Nvidia GeForce 7800 GT - BeHardware
>> Graphics cards

Written by Damien Triolet

Published on August 11, 2005

URL: http://www.behardware.com/art/lire/584/


Page 1

Introduction, architecture




We were expecting a reaction from ATI in response to the release of the GeForce 7800 GTX, but NVIDIA made the next move with a new high end graphic card, the GeForce 7800 GT.


The GeForce 7800 GT is based on a G70 chip (same as the 7800 GTX), but with one quad engine (4 pixel shader pipelines) and a vertex engine deactivated. Frequencies are also slightly lower at 430 instead of 600 MHz.

You probably noticed the GPU frequency in quotation marks and there is a reason for that. We didn’t see it in June in the first GeForce 7800 GTX test, but the GeForce 7800 uses quite a complex frequency system. Two different parts of the CPU use different frequencies, one close to the announced figure (but not identical) for pixel shader pipelines and ROPs and a much higher one for the other parts, mainly vertex shaders.

For NVIDIA the release of the 7800 GT is the opportunity to reorganise their high end graphic cards and the GeForce 6800 Ultra will become of lesser interest and eventually disappear. Official prices of the GeForce 6800 GT and 6800 are lower and this corresponds more or less to price cuts applied in June. So, the price of the GeForce 7800 GT should be around $400, $299 for 6800 GT and $199 for 6800.


CineFX 4.0: gains in practice
Overall the GeForce 6800 and 7800 feature quite similar functions: Shader Model 3.0, HDR FP16, and SLI. The power of three as seen by NVIDIA.

We already described in detail the GPU architecture in the 7800 GTX test, here. We only remind you that with the GeForce 7800, NVIDIA has slightly optimised CineFX architecture to increase efficiency, related to an increase in IPC (number of instructions processed per cycle). This improvement is difficult to measure as it varies from 0 to 100%. We estimate that it’s generally from 5 to 20% and much more in very specific cases. We tried to verify this with the comparison of a 16 pipe GeForce 7800 and a GeForce 6800 GT clocked at the same frequency. The GeForce 7800 with 16 active pipes was clocked at "346 MHz", 351 MHz for pixel shaders and ROPs and 387 MHz for vertex shaders. This difference in vertex shading has no influence on games as they are rarely restricted on this level, and so the comparison is valid.


There are indeed performance gains and they are due to several NVIDIA optimisations. First, antialiasing and anisotropic filtering efficiency have improved (unfortunately with a slightly lower quality for the latter) with performance gains from 6 to 33%.

For Doom3, the gain is almost 30% in 1920x1200, showing an optimisation in management of high resolutions. Some optimisations, which are meant to avoid useless calculations have an integration cost directly related to the higher resolution they accelerate. With the 7800, NVIDIA extended some of these optimisations beyond 1600x1200, which explains this type of gain.


The HDR mode is still costly to performance, but is slightly more efficient with the 7800. The gain due to the architecture is 16% in Farcry.

Finally, you will see that there has been an amazing progression in performance in Colin McRae 05. It’s hard to explain and is probably due more to a software optimisation rather than architectural modifications.


Page 2
The card, power consumption, Overclocking

The graphic card


The 7800 is single slot and a bit shorter than its big sister. Its dimensions are actually similar to the 6800 GT, making it 1.5 cm shorter than the 7800 GTX.

The NVIDIA standard card features 2 ns Infineon memory clocked at 500 MHz. All chips are located on the same side sot the PCB isn’t designed to accomodate 512 MB of DDR.

The cooling system is new, but NVIDIA could have found something a little quieter (like the 6800 GT’s). The fan produces alot of air flow, but noise isn’t as disturbing as that of the turbine on the Radeon X850 XT PE’s. Some manufacturers like Asus, whose chip is pictured, have decided to use another system similar to the 6800 GT’s.


Power consumption
We measured the power consumption of different graphic cards used in this test. Measurements were taken directly at the power outlet, which therefore represents the computer’s entire power consumption, here an Enermax 535W. We reported figures obtained under Window’s desktop and in use with a 3D scene and Prime95. Prime95 makes it possible to have constant CPU use regardless of a graphic card’s performance.


NVIDIA indicated in their latest documentation that the 7800 GTX’s power consumption is 100 Watts as compared to 85 Watts for the 7800 GT. In practice, we measured a bigger difference, and it seems that our 7800 GTX’s consumption is higher than 100 Watts. This isn’t that surprising as NVIDIA changed to 110 Watts for the 7800 GTX. The 27 Watts gap measured between the two graphic cards is indicative of a 110 Watts power consumption for the 7800 GTX. At any rate, the 7800 GT’s level in 2D is the lowest of this roundup and very good in 3D.


