Radeon 3850 AGP Plus Single-Core CPU

Test System and Configuration

Once again, we will be benchmarking on the unique ASROCK 939Dual-Sata2 board, because of its ability to offer both a full-speed AGP 8x slot and a full-speed PCIe 16x slot. This allows us to test the difference the bus speed makes on similar cards, which in the past has been very useful in showing us that there isn’t a heck of a lot of real-world difference between AGP and PCIe bandwidth.

We will be testing Powercolor’s Radeon 3850 AGP against a variety of cards that represent a good cross section of what’s available out there, both new and what’s already in older systems.

For a baseline reference against older AGP cards, we’ve included an AGP Geforce 6600 GT. This card used to be the mainstream leader, and has comparable performance to the older AGP cards like the Radeon 9800 XT, 9700 PRO, and X700, as well as the newer X1600 PRO.

We included the Radeon 2600 XT because it is probably the best price/performance card available for AGP right now, and it performs very closely to other AGP options such as the Geforce 7600 GT, Geforce 6800 Ultra, and Radeon X800 XT. Although we couldn’t get our hands on an AGP version of the 2600 XT, our previous AGP analysis showed absolutely no performance difference between the similarly performing 7600 GT AGP and 7600 GT PCI Express, so we are satisfied that the AGP bus will not noticeably slow down this class of card.

Of course, the review wouldn’t be complete without the previous king of AGP, the Radeon X1950 PRO 512MB. This is still a very good gaming card, and it will be interesting to see if the 3850 has enough legs to leave the X1950 behind before the CPU bottlenecks the system. The X1950 PRO performs closely to the Geforce 7900 GS, another of the fastest AGP cards available, so this is another indicator we can use when considering how alternatives fit in the broad spectrum of performance.

We will note that our X1950 PRO sample had overheating problems when playing the newest titles, and the Catalyst driver would report that it was lowering clock speeds to manage this problem. We solved the issue with a Thermaltake Cu-900 aftermarket GPU cooler that made the heat issue go away in short order.

The final card we are testing is a Radeon 3870 PCI Express model. The idea here was to demonstrate if there is a difference in performance between the AGP and PCIe bus with a similar card. We couldn’t get a hold of a 3850 PCIe, so we underclocked our 3870 PCIe to the same core 668 MHz clock speed of the AGP 3850, in the Catalyst Control Center’s Overdrive panel. Unfortunately, Overdrive wouldn’t support memory underclocking, so we were stuck at a much higher 1125 MHz memory speed than the AGP 3850’s 828 MHz. As you will see in the benchmarks, however, this had almost no impact.

We chose Windows Vista 32-bit over Windows XP for testing, and we had a few good reasons for doing so. First, we wanted to see if Vista features like DirectX 10 worked with the AGP versions of the Radeon 3850 (it did). Second, we wanted to see if Vista was a viable operating system on older hardware (it was). Third, and most important, we like to choose the worst-case-scenario for our benchmarks, so that we can deliver a realistic expectation of the kind of performance the end user will see.

Because we chose Vista, we wanted at least 2 GB of RAM. Unfortunately, with older DDR being quite scarce nowadays, we had to use different memory with different timings to make this happen. This bit us in the butt later when we tried to overclock, but once again, it’s probably a good representation of the kind of long-standing previously upgraded systems out there in the wild, and what their owners have to deal with.

With all of these factors in play, we proceeded to benchmark the cards on the newest and most demanding games we could find. Is your old AGP platform up to snuff with nothing but a video card upgrade? Let’s find out!