Ex AMD designer: Bulldozer to disappoint

**demonkevy666** · 06-17-2010, 01:17 PM

Originally Posted by BeepBeep2

And you milk out ~80% more operating efficiency?

no you miss understand.
I'm not milking anything out it's common scaling from one to two.
just like in gpu's.

example
1 cards 100% vs 1 card
2 cards 190% vs 1 card you'd like 200%
3 cards 260% vs 1 card you'd like 300%
4 cards 320% vs 1 card you'd like 400%

since it can't be perfect you get that first number instead.

(that's not bad scaling when compared to one card for crossfire)
scaling on cpu's can be over came by clock speed look at cinebench 11.5 for examples.

**BeepBeep2** · 06-17-2010, 05:16 PM

Oh.

**Donnie27** · 06-18-2010, 07:02 AM

Originally Posted by demonkevy666

I think those that conclusion is form the comment that it 1.8 times performance for dual thread instead of the 1.9 times scaling is with normal added cores.

1.8 times over what? Please explain where they started? Doesn't matter if they scale 1.8 or 1.9 times if each core is 25 to 40% slower depending on the app. NOTE, the 25 to 40% is JUST an example!

I gave up on AMD (add to that Intel, nVidia and everyone else) estimates and models. And this BS about hitting on all cylinders is a crock IMHO!

**demonkevy666** · 06-26-2010, 09:13 AM

Originally Posted by Donnie27

1.8 times over what? Please explain where they started? Doesn't matter if they scale 1.8 or 1.9 times if each core is 25 to 40% slower depending on the app. NOTE, the 25 to 40% is JUST an example!

I gave up on AMD (add to that Intel, nVidia and everyone else) estimates and models. And this BS about hitting on all cylinders is a crock IMHO!

Over single core to two cores. what was so hard to get ?
the only way to get above 1.9 is higher IPC per core.

for a the only real example I can give is gulftown 12 threads vs MC 12 cores on cinebench 11.5 :shurg:

MC scores 14.85 @ 2.2ghz.
gulftown scores 12.04 @ 4.5ghz

they MC has lower IPC Core core but scales better because it's more real cores.

how ever the single thread score can be throwing that off completely, since clock speeds are far apart.

I couldn't find any encoding test with the Opteron 6174 :/ or else I would have used that.

**Chumbucket843** · 06-26-2010, 04:22 PM

Originally Posted by demonkevy666

Over single core to two cores. what was so hard to get ?
the only way to get above 1.9 is higher IPC per core.

not true.

superlinear speed ups are possible especially with the memory wall.

**~~pcsky3~~** · 06-27-2010, 07:04 PM

Originally Posted by Chumbucket843

not true.

superlinear speed ups are possible especially with the memory wall.

yeah, you might right

**Donnie27** · 06-29-2010, 11:13 AM

Originally Posted by demonkevy666

Over single core to two cores. what was so hard to get ?
the only way to get above 1.9 is higher IPC per core.

for a the only real example I can give is gulftown 12 threads vs MC 12 cores on cinebench 11.5 :shurg:

MC scores 14.85 @ 2.2ghz.
gulftown scores 12.04 @ 4.5ghz

they MC has lower IPC Core core but scales better because it's more real cores.

how ever the single thread score can be throwing that off completely, since clock speeds are far apart.

I couldn't find any encoding test with the Opteron 6174 :/ or else I would have used that.

http://www.bit-tech.net/hardware/cpu...x5650-review/1