AMD Zone heeft een interview gepost met Sean Stanek, ontwikkelaar van de geoptimaliseerde Athlon RC5 core die over enkele dagen gereleased wordt. Met de huidige verbetering trekt deze client een circa 10 tot 15% hogere keyrate uit een Athlon. Met toekomstig geplande verbeteringen hoopt Sean de 3 integer units van de Athlon nog optimaler te benutten:
Q: What kind of performance increase can we expect from the new client?
A: From the soon-to-be-released distributed.net client, I only got about a 10%-15% increase in keyrate. After I implement the next two stages and get it working fine, expect much more keyrate increase OVER the soon-to-be-released client.
Q: Will you be working further to improve the optimizations?
A: Definitely. As I said before, I've already got two more stages planned. As it is, I'm getting two or less instructions executing per clock on the Athlon core, but the processor itself can execute three. The two stages I have planned should get more towards getting three instructions per clock cycle. Although one of my new ideas should help many x86 processors do a little better, Intel processors at the moment can only execute two instructions per clock as a maximum, whereas the Athlon can do three - so there's much room for improvement on the Athlon. In fact, most applications are so poorly written (i.e. by compilers) that they can't even get 1 instruction per clock cycle. I've heard as bad as 0.2 for some commercial products. Since the Athlon can physically do 3, there's a lot of things that could use optimization besides just RC5.