The Chip

It's interesting how the wheel of reincarnation turns when it comes to naming, since Cypress' first implementation bears a moniker rather close to a rather notorious graphics product, the NVIDIA GeforceFX 5800, by virtue of the ATI Radeon HD 5800 series naming scheme. What we have on our hands is the 5870 SKU, which will spearhead ATI's initial offensive, being the highest end iteration available today. It's also mind-bogglingly complex, and impressively dense, as you can see on the left.

Fitting 2.15B transistors into a 334mm2 die is quite a feat, and using TSMC's 40nm process to do it gets you extra bonus points, since it has had a troubled life-cycle, with a pretty difficult ramp-up. Doing some quick, simple math, we can see that density sits nicely at around 6.43M transistors per mm2, almost doubling what the RV770 achieved...and that one already blew our minds a bit with its density! All of those petite transistors join forces and conspire to make up a fully DX11 compliant GPU that's roughly double the RV770 in everything but the memory interface and the appetite for power. Focusing on the latter, Cypress supports both engine and memory down-clocking and downvolting (RV770 didn't down-clock or down-volt its GDDR5), and ATI claims it'll only suck about 27 W at idle and 188 W under load, which is actually slightly less than a similarly clocked RV790-powered Radeon HD 4890 – not bad at all!

In terms of the memory interface, we're not sure if GDDR3-using derivatives will ever see the light of day, but we'd lean towards saying no given current GDDR5 pricing. Having said that, let's see what the ATI Radeon HD 5870 is all about (note, that's the last time we'll use its naming in extenso!).