It’s pretty fair to say that AMD’s Bulldozer hasn’t had the smoothest or most welcoming release. After a fair amount of hype and patience, reviews on release day gave the architecture quite a thrashing claiming that the single core performance was not exactly worth bothering with over Intel’s 1155 based competition. Thanks to its greater core count, Bulldozer was, unsurprisingly, said to pull ahead here but how does it stack up in the real world? I wanted to find out what is was like to use day to day, what the single core performance was really like and maybe the most important to me was would it rid of my bottleneck in Battlefield 3?

As is fairly normal now, I will be performing some synthetic benchmarks along with some inbuilt game benchmarks. This will show not only how the FX-8150 compares with my Phenom II but where it excels or falls behind in productivity tasks and gaming. The FX-8150 is up against my Phenom II x4 805. At stock this is a 2.5GHz quad core with 4MB of L3 Cache. It was, however, run at 3.2GHz with the L3 cache unlocked to 6MB making it exactly the same as a stock Phenom II x4 955. This was seated in an Asus Crosshair III Formula while the FX-8150 will be sat in a Gigabyte GA-990FX-UD7. Other than that, the other hardware and software specifications will remain exactly the same – with the other key two being that I am using 8GB of memory (4x2GB sticks) and a Gigabyte GTX570. Unfortunately, despite having an unlocked multiplier and  so begging to be overclocked, due to being an engineering sample from when AMD were struggling with yields I have been told not to overclock the 8150 as it won’t be able to take it. It is a shame, especially when up against an overclocked chip, but there are many who do not overclock through either not wanting to or knowing how to safely. Even then, it still gives a good baseline as to how the chip performs.

The first synthetic benchmark I use will be Cinebench. Developed by MAXON, it’s based on their award winning animation product CINEMA 4D which is used worldwide for creating 3D content. A score is given based on how long the CPU takes to render a photo-realistic scene involving shadows, reflections (blurred and sharp), anti-aliasing and procedural shaders to name a few. This scene can be set to be rendered by only a single core or by all available cores with the performance being given a score in points. Rendering the scene with only a single core being utilised, the Phenom II x4 805 managed to score 0.93. With a similar number of background processes running but this time with the FX-8150, rendering the scene returned a score of 0.91. Now using all of the available cores/threads to render the scene, the Phenom II was able to score 3.55 and the Bulldozer a score of 5.62. To the 8150’s advantage, it can dynamically clock up and clock down based on the heat and load. This means that with a stock clock speed of 3.6GHz, under low load/low thermal output conditions it can clock itself to 3.9GHz or 4.2GHz. I will touch more on this later but even when doing this it was just about able to keep up with the Phenom II. Multi-threaded performance is where the 8150 clearly excels with it clearly beating the x4 805 but not scaling anywhere near perfectly – e.g. in the simplest sense, many would expect twice the number of threads to equal twice the score.

Moving on, we come to the greatly popular 7zip which has an inbuilt benchmark to measure a CPU’s compression and decompression capabilities. Similar to the Cinebench tests, I will compare single threaded performance on both CPUs and the performance of them both utilising as much power as they can muster. To give a greater depth to this test, I will also use both 32MB and 256MB dictionary sizes. The results from these tests can be seen in the table below:

 

32MB, 1 thread

32MB, max threads

256MB, 1 thread

256MB, max threads

 

Compression (KB/s)

Decompression (KB/s)

Rating (MIPS)

Compression (KB/s)

Decompression (KB/s)

Rating (MIPS)

Compression (KB/s)

Decompression (KB/s)

Rating (MIPS)

Compression (KB/s)

Decompression (KB/s)

Rating (MIPS)

Phenom II x4 805

2667

35076

3172

9522

132253

11654

2109

33616

3120

7794

124734

11557

Bulldozer FX-8150

2859

33952

3229

15974

218637

19399

2200

32667

3138

7947*

119518*

11410*

* Due to memory requirements this test was run with 4 threads. 8 threads required 12GB of memory which my setup does not have and I do not believe the majority of setups to have. It does provide a good comparison of four threads on both CPUs though.

