Test Setup And Methods

Whilst we did strive to ensure optimal accommodations for Cypress, our testing rig isn't on the bleeding edge, falling a bit short of the latest and greatest Core i7 magic. However, at least for the very specific tests we use to outline general architectural traits, it should be more than sufficient, although games may be a different story altogether. We've toyed both with Windows 7 and with Vista SP2, in their X64 flavours, but we've finally opted for Vista since that played nicest with our tools. The drivers we used are incredibly fresh, so fresh that their compile date is less than 2 weeks before launch, and we actually had to wait for them to come out before we could use the card, which was pure torture!

As for testing methods themselves, they're roughly placed on the same coordinates you've come to love and know many moons ago, when we authored new binaries for G80, with focus being placed in illustrating if the hardware hits its theoretical rates and on seeing how it behaves when faced with certain workloads. Make no mistake here, the tests included in an architecture analysis are highly synthetic ones, meant to show us how the cogs and gears interact, and if everything is in working order, so if you see Cypress having twice the instruction issue rate don't immediately assume that Crysis FPS will be doubled too.

We're using a wide variety of tools, most in-house developed (and in need of a refresh!), but also freely available tools written by many great guys involved in this industry. We've not used any DX11 code in this outing, in spite of it being alive and well both in Windows 7 as well as in Vista, and drivers properly exposing support, mainly because we're deferring that to somewhere in October to coincide with its official launch. We have however checked some of the new SDK samples, so you may see them mentioned once or twice. Now cry havoc, and let loose the dogs of analysis (don't do it in a very loud fashion though, your neighbours may not be amused).