HD3870 - Overclocking Odyssey - Overclock.net

darcness · 12-05-07

The Mission Statement

I've been watching some threads around the net pretty closely lately and I got a few ideas on a comprehensive test for those who want to know synthetic vs. real world benchies on different hardware components. I decided there had to be a way to do this fairly easily. Using this test and the results I not only help those looking to buy a HD3870, but people who want to know how much over clocking can actually do for the average enthusiast.

Keep in mind I say average. I see a bunch of benchmarks around, but not a lot are catered for the "budget" conscious buyer/builder. With the HD3870 and a relatively cheap dual core CPU I think I have the ideal test bed for this very purpose.

I came up with a way to compare CPU overclocks, GPU overclocks, and any combination between. I'll be summarizing it and probably adding to it with more things I think of and possibly on a user request basis. For the test I wanted to eliminate ALL other factors so I ran on a clean install of XP and made sure to turn off all extra programs other than the mandatory ones. First I'll give details on hardware/software and then I'll list results, then summarize.

Test Setup

OS and Software
Windows XP SP2 Clean Install
Gigabyte P35 Mobo Drivers (from install CD)
Catalyst Crysis Hotfix Drivers
RivaTuner 2.06 (for overclocking)
CSS: Source (latest update)
Crysis Benchmarking Tool (with Crysis full version)
3DMark06 BE 1.1.0

Hardware
E2180 Processor
Gigabyte P35-DS3L (with F5 bios)
Sapphire HD3870 (bios unlocked)
2GB G.Skill DDR2 800 (red heat spreaders)
Thermaltake Purepower 500watt PSU

All other hardware should be redundant. It doesn't affect the results in any way and therefore will be omitted.

The Budget Enthusiast At the Fore

Before I get started let me say this. I'm using overclocks here that cater to the budget/mid-range builder. They are achieved using stock cooling and hardware setups with no real modifications needed. Also, they are tested and 24/7 stable and scaled down some from the max settings. This gives us a TRUE look at the gains obtained because these will be stable settings that can be used over the life of the hardware in the test, giving us a more realistic result of EVERY day use.

Notes

For all the tests I ran the CCC defaults except Vsync was always off. Source stress test was ran at 1280x1024 with 8xAA and 16xAF. All other options completely maxed. Crysis details are listed with results. Memory was ran at stock settings while the E2180 was running at 2.0GHz. All the tests with the CPU running at 3.0GHz had the corresponding memory speed of 400MHz but timings were increased from 5-5-5-15 to 4-4-4-12.

The Test

CPU is at 2.0GHz (E2180) and the HD3870 runs at 776/1125. Here's how things panned out...

3DMark06
3DMark Score - 8284 3DMarks
SM 2.0 Score 3601 Marks
SM 3.0 Score 4280 Marks
CPU Score 1742 Marks

Crysis
NEXT BENCH RUN- 12/5/2007 3:38:07 AM - XP
Beginning Run #1 on Map-island, Demo-benchmark_gpu
DX9 1280x1024, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=3, Time Of Day= 9
Global Game Quality: High
================================================== ============
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 79.15s, Average FPS: 25.27
Min FPS: 7.16 at frame 138, Max FPS: 35.06 at frame 993
Average Tri/Sec: 24987412, Tri/Frame: 988926
Recorded/Played Tris ratio: 0.93
!TimeDemo Run 1 Finished.
Play Time: 65.41s, Average FPS: 30.58
Min FPS: 7.16 at frame 138, Max FPS: 36.14 at frame 975
Average Tri/Sec: 30593622, Tri/Frame: 1000525
Recorded/Played Tris ratio: 0.92
!TimeDemo Run 2 Finished.
Play Time: 64.88s, Average FPS: 30.83
Min FPS: 7.16 at frame 138, Max FPS: 36.18 at frame 975
Average Tri/Sec: 30857132, Tri/Frame: 1000957
Recorded/Played Tris ratio: 0.92
TimeDemo Play Ended, (3 Runs Performed)
================================================== ============

Completed All Tests

Source Stress Test
Css = 148.75fps

Nothing really sticks out as the whole system is stock. However, let's up the ante a bit and do some mild overclocking. First up, let the E2180 shine. These E2xxx cpu's are known for not only being VERY cost effective but overclocking extremely well. I've got mine at 3.0GHz for this test. While this isn't the top of the line OC, it does give a great baseline for the testing. Should be attainable by most E2xxx users. The GPU remains untouched, stock speeds. Here's the run down...

