MSI GeForce 9800GTX-SLI

Today, we've got another video card GeForce 9800GTX, this time one made by MSI. As we expected, MSI N9800GTX-T2D512 makes almost no difference from its "congeners" GeForce 9800GTX made following NVIDIA's reference design. Making use of that, we decided to test this MSI's product versus the same card by Foxconn which we already described. Before we move on to getting familiar with the test results for the SLI-combination MSI N9800GTX+Foxconn 9800GTX, we'll be taking a closer look at the new product by MSI.

Exterior and package bundle

The package of N9800GTX is made in a style customary for MSI. The designers at the company decorated the box with quite a large image of a girl and found enough space for an impressive list of specifications.

The reverse side of the box, unlike the front side is far from being informative. Most of the data specified over there is standard for many video cards by MSI.

The package bundle of MSI N9800GTX is as follows:

• two power supply adapters for PCI-E video cards;
• DVI -> D-Sub adapter;
• DVI -> HDMI adapter;
• a cable for internal connection of S/P-DIF signal (required for sound transmission via the HDMI);
• s-video > s-video splitter + component out adapter;
• drivers CD;
• user's guide.

The MSI N9800GTX, as the formerly reviewed GeForce 9800GTX made by Foxconn, is made following the reference design and stands out among the mass of similar products with stickers only, and of course with frequencies raised versus the recommended clock speeds.

The reverse side of the video card nor brings any surprises.

There is nothing new "under the hood". N9800GTX is cooled with a standard cooling system made of aluminum and copper.

Over the center of the PCB, there is the G92 graphic processor, with eight GDDR-3 memory chips making up 512 MB positioned in half-moon around.

The graphic processor has the G92-420 marking and belongs to revision 2 of the chip. The chip offers 128 universal processors, therefore with the specifications for GeForce 9800GTX.

The memory chips installed are made by Samsung and offer 0.8 ns access time, which is equivalent to the effective clock speed 2400 MHz. The real operating frequency of the memory is a bit lower - 2200 MHz

Upon removal of the thermo paste off the copper base of the cooling system we found scratches. It is known that almost all the "expensive" video cards by NVIDIA are manufactured at the same partnering factories of NVIDIA and then distributed to end manufacturers, so MSI is unlikely to blame for anything. Since the formerly reviewed Foxconn 9800GTX was free from these defects, that is most likely a special case.

Efficiency of the cooling system and overclocking

This time, we'll be testing the efficiency of the cooling system of MSI N9800GTX using the Firefly Forest test from the 3DMark 06 suite. The test conditions: the resolution 1600x1200, 4-X FSAA, and 16-X AF. After nine runs of the test we produced the following results:

When under load, MSI N9800GTX heated up to 73 C (60 C - at rest), which is a good result. In terms of overclocking, the new product demonstrated approximately the same results as the formerly tested one by Foxconn. Note that MSI N9800GTX was originally overclocked to 720/1728 MHz for the graphic processor, and the frequency of the video memory remained at its default level - 2200 "efficient" MHz. During further overclocking, the new product by MSI overcame the 756/2000 MHz "bar" for the GPU and 2406 MHz for the video memory. Recall that the frequencies recommended by NVIDIA for GeForce 9800GTX are 675/1688 MHz for the graphic processor and 2200 MHz for the video memory. At the same time, the real operating frequency of the GPU is 675/1674 MHz. Such a discrepancy in frequencies set in the BIOS and the real frequencies of the video card are caused by the minimum increment of the clock speed which is not always taken into account. For instance, the frequency of the shader domain varies in ~50 MHz increments, and it most frequently does not coincide with the frequency set in the BIOS of the video card.

Test results and conclusions

Since this MSI N9800GTX was delivered to us promptly enough, we had a chance to test in a combination with Foxconn 9800GTX. Despite the different clock speeds, the two video cards by different manufacturers started up successfully in the SLI mode. However, to get more correct results, we updated the firmware of Foxconn 9800GTX with the BIOS taken from MSI N9800GTX. Therefore, we got kind of two MSI N9800GTX video cards. This is the test-bench configuration:

To test the video cards, we used ForceWare 174.74 drivers.

Traditionally, we are examining the results starting with 3DMark tests.

At 3DMark, the tandem of two MSI video cards proved to be the fastest, however, the performance boost as compared to the single video card is minimum. The thing is, in the "default" mode the performance of such powerful novelties is bounded by the CPU, which was just the case in this test. We'll be running further tests of the video cards in high-quality modes with 4X FSAA and 16X AF enabled, which will let this tandem of MSI N9800GTX show its full worth in the SLI mode.

At Call of Duty 4, the performance boost attained from using it is seen to the named eye. However, let's recall our recent tests of GeForce 9800 GX2 which is based on two G92 video chips running in the SLI mode. That time, the result was much better, but the tests were run on a motherboard with the newer chipset nForce 790i Ultra SLI versus the current nForce 680i SLI. Hence, we can assume that if we had used a motherboard based on the nForce 790i Ultra SLI chipset, we would have got a better result.

At Call of Juarez, the picture is entirely different - the combination of two MSI N9800GTX video cards leaves the single video card well behind, and with the rise in resolution the performance boost attained from using the second cards turns almost two-fold.

On migration to Windows Vista, the advantage from using the second MSI N9800GTX video card at Call of Juarez came up almost close to the two-fold. In terms of percentage, the performance boost amounted to 83% – 92.9%.

We were in for the first issue with the SLI at the Crysis game. All was going smooth, but at 1600x1200 the tandem of two MSI N9800GTX video cards failed to pass a single test. At best, all ended up in a hang of the game, and in the worst case - in the "blue death screen" and further PC reboot. Again, our suspicions fell on the nForce 680i SLI chipset used since while testing the dual-processor GeForce 9800GX2 video card on a motherboards based on the nForce 790i Ultra SLI chipset all was fine.

On migration to Windows Vista, the situation aggravated substantially. While at 1024x768 the combination of two MSI N9800GTX video cards was able to pass the test, although scored as a single video card, then at 1280x1024 and 1600x1200 soon after the start of the test there occurred a serious deceleration as if it was short for video memory. We only have to recall our recent test at which there were fewer issued with Windows Vista, although there were few anyway.

At Need for Speed Carbon, the advantage from using the second video card grows with the resolution. For instance, at 1600x1200 the speed boost amounted to 46.7%, although at 1024x768 there was almost none. This is another proof of the fact that multiprocessing technologies are biased to hard graphic modes.

At Need for Speed Pro Street Racing, the advantage from using the second video card is seen only at 1600x1200, albeit negligible - merely 13.8%. The issue is most likely related to the drivers since in our recent tests of GeForce 9800 GX2 and a motherboard based on the nForce 790i Ultra SLI chipset the performance gain was almost twofold.

At Prey, the advantage of using two MSI N9800GTX video cards rises with the resolution. On the other hand, this test is not much of being indicative, since even a single GeForce 9800GTX is quite enough for this game.

Final Words

Our today's tests have shown that issues with drivers may come up in the produce by any company, and NVIDIA is no exception at that. Today, we witnessed the poor operation of the SLI technology in some games powered by Windows XP, with the game Need for Speed Pro Street Racing being the most indicative example of that. Hopefully, in the nearest future the developers of drivers will fix these issues (if the issue is rooted in the drivers), and we can again enjoy the impeccable operation of the SLI technology on not only the top-end motherboards like that recently reviewed XFX nForce 790 Ultra SLI but also on older boards. Somehow or other, for completeness of picture, in the nearest future we'll be running a roundup test of the SLI technology on various platforms, and then we make the final conclusion.