(Early) Conclusions

This concludes our first date with the first DX11 GPU ever, and it's been an interesting one, concluded with a hasty departure of the HW that left us heartbroken...well, not really, but many unanswered questions were left behind. Cypress is definitely an impressive bit of manufacturing and design, but it's not by any means revolutionary. It's a refinement of an architecture that worked well enough up to now...and we can't fault that, really. Adding full DX11 support and rounding off some pretty sharp corners all-over the chip is definitely nothing to sneeze at.

However, we are convinced this is just a first step -- an introduction to a story that will unfold across many months, with many interesting pages left to be written. We're eagerly looking forward to continuing our exploration of the hardware, since we feel there's still quite a bit we haven't seen.

Where there may be some disappointment, based on what your expectations were,  is in the fact that all that hardware almost never doubles the 4890 in practice, although the specs would appear to promise it. At least that's the case now, going by (the few) performance tests we're run ourselves, and by what reputable outfits like TechReport or Hardware.fr are showing. Will it change in the future, with driver growth? It's possible, although we can see some obvious hardware limits that drivers won't fix, unless some hidden magic is yet to be revealed. On the flipside, a 5870 is probably the best card you can buy now. Single card multi-GPU solutions may defeat it at times, but those come with their own baggage. For how long will the 5870 be the best card you can buy? Well, to find out about that we'll all have to wait a bit longer.

Finally, in closing, we'd like to thank Joe Macri, Richard Huddy, Jon Story and Mike Houston for putting up with our (numerous) questions and providing invaluable insight into the inner workings of the chip.