Outline ·
[ Standard ] ·
Linear+
AMD Bulldozer & Bobcat
|
dstl1128
|
Oct 11 2011, 09:27 AM
|
|
Any benchies on multiple VM doing stuff (eg. compiling kernel source)?
The 4x2 integer cores might be helpful here. Curious to know 4x2 int cores vs 4 int cores with HT. If 4 + HT wins in this department then AMD need to try harder again.
p/s: I don't believe 4 BD floating-point cores could win 4 SB floating-point cores anyway.
|
|
|
|
|
|
dstl1128
|
Oct 13 2011, 07:33 AM
|
|
Bulldozer actually looks good as it inline to the kind of workloads I do often - lots of integer calculation.
The only disappointment is the power consumption.
|
|
|
|
|
|
dstl1128
|
Oct 13 2011, 10:25 AM
|
|
Better than Intel: » Click to show Spoiler - click again to hide... « About same as Intel: » Click to show Spoiler - click again to hide... « It is a heavily threaded integer monster. Goes floaty things turns out bad. Gamers shouldn't get it. It is priced between 2600k & 2500k, so performance seems ok. But the power consumption for such new chip... is too high.
|
|
|
|
|
|
dstl1128
|
Oct 18 2011, 08:24 AM
|
|
QUOTE(dma0991 @ Oct 17 2011, 11:34 PM) Even when Windows 8 comes, not much improvement can be seen in MT tasks but might improve ST performance slightly. The current Windows 7 does not recognize BD architecture that well so if a program wants a thread/core, it will go in the order of Core 0,1,2,3,4,5,6,7. This is not what the AMD engineers wanted because if a thread were to use Core 0 & 1 for example, it will share the front end resources and create a bottleneck. What AMD engineers wanted with Windows 8 it that will schedule threads better so it will use in the order of Core 0,2,4,6,1,3,5,7 so there will be no sharing of front end resources and improve ST performance slightly. There might not be any MT performance increase from this newer thread scheduling because all 8 cores are used unless Windows 8 somehow makes use of the new instruction sets that come with BD. They should advertise, get 8 cores in your system with Windows 8! Chinese will love the string of lucky numbers.  Not as easy as it sounds. Windows (since the P4 HT days) already have affinity on even cores (i.e. priority on CPU0 and CPU2, and then CPU1 & CPU3). By right it should work in BD as well. Just that if two threads were sharing the same data but being schedule to different module, then syncing the two modules' cache would have very high performance penalty and could just drag everything down the drain. I don't think Win8 would help much in BD score (maybe I need to eat back my words later  ). It just need wider front end and the internal ring bus like SB.
|
|
|
|
|
|
dstl1128
|
Oct 18 2011, 08:54 AM
|
|
Well if two threads have a lot of shared data and end up falls in the same module then this should be intended.
It is only ok to be schedule to another BD module if they have separate data.
We will see how it goes. Seems like I need to wait for IB.
|
|
|
|
|
|
dstl1128
|
Oct 26 2011, 03:29 PM
|
|
QUOTE(dma0991 @ Oct 26 2011, 02:24 AM) Why the Cinebench 11.5 multi-threaded result so much difference from the rest? 2600k just 0.01 point more than stock FX8150?
|
|
|
|
|
|
dstl1128
|
Oct 27 2011, 08:25 AM
|
|
QUOTE(kingkingyyk @ Oct 26 2011, 03:36 PM) Cinebench fully utilize the whole processor. Result for "8 cores" is definitely normal. What I mean was the result is 'different' from other review sites Cinebench 11.5 benchmark result.
|
|
|
|
|
|
dstl1128
|
Nov 2 2011, 10:23 AM
|
|
With that high frequency, it probably beats its own record of power consumption too.
|
|
|
|
|