AMD vs. Intel
21 posts
• Page 2 of 2 • 1, 2
by Bobby » Tue Jul 17, 2007 11:29 am
I'm not saying more than two cores are bad, but for most applications they're not needed.
The Cell is a very specialized CPU, and quite different from other designs. It has one general
purpose core, the PPE, part of the PowerPC family, and eight SPEs which aren't general purpose
cores - they're independent engines with no branch prediction, and with their own local memory. If
the SPEs want to access main memory, they have to do it through the PPE.
Also, the Cell executes instructions in-order. This is a key point, because most popular platforms
(x86) are out-of-order execution. Those CPUs have hardware (the instruction window) that schedules
and reorders instructions to keep its execution units busy - this is ILP, instruction level parallelism.
The programmer doesn't need to do much to get this benefit.
In the Cell and Xenon (also in-order,) that task falls to the programmer and compiler - this is TLP,
thread-level parallelism. Suddenly the programmer has a lot more low-level stuff to worry about.
I'd think the Xenon would be more flexible, since you have three identical cores to work with,
whereas the Cell has more raw power, but less flexibility in certain cases where you need the
PPE. It's also interesting to note that the Xenon's cores and the Cell's PPE are based on the same
design.
Three cores seems like a good number, because you've got at least two big subsystems that don't
thread well - AI and physics. Run each on a core, and the third core for the rest... but there's the
downside that AI and physics are branchy code, and would do best on an out-of-order chip.
I'm pretty sure I said this in another thread, but the three next-gen consoles stack up like this:
Wii: easiest to code, uses a standard out-of-order PPC, but also lowest ceiling.
XB360: harder to code, higher ceiling. It'll take a couple of years to fully exploit the hardware.
PS3: hardest to code, highest ceiling. Will they take full advantage of it in 5 years?
Also, much of the ceiling is in visual richness, not in the things that make games fun to play (AI,
physics.) It's an era of style over substance.
There's also development cost. The more your programmers have to learn to support a new
architecture, the more it's going to cost to produce that game. Can you make as big of a profit
writing games for the Cell vs. the Wii?
-b
p.s. If you haven't read it, here's Jon Stokes on gaming performance on the Cell and
Xenon, based on the hardware, and here are some from developers.
The Cell is a very specialized CPU, and quite different from other designs. It has one general
purpose core, the PPE, part of the PowerPC family, and eight SPEs which aren't general purpose
cores - they're independent engines with no branch prediction, and with their own local memory. If
the SPEs want to access main memory, they have to do it through the PPE.
Also, the Cell executes instructions in-order. This is a key point, because most popular platforms
(x86) are out-of-order execution. Those CPUs have hardware (the instruction window) that schedules
and reorders instructions to keep its execution units busy - this is ILP, instruction level parallelism.
The programmer doesn't need to do much to get this benefit.
In the Cell and Xenon (also in-order,) that task falls to the programmer and compiler - this is TLP,
thread-level parallelism. Suddenly the programmer has a lot more low-level stuff to worry about.
I'd think the Xenon would be more flexible, since you have three identical cores to work with,
whereas the Cell has more raw power, but less flexibility in certain cases where you need the
PPE. It's also interesting to note that the Xenon's cores and the Cell's PPE are based on the same
design.
Three cores seems like a good number, because you've got at least two big subsystems that don't
thread well - AI and physics. Run each on a core, and the third core for the rest... but there's the
downside that AI and physics are branchy code, and would do best on an out-of-order chip.
I'm pretty sure I said this in another thread, but the three next-gen consoles stack up like this:
Wii: easiest to code, uses a standard out-of-order PPC, but also lowest ceiling.
XB360: harder to code, higher ceiling. It'll take a couple of years to fully exploit the hardware.
PS3: hardest to code, highest ceiling. Will they take full advantage of it in 5 years?
Also, much of the ceiling is in visual richness, not in the things that make games fun to play (AI,
physics.) It's an era of style over substance.
There's also development cost. The more your programmers have to learn to support a new
architecture, the more it's going to cost to produce that game. Can you make as big of a profit
writing games for the Cell vs. the Wii?
-b
p.s. If you haven't read it, here's Jon Stokes on gaming performance on the Cell and
Xenon, based on the hardware, and here are some from developers.
by [spoon] » Tue Jul 17, 2007 11:04 pm
is it true that the power of the PS2 remains largely untapped because of the unique programming required to fully utilize the design?
both articles were interesting reads and personally i enjoy the wii because of its unique interface. i cannot for the life of me properly use an xbox controller. a ps2 controller i have had a little more luck with but its still far from a keyboard and mouse.
again personally i see consoles as party systems. mario kart style. for first person shooters i want more than 3 buddies. (maybe 20+ player online games may have ruined this for me) but looking around for your 2 or 3 opponents on a map that you could have 16 v 16 is just lame.
one more question...
could you have specialized processors for tasks requiring out-of-order operations, and then another set for ordered operations?
both articles were interesting reads and personally i enjoy the wii because of its unique interface. i cannot for the life of me properly use an xbox controller. a ps2 controller i have had a little more luck with but its still far from a keyboard and mouse.
again personally i see consoles as party systems. mario kart style. for first person shooters i want more than 3 buddies. (maybe 20+ player online games may have ruined this for me) but looking around for your 2 or 3 opponents on a map that you could have 16 v 16 is just lame.
one more question...
could you have specialized processors for tasks requiring out-of-order operations, and then another set for ordered operations?
21 posts
• Page 2 of 2 • 1, 2
Who is online
Users browsing this forum: No registered users and 162 guests