by John Reynolds
Introduction
As demonstrated in SimHQ’s Radeon HD 4850 Performance Preview, AMD has a winner with its new graphics chip, the RV770. But the HD 4850 was only the first round in the salvo being fired at NVIDIA, and today we take a look at the 4850’s big brother: the Radeon HD 4870. Like the HD 4850, the HD 4870 is also based on AMD’s new RV770 chip, though with a higher clock speed: 750MHz compared to the 4850’s 625MHz. The HD 4870 is also the first graphics board that brings GDDR5 to the market. If the HD 4850 gave the GeForce 8800 GTX a run for its money at $200 USD, what will the HD 4870 do to its competition? How would two HD 4870s in CrossFire hold up to one of NVIDIA’s new GTX 280s? Today we find out, but first let’s take a quick look at what GDDR5 memory brings to the table.
The rendering of high resolution, highly textured graphics with complex shaders at fast frame rates demands massive amounts of memory bandwidth. The graphics firms have pursued various approaches to giving their latest GPUs sufficient bandwidth, such as using wider memory interfaces, faster modules of an existing memory standard, or, as with the HD 4870’s GDDR5, quickly moving to a new memory type that supplies more bandwidth per clock cycle compared to older types. One of the main features new with GDDR5 is something called Clock Data Recovery, which is essentially the training process that occurs upon power-up between the memory and the host device (the RV770 chip in this case). This training process syncs the command and address clocks, with the data stream generating the memory’s clock speed, all of which results in fewer signaling issues. This in turn makes the life of board designers a bit easier, since the complexity of the actual PCB of a graphics board using high speed memory makes the layout of trace wires a chaotic mess. With the use of GDDR5, these traces no longer have to be identical in length since the new memory standard and its cleaner signaling is more tolerant of different trace lengths. There are also other features of GDDR5 that help it deliver more bandwidth than GDDR3 or GDDR4, but the main concern for consumers is cost. GDDR5 is currently seeing three vendors ramping up production and prices should start falling in the coming months, giving AMD better price flexibility with the HD 4870s if NVIDIA’s 9800 GTX+ turns out to be a real threat to the Radeon 4800 series.
For this performance preview, SimHQ decided to place the HD 4870 against the HD 4850 to allow our readers to determine whether or not its extra $100 price tag is worth the cash outlay. Because AMD sent us two Sapphire HD 4870s, we also decided to include a GeForce GTX 280 in our benchmarks. This pits the two graphics firms’ current best against each other, though obviously the GTX 280 scores are of interest only against HD 4870 CrossFire numbers. The basic specifications of these various GPUs are as follows:
ATI HD 4850
|
ATI HD 4870
|
NVIDIA GTX 280
|
|
Core speed |
625MHz
|
750MHz
|
621MHz
|
Shader Units |
800
|
800
|
240
|
Shader Units Speed |
625MHz
|
750MHz
|
1.35GHz
|
Onboard Memory |
512MB
|
512MB
|
1GB
|
Memory Speed |
2.0GHz
|
3.6GHz
|
2.27GHz
|
Memory Interface |
256-bit
|
256-bit
|
512-bit
|
Memory Bandwidth |
64GB/sec
|
115GB/sec
|
145GB/sec
|
Texture Units |
40
|
40
|
80
|
ROPs (render back end) |
16
|
16
|
32
|
Price |
~$200
|
~$300
|
~$660
|
The switch to GDDR5 for the HD 4870 gives the faster board an 80% memory bandwidth increase over the HD 4850’s GDDR3. The GTX 280 used for this article was a slightly overclocked board that requires both a 6-pin and 8-pin power connectors and is 10.5” in length. The HD 4870, unlike the HD 4850, is a dual slot board that requires two 6-pin connectors and is 9.5” long, a slight .25” increase over the 4850’s length.