cleeve :
Desktop Piledriver (FX) will use socket AM3+, APU Piledriver (A6 to A10) will use socket FM2 on desktop
Cleeve could you get in contact with the benchmarking crew and answer a few questions. I posted these on the forums.
Can any of the mod's here contact the THG crew and see if they'll answer a few questions.
#1 Was any of those tests done while using processor affinity to force Turbo Boost?
Doing all sorts of tests on my Llano I can say with absolute authority that TB will not activate if you let Windows handle your process assignments. Windows will keep moving a thread around to different "under utilized" CPUs which prevents turboboost from kicking in. Putting my CPU at factory settings, I was able to force 1~2 cores to max speed using processor affinity, setting affinity to auto it would never hit max speed.
#2 Was there any tool provided from AMD or off the intarwebs that allows them to play with P / B states?
This would be good for analysis and seeing how much headroom there is on each CPU.
#3 Did someone check with a different set of memory?
The A8-3500M only supports DDR3-1333, that is why the DD3-1600 memory would only run at 1333 speeds. The fact that your getting IDENTICLE bandwidth measurements on a 3500M and a Trinity tells me something is wrong in the memory subsystem. The A8-35xxMX series was the Sabine line that supported DDR3-1600 memory, and it's more bandwidth then a 3500M. Memory bandwidth is incredibly important for the graphics system that your benchmarking.
Finially a comment, a 2.05 is still horrible for CB11.5. Can a quick test be done with CB11.5 on Single Threaded, then affinity lock the CB process and do another Single Threaded test and check the results. Will let you know what kind of improvement to expect in single / dual threaded applications.
Add to the list of quoted messages Quick edit Edit this message
I noticed a few things not quite right (not faulting the THG crew at all). First is that the A8-3500M only supports DDR3-1333, its only the A8-35xxMX models that support DDR3-1600. Thus any benchmarks run with the 3500M would be set at 1333 speeds. The Trinity is supposed to support DDR3-1600 memory yet it scored the same as the 3500M for memory bandwidth which is something I've seen on several other sites. This has lead me to believe the reference box that was sent out might not actually be using the memory at 1600 speeds. My HP-DV6 has a 3550MX and two sticks of DDR3-1600 but it's refusing to use them at that speed, it's really picky that way.
Also I believe turbo-core (boost whater) is not working on that laptop. Actually I know it's not working, it never works with how Windows 7 schedules tasks. Windows likes to move processes from once CPU to another constantly, this prevents the turbo boost from engaging. You have to use processor affinity to prevent it from being moved, doing this immediately engages turbo-boost and your scores will fly up.
I ran my own CB11.5 bench's on my 3550MX and here is what I got.
http://www.tomshardware.com/forum/page-316983_28_4250.html#t2526321
3550MX stock, no modification
B0: 2700 @1.3500
P0: 2000 @1.1125
P1: 1700 @1.0875
P2: 1600 @1.0625
P3: 1400 @1.0250
P4: 1200 @1.0000
P5: 1000 @0.9625
P6: 0800 @0.9375
CB 11.5
CPU:2.57
Single Thread:0.64
MP Ratio: 4.03
Watching HWinfo64 I could see all four cores at 2.0Ghz the entire time, temps were 60~65c
That would be a 25.7 on Toms' charts which is higher then the Trinity and smacking right into the i5.
OC / UV
3550MX Overclock / Undervolt modification with K10
B0: 3000 @1.3500
P0: 2700 @1.3500
P1: 2000 @1.1125
P2: 1600 @0.9250
P3: 1400 @0.9125
P4: 1200 @0.9000
P5: 1000 @0.8875
P6: 0800 @0.8750
Up: 200ms @ 60%
Down: 2000ms @ 20%
CB 11.5
CPU:3.22
Single Thread: 0.75
MP Ratio:4.29
This is allowing Windows to dynamically move the thread around. This resulting in the four cores constantly clocking at 2.7Ghz, they never made it to 3.0 due to the thermal headroom being hit. Temps were 89~94c but holding steady and not rising. When I set the affinity mask to core 3 I got the following.
I just beat the I5, and beat it badly.
Finally I wanted to test how much of an impact Windows moving threads around was having.
When I set the affinity mask to core 3 I got the following.
Single Thread: 0.86
HWinfo64 had Core 3 pegged at 3.0Ghz and temps at ~64c and holding. The other cores clocked down to their 800 idle speed.
From 0.64 @ 2.0 to 0.75 @2.7Ghz and from 0.75 @2.7 to 0.86 @3.0Ghz.
Forcing turbo boost alone would of gained a 17% single threaded performance increase from stock and another 14.6% from OC'd. Going from a P0 state of 2.0 and a B0 state of 3.0 would of netted me a 34.3% performance increase.
Turbo boost works as promised when you force a thread to stay on it's assigned core. I've tested that you can do this with up to two cores engaged, the other two will be forced to idle down which free's up TDP for the first two to surge up. Windows random placement of threads and reshuffling them prevents cores from idle down to allow others to boost up.
Can tom please rerun some of those single threaded tests (and CB 11.5 Single Threaded test) with the process being forced onto one to two cores. This would demonstrate what the Trinity APU is actually capable of should an intelligent individual get their hands on one.
Facebook
Twitter
Instagram