3DMark06
3DMark Score - 10432 3DMarks
SM 2.0 Score 4370 Marks
SM 3.0 Score 5004 Marks
CPU Score 2573 Marks

Crysis
NEXT BENCH RUN- 12/5/2007 3:55:48 AM - XP
Beginning Run #1 on Map-island, Demo-benchmark_gpu
DX9 1280x1024, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=3, Time Of Day= 9
Global Game Quality: High
================================================== ============
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 70.14s, Average FPS: 28.51
Min FPS: 15.36 at frame 153, Max FPS: 37.41 at frame 984
Average Tri/Sec: 28205744, Tri/Frame: 989237
Recorded/Played Tris ratio: 0.93
!TimeDemo Run 1 Finished.
Play Time: 62.22s, Average FPS: 32.15
Min FPS: 15.36 at frame 153, Max FPS: 37.41 at frame 984
Average Tri/Sec: 32168644, Tri/Frame: 1000728
Recorded/Played Tris ratio: 0.92
!TimeDemo Run 2 Finished.
Play Time: 62.17s, Average FPS: 32.17
Min FPS: 15.36 at frame 153, Max FPS: 37.93 at frame 996
Average Tri/Sec: 32203940, Tri/Frame: 1001138
Recorded/Played Tris ratio: 0.92
TimeDemo Play Ended, (3 Runs Performed)
================================================== ============

Completed All Tests

Source Stress Test
Css=165.81

In summary, we see that all the scores improved some what. You'll notice that the 3DMark06 and CSS scores improved exponentially compared to the Crysis scores. I think this shows a clear indication of CPU dependency with these particular programs. I'll get more into this later in the overall summary of the testing. For now, let's move on. Up next we're going in reverse a bit. Putting the E2180 back to a lowly 2.0GHz (stock) and ramping up this HD3870 to unlock it's full potential. I settled on clocks of 860/2300 for this test. It's easily achievable even for those who don't want to mod the bios. Here's how things look with the stock CPU and overclocked GPU...

3DMark06
3DMark Score - 8509 3DMarks
SM 2.0 Score 3682 Marks
SM 3.0 Score 4502 Marks
CPU Score 1743 Marks

Crysis
e2189 @ stock with HD3870 860/2300

NEXT BENCH RUN- 12/5/2007 4:32:56 AM - XP
Beginning Run #1 on Map-island, Demo-benchmark_gpu
DX9 1280x1024, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=3, Time Of Day= 9
Global Game Quality: High
================================================== ============
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 76.07s, Average FPS: 26.29
Min FPS: 3.88 at frame 138, Max FPS: 38.44 at frame 965
Average Tri/Sec: 25992474, Tri/Frame: 988674
Recorded/Played Tris ratio: 0.93
!TimeDemo Run 1 Finished.
Play Time: 62.09s, Average FPS: 32.21
Min FPS: 3.88 at frame 138, Max FPS: 39.97 at frame 974
Average Tri/Sec: 32203802, Tri/Frame: 999846
Recorded/Played Tris ratio: 0.92
!TimeDemo Run 2 Finished.
Play Time: 61.87s, Average FPS: 32.33
Min FPS: 3.88 at frame 138, Max FPS: 39.97 at frame 974
Average Tri/Sec: 32366426, Tri/Frame: 1001200
Recorded/Played Tris ratio: 0.92
TimeDemo Play Ended, (3 Runs Performed)
================================================== ============

Completed All Tests

Source Stress Test
Css= 155.48

Now I know, some people may say... "What's the point?!?" Well, here's my reasoning... I wanted to see if some one who bought a CPU and wasn't comfortable or knowledgeable enough to OC it could get real gains by tweaking JUST the GPU. Also, it gives me a gauge of if such a person would be limited by that same CPU. Again, I'll summarize more in the final, but as you can see simply overclocking your GPU really doesn't scale as well as the CPU in this case. Gains were there, but not as evident.

Now for the test we've all been waiting for. For this test I overclocked the CPU to 3.0GHz like above, but also left the GPU overclocked at the same 860/2300 speeds. Here's the rundown for the total system overclock...

