Earlier this month we posted the first part of our “Revisiting AGP” article. Today we have posted the second and final part.

For the most part, in the first part of the last article the AGP cards performed on par with their PCI Express counterparts. In fact, in some cases the AGP parts actually led the tests in terms of performance. The X1950PRO cards, whether they were AGP or PCI Express, performed basically the same.

We continue our tests today starting off with Half-Life 2 Episode 1 and Lost Coast. From there, we will move on to Prey, X3 Reunion, Splinter Cell Chaos Theory, and Serious Sam 2. We will end the article with noise, consumption, temperature tests, and finally, some brief final words from us. The first part of this article covered various synthetic tests, F.E.A.R., and Oblivion, and can be found here.

Half Life 2: Episode 1 and Lost Coast

In the first Half-Life 2 test it is clear that the game performs better with the PCI Express interface at higher resolutions. In total, the PCI Express video cards hold a 10% lead over their AGP counterparts. The PCI Express video cards are also faster at lower resolutions, but the AGP cards are in the ballpark.

With HDR, the reference ATI board and the Sapphire X1950PRO AGP both experience significant performance diminishment. In this test, however, the GPU frequency is what makes the difference, not the card interface, as we see the overclocked X1950PROs, one of which is AGP and the other PCI Express, top our chart.

In the last Half Life 2 test memory plays a fundamental role in performance, and it gives the Sapphire Radeon X1950PRO AGP the edge to outperform all other cards. Interestingly enough, the HIS IceQ3 AGP experiences a very large drop in FPS as the resolution is increased to 1600x1200. We are not sure what the reason behind this huge drop is, however, it might be the result of a combination of smaller memory and slower AGP interface.

Prey is a video game that is based on the now aging Doom 3 engine. We chose this game in place of the conventional, yet very popular, Doom 3 primarily because we wanted to use a more recent game in our tests. We used a retail copy of the game and benchmarked using HOCBench with the hwzone demo. We tested at a variety of resolutions, including 1024x768, 1280x1024, and 1600x1200, with anti-aliasing set at 4x and anisotropic filtering at 16x.

Again we see a repeat of what we saw in the last two Half Life 2 tests. The cards equipped with 256MB of memory end up performing slower than the 512MB Sapphire Radeon X1950PRO AGP. Performance is very similar at 1024x768 and 1280x1024 between all X1950PRO cards, however, right when we move to 1600x1200 the Sapphire AGP X1950PRO easily takes the lead. In addition, in this test we again see the HIS Radeon X1950PRO perform lower than the average frame rate of the X1950PRO cards. The performance drop, though, isn’t as drastic as the one we saw with Lost Coast test.

We also tested using the rolling demo of X3 Reunion. The benchmark allows us to test performance at a variety of resolutions using an integrated utility. We chose to test at resolutions of 1024x768, 1280x1024, 1600x1200 with 4x anti-aliasing and 16x anisotropic filtering.

Similar to the earlier benchmarks we ran, the performance displayed by the X1950PRO cards is practically the same.

The SC tests were run on a freshly installed copy of Splinter Cell Chaos Theory patched to version 1.05. We ran the benchmarks using the Lighthouse demo released by Ubisoft. Our test resolutions were at 1024x768, 1280x1024 and 1600x1200 and we always had 16x anisotropic filtering enabled along with High Dynamic Range (HDR).

This particular test rewards higher operating frequencies, as we see the overclocked HIS X1950PRO AGP take the lead. The performance difference between it and the Sapphire AGP X1950PRO, though, is very slight. Interestingly enough, the HIS AGP X1950PRO performs 10% better than the PCI Express ATI Radeon X1950PRO reference board.

Again, in Serious Sam 2 operating frequencies has a large impact on performance and the higher clocked X1950PRO cards dominate the tests.

To provide a complete picture of the products we tested today, we decided to include noise, temperature, and consumption tests in the article. For our consumption tests, we used the same power supply unit, the same CPU, and the same memory all in an effort to have the AGP and PCI Express cards on the same playing field.

