultra is slower than the RADEON X850 XT Platinum Edition here. Of course, neither of these two cards can challenge the GeForce 7800 GTX. The overclocked e-GeForce 7800 GTX is no worse than the SLI configuration of two GeForce 6800 Ultra in the “eye candy” mode.
The e-GeForce 7800 GTX clocked at 490/1380MHz frequencies overtakes the GeForce 6800 GT SLI in 1280x1024 resolution of the “pure speed” mode, but the latter platform goes far ahead in the “eye candy” mode. The 3DMark05 tests also confirm that the EVGA card is a little faster than the standard GeForce 7800 GTX.
The GeForce 7800 GTX SLI shows an exceptional speed in this terribly complex test from 3DMark05 – over 50fps in 1600x1200 with enabled full-screen anti-aliasing. Maybe a CrossFire configuration of two graphics cards on ATI’s upcoming R520 chips will do something like that, but so far the SLI configuration of two GeForce 6800 GTX is unrivalled.
So, here’s a summary of our experience with the e-GeForce 7800 GTX graphics card from EVGA. First we investigated such an important feature of the NVIDIA G70 processor as independent clocking of different GPU subunits. Our experiments suggest that the vertex processors of the G70 chip do work at a higher frequency than the pixel pipelines and the ROPs. We also found out that the frequency of the vertex processors is always 40MHz higher, irrespective of the frequency growth at overclocking.
This fact has one unfortunate consequence. Some manufacturers may and do declare the frequency of the vertex processors as the main one, thus misleading the user into thinking that 470MHz refer to the pixel pipelines. It’s not a serious problem, but you should be aware of that. You can check the real frequencies of the GPU with the help of the RivaTuner utility even though its 15.6 version can only monitor the frequency of the vertex processors: if you see the same number in RivaTuner’s frequency monitoring module and in the ForceWare control panel, then this is the frequency of the vertex processors of your GeForce 7800 GTX. You can subtract 40MHz from this number to get the frequency of the pixel pipeline and ROPs. It is going to be 430MHz, like on the reference sample of the GeForce 7800 GTX.
We can also disprove the information that the new GPU from NVIDIA can only change its frequency discretely, with 27MHz and 13MHz steps for pixel and vertex processors, respectively. Every time we overclocked the card by some value which was certainly smaller than the discrete step, we got an appropriate performance gain. By adding 9MHz to the GPU frequency we enjoyed a small but stable speed gain in almost all games and benchmarks excepting those where the speed gain was smaller than the measurement error range. The next version of RivaTuner may show these things more clearly, but so far we can say that the GeForce 7800 GTX seems to overclock in exactly the same way as the GeForce 6800 Ultra does, except that the frequency of the vertex processors of the G70 chip is always 40MHz above the main one.
Another thing we want to mention in this summary is SLI technology. We have been including SLI configurations into our tests for some time now. NVIDIA’s multi-GPU technology has grown mature, but is still not free from problems. At least, some games still run slower on SLI platforms than on a single card. It means NVIDIA’s programmers have much to do yet, and we have to wait for ForceWare Release 80.
As for the subject matter of this review, the EVGA e-GeForce 7800 GTX graphics card really has higher default frequencies than the reference GeForce 7800 GTX: 450/490/1200MHz against 430/470/1200MHz. This is of course a small increase, barely noticeable in real applications, but we just want to set this case apart from those situations when the vertex processors frequency is declared as the main one.
It is hard to say anything special about the e-GeForce 7800 GTX just because it doesn’t differ from the reference GeForce 7800 GTX other than in the GPU clock rates and in the design of the cooler’s cover. Our tests revealed no defects; the card remained stable throughout the test session. It also boasted excellent overclockability (we were especially pleased with the memory frequency gain – 180MHz (90MHz DDR) for 1.6ns memory was a highly satisfying result).
Thus, the EVGA e-GeForce 7800 GTX is one of the fastest versions of the GeForce 7800 GTX available. It is overclocker-friendly, but comes with a rather limited set of accessories. Buying this card today you get the fastest performance in games, good overclocking opportunities, support of the multi-GPU SLI technology, HDTV and VIVO. Again, even if you are not into overclocking, the EVGA e-GeForce 7800 GTX is no worse or better than any other GeForce 7800 GTX. But remember that you need a high-quality 400W or better power supply for your GeForce 7800 GTX card to work well.
Highs:
Lows: