AMD FX-8150 Bulldozer CPU Review

CPU by leeghoofd @ 2011-10-12

It has been a while since AMD has revamped their CPU lineup. New Graphics Cards are introduced on a regular basis, though mostly their silicon motherboard processor variants are not. Mostly a brand new breed of CPUs goes hand in hand with the launch of a new CPU socket. And if the end user is really unlucky the RAM and CPU cooler need a swap too. It has been over 18 months since AMD introduced their hexacore Thuban CPU. So it was about time to give an answer to Intels SandyBridge lineup. Or why not even aim for Intels high end socket 1366 Gulftown lineup. With every new CPU release, speculations rule the various forums. How the design is gonna be, performance expectations, die size, TDP,.. you name it and has been dealt with on the tech sites. On the 12th of October AMDs new born CPU core baptized Zambezi, for the desktop PCs and for the servers Interlagos and Valencia will see the daylight. Let's open the press kit.

  • prev
  • next

Results at stock clocks

Time to see how this new born performs. The hardware used was the following :

 

  • Asus Crosshair V Formula 990FX chipset (bios 0083)
  • AMD FX-8150 CPU cooled by Corsair H80
  • 4GB of Corsair Dominator GT 1600C8
  • Asus GTX480 GPU WHQL driver 280.26
  • Western Digital Green Caviar 1tb Hard drive
  • Corsair AX1200 PSU

 

The other setups were equipped with the same ram kit ; 1600Mhz CL8-8-8-24 1T and GTX480 GPU. Only the i7 970 had triple channel running.

 

First up the SuperPi and WPrime tests. The Pi test is purely single threaded, Wprime will load all cores.

 

 

Okay this is not what I expected from a new generation CPU. Something is not right. The single threaded performance is sub par. In fact even worse then the previous generation Thuban CPU, and the latter running at far lower clocks (3.6Ghz Turbo). Via the Tmonitor we spot that the cpu stays at 4.1Ghz, even after setting affinity to core 0. Not entirely the 4.2 we expected. But this software has been buggy before with Sandybridge readouts. So to countertest we upped the CPU frequency to 4.2Ghz, disabling Turbo. With no affinity set we achieve a similar similar result as above. When setting affinity, we gain a bit of performance (20secs 502 iso 20.882) But still being far slower in Pi 1M then the Thuban. Kindda weird as you expect a nice boost from a new generation of CPUs. Apparently I was wrong. Weirdly enough not only the single threaded performance is not so stellar. The Wprime32 run is also at least half a second too slow. Seems special tailored or more modern programs are required.

Let's increase the strain over a longer period.

 

 

 

Darn, similar outcome as above. The new technolgy failing to deliver here. The single threaded SuperPi has never been an AMD benchmark, always having to give in to the blue team. But this is no way to be competitive. Loading all cores with Wprime 1024 makes me wonder if we are really dealing with an 8 core CPU or more sort of a 4 core and something else CPU. Let's see if we can spot a trend.

 

 

Okay this looks a lot better, spotting a nice improvement over previous AMD brethern. The Crosshair V Formula latest bios allows ram dividers up to 2400mhz. We will test ram dividers a bit later in the review. At the moment 1600C8 will have to do. Looking at the output, we have close to X58 bandwith now. So a big step forward there.

 

 

Cinebench Release 10 64bit can test either single core or in a multithreaded environment. The Single core test is barely faster then the lower clocked Thuban CPU. In fact we are not much better then the Deneb 965BE either. A really bad IPC (instruction per clock ) ratio. Luckily the CPU starts to blow fire once we opt to go for the full monty test. And yes we are finally in 2500K territory people. Reminder though : thats 4 cores for the 2500K versus the "8" cores of the FX-8150. AMD's FX-8150 manages to beat all of the Bloomfield CPUs on test, when going multi threaded. Good job there by this CPU.

 

 

X264HD test compiles a 30 sec High Definition video into a X264 encoded clip. Being highly multithreaded one of the ideal tests for multi core monsters. Test 1 is again not better than the Thuban CPU, yet Test 2 is competitive with the current SB flasghip, the 2600K. A new improved version of this software has been released on the 10Th of October, supporting the new instructions of the FX CPU. Seems this CPU has some decent compression/encoding  power. Time for some synthetic 3D action :

 

 

Futuremark 3DMark fun.  We always include the aging 3D01 benchmark as it's purely raw Mhz that count and this 3D test is not heavily dependant on multiple cores. To test the latter, we use 3DMark06 and later on in the review 3DMark Vantage. Since we no longer own some CPU/setups, 3DMark06 has to used on this page just for comparative reasons. In 3DMark01, the Superpi 1M fiasco repeats itself. This Zambezi is absolutely not brilliant in single threaded stuff. 3Dmark06 is near Thuban performance, but we still got 2 cores less on that previous gen CPU. Time to give the setup a spin in some games. Here similar remark as before, only a few games tested to be able to quickly compare the FX-8150 with older CPUs. Test resolution is at 1280 x 1024 with high detail set to avoid the GPU limiting.

 

 

 

Far Cry 2 hardly benefits from the newer architecture. Probably more from the higher clock speeds compared to other AMD offerings. Mafia II, which is a more recent title seems more multi core optimised and is neck to neck with the Thuban series. Compared to the Intel CPUs, out of the box this FX CPU really doesn't stand a chance at this low  resolution.(specially picked to show the differences between the CPUs) Take note however that if you cranck up the detail and go heavy on the GPU side things might look a bit different, as the CPU plays a lesser part. Also neitther of these twol games are optimised for multithreaded CPUs. Sadly due to time constraints that will be part of a follow up.

 

 

 

  • prev
  • next
Comment from petervandamned @ 2011/10/12
Nice one

U did a great job bro !
Comment from Teemto @ 2011/10/12
Nice review. But still I feel FX is a dissapointment as I was hoping for it being at least on par with the 2600K.

Now the question arrizes:

What do I choose as next shrimp bench setup :

1. Keep current Asus Z68 Gene-Z with
a) better 2600K (should ask for a Tones binning day )
b) new 2700K (and hope for the best)

2. Wait for Sandy Bridge-E (most costly upgrade -> wife factor comes into play )

3. Formula V + FX-8150 (hoping that with my new cascade I can manage a stable 5.5+GHz system). It'll look nice but will it be an improvement over my current 5.3 sandy bridge? I have my doubts...
Comment from leeghoofd @ 2011/10/12
I sold your CPU btw today... so something better is on the way if you are game...

For benching sorry but this CPU is only good at CPU-Z... it's a nice game platform or encoding machine as long as the apps support it. for benching plz look elsewhere...
Comment from larkin @ 2011/10/14
Nice NB overclocking test but you should really have high speed memory in there to see a bigger difference. I'm pretty sure PC1280 is a bottleneck at that OC.
Comment from leeghoofd @ 2011/10/14
thanks ir.

Yes i just wanted to have a reference. Sin at XS has tested NB scaling with 1866Mhz Link here : http://www.xtremesystems.org/forums/...locking-Guide- ... His results were not scaling to much.

Too bad we need to go subzero to go over 2800Mhz... I hope retail silicon will be a better as steppings progress...
Comment from leeghoofd @ 2011/10/14
thanks sir.

Yes I just wanted to have a reference. Sin at XS has tested NB scaling with 1866Mhz Link here : http://www.xtremesystems.org/forums/...locking-Guide- ... His results were not scaling to much.

Too bad we need to go subzero to go over 2800Mhz... I hope retail silicon will be a better as steppings progress...

 

reply