3DMark06
3DMark Score - 11076 3DMarks
SM 2.0 Score 4767 Marks
SM 3.0 Score 5387 Marks
CPU Score 2573 Marks

Crysis
NEXT BENCH RUN- 12/5/2007 5:04:23 AM - XP
Beginning Run #1 on Map-island, Demo-benchmark_gpu
DX9 1280x1024, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=3, Time Of Day= 9
Global Game Quality: High
================================================== ============
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 65.36s, Average FPS: 30.60
Min FPS: 12.63 at frame 138, Max FPS: 38.87 at frame 993
Average Tri/Sec: 30250852, Tri/Frame: 988623
Recorded/Played Tris ratio: 0.93
!TimeDemo Run 1 Finished.
Play Time: 56.53s, Average FPS: 35.38
Min FPS: 12.63 at frame 138, Max FPS: 41.40 at frame 981
Average Tri/Sec: 35358676, Tri/Frame: 999497
Recorded/Played Tris ratio: 0.92
!TimeDemo Run 2 Finished.
Play Time: 56.28s, Average FPS: 35.54
Min FPS: 12.63 at frame 138, Max FPS: 42.62 at frame 987
Average Tri/Sec: 35566900, Tri/Frame: 1000789
Recorded/Played Tris ratio: 0.92
TimeDemo Play Ended, (3 Runs Performed)
================================================== ============

Completed All Tests

Source Stress Test
Css= 176.43

And there you have it. From the ordinary to the not so extraordinary, and everything in between. The last test really shows an overall affect that a total system overclock can achieve. We'll examine some of this in the summary below. For now, I'm spent on testing.

Final Verdict and Observations

Summary of scores (all gains are in PERCENTAGE points):
e2180 stock with hd3870 stock
3dmark: 8284 (3601, 4280, 1742)
Crysis: 30.83
Css: 148.75
Gains: N/A

e2180 @ 3.0G with HD3870 stock
3dmark: 10432 (4370, 5004, 2573)
Crysis: 32.17
Css: 165.81
Gains: 20.59, 4.16, 10.28

e2189 @ stock with HD3870 860/2300
3dmark: 8509 (3682, 4502, 1743)
Crysis: 32.33
Css: 155.48
Gains: 2.64, 4.63, 4.32

e2180 @ 3.0G with HD3870 @ 860/2300
3dmark: 11076 (4767, 5387, 2573)
Crysis: 35.54
Css: 176.43
Gains: 25.20, 13.25, 15.68

As you can clearly see, overclocking the CPU and GPU in tandem provide the absolute most to the average user. All programs showed significant gains of at least 13% when the GPU and CPU were both overclocked. However I have noticed some things to think about.

CPU or GPU?
Judging by the results one may say that simply overclocking the CPU will give you the most net gains. By the numbers, they may be right. But is this correct? I think not. If you look at the newer games that are well balanced between CPU and GPU load, things get a little less clear. In Crysis we saw gains with GPU ONLY score better than CPU ONLY OC's. The reason why CSS and 3DMark scale so different could only be attributed to the fact that they are indeed CPU bound/dependent. While CSS actually showed good gains with both CPU AND GPU overclocks, 3DMark06 was a different story. This lends credence to those who say 3DMark shouldn't be used as any real world indication of gaming performance. Where 3DMark showed huge gains in CPU overclocks, it showed little increases from the GPU tweaking. Yet when applied to a real world scenario such as Crysis, the results were contradicting. It actually scaled better with the GPU only overclock. Therefore if you have to overclock only 1 component I would recommend the GPU itself. This is a matter of preference and what uses you have for your PC of course, but for modern gaming with balanced games, GPU overclocks will net you more gains if you don't want to or can't manage any CPU OC's.

3DMark06: Bragging Rights or Real World?
I think this test puts a definitive nail in the coffin of 3dMark06 as a real world benchmark. It is what it is, a synthetic benchmark. Not good for much outside of bragging rights and a baseline measurements. If you look at the summary you'll see that almost no gains were shown from overclocking only the GPU. However, any time the CPU was at a speed of 3.0GHz the scores remained relatively stable. It explains why Quad core and the like systems get such gaudy numbers on this benchmark while the lowly dual cores are stuck playing second fiddle. However, no need to worry dual core owners, CPU scores on 3DMark don't mean much for us gamers.

To Clock or Not to Clock?
A resounding YES! As an avid budget enthusiast I recommend overclocking with great confidence. As shown, gains can be good, even on cutting edge gaming. In many cases it could be the difference from playing a game in medium settings, vs. higher settings. Also, it makes a big difference in overall PC use. From loading windows to browsing the internet, to encoding video you can tell a difference. The test system documented above clearly shows that you can game on a budget and still have a great experience doing so with a little help from an OC.