Overclocking
As we said before, the 7800 uses two GPU frequencies and none of them correspond to that which is displayed nor is it possible to change them in current overclocking tools (including the one integrated in NVIDIA drivers). Here is a table which shows the real frequencies compared to the frequency selected:


Frequency 1 is the most important, and is the one which influences most performances in games. Vertex shaders do not restrict any games in practice (but they do restrict 3Dmark!). This table clearly shows that we need to look for overclocked cards in shops. For example, a GeForce 7800 GTX clocked at 445 MHz instead of 430 MHz looks good on the paper, but in practice in both cases, the higher is clocked at 432 MHz. We measured a 0.2% difference between these two frequencies, and it’s useless to spend more money for a card of this type.

Our reference 7800 GT was stable at 496 MHz for the core (486 MHz for PS and 535 MHz for VS) and 575 MHz for memory. The gain is 20% and 15%, respectively.

Our attempts to unlock the deactivated quad engine and vertex engine failed. We successfully deactivated additional units but did not re-activated some. Could NVIDIA have found a way to stop re activations?


Page 3
Antialiasing, filtering, tests

Antialiasing
Three weeks ago, NVIDIA released SLI anti-aliasing, which calculates an image with each GPU and then combines them before displaying them. You can find all details on SLI AA and the different performances of anti aliasing modes here.

In the tests we saw a very low SLI AA performance and couldn’t find an explanation. Now, however, we found the reason. In SLI if the second card’s data is most often brought back through the SLI connector to the first card, it doesn’t seem to be usable when the two images have to be recombined. They then have to use PCI Express. GPU2 can’t calculate a second image as long as the first one isn’t completely transferred. The effective time to calculate one picture = real calculation time + time to transfer. So because of the relatively low PCI Express bandwidth in transfering high resolution images, performances collapse.

The new SLI 16x, which give the possibility of including two real PCIE 16X to motherboards should help to solve the problem, but it may not be enough. The data of the second GPU has to travel through GPU2 memory -> PCI Express -> chipset -> HyperTransport -> CPU -> central memory bus -> central memory -> central memory bus -> CPU -> HyperTransport -> chipset -> PCI Express -> GPU1 memory. This is quite a long way to go! We verified that the central memory bandwidth also restricted performances and that the HyperTransport bus could be restrictive if performances increased. Unless NVIDIA finds a way to use the SLI connector with the SLI AA, it won’t be really of use in complex games.


Filtering
As we explained in the GeForce 7800 GTX test, NVIDIA slightly changed the anisotropic filtering algorithm but in the wrong way. The result is lower quality. There is now a twinkling effect in textures in the form of interference artefacts. It is difficult to show them on a fixed screenshot and, unfortunately, non-compressed full size videos are too big.

Quality reduction varies depending on the texture displayed. For some games, the difference is almost invisible whereas with others it’s obvious. We personally feel this is unacceptable to find this type of optimisations for 2005 high end products. If the optimisation is on the software level we hope that it will be possible to deactivate it in future drivers. If unfortuneatly it’s a hardware optimisation, we hope that ATI will provide better quality with the R5x0 and force NVIDIA to provide a better result and not a pre-3dfx sampling point with the future GeForce 10000…

Tests
- ASUSTeK A8N-SLI Premium (bios 1005)
- AMD Athlon 64 4000+
- 2x512 MB of PC3200 2-2-2 memory
- Enermax 550W power supply
- ForceWare 77.77
- Catalyst 5.7
- Windows XP SP2

For this test we measured performances in 1280*1024, 1600*1200, and 1920*1200 (or 1920*1440 when this mode was unavailable as in Act Of War), with different graphic settings: standard, 4x anti aliasing and 8x anisotropic filtering and HDR if available. Testing this type of high end graphic cards in 1024*768 isn’t really of interest. We preferred to not include this resolution and chose a higher one, which should please users of bigger monitors.


Page 4
Half-Life 2, Doom 3

Half-Life 2




GeForce 7800 cards, especially in SLI, are a little too powerful for the Athlon 4000+ in Half-Life 2. Logically, the GeForce 7800 GT is slightly behind compared to the 7800 GTX and the Radeon X850 XT PE, which provide very good performances with Half-Life 2.

Doom 3



With Doom3, NVIDIA is in the lead. The 7800 GT is, of course, behind the 7800 GTX but is way in front of the 6800 Ultra. Once AA is activated, SLI shows its power. Performance gaps are smaller, however, because the 6800 and 7800 have the same amount of ROPS, which has an effect on results.


Page 5
Far Cry, Splinter Cell CT

Far Cry




As the Athlon 64 4000+ is restricted to +/- 70 FPS, in standard mode, all graphic cards provide similar performances. Once antialiasing and anisotropic filtering are activated, performance gaps increase. Once again, the Radeon X850 XT PE is slightly in front of the GeForce 7800 GT

In HDR mode, which fully takes advantage of the FP16 surfaces in the GeForce 6800 and 7800, results of NVIDIA graphic cards are similar. You will notice that the SLI doesn’t always significantly improve performances in this mode. This is unfortunate because it could have used it!


Splinter Cell Chaos Theory

A new patch just came out for SCCT. The 1.04 version brings PS 2.0 support for ATI graphic cards, whereas it only previously supported PS 1.1 and 3.0. All 3.0 mode effects are available in 2.0, however HDR is much lower in PS 2.0 and even of no use.




