AMD FX "Bulldozer" Review - (4) !exclusive! Excuse for 1-Threaded Perf.

[dess]

chew* I think games with only two threads may run best in 2CU/2C mode, instead of 2CU/4C, where sharing is present but not needed, as there are only two demanding threads. And with two cores disabled, there should be a little more room for OC, as well.

[demonkevy666]

chew* I think games with only two threads may run best in 2CU/2C mode, instead of 2CU/4C, where sharing is present but not needed, as there are only two demanding threads. And with two cores disabled, there should be a little more room for OC, as well.

I think you'd want a game with three threads game where the first thread is 100% set to a single CU
and the other two that are around 50-60% for the other threads both on the same CU.

set threw taskmanager or over drive.

[chew*]

chew* I think games with only two threads may run best in 2CU/2C mode, instead of 2CU/4C, where sharing is present but not needed, as there are only two demanding threads. And with two cores disabled, there should be a little more room for OC, as well.

Possibly but I'm using the other threads to offload other tasks.

[Oliverda]

Hi Chew,

It's a little bit OFF topic but have you been able to run four modules at 1866 MHz on BD?

[Barr3l Rid3r]

hi chew,

it's a little bit off topic but have you been able to run four modules at 1866 mhz on bd?

@ 6.6.6.18? :d

[Cecil]

Is this exclusive to the CHV, or does it work on the Sabertooth and M5A99X-EVO?

[chew*]

Hi Chew,

It's a little bit OFF topic but have you been able to run four modules at 1866 MHz on BD?

I haven't tried with this set since it's mismatched, honestly high speed 16g sets are not as fast as a set that can do 7-8-7 1600....if i went 9-11-9 i'm sure 1866 would be fine though,

[chew*]

How's this for almost apples to apples

Sandybridge ( The only diff is this system has a 120gig boot/OS SSD higher ram but a tad looser)


BullDozer optimized NB and ram clocks now.

[mR Yellow]

Nice, would be great once you have done testing to share your optimal bios setup

[Swanie]

Wow nice comparison there Chew. So close between BD and SB. Keep it going

[Hondacity]

Brian

that is crazy!!!

[haylui]

I have a question:
when BD is running 1 thread per module, would the shared Fetch/Decoder works 4 instruction per cycle without switching in every clock cycle??

IMO, if BD still switching every clock cycle even one of the core is idle/disable, then the 20%~35% performance gain while BD is running on 1C/1M configuration is very impressive.
Imagine, if a BD design with full time Fetch/Decoder feeding to only 1 core like conventional design.

[Oese]

hehe here comes the sandy killer finally lol

turbo involved?

@haylui: I think the shared Frontend will now work fully for only 1 core. I guess its the frontend rather then the FPU thats limiting BD by sharing ressources...

[chew*]

hehe here comes the sandy killer finally lol

turbo involved?

No turbo, full clocks on both rigs.

[son14]

Is it possible to post a benchmark and settings for a lower oc at around 4 GHz?

Does the core deactivating is a Crosshair V Formula only feature, or is there a chance to find it on a cheaper board???
I would like to start my own tests, but i dont know.

To be an alternative for i5 or i3, it is not very cool, to take a >200 $ board and pair it with a 100 $ FX-4 CPU and oc it so much. The bill from the energy supplier is part of the package too.
So your smart nb oc and 4 GHz on 3C/3T or 2M/2T, could be an alternative in dirt 3 for example. A test would be very helpful against i7 with resolution and quality specs and with simulated FX-4 @ stock. The results would be very, very helpful to compare it to other reviews, with more games testet.

Thanks for the great testing.

Here some results from rage3d => FX@nb oc:

Battlefield
Min fps Avg fps
109,50% 108,9 %

Civilization 5
175% 110%

Crysis 2
100% 106%

Medal of Honor
114% 117%

Add the 3CU/3C performance gain of 10-20 % and in the end, FX IS a gamer CPU.

[Oese]

No turbo, full clocks on both rigs.

aha cool

[boxleitnerb]

What graphics card are you using? And what settings?

[The Stilt]

Asrock boards have the "core leveling" feature available too atleast (dunno if it is enabled thou).
Other manufacturers can easily add it too, if they only want to.

[chew*]

What graphics card are you using? And what settings?

1920x1080 maxxed out settings with a 6990 in each rig.

Low quality settings might change things but why test at settings I won't use in real world

[haylui]

hehe here comes the sandy killer finally lol

turbo involved?

@haylui: I think the shared Frontend will now work fully for only 1 core. I guess its the frontend rather then the FPU thats limiting BD by sharing ressources...

do you think the frontend has very less effect on the performance difference?
but I think FPU really affected much since most of the desktop apps are FPU intensive

[EniGmA1987]

BullDozer optimized NB and ram clocks now.
Attachment 121415

Is this 2cu/4core or 4cu/4core for the bulldozer test?

[chew*]

Is this 2cu/4core or 4cu/4core?

Thats sandy bridge with HT disabled so 4 core.........

[boxleitnerb]

What graphics card did you use and what resolution, AA/AF was applied? You know you are most likely GPU-limited unless you have a CF-rig.

[chew*]

1920x1080 maxxed out settings with a 6990 in each rig.

Low quality settings might change things but why test at settings I won't use in real world

What graphics card did you use and what resolution, AA/AF was applied? You know you are most likely GPU-limited unless you have a CF-rig.

6990 is 2 6970's crossfire last I checked

Not to mention if you look further back in thread you can see that I posted results from same game with diff settings and the total system results are improving as I optimize the system AKA i'm cpu bottlenecked

[ohms]

6990 is 2 6970's crossfire last I checked

You sir are correct!



So is the general consensus 4 cores disabled to get better single thread IPC performance? But with 4 cores disabled you less overclocking headway?

Sorry for my ignorance but I'm not very good when it comes to all the CPU technical jargon.