Coming To an OCN Near YOU!!
In the coming days I will be doing a cost analysis on this exact test setup. I will be comparing the amount of money spent on the actual system to the penny. I will then "build" a comparable system from newegg (where I ordered all these components) that equals the specifications of the overclocked system. It will be interesting to see what kind of money was saved with the moderate OC results. Also, if any one would like to see another game/program benchmarked, more information, or anything of the like feel free to contact me or post. I'm limited to 1280x1024 on my LCD but I could hook up my old CRT and run tests at 1600x1200 if it's a hot topic among the replies I receive.

Now discuss amongst yourselves...

(This started as a simple post, but ended as quite a bit more. If I should put this in another forum/section please, feel free to say. I just think where ever it would get the most view from like minded PC users would be best.)

agntallen · 12-06-07

i think this was a very nice, detailed "odyssey" i like it.

Lxcivic2k1 · 12-06-07

Very well put together! Very detailed too.

I'm curious....have you tried to push the HD3870 to its max yet? and if ya did what was it since you have the modded BIOS.

darcness · 12-06-07

I haven't really found the max yet. I simply removed the 860 lock on the bios but haven't done much testing beyond that. I know I can run 860/2300 on ATI Tool artifact free, but ATI Tool messes with my fan settings so I try not to use it. I suppose I could get more outta the card but right now don't think it's needed. I'll wait and see how the volt mods and hopefully volt modded bios' go. Then I'll do some more clocking. For now it's plenty fast enough for any game. Only game I've played that really pushed it at my max res. of 1280x1024 is Crysis of course.

Antho · 12-06-07

Nice write-up. Have flashed my cards to unlock them, but not getting past 860

Good point about the whole bragging rights issue with 3dmark06

darcness · 12-09-07

Quote:

Originally Posted by Lxcivic2k1

Very well put together! Very detailed too.

I'm curious....have you tried to push the HD3870 to its max yet? and if ya did what was it since you have the modded BIOS.

Checked it out today and got a max of 385/1334. That's stable via ATITool and runs through benchmarks just fine. Anything higher yields artifacts or VPU recoveries. Still not bad with the stock cooler and hitting 60's on FULL load. Here's an update on the 3dMark06 and Crysis scores @ those settings.

3DMark06 - 11195
compare url: http://service.futuremark.com/compare?3dm06=4097127

Crysis:

NEXT BENCH RUN- 12/9/2007 5:43:37 AM - XP
Beginning Run #1 on Map-island, Demo-benchmark_gpu
DX9 1280x1024, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=3, Time Of Day= 9
Global Game Quality: High
================================================== ============
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 64.08s, Average FPS: 31.21
Min FPS: 13.92 at frame 138, Max FPS: 41.76 at frame 977
Average Tri/Sec: 30861706, Tri/Frame: 988796
Recorded/Played Tris ratio: 0.93
!TimeDemo Run 1 Finished.
Play Time: 54.63s, Average FPS: 36.61
Min FPS: 13.92 at frame 138, Max FPS: 43.55 at frame 991
Average Tri/Sec: 36618144, Tri/Frame: 1000141
Recorded/Played Tris ratio: 0.92
!TimeDemo Run 2 Finished.
Play Time: 54.69s, Average FPS: 36.57
Min FPS: 13.92 at frame 138, Max FPS: 43.55 at frame 991
Average Tri/Sec: 36609472, Tri/Frame: 1001005
Recorded/Played Tris ratio: 0.92
TimeDemo Play Ended, (3 Runs Performed)
================================================== ============

Not too bad if I do say. However, I'm thinking I'll probably back it down for 24/7 use. I'm going to do some artifact testing with ATITool and whatever the highest settings I can run for @ least 4-5hours without errors will be what I stick with.

swatsor · 12-09-07

GREAT post.

darcness · 12-11-07

I'm in the process of installing Windows Vista 64 and 32 bit @ the moment. Once I have both installed I'll update with some results from Vista as well. I've seen many reviews of the 8800gt around and my goal is to push this card until I can match the Crysis performance of the 8800gt. Alas, all of the benchmarks I've found are run via Vista, even the Dx9 ones. So in order to compare things properly I'll do so with Vista. As it stands now, I'm on par (basically dead even) with the 8800gt's I've seen benched on Crysis in reviews. And they are using quads...

darcness · 12-12-07

Windows Vista x64 update!!!

Well I ran some benchmarks on Vista 64 and came up with these numbers for Crysis and 3DMark06.