Splinter Cell CT is one of the most interesting games to test graphic cards with as it is not much restricted by the CPU. SLI solutions can express all their potential and performances are almost doubled. The 7800 GT is between the 6800 Ultra and 7800 GTX, but closer to the latter. The Radeon X850 XT PE surpasses the 6800, but it isn’t at the same level as the 7800 GT.

The situation is similar with AA and Aniso and in HDR. It’s interesting to note that ATI’s HDR is far from NVIDIA’s. Effects such as Soft Shadows and Parallax Mapping, however, are rendered similarly.


Page 6
Colin McRae 05, NFSU 2

Colin McRae 05



With Colin McRae, 7800 cards are clearly in front of the 6800 (as the comparison in architecture explains), which were way behind the Radeon. The Radeon X850 XT PE is, in spite of this progress, in front of the 7800 GT in standard mode and the 7800 GTX with antialiasing and aniso activated. SLI makes a big difference as everyone will agree.

Need for Speed Underground 2



The GeForce 6800 suffers from a strange bug with NFSU 2 in 1600x1200. After a couple of seconds the number of frames per second falls from 60 to 0.1, making simple navigation through the menu impossible. The X850 XT PE is once again neck and neck with the 7800 GT and sometimes takes the lead depending on the resolution. Once antialiasing is activated, ATI clearly dominates and SLI only slightly improves 7800 performances.


Page 7
Act of War; Pacific Fighters

Act of War



Without AA and aniso, NVIDIA clearly dominates. The two SLI solutions, however, are restricted by the CPU. With AA and aniso activated differences become more apparent between NVIDIA graphic cards. This time, NVIDIA is back in the game and the X850 XT PE is at the same level as the 6800 Ultra.


Pacific Fighters



The 7800 is on top again and SLI is CPU restricted, at least without AA and aniso. For once, ATI provides better results without these optimisations than with them, but in both cases they aren’t as good as the 7800’s.


Page 8
In a nutshell

Average

For standard display, the two SLI solutions provide similar performances and the 7800 GT lies between the 7800 GXT and 6800 Ultra. On average, the Radeon X850 XT PE easily surpasses the 6800 Ultra but isn’t much better.


AA 4x and Aniso 8x average

With AA and aniso, ATI provide better results probably because of the higher sample transfer rate. The Radeon X850 XT PE’s frequency helps to set it apart from NVIDIA graphic cards on this point as the 7800 only have 16 ROPs. This time, the X850 XT PE is very close to the 7800 GT. Once again, whether it’s with the 7800 GT or the 7800 GTX, SLI performances are way above the rest. It is really time for ATI to respond to this.

Average in HDR

HDR averages only concern Far Cry and Splinter Cell CT. There is a noticeable progress with the different NVIDIA solutions. SLI gains are relatively low as it’s the combined average with Far Cry (not good) and Splinter Cell (very good).


Page 9
Conclusion

Conclusion
With the 7800 GT, NVIDIA makes its new high end GPU more affordable with a lower price proportional to the cut in performance (more or less 20%). NVIDIA confirms its undisputed leadership in high end graphic cards, while ATI doesn’t seem have any product to match.

7800 cards aren’t flawless, however. Indeed NVIDIA had the wrong idea in reducing anisotropic filtering quality, which led to a sub-filtering effect and twinkling in textures. We would definitively have preferred avoiding this optimisation. In many cases the 7800 will be restricted by less complex games or monitors restricted to 1280 x 1024. What is the interest of reducing quality to gain a couple of performance points? We still hope that NVIDIA will correct this via future drivers, but if the optimisation is 100% hardware we can only say it’s unfortuneate to see this type of thing on high end GPUs.

With a relatively silent 7800 GTX, we didn’t expect the new cooling system. NVIDIA equipped the 7800 GT with a noisier one than the 6800 Ultra. Some of NVIDIA’s partners should fortunately replace it with more silent systems but only after the first graphic cards will be sold. You will have to be patient if you like quiet systems or replace it yourself.


The Canadian R5x0 range will be unveiled at the end of September with a completely unknown performance level. Is it best to wait before buying? This is the eternal question. If we had mixed feeling at the time of the 7800 GTX’s release, now ATI’s release date is closer and we would be tempted to tell you to wait. But who knows. Since ATI’s promises in May about CrossFire advantages, we haven’t seen anything concrete. NVIDIA has taken all of ATI’s good ideas and they are already functional, even if SLI Antialiasing is unusable in recent games due to the cost to performance. If you can be patient a few more weeks, we advise you to do so to make a better choice. In the worst case scenario you will have saved a bit of money. If you are in a rush or if you have lost all hopes of seeing ATI releasing a new high end product or the CrossFire, the 7800s are excellent graphic cards-if you aren’t too particular about filtering quality.


Copyright © 1997-2010 BeHardware. All rights reserved.