On this page, all performance was measured in two states; idle and stress. In our stress tests, we effectively placed a heavy burden on the graphics subsystem by running Orthos and Shadermark.

To measure the noise of the cards, we placed a noise meter 10cm away from the cards’ cooling fans. To ensure consistency, we also made sure to place the noise meter in the same position every time for each video card. It is important to remember that these tests are mainly useful for comparing the noise between the cards, and not for measuring how loud the cards would be in a normal environment. Our testing was done without the components installed in a PC case, meaning they can’t be used to effectively assess fan noise with the components installed in a case since it is safe to assume that noise is reduced with components in a case.

The GeForce 7600GT turned out to be the loudest card. This attribute can be accredited to its smaller fan, which spins at a higher RPM. It is also interesting to note that the Sapphire X1950PRO gives a low value in idle, however, under stress the noise level raises considerably.

The HIS offerings are much less sensitive to the stress level. The use of a rather large heatsink along with heatpipes is what makes the difference here and they maintain practically the same level of noise going from idle to full load.

We measured GPU temperature using Riva tuner, and our results turn out to be quite interesting. In this test we again see the Sapphire card perform quite well at idle, however, when placed under stress GPU temperature nearly doubles. The HIS cards end up being much more efficient in terms of temperature, thanks to the large cooling system the cards have.

From the aspect of power consumption, for the most part the AGP parts are more demanding than the PCI Express ones. We measured consumption by recording the overall power draw of the system. This means that there might be slight performance variations due to different motherboards being used. The RAM and the CPU, however, were the same across the test platforms.

The largest difference between the PCI Express and AGP interface is the maximum bandwidth that can be obtained. AGP 8x can only offer 2.133 GB/s peak, while PCI Express x16 can offer 4 GB/s. As has been beaten into our heads ferociously by marketers for the past couple of years, this means that PCI Express is able to theoretically offer double the performance of AGP cards. It is important to remember, though, that this is all in a theoretical situation where there is 100% efficiency. For the most part, the benchmarks that were run today show us that AGP and PCI Express perform practically the same in the mainstream market segment.

What we are seeing in the video card market is similar to the situation of hard drives. Although the SATA 3.0 Gbit/s interface has a bandwidth three times that of Ultra DMA ATA 133, the actual performance difference between the interfaces is quite small, meaning we still haven’t been able to fully exploit all that the SATA interface has to offer.

This situation is being mirrored in the video card market creating a dilemma for users with older AGP systems. They are left to make the choice to buy a completely new system or just upgrade the video card to a newer AGP one. In most cases, we saw very little performance difference between AGP and PCI Express cards. There were some tests where there were small performance differences; however, the PCI Express interface’s victory margin wasn’t large enough to actually draw a conclusion off of. In fact, there were also a handful of tests where AGP came out to be the victor.

The final decision really falls on budget. It is a known fact that upper end performance AGP cards are scarce. We did see some at CeBIT this year , but we still need to get some more information on when products like these will start hitting the market along with their pricing.

There are a number of things to keep in mind before actually making a decision. First and foremost, upper end cards are always being released in PCI Express 16x first, meaning you won’t be able to find a GeForce 8800 GTX AGP right now or anytime soon. This means that if you really want the latest cards immediately in your computer, you might want to look into a new system.

The cost of a new system is also something to keep in mind. Because of the market transition from DDR to DDR2, there won’t be many components in your system you’ll be able to reuse. This fact alone will hike up the price of a new computer. The AGP cards we reviewed today, on the other hand, come for less than $200, much less than the cost of getting a completely new PC.

The bottom line is this. If you have the choice of upgrading your old AGP system, or instead making a completely new system on a tight budget that only has a mainstream card like the X1950PRO, from what we’ve seen today you would be better off just upgrading to a new AGP card. Spending the money for a completely new PCI Express system that has a midrange video card is pointless considering that there isn’t much of a performance benefit at the moment going from AGP 8x to PCI Express. This will allow you to, at least for right now, enjoy the best of both worlds. You’ll be able to enjoy better gaming performance at a low cost and successfully extend the life of an older AGP system.