Both of the single threaded tests show the FX-8150 winning in the compression test and overall rating but losing out on the decompression. None of these were large victories so as has been demonstrated before I’d draw the conclusion of there not being much between them. Comparing four threads on the Phenom II with 8 threads on the Bulldozer when the dictionary size is set to 32MB really allows the 8150 to stretch its legs and win comfortably. Just like the Cinebench tests, however, it does not scale perfectly but upon average the 8150 is 66% faster than the x4 805. Finally, with a dictionary size of 256MB meaning that the Bulldozer chip wasn’t able to use all 8 of its threads, I set it to only use four as an apples to apples test against the Phenom II. Continuing the pattern from earlier, neither CPU manages to win outright but this time the FX chip won on the compression test and the Phenom II on the decompression test and overall rating. Given that in the tests where both CPUs are utilising the same number of threads we’re seeing differences of less than 5%, I’d be inclined to call them the same based on external factors which are hard to control – e.g. background processes.

3DMark 11 now and in all honesty there isn’t much to see here. Using the performance preset, the Phenom II returns a score of 5179 while the Bulldozer chip manages 5886. Breaking this down a bit, the graphics scores were 5858 and 5897 respectively which shows that the graphics card isn’t held back by either CPU. The physics scores saw an increase for obvious reasons but remained consistent in being an increase of about 62% - from 3844 to 6222. Due to the increase in physics score the combined score obviously improved which explains the overall increase in score. With these explained, it’s no surprise that graphics bound tests stayed around the same while tests solely for or including the CPU heavily improved. Running the tests with the extreme settings saw near enough the same results produced but this had a marginal effect on the overall score. All four graphics based tests were around the same, with the graphics score also telling the same story. As you’d expect, the physics test and physics score saw improvements of 61 or 62% but the combined test saw a fall from 10.67 to 10.43 – this was also reflected in the combined score. Why exactly this is I’m afraid I’m not too sure as it was present in multiple runs of the benchmark meaning it wasn’t just a one off odd result.

Onto games and there’s not as much to talk about as I expected there would be either. I used the built-in benchmarks for Metro 2033, Batman: Arkham Asylum, Mafia II and Just Cause 2 with PhysX enabled where possible and all other settings at their highest with a resolution of 1920x1080. Metro, Mafia II and Batman all saw negligible differences between using the x4 805 and the 8150. The only difference greater than one frame per second was Just Cause 2 which saw nearly a 3.5 FPS increase from 39.92 to 43.31. If being really picky then you will notice that the results for the Mafia II benchmark and the Arkham Asylum benchmark were ever so slightly worse with the 8150 but in my opinion it’s not something noticeable or worthwhile in the real world so not worth getting caught up on. The full results table can be seen below for you to have a look over but, unfortunately, there’s really not much to speak about.

Metro 2033

Batman: Arkham Asylum

Mafia II

Just Cause 2

Min

Avg

Max

Min

Avg

Max

Avg

Avg

Phenom II x4 805

9.92

27

63.52

43

70

107

29.96

39.92

Bulldozer FX-8150

4.99

27.67

114.9

40

69

104

29.8

43.31

I said earlier that I would come back to the CPU being able to dynamically overclock itself based on the current load conditions and this is because I noticed its behaviour as being fairly sporadic. To demonstrate this I had hoped to show a screenshot of a program called TMonitor which is from CPUID – the same people who make HWMonitor, CPU Z and PC Wizard – but the 8150 is not supported by it. TMonitor shows each core’s clock speed in real-time allowing you to see the processor clocking up and down when it needs to or can. Anyway, what I did notice was the 8150 struggling to settle on one clock speed for any reasonable amount of time. Under low load conditions it would constantly be flicking from 3.9GHz to 4.2GHz and either jump back to 3.9GHz or down to its power saving clock speed of 1.4GHz before being off again to 3.9 or 4.2GHz. The CPU voltage would try to follow along to allow the processor to save power when it could but due to not staying still it didn’t really achieve much. Thankfully, this technology is designed and implemented so that the end user never feels or notices this so it’s just a minor observation.

It was here that I believed I had everything I needed to complete the review and fairly give my opinion and feedback on Bulldozer. I was proven wrong, however, when with nothing to do one evening I decided I’d play some Saints Row The Third. This was the first time I had gone to play it as I’d only purchased it in a sale in the past few days. Having let Steam do its thing of installing DirectX updates and anything else the game may need, I was presented with the choice of playing the game in DX9 or DX10/11. I obviously wanted DX11 so picked the associated option and much to my disappointment was presented with a BSOD. The stop code for this BSOD was 0x101 which, having looked it up on MSDN, is caused when the CPU becomes nonresponsive or deadlocked. Thinking this was just a one off affair I allowed my computer to reboot and tried again. This caused another BSOD and made me wonder if there were issues with my installation of the game or the DX10/11 version. Having rebooted again I tried the DX9 version and got yet another BSOD but this time with the stop code 0x124. 0x124 is caused by a machine check exception meaning that the CPU has detected a hardware problem. Feeling quite put off, I decided I’d delete my installation of Saints Row The Third and allow Steam to reinstall it while I investigated the problem a bit more to see if others had come across the problem.

It turned out that I was not alone and typing “Saints Row 3 BSOD” into Google came back with multiple forum topics stating that it was an issue with Bulldozer. Some of these forum topics also picked out Total War Shogun 2 and Deus Ex Human Revolution as two other games which see this behaviour from Bulldozer. Luckily I own both of these and like Saints Row I couldn’t even get to the game’s main menu before being faced with a blue screen. I did some more searching to see if I could find a solution to this and this page came up on AMD’s website. It was an unknown issue to me but as had been proven by there being forum topics and now a page devoted to it on AMD’s site, it was indeed a known issue. AMD’s recommended solution is to update the BIOS but the BIOS they recommended in the above article was the one I was using. Having seen that there were two more BIOS releases after that article I decided to try the latest beta BIOS (F8a). Fortunately, this was able to solve the issue and I was again able to play Deus Ex Human Revolution and Shogun 2 as well as start Saints Row The Third. It had left me greatly disappointed though; both Deus Ex HR and Shogun 2 are recent games and, more to the point, they are AMD sponsored games with the beginning having an “AMD Gaming Evolved” badge. Yet, despite this, AMD couldn’t get them working out of the box on something which is meant to be fairly universal, instead leaving users having to wait roughly two months for a BIOS update to fix it.

To conclude, I’m not in the crowd of people who think AMD’s effort with Bulldozer as terrible and worthless. While it did relieve my CPU bottleneck in Battlefield 3 where my Phenom II would sit at 100% load completely choked, completely ruining the fluidity of gamplay and also having had a chance to take full advantage of all 8 threads when using encoding programs such as Handbrake, I still can’t help but feel disappointed. For the majority of enthusiasts the single threaded performance is a sidestep, or maybe a trade off for having more cores, but when the incompability I suffered is taken into account – hopefully no others crop up for myself or other Bulldozer users – it leaves a lot to be desired. I can understand AMD wanting to move onto something new and that is what they’ve spent a lot of time and money over with Bulldozer but it would have been interesting to see how a die-shrink of K10 performed. Unless an AMD FX chip is a logical upgrade path for you where a sidestep in or possible lowering of single threaded performance is a worthwhile trade off for the benefit of encoding and other applications which can make decent use of 8 threads then by all means Bulldozer is for you. Otherwise, it has to be said that the Intel camp is where you want to be.

Share this