FX Bulldozer VS Piledriver, 8120 VS 8320, Gaming comparrison

[ssjwizard]

Alright so the reviews flying around out there do a good job of demonstrating some of the differences between these two generations of FX but none have done head to head clock for clock comparisons that I'm aware of.

The tests I will be running my 8120 against my 8320 both using my GTX 460 at stock speeds. Comparisons will be made using *identical settings, in gaming benchmarks. I will run each benchmark at various CPU speeds three times each and post the middle result.

The tests:

3DMark Vantage - performance
3DMark11 - basic
HWBot Haven basic and xtreme
Metro 2033

**Dirt 2/3

I may be willing to run additional tests as I have a bit of free time in the next few weeks. If there's a particular test you'd like to see let me know and I will consider it. My biggest limiting factor on actual game tests is what I own. Im happy to run games with included benchmarking tools but I will not be doing anything with FRAPS or similar software.

Note* My 990FXa-UD3 does not report "Rated FSB" in CPUz, it never has all tests will be run with an x11 NB and HT link multiplier
Note** will be run if I have time.

 

[ssjwizard]

3DMark Vantage - Performance:

3.1Ghz

Spoiler

3.4Ghz

Spoiler

3.8Ghz

Spoiler

[B]CPU Score:
Speed        FX8120      FX8320[/B]
@3.1Ghz      15,589      17,377
@3.4Ghz      16,470      19,128
@3.8Ghz      16,951      21,044


[B]Difference CPU score:[/B]
[B]Speed      Difference    8120/8320[/B]
@3.1Ghz       1,788        87.91%
@3.4Ghz       2,658        86.10%
@3.8Ghz       4,094        80.55%

At the 8120s stock speed the scores are less than 2k apart, as the clockspeed goes up the 8320 pulls much further ahead. As you can see the clock scaling on the 8320 is vastly superior in this benchmark. Aside from the runs done at 3.1Ghz the GPU score is within 500pts of each other.

Edit. I just noticed that the 8120 is running in single channel ram at 1866 so I will rerun these tests at 1600 dual channel on both rigs later to fix this.
Edit2. I reran the tests at 3.4 and 3.8 the results are well within margin of error from the original single channel scores so I did not rerun @ 3.1Ghz.

3DMark11 - Performance:

3.4Ghz

Spoiler

3.8Ghz

Spoiler

[B]Physics Score:
Speed        FX8120        FX8320[/B]
@3.4Ghz      6,020         6,714
@3.8Ghz      5,702         7,164


[B]Difference CPU score:[/B]
[B]Speed      Difference    8120/8320[/B]
@3.4Ghz       0,694        89.66%
@3.8Ghz       1,462        79.59%

After rerunning the Vantage results I decided to skip running further tests at 3.1Ghz. I'm having a hard time explaining this one, again we see the clock speed scaling diminish on the 8120 around the 3.8Ghz mark, but more than that the scores in this bench are actually lower. Of the 3 runs at 3.8Ghz this was the best result. After the first run was so much lower I did try bumping the vCore up but I am running the identical settings as I put vantage through. The only thing I can think of is that around this speed the loose edged resource manager just starts choking the CPU. Changes in the way resources are allocated and released to a much finer degree were one of AMDs published improvements with PD and I think this test has demonstrated that.

Again we also see a minimal variance in GPU scores with the 8320 coming out on top by a very small margin(~0.8%) across the board.

 

[ssjwizard]

HWBot Unigen Heaven - Basic

3.4Ghz

Spoiler

3.7Ghz

Spoiler

4.0Ghz

Spoiler

[B]Score:
Speed        FX8120       FX8320[/B]
@3.4Ghz      1,731.5      1,876.5
@3.7Ghz      1,816.6      1,918.9
@4.0Ghz      1,865.7      1,965.2


[B]Difference score:[/B]
[B]Speed      Difference    8120/8320[/B]
@3.4Ghz       145.0        92.27%
@3.7Ghz       102.3        94.66%
@4.0Ghz       099.5        94.93%

HWBot Unigen Heaven - Xtreme

3.4Ghz

Spoiler

3.7Ghz

Spoiler

4.Ghz

Spoiler

[B]Score:
Speed        FX8120       FX8320[/B]
@3.4Ghz      729.94       809.21
@3.7Ghz      798.35       842.90
@4.0Ghz      779.52       846.57


[B]Difference score:[/B]
[B]Speed      Difference    8120/8320[/B]
@3.4Ghz       79.27        90.20%
@3.7Ghz       44.55        94.71%
@4.0Ghz       67.05        92.08%

 

[ssjwizard]

I will be running Metro 2033 on a variety of configurations. 3.3Ghz, 3.8Ghz, 4.0Ghz, DX9, and DX11 are our test parameters.

#1 I call CPU Crushing is 1024x768, high quality, PhysX on, nVidia drivers PhysX - CPU ONLY. This setting is not going to give particularly real wold FPS, but what it will do is allow us to see the difference in CPU performance. Scores are ranked at 10 runs per clockspeed.

#2 720P is 1280x720, high quality, PhysX off Scores based off of 3 runs per clockspeed.

#3 1080P is 1920x1080, high quality, PhysX off, Score based off 3 runs per clockspeed.

Metro 2033 CPU CRUSHING:

DirectX 9:

Spoiler

3.3Ghz

Spoiler

3.8Ghz

Spoiler

4.0hz

Spoiler

DirectX 11:

Spoiler

3.3Ghz

Spoiler

3.8Ghz

Spoiler

4.0hz

Spoiler

Metro 2033 720P

Metro 2033 1080P

DirectX 9:

Spoiler

3.8Ghz

Spoiler

4.0hz

Spoiler

DirectX 11:

Spoiler

3.8Ghz

Spoiler

4.0hz

Spoiler

 

[ssjwizard]

Duplicated post for editing.

 

[ssjwizard]

Duplicated post for editing.

 

[RGone]

Nice work 'ssjwizard', that seems to tell the tale of the tape. Thanks.

 

[Mandrake4565]

Keep them coming wizard and nice work.

 

[caddi daddi]

this is some handy dandy info.

 

[RGone]

Good work and Kudos to 'ssjwizard'...

Over the course of the last couple of months I had worked on testing my FX-8120 and was getting 'results' very very similar to what "ssjwizard" is seeing during his test comparison of the FX-8120 to his newer FX-8320.

He is seeing a leveling off of performance results to cpu speed scaling. If we can grip this phenomenom, it may become much less of an expensive and frustrating experience for those that have not spent well on heavy duty motherboards to hold-up to the rigors of supplying 8 core 81xx cpus with the power they need under heavy loads.

If we can come to realize there is a sort of flat-lining of the performance increase to cpu scaling we could reduce power consumption and heat to be removed from an overclocked FX-81xx type processor.

Let me interject something here that I have seen exhibited to me as well. IF you take a four core FX-41xx processor set it to the same CPU Mhz and run it in Cinebench R11.5, it will have a CPU score of almost *dead-on* half of what the 8 core FX-8100 cpu will get for a CPU test result score at identical CPU Mhz. By inference, I would imagine that limiting excess CPU Mhz on all BD type FX-series will result in a more pleasurable experience as a result of not having to battle huge current draws and resultant heat to be dealt with.

In short I decided that rather than RE-invent the wheel as is so often done, I would forget posting all of the cpu tests as "ssjwizard" has so capably and kindly done for those whom can reason; I decided to simply post one graph that seems to clearly exhibit the leveling-off of real performance gains as he has already found.

Again I say Kudos to you "ssjwizrd". If your hard work and effort is truly taken to heart, there should be a real lessening of some frustrations about heat and power consumption of the BD for those just now coming to the overclocking stage. Thanks man. RGone...ster.

 

[ssjwizard]

RGone thanks for posting that graph.

I am preparing metro 2033 results as I type this. The testing is complete on the 8120 and Ive had all I can stand to watch that bench run so I will finish the 8320 results tomorrow. Its going to be ALOT of data to sift thorough as I ran it at many different settings to try and get a feel how the CPU handles different types of workloads. So far at my floor clockspeed for this test(3300Mhz) in my most CPU intense configuration the 8320 was almost 2FPS ahead on a scale of 5.... and at 3100 the 8120 failed to provide >0 FPS at all times so it was not run for more than a single test and will not be used. This one was a bit more generous on CPU work even when I maxed out the CPU load and I was able to run this test at 4,000 with the 8120 on the M5A78*.


Note* I have still not decided to disable the OCP, and surge protection on the M4A so my tests are limited to whatever speeds the 8120 will do on that board.

 

[Anonaru]

Over the course of the last couple of months I had worked on testing my FX-8120 and was getting 'results' very very similar to what "ssjwizard" is seeing during his test comparison of the FX-8120 to his newer FX-8320.

He is seeing a leveling off of performance results to cpu speed scaling. If we can grip this phenomenom, it may become much less of an expensive and frustrating experience for those that have not spent well on heavy duty motherboards to hold-up to the rigors of supplying 8 core 81xx cpus with the power they need under heavy loads.

If we can come to realize there is a sort of flat-lining of the performance increase to cpu scaling we could reduce power consumption and heat to be removed from an overclocked FX-81xx type processor.

Let me interject something here that I have seen exhibited to me as well. IF you take a four core FX-41xx processor set it to the same CPU Mhz and run it in Cinebench R11.5, it will have a CPU score of almost *dead-on* half of what the 8 core FX-8100 cpu will get for a CPU test result score at identical CPU Mhz. By inference, I would imagine that limiting excess CPU Mhz on all BD type FX-series will result in a more pleasurable experience as a result of not having to battle huge current draws and resultant heat to be dealt with.

In short I decided that rather than RE-invent the wheel as is so often done, I would forget posting all of the cpu tests as "ssjwizard" has so capably and kindly done for those whom can reason; I decided to simply post one graph that seems to clearly exhibit the leveling-off of real performance gains as he has already found.

Again I say Kudos to you "ssjwizrd". If your hard work and effort is truly taken to heart, there should be a real lessening of some frustrations about heat and power consumption of the BD for those just now coming to the overclocking stage. Thanks man. RGone...ster.

Man, I knew they hit a roof at 4.2, but I didn't know it was anything like this :S. The info in this thread has already been passed to helping a new OCer on this forum, this is golden! Gives us a ceiling to aim for when OC'ing these parts (A very easy one to reach, mind you).

I have yet to find a hard ceiling on the PD chip (4.6-4.8 seems to be where scaling declines significantly).. Do you have any data regarding piledriver and hitting a scaling roof as well?

 

[RGone]

The info in this thread has already been passed to helping a new OCer on this forum, this is golden! Gives us a ceiling to aim for when OC'ing these parts (A very easy one to reach, mind you).

I have yet to find a hard ceiling on the PD chip (4.6-4.8 seems to be where scaling declines significantly).. Do you have any data regarding piledriver and hitting a scaling roof as well?

Yes 'anonaru' I followed the trail of "djjmatexxe" as you and C_D walked him up to about 4.25Ghz and the most awesome part of that entire thread to me was his statement > "thanks to you guys, I now know a bit more about my setup". For us older heads that have already been there and done that on most configs and the principle is 'always' the same, the high cpu mhz is no longer the end all be all of a computer. Getting a reasonable and cost effective clock that has reduced the outlay for a cpu is the goal today in my mind. Truth is getting more for less was what started the majority of us our overclocking journey and "then" as always, it became some sort of a sport and in many respects overclocking became just a game. Everyone was trying to do it. Overclocking became a right and not a hard-one privilege just like so many other things in life. Shut up RGone.

As regards the actual 'do work' scaling of the FX-Vishera series, I have no real clue. Honestly I have not even tried to infer from 'ssjwizards' great work above. I think he might comment on that more accurately. I will say; am as I type this, fishing for an FX-8350 for just such test purposes as you are quizical about. Hehehe. RGone...ster.

 

[Anonaru]

Yes 'anonaru' I followed the trail of "djjmatexxe" as you and C_D walked him up to about 4.25Ghz and the most awesome part of that entire thread to me was his statement > "thanks to you guys, I now know a bit more about my setup". For us older heads that have already been there and done that on most configs and the principle is 'always' the same, the high cpu mhz is no longer the end all be all of a computer. Getting a reasonable and cost effective clock that has reduced the outlay for a cpu is the goal today in my mind. Truth is getting more for less was what started the majority of us our overclocking journey and "then" as always, it became some sort of a sport and in many respects overclocking became just a game. Everyone was trying to do it. Overclocking became a right and not a hard-one privilege just like so many other things in life. Shut up RGone.

As regards the actual 'do work' scaling of the FX-Vishera series, I have no real clue. Honestly I have not even tried to infer from 'ssjwizards' great work above. I think he might comment on that more accurately. I will say; am as I type this, fishing for an FX-8350 for just such test purposes as you are quizical about. Hehehe. RGone...ster.

Aye, I saw that, and it gave me the good feelin'

I can start compiling scaling data with my 8320 and 8350 and see if I can give you guys some numbers. Never tried to organize my benchmark information for structure like this before-- it will be an adventure

 

[ssjwizard]

The scaling on vishera is significantly smoother but I havnt had a chance to run enough tests high up to see where the returns start diminishing. I mean if you look at my 8320@5Ghz wPrime vs the [email protected] it paints the picture pretty clearly just how bad the clock scaling chokes on the BD but doesnt contrast vishera to well. Ill probably run some tests for this in a while but I need to finish up this comparison first.

 

[EarthDog]

Awesome, bring on what really matters, GAMES!

 

[RGone]

Never tried to organize my benchmark information for structure like this before-- it will be an adventure

I know one thing and that is when you want to make a graph...I don't know squat. I made that graph late last night to put together what I had benched over a pretty good period of time. Like I said no sense in REinventing the wheel but rather to bolster what 'ssjwizard' so capably was finding.

Well tht graph took me the better part of 2 hours to make. Numb nutts me for sure. I finally did it in OO calc. Then because it and one more would make two, I finally took what OO calc would give me and finished the rest in a vector grarphics application. Can you say shett far and call the dogs? Hehehe.

C_D had bench results that show much the same scenario and I was going to tabulate them for him, but 'ssjwizard' has raced to all of our rescues. I know when I see real work done and just wanted to assist 'ssjwizard' and hopefully did and will.

RGone...ster

PS: C_D has that ginormous Linux based application that takes days to run. Not just one of these little benches that I can run with almost a 10th less Vcore than it takes to become P95 Blend Stable or 20 runs of Lin-x testing. By the way Lin-x is far less demanding than P95 Blend mode on my FX-8120. Hehehe.

Anyway C_D has run his 5 day long application on an FX-4170 and is n0w running the same application on a 965BE on the same board with the same ram and well you get the idea. FX-4170 at 4.8Ghz and the 965BE at 4.03GHz. Going to be intersting to see which one of the two cpus can finish the application first. Stand-bye for results. RGone...again.

 

[ssjwizard]

Heaven results up, partial Metro 2033. I am still trimming screenshots for 720P and 1080P configurations. I realise I also only ran 5x Runs on the Metro2033 @ 4Ghz on my 8120 but my wife needed the system and I think 5 still gives a decent average result. Ill try and have the Metro2033 segments done tomorrow sometime. At this point I have only briefly glanced at the data and dont have any figures calculated, but the 8320 is a few FPS ahead at all times.

After that Ill be looking at Dirt 2 since its a touch less graphically intense than Dirt3 but has the same CPU loads. Following the tests in Dirt 2 Im done with this article unless someone has a request for a particular bench.

 

[Hicksimus]

After that Ill be looking at Dirt 2 since its a touch less graphically intense than Dirt3 but has the same CPU loads. Following the tests in Dirt 2 Im done with this article unless someone has a request for a particular bench.

Good stuff, and Dirt 2 has better music to benchmark to as well.

 

[gnusounduave]

A little sidetrack here but I've noticed the same thing on my 6100. At 4.0Ghz the chip comes alive and then at 4.2 it becomes a beast but anything after that the excitement is gone.

From my Cinebench and 3DMark 11 tests I've seen the numbers start to go down around 4.4ghz and the diminishing returns begin. The plateau is there between 4.2 and 4.4 for sure.