AMD Strategy, what's in a name, twin-GPU monster
Let's delve into AMD's GPU roadmap to see where the new GPU fits in.
This slide is part of a presentation given by AMD before the launch of the Radeon HD 5870.
AMD wants to position the DX11 architecture from $75 through to $599, and it's using the same basic design to do so. Please head on over to here to see how the architecture provides a scalable solution.
We will see sub-$100 5000-series cards in a couple of months' time, codenamed Redwood and Cedar. They will likely replace the current Radeon HD 4600/4700/4830/4850 cards.
Pertinent to today, at the very top of the picture is the Radeon HD 5970. It is a dual-GPU card that uses a mixture of HD 5870 and HD 5850 technology.
Radeon HD 5970 - what's in a name?
Previous dual-GPU cards from AMD have been identified with the suffix 'X2' - Radeon 4870 X2, 4850 X2, and 3870 X2 being obvious examples.
In the numbering game that's been played by three major PC-related manufacturers - Intel, AMD, NVIDIA - the suffix has been dropped in favour of a higher model number. We know that AMD was toying with this idea last year, almost calling the HD 4870 X2 an HD 4970, but now it's official.
Twin-GPU monster
The following slide shows how AMD has created this juggernaut of a card, codenamed Hemlock.
As we alluded to earlier, the new card is a pastiche of the two single-GPU Radeon HD 58xx cards. The basic setup is double that of a Radeon HD 5870, evinced by the (combined) 3,200 stream processors, 160 texture units, and 64 ROPs.
However, the frequencies of the card are those of the lower-clocked Radeon HD 5850. The vagaries of compute performance (frequency x shader) and texturing mean that the 4.64TFLOPs and 116 GTexels/s are more than double a HD 5850's, thanks to the underlying HD 5870 architecture, but pixel fillrate, Z/Stencil, and memory bandwidth (all based on speeds) are exactly 2x.
Somewhat tenuously, the mix-and-match card doesn't deserve the HD 5870 X2 nomenclature, says AMD, because it isn't 2x that card, for the reasons detailed above.
