AMD RADEON HD 4890 Print E-mail
Written by Michael Schuette   
Apr 09, 2009 at 12:03 AM

The longer I am in the review business, the more the evolution of personal computer performance strikes me as nothing short of breathtaking. Aside from the raw CPU performance, the main improvements are in the world of computer graphics. Needless to say that neither one of the two big survivors in today’s graphic scene was even around when we played the first computer games on Amiga and Atari ST machines. Today we have AMD formerly known as ATI and nVidia battling it out over and again. Missing in action are Hercules, Tseng Labs, ALi, SiS, Kyro and Trident. Everybody else, including Intel, Via and Matrox, is just playing a wallflower role with the only realistic chance to break into the duo-poly coming from Intel’s mystery architecture called Larrabee.

For the time being, the market is ruled by the red and the green teams pursuing somewhat different strategies with respect to clock speed vs. parallelism. At this point we are not too concerned with nVidia, PhysX or Cuda - courtesy of their PR department - instead we are focusing on AMD’s latest and greatest product available as RADEON 4890 and/or RADEON 4890 OC with the moniker designating higher clock speed for core and memory. Most importantly, the RADEON 4890 will also be the last high end GPU released by AMD that is manufactured on a 55 nm process before the 4700 series based on a 40 nm design kicks in.

What’s New?

For those who have followed the 4800 series, there are no big surprises, the RADEON 4890 is another iteration of the same concept that was already present in the 4850 and 4870. That said, the devil is always in the detail or else, it is 10% genius and 90% hard work. In terms of hardware, the design is 10% of the way, the rest is fine tuning and tweaking to get the maximum performance. Looking at the RADEON 4890, certain parallels to AMD's Thoroughbred processor and its re-spin come to mind. In fact, this is one of the cases where the symbiosis between AMD and (former) ATI seems to bear some fruit in that some solid optimization expertise brought in from Austin meets the ingenuity from Markham and the result is…

Who would have thought, there is a ring of decoupling capacitors around the GPU die. Arguably, putting it like this is a very simple-minded view of the effort resulting in the RADEON 4890. If it was as simple as throwing in a bunch of capacitors, anybody could optimize their dies to warrant the best performance. In real life, it is just not quite as simple as anybody who has ever designed ICs will be able to confirm. Details, though, of what AMD did to their 4870 die are not publicly available and there is probably a reason for that. Typically, those reasons are called trade secrets and we won’t dig any further – otherwise potentially Rambus would call us and try to get a slice of the pie. Fact of the matter , however, is that the tuning of the die consumed another three million transistors in various places, and as a result of the entire make-over, the RADEON 4890 not only clocks higher but it also consumes relatively less power.

Last Updated ( Oct 23, 2009 at 05:36 AM )
<Previous Article   Next Article>