Test Setup

Software
Windows Vista 64-bit (newest updates)
Gigabyte P35 64-bit drivers
ATI Crysis hotfix drivers (vista 32/64 version)
Riva-Tuner 2.06
3DMark06 BE 1.1.0
Crysis Full Game
Crysis Benchmark Tool

Hardware
Same hardware as previous tests

Overclocks

E2180 @ 3.0GHz
Sapphire HD3870 @ 885 core, 2668 memory

Results

3DMark06
3DMark - 10967 3DMarks
SM 2.0 Score 4746 Marks
SM 3.0 Score 5330 Marks
CPU Score 2532 Marks

Crysis DX10
NEXT BENCH RUN- 12/12/2007 5:44:32 AM - Vista 64
Beginning Run #1 on Map-island, Demo-benchmark_gpu
DX10 1280x1024, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=2, Time Of Day= 9
Global Game Quality: Custom
Custom Quality Values:
VolumetricEffects=High
Texture=High
ObjectDetail=High
Sound=High
Shadows=High
Water=High
Physics=High
Particles=High
Shading=High
PostProcessing=High
GameEffects=High
================================================== ============
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 97.60s, Average FPS: 20.49
Min FPS: 9.15 at frame 138, Max FPS: 30.07 at frame 1638
Average Tri/Sec: 20087674, Tri/Frame: 980312
Recorded/Played Tris ratio: 0.94
!TimeDemo Run 1 Finished.
Play Time: 76.23s, Average FPS: 26.24
Min FPS: 9.15 at frame 138, Max FPS: 33.20 at frame 1741
Average Tri/Sec: 26009830, Tri/Frame: 991373
Recorded/Played Tris ratio: 0.92
TimeDemo Play Ended, (2 Runs Performed)
================================================== ============

Crysis DX9
NEXT BENCH RUN- 12/12/2007 6:27:29 AM - Vista 64
Beginning Run #1 on Map-island, Demo-benchmark_gpu
DX10 1280x1024, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=2, Time Of Day= 9
Global Game Quality: High
================================================== ============
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 94.54s, Average FPS: 21.16
Min FPS: 4.42 at frame 1167, Max FPS: 33.46 at frame 871
Average Tri/Sec: 20734542, Tri/Frame: 980109
Recorded/Played Tris ratio: 0.94
!TimeDemo Run 1 Finished.
Play Time: 72.94s, Average FPS: 27.42
Min FPS: 4.42 at frame 1167, Max FPS: 33.46 at frame 871
Average Tri/Sec: 27180860, Tri/Frame: 991277
Recorded/Played Tris ratio: 0.92
TimeDemo Play Ended, (2 Runs Performed)
================================================== ============

NEXT BENCH RUN- 12/12/2007 6:32:52 AM - Vista 64
Beginning Run #1 on Map-island, Demo-benchmark_gpu
DX9 1280x1024, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=2, Time Of Day= 9
Global Game Quality: High
================================================== ============
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 74.41s, Average FPS: 26.88
Min FPS: 13.58 at frame 138, Max FPS: 39.42 at frame 847
Average Tri/Sec: 26355886, Tri/Frame: 980550
Recorded/Played Tris ratio: 0.93
!TimeDemo Run 1 Finished.
Play Time: 58.12s, Average FPS: 34.41
Min FPS: 13.58 at frame 138, Max FPS: 40.78 at frame 1009
Average Tri/Sec: 34086232, Tri/Frame: 990508
Recorded/Played Tris ratio: 0.93
TimeDemo Play Ended, (2 Runs Performed)
================================================== ============

Notes

DX10 and Vista 64
Well, pretty much DX10 eats this card alive. About an 7fps hit on Vista 64 (DX9 vs DX10). And approximately a 9-10fps hit going from XP and DX9 to Vista and DX10. Really re-affirms my belief that most hardware isn't even ready to run DX10 still, at least on newer games. I really hope both ATI and Nvidia come out with some great next gen cards soon because I would like to play the new games on DX10 but it seems it's just not feasible with this current setup.

DX9 and Vista 64
Really suprised that it didn't take more of a hit on DX9. I saw about a 2-3fps hit from XP and DX9 to Vista 64 and DX9. Still that is 2-3fps I won't live without. For the time being I'm sticking with XP. This does go to show that drivers have improved by leaps and bounds however. Vista performance with DX9 is nearly on par with XP and with a few driver tweaks I believe ATI can close that gap even farther, making us user who prefer XP because of the higher frame rates, able to switch over to Vista without our games taking any real performance hits.

dr4gon · 12-12-07

Very nice comprehensive organized review...

From my own testing I have been able to pull off 890/1300 (probably could go a bit higher, not sure how much) stable for crysis bench looping

		Overclock.net - Overclocking.net > Graphics Cards > ATI
HD3870 - Overclocking Odyssey

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)