AMD Radeon HD 3870 X2 Performance Preview

For years processor manufacturers AMD and Intel were locked in a brutal clock speed war. The Athlon and Pentium III raced to hit speeds of 1GHz and beyond, all in an effort that was as much about bragging rights as it was about actual CPU performance. Eventually as clock speeds ratcheted ever higher, AMD, and eventually Intel, realized that this strategy wasn’t sustainable – as clock speeds increased, so did heat. AMD and Intel’s focus shifted from sheer clock speeds to instructions per clock cycle (IPC); thus bringing us more efficient CPUs that didn’t run quite as fast as their predecessors in terms of clock speed, but still delivered better overall performance thanks to improved IPC.

Ultimately this approach led us to CPUs with multiple processing cores. The first dual-core CPUs were released in 2005, and today’s latest Core 2 and Phenom CPUs sport four processing cores.

This leads us to graphics.

Today’s latest high-end graphics processors contain over 650M transistors. In comparison, a quad-core Penryn CPU contains 820M transistors. Unlike a high-end GPU, the majority of those 820M transistors inside Penryn consist of its massive 12MB L2 cache, so you can argue that a GPU is a more complicated design. In creating the G8x GPU inside the GeForce 8 series, NVIDIA has stated that they spent half a billion dollars in R&D and that the GPU took four years to develop.

Quite simply, because these devices are growing more complicated, developing a cutting edge GPU like R600 or G80 is taking longer than ever and both AMD and NVIDIA are spending hundreds of millions of dollars in R&D to bring them to market. Because of this, the days of the 6-month product cycle didn’t occur this generation, and it’s possible that other than die shrinks, we may not see them again.

This puts AMD and NVIDIA in a difficult situation: how do you continue to deliver dramatic breakthroughs in performance if it takes longer and is more expensive to develop a next-generation high-end GPU? Simple, you add more GPUs to the graphics card itself.


This is exactly what NVIDIA did with their GeForce 7900 GX2 and GeForce 7950 GX2. By integrating two G71 GPUs onto one board, NVIDIA was able to deliver performance levels that were similar to two comparable graphic cards while only using one PCI Express graphics slot. Back in 1999, ATI did something similar with their Rage Fury MAXX card, which consisted of dual Rage 128 Pro chips.


The AMD Radeon 3870 X2 follows this same philosophy. As its name implies, the card essentially consists of two Radeon 3870 GPUs that have been grafted together onto one PCB and cooler. But AMD has made a couple of tweaks to enhance the performance of each GPU…