Yes, worth the wait.
Look, from the way the fabs were shaping up a lot of us who were not over projecting were looking at a 1:8 cell under 4GHz. 3.2GHz is surprising a little, but a 1:7 makes sense (and I believe nAo even had mentioned that a few months ago as a possiblity due to yields).
Still, you are looking at 218GFLOPs and an INSANELY fast memory architecture/bus.
What I think is more surprising is how the xCPU outdid itself. Many of us were skeptical that it would break 3GHz. In fact they hit 3.2GHz and hit 115GFLOPs. Impressive indeed.
Not as impressive as 218GFLOPs, but these are just all paper numbers. I tend to think the CELL, once gotten under control (especially the SPE cache size) it will be better because it is designed from the ground up to be multithreaded/stream.
The CELL also is totally scalable--to judge CELL this gen misses the big picture. PS is now officially a PLATFORM. I suspect we may see something like a 4:32 CELL @ 8GHz in 6 years, maybe even 8:64. The tools will already be in place, and progammers will have the basics handily whipped. And since it is built to scale I see the PS4 as potentually a "dream machine". Not sure what MS will have an answer for this... there will be something, but I doubt anything as high performing as CELL. Call me pessimistic. But we have already begun to see diminishing returns soo...
I think CELL lived up to its expectations--at least mine.
xCPU over achieved. It should be fairly easy to program for also, it has 1MB of L2 cache (compared to 512K for the CELL). And I really see the 3 PPC cores downplayed.
Maybe someone here can answer this: The PPC cores take up a ton of realestate, much more than a SPE. But we tend to focus on the SPE/VMX units. Obviously the PPC core is valiable, if not MS could have gone with a PPC with 8 VMX--but they did not. The PPC is good at Integer and basic processing.
True, it may not be great at physics (VMX for those) or vertex shaders (VMX and VS units on the GPU), but it can do a lot right.
So, what am I missing? If the PPC cores are useless why use all the realestate for them? Why do we rarely mention the performance of the PPC cores? If the PPC cores are usless for gaming why did not MS just go with more VMX units or something? PPC are useful for something, right?
Look, from the way the fabs were shaping up a lot of us who were not over projecting were looking at a 1:8 cell under 4GHz. 3.2GHz is surprising a little, but a 1:7 makes sense (and I believe nAo even had mentioned that a few months ago as a possiblity due to yields).
Still, you are looking at 218GFLOPs and an INSANELY fast memory architecture/bus.
What I think is more surprising is how the xCPU outdid itself. Many of us were skeptical that it would break 3GHz. In fact they hit 3.2GHz and hit 115GFLOPs. Impressive indeed.
Not as impressive as 218GFLOPs, but these are just all paper numbers. I tend to think the CELL, once gotten under control (especially the SPE cache size) it will be better because it is designed from the ground up to be multithreaded/stream.
The CELL also is totally scalable--to judge CELL this gen misses the big picture. PS is now officially a PLATFORM. I suspect we may see something like a 4:32 CELL @ 8GHz in 6 years, maybe even 8:64. The tools will already be in place, and progammers will have the basics handily whipped. And since it is built to scale I see the PS4 as potentually a "dream machine". Not sure what MS will have an answer for this... there will be something, but I doubt anything as high performing as CELL. Call me pessimistic. But we have already begun to see diminishing returns soo...
I think CELL lived up to its expectations--at least mine.
xCPU over achieved. It should be fairly easy to program for also, it has 1MB of L2 cache (compared to 512K for the CELL). And I really see the 3 PPC cores downplayed.
Maybe someone here can answer this: The PPC cores take up a ton of realestate, much more than a SPE. But we tend to focus on the SPE/VMX units. Obviously the PPC core is valiable, if not MS could have gone with a PPC with 8 VMX--but they did not. The PPC is good at Integer and basic processing.
True, it may not be great at physics (VMX for those) or vertex shaders (VMX and VS units on the GPU), but it can do a lot right.
So, what am I missing? If the PPC cores are useless why use all the realestate for them? Why do we rarely mention the performance of the PPC cores? If the PPC cores are usless for gaming why did not MS just go with more VMX units or something? PPC are useful for something, right?
