June was an exciting month in the discrete GPU market. Impressive releases from both AMD and NVIDIA left the companies battling for your hard-earned dollar. As a result, we’ve seen blistering performance, fierce competition, and unprecedented price cuts. Today we’ll be looking at AMD’s current top-end single-GPU card, the Radeon HD 4870.
The AMD Radeon HD 4870
Released on June 25th, the HD 4870 is based on the new RV770 core. While the new GPU owes much to ATI’s previous R600 core, it has received numerous improvements to produce a super-efficient GPU. AMD has managed to harness the power of 800 stream processors to produce 1.2 teraflops in a 260mm^2 package, all with no die shrink. Let’s take a look at the details for this series:
Under the hood
The HD 4870 comes equipped with GDDR5, which is a massive improvement over GDDR4 and prior specs. There are a plethora of improvements for GDDR5 that range from error detection, to an entirely new clock structure. The simple beauty of GDDR5 is that it offers insane memory bandwidth at low latencies.
You could even say that GDDR5 operates at a quad data rate due to the fact that it has twin I/O links in parallel. That’s right, quad data rate. The sheer power of GDDR5 allows the 900MHz memory to run at an effective speed of 3.6GHz which provides 115.2GB/s of theoretical bandwidth. This also means that, clock for clock, AMD’s 256 bit memory bus will offer similar bandwidth to NVIDIA’s 512 bit bus using GDDR3. The card we’ll be looking at today features ICs from Qimonda, which offers a white paper (PDF) on GDDR5 if you’d like the nitty-gritty.
In the end it’s about performance and value, so let’s dispense with the theory and see what the HD 4870 has to offer. Note that we will be running benchmarks but not comparison benchmarks, so this is not a full technical review.
